![Biological neuron [41]](https://www.researchgate.net/profile/Nitin-Shrivastava-4/publication/351300249/figure/fig1/AS:1019605107625984@1620104209079/Biological-neuron-41_Q320.jpg)
![A schematic diagram of the ANN [54]](https://www.researchgate.net/profile/Nitin-Shrivastava-4/publication/351300249/figure/fig2/AS:1019605107613696@1620104209127/A-schematic-diagram-of-the-ANN-54_Q320.jpg)
Content uploaded by Nitin Shrivastava
Author content
All content in this area was uploaded by Nitin Shrivastava on May 04, 2021
Content may be subject to copyright.
Content uploaded by Nitin Shrivastava
Author content
All content in this area was uploaded by Nitin Shrivastava on May 04, 2021
Content may be subject to copyright.
Vol.:(0123456789)
1 3
Archives of Computational Methods in Engineering
https://doi.org/10.1007/s11831-021-09596-5
REVIEW ARTICLE
Application ofArtificial Neural Network forInternal Combustion
Engines: AState oftheArt Review
AdityaNarayanBhatt1· NitinShrivastava1
Received: 16 December 2020 / Accepted: 23 April 2021
© CIMNE, Barcelona, Spain 2021
Abstract
The automotive industry is facing a crucial time. The transformation from internal combustion engines to new electrical
technologies requires enormous investment, and hence the IC engines are likely to serve as a means of transportation for
the coming decades. The search for sustainable green alternative fuel and operating parameter optimization is a current
feasible solution and is a critical issue among the scientific community. Engine experiments are complicated, costly, and
time-consuming, especially when the global economy is drastically down due to the COVID-19 pandemic and putting the
limitation of social distancing. Industries are looking for proven computational solutions to address these issues. Recently,
artificial neural network has been proven beneficial in several areas of engineering to reduce the time and experimentation
cost. The IC engine is one of them. ANN has been used to predict and analyze different characteristics such as performance,
combustion, and emissions of the IC engine to save time and energy. The complex nature of ANN may lead to computation
time, energy, and space. Recent studies are centered on changing the network topology, deep learning, and design of ANN to
get the highest performance. The present study summarizes the application of ANN to predict and optimize the complicated
characteristics of various types of engines with different fuels. The study aims to investigate the network topologies adopted
to design the model and thereafter statistical evaluation of the developed ANN models. A comparison of the ANN model
with other prediction models is also presented.
Abbreviations
ANN Artificial neural network
BFGS Broyden–Fletcher–Goldfarb–Shanno method
BP Brake power
BPNN Back-propagation neural network
BSFC Brake specific fuel consumption
CFD Computational fluid dynamics
CGP Pola–Ribiere conjugate gradient
CO Carbon monoxide
CO2 Carbon dioxide
CPO Crude palm oil
ENN Elman neural network
EGT Exhaust gas temperature
EGR Exhaust gas recirculation
FIP Fuel injection pressure
FES Fuzzy expert system
GA Genetic algorithm
GEP Gene expression programming
GRNN General regression neural network
H2 Hydrogen
HC Unburned hydrocarbon
HCCI Homogeneous charge compression ignition
HCNG Hydrogen-enriched compressed natural gas
engine
HORD Hyper-parameter optimization-radial basis
function-dynamic coordinate search
HRR Heat release rate
IC Internal combustion
IMEP Indicated mean effective pressure
LM Levenberg–Marquardt
Logsig Logistic sigmoid
LP-EGR Low pressure cooled exhaust gas recirculation
MAPE Mean absolute percentage error
MIMO Multi-input multi-output
MISO Multi-input single-output
MLP Multi-layer perceptron
MRE Mean relative error
MSE Mean square error
NSGA Non-dominated sorting genetic algorithm
NOx Oxides of nitrogen
* Nitin Shrivastava
nitins@rgtu.net
1 Department ofMechanical Engineering, University Institute
ofTechnology, Rajiv Gandhi Proudyogiki Vishwavidyalaya,
Bhopal, India
A.N.Bhatt, N.Shrivastava
1 3
O2 Oxygen
PM Particulate matter
PPM Parts per millions
PSO Particle swarm optimization
PUNN Product unit neural network
R Correlation coefficient
R2 Coefficient of determination
RBFNN Radial basis function neural network
RMSE Root mean square error
RSM Response surface methodology
SCG Scaled conjugate gradient
SFC Specific fuel consumption
SPL Sound pressure level
SVM Support vector machine
Tansig Tangent sigmoid
TOPSIS Technique for ordering preferences by similar-
ity to ideal solution
TPE Tree-structured parzen estimator
1 Introduction
The 1.1 billion light-duty vehicles and 380 million trucks
are on-road vehicles worldwide. These numbers are likely
to cross the 1.7 to 1.9 billion figure by 2040. China and
India are likely to see most of the rise. The current internal
combustion-based automotive industry is uncertain about
future technology. Hydrogen, electricity, and biofuels are
the future of transport. The complete transformation of the
internal combustion engines to new technologies is chal-
lenging due to socio-technological issues with tremendous
capital investment. Transport electrification alone is unable
to resolve the environmental challenges [1–4]. The safety
reliability and the long term maintenance of electric vehi-
cles is also a new concern [5, 6]. Thus internal combustion
engines are expected to serve the transportation sector for
the coming decades [1, 7].
The future internal combustion engines have to deal
with the legislative emission requirements, energy security,
engine efficiency, affordability, and performance require-
ments of consumers. Internal combustion engines running
on next-generation biofuels is the thrust area amongst the
research community. Studies on bio fuelled engines showed
improved performance, significant drop in CO emissions,
unburned hydrocarbons, and particulate matters [8–18].
There can be no single best suggestive biofuel for the engine.
The sustainable and affordable solution to use the biofuel
depends upon the availability of the local feedstock or waste
biomass, the properties of biodiesel, and the enhancement
methods of improve the properties [2, 19–22].
Performance, combustion, emission, and other charac-
teristic studies of the IC engine involve lots of parameters.
Combustion phenomenon and exhaust emissions are very
complex to understand and hence require several tests to
find the relation between various parameters. High-precision
instruments and equipment are required to measure the dif-
ferent operating parameters with these biofuels. These kinds
of tests are expensive and time-consuming [23–25].
The recent economic crisis due to the global COVID-19
pandemic led the industry to huge revenue loss [26–28].
Social distancing, restricted research fundings has put the
limitation on traditional research. So there is a need to
develop a different alternative solution, which saves experi-
mentation cost and time. In the current and long uncertain
COVID-19 scenario, improved artificial intelligence-based
research may dominate traditional research to reduce the
cost and assist the social distancing by work from home
culture. In the last few years, some computational modeling
techniques are used to relate the different parameters of the
engine and predict the different characteristics of the IC
engines like performance, combustion, and emission and
seem to be helpful due to its quite accurate predictions [29,
30]. In this approach, the artificial neural network is one of
the possible techniques to predict the different characteristics
of the IC engine with different sets of inputs.
2 Articial Neural Network
The artificial neural network is the mainstream of artificial
intelligence. Artificial intelligence is the development of
intelligence and its analysis in the machine. An intelligent
machine is a whippy noetic agent that perceives its envi-
ronment and takes action that maximizes its opportunity of
success at an arbitrary goal [31]. In the past decade, much
work carried out using the ANN [32, 33]. It can relate com-
plex and non-linear problems, which may be complete or
incomplete.
ANN finds its wide application in different streams such
as engineering, science, pharmaceutics. Some of the impor-
tant fields include sound and pattern detection, the predic-
tion of the market trends, the bankrupt-ion, military targets,
and mineral exploration sites [34–37]. Neural networks
fend off the need for costly and impractical physical mod-
els, complex mathematical formulas, and computer models.
Numerical or analog data that are difficult to handle due to
the presence of many variables are easily handled by the
neural network [38, 39].
ANN uses artificial neurons as that of biological neu-
rons present in the human brain to process the data. Fig-
ure1 presents the structure of a biological neuron. Coded
data transfers in the brain from synapses to the axon via
electrochemical media known as neurotransmitters. This
coded information transfers to the set of neurons. A set of
neurons arranged it into sub-systems; these sub-systems
constitute the brain. Approximately there are around 100
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
billion interconnected neurons present in the human brain.
The artificial neural network has an inherent tendency like
the human brain in the following respect [40]:
(i) Gaining knowledge from learning through experi-
mental data.
(ii) Assign the required weights for storing them.
A schematic diagram of the ANN model is present in
Fig.2 that contains an input layer, hidden layers, an output
layer, and a set of neurons in each layer. With the help of
training, ANN gets the required information about the base
problem by learning possible non-linear relationships, which
are buried in the problem domain. It takes input from exter-
nal sources such as experimental data and inter-relates them
to get the required output [42]. It tends to relearn if any new
set of data comes, and executes according to it, and predicts
the new set of output [40, 43]. For the prediction and analy-
sis in engineering problems, ANN may be used as an alter-
native method [44]. To predict the multiple output variables,
it accommodates multiple input variables. It has a different
approach than the conventional approach of understanding
the system without any information. A well-trained artificial
neural network is more rapid than the established method to
predict the result because it does not need to solve lengthy
and complicated mathematical problems [45].
Neural network analysis uses a ‘black box’ technique to
solve the problem so that the user needs not worry about
sophisticated knowledge of mathematics [46]. The neu-
ral network stores the knowledge and information within
the trained network, which can be easily accessed by the
user. ANN provides a high level of exactness for earlier
Fig. 1 Biological neuron [41]
Fig. 2 A schematic diagram of the ANN [54]
A.N.Bhatt, N.Shrivastava
1 3
unobserved data sets of the base problem, which are not
required in the ‘training’ process [47]. There could be some
possible shortcomings comes when the neural network is
used for solving complex problems. Most important of them
is the data present in the base problem, which contains lots
of information to train ANN, which may uniformly scatter
all over the range of the system [48].
Different algorithms are available for training purposes
such as Levenberg–Marquardt, conjugate gradient, and
quasi-Newton [49]. The gradient descent algorithm is a slow
learning process, whereas LM, quasi-Newton, and conjugate
gradient algorithms are quicker and use standard numeri-
cal optimization techniques [50, 51]. The complex nature
of ANN may lead to computation time, energy, and space.
Recent studies are centered on changing the network topol-
ogy and design of ANN to get higher performance [52, 53].
3 Modeling ofInternal Combustion Engines
The various studies adopting ANN as a tool in the field of
the internal combustion engine are available. Most of these
studies investigate the effect of alternative fuels, and diverse
operating conditions on the performance, emission, noise,
and other characteristics of the engine. The investigations
were carried out experimentally or through computational
fluid dynamics software. The generated data used to train
the neural network-based models and, consequently, to pre-
dict the various characteristics of IC engines. The literature
showed different approaches to predict the characterisitics.
The present study compiled most of the literature and pre-
sented the applications of neural network modeling in IC
engines, and the methodology adopted to develop the ANN
model with their statistical efficiency. Studies carried out on
different engines with various fuels are broadly categorized
into three types of engines.
1. Spark Ignition engine
2. Compression ignition engine
3. Homogeneous charge compression ignition engine
The first two presented groups of the engine based on the
traditional classification, i.e., the type of ignition, and the
third is the newly developing HCCI engine.
3.1 Spark Ignition Engine
Studies on spark ignition are presented here in three catego-
ries. The first category deals with modeling studies using
standard gasoline as fuel; the second deals with the alterna-
tive SI engine fuels, and the third involves the compara-
tive studies of neural network modeling with other mod-
eling techniques along with different operating parameter
optimization studies. Various ANN models developed to
forecast the performance, pollutants, misfire event, effect of
combustion chamber coating, etc. are presented.
3.1.1 Models withStandard Fuel
This section presents the models using standard gasoline
as fuel. The researchers investigated and predicted various
engine operating characteristics. Initially, studies adopting
the Multi-input single-output models are presented, and the
later section deals with the multi-input multi-output mod-
els. In a study by Togun and Baysec [55], the authors pre-
dicted the SFC and torque of a gasoline engine as the output
of ANN. Effect of Ignition advance, throttling status, and
the engine speed investigated and considered as the input
parameters. Each output parameter was predicted by a sepa-
rate network. A 3-13-1 network architecture was opted to
predict the engine torque, which consists of 3 input param-
eters, one hidden layer having 13 neurons, and one output
layer. BSFC predicted through 3-15-1 network architecture.
The study involves the use of logistic sigmoid as transfer
functions, 10,000 epochs, and Levenberg–Marquardt as a
learning algorithm. It includes the 63 experimental data-
sets to train and 18 to test the model. The torque predic-
tion model resulted in the correlation coefficients of around
0.99 for training and testing, whereas for the prediction of
BSFC, the values were 0.9971 and 0.98331, respectively.
The mean absolute percentage error for torque measurement
was 0.2912 for training and 1.74 for the testing. BSFC pre-
diction involves MAPE of 1.0186 and 2.7588, respectively.
Cay [56] developed three separate ANN models for pre-
dicting the BSFC, EGT, and effective power as an output
parameter to examine the effect of three input parameters.
The input of ANN consists of fuel flow, speed, inlet manifold
temperature, torque, and water temperature. The algorithm
used for the training of the network was back-propagation. A
Levenberg–Marquardt and scaled conjugate gradient learn-
ing algorithms with 3 to 15 neurons in a hidden layer were
compared. It concluded the use of SCG with 7 neurons. For
the testing data, the mean error percent found less than 2.7%,
and RMSE values less than 0.02. For both training and test-
ing, the R2 value was found close to 0.99.
Excess air coefficient affects the engine performance and
emission drastically, and its determination is necessary for
the efficient control of the air–fuel ratio. The widely used
Step type and wideband lambda sensors, located on the
exhaust pipe selected for this measurement. However, these
sensors are very costly. Sahin [57] suggested the estima-
tion of this coefficient with ionization current data using
the secondary spark plug and different engine variables. The
developed ANN model involves the excess air coefficient as
an output and ignition angle, engine speed, and the peak ion-
ization current with the location were the input parameters.
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
Two hidden layers were selected. The network trained with
two different training algorithms, and the result showed that
the Scaled Conjugate Gradient algorithm was better than the
Levenberg–Marquardt algorithm in both training and test-
ing. The values of MAPE, R2, and RMSE, for testing, were
12.34723, 0.99508, and 0.04063, respectively.
Many researchers predicted the characteristics of the
engine using Multi-input multi-output models. Golcu etal.
[58] studied the effect of variable valve timing on perfor-
mance and fuel economy. The authors advance and retard
the crank angle by 10 to 30 degrees and develop the ANN
model using valve-timing and speed as inputs. Fuel flow and
torque were chosen as the output data. A single hidden layer
with 15 neurons opted. The back-propagation algorithm with
logsig and purelin adopted as the transfer functions. The
torque and fuel flow testing showed the RMSE of around
0.9% and 0.28%, respectively.
Hazar and Gul [59] studied the result of piston and valves
coating on an engine. 300-µm Cr3C2 plasma spray coating
employed to examine the performance and emissions of the
coated and uncoated engine. Experiments were performed at
different engine speeds to train the ANN network. SFC, HC,
CO, NOx, and EGT values of both the engine were recorded.
A single hidden layer with 10 neurons was adopted. Tangent
Sigmoid function selected for the hidden layer and LM as the
learning algorithm. The regression value of output param-
eters found very close to 1.
3.1.2 Models withAlternative Fuels
In search of the alternative fuel for better performance and
improved emissions, researchers developed models to inves-
tigate the various alcohol blended fuels, different octane
number fuels, hydrogen, etc. Initially, MISO based mod-
eling studies are presented. Cay etal. [60] analyzed the ANN
model for a methanol engine. Torque, speed, water tempera-
ture, inlet air temperature, fuel flow parameters were used
as inputs while SFC, effective power, pressure, and EGT
were predicted separately as outputs. The back-propagation
algorithm with five neurons in a single hidden layer and
logistic sigmoid function was adopted. Experimental data
of a methanol fuelled engine was opted to train the ANN
model. The coefficient of determination was found near to
one, for training and testing data. RMSE and MAPE were
less than 0.015 and 3.8% respectively for the testing data.
In a similar study Cay etal. [61] developed a three-layered
model to forecast SFC, air–fuel ratio, and emissions like CO
and unburned hydrocarbon of the engine. One hidden layer
with 5 to 15 neurons with BFGS, LM, RP, and SCG algo-
rithms tested to forecast BSFC, AFR, HC, and CO respec-
tively. The best network structures for these parameters were
4-7-1, 4-11-1, 4-14-1, and 4-7-1, respectively. In comparison
with the experimented data, the correlation coefficient was
0.998621, 996,075, 0.977654, and 0.998382 respectively.
Kapusuz etal. [62], in their study, use methanol and etha-
nol blends with gasoline together and separately. The study
was to optimize the use of methanol and ethanol share for
the minimum SFC, maximum torque, and maximum power.
Seventy data sets were used, which includes each ratio of
ethanol and methanol in increments of 5 percent. These three
parameters formed as inputs to ANN. Power, torque, fuel
consumption rate, and BSFC were obtained as the separate
outputs intended to achieve high success. The LM algorithm
was used for this purpose. The weights and biases were
update using the gradient descent with momentum back-
propagation algorithm. The tansig function opted for the hid-
den layer due to the high success rate. The architecture for
torque, power, fuel consumption per hour, and BSFC were
3-10-1, 3-10-1, 3-7-1, and 3-18-1, respectively. ANN gave
the regression values of 0.9906, 0.997, 0.9974, and 0.9312
for the torque, power, fuel consumption per hour, and BSFC,
respectively. The study showed a mixture of 11% methanol
with 1% ethanol gave the best results, whereas the mixture
containing 2% ethanol gave the minimum BSFC.
Many researchers studied MIMO models; in this context,
Kiani etal. [63] worked with four strokes, four-cylinder SI
engine, and use ethanol-gasoline fuel. Torque, BP, and pol-
lutants such as CO2, NOx, HC, and CO predicted with three
input parameters viz; load, speed, and ethanol share. Two
hidden layers with 25 neurons showed the highest correla-
tion coefficient. The authors used the back-propagation tech-
nique with the Widrow-Hoff learning rule and compared the
trainrp, traingdx, trainscg, and trainlm as training functions
whereas the trainlm function was selected. The correlation
coefficients were 0.96 and 0.99 for BP and torque, whereas,
for NOX, HC, CO2, and CO, values were 0.71, 0.90, 0.96,
and 0.98, respectively. Najafi etal. [64] evaluated the use
of bioethanol-gasoline blends in an engine. A 2-20-9 con-
figured model developed to forecast the relation between
volumetric efficiency, thermal efficiency, power, torque,
BSFC, CO, CO2, NOx, and HC emissions, using different
combinations of ethanol and speed as inputs. 20 neurons in
one hidden layer were selected. The Back-propagation algo-
rithm opted to predict these results. The gradient descent
rule was adopted to minimize the error. Amongst the trainrp,
traingdx, trainscg, and trainlm functions, the trainlm func-
tion was adopted to train the network. The R values of the
predicted results lie between 0.97 to 1. MRE values were
found between 0.46% to 5.57% and RMSE was found very
low.
In another study, Yu and Arcakliog [65] tested ethanol
blends with unleaded gasoline. The author performed the
test on different ignition timing, compression ratio, density,
and the relative air–fuel ratio. These parameters were fed as
inputs to ANN architecture. Torque and SFC were calculated
A.N.Bhatt, N.Shrivastava
1 3
and taken as output parameters for the network. One hidden
layer containing five neurons was selected. LM and SCG
algorithms compared to train the model. LM found to give
the least error. The logsig transfer function was selected.
After testing, the R2 data for SFC and engine torque were
0.999915 and 0.999977, respectively. Similarly, Danaiah
etal. [66] used a tert butyl alcohol–gasoline blend and
developed a 3-1-10 network to forecast the various engine
characteristics and found an excellent correlation among the
experimental and forecasting results.
Sayin etal. [67] predicted thermal efficiency, SFC, HC,
CO, and exhaust temperature using fuels of three different
octane numbers. 96 engine test run data generated with 95,
93, and 91 octane number gasoline. The lower heating value
of fuel, air inlet temperature, torque, and speed opted as the
input parameters. The model involves the selection of one
hidden layer with 15 neurons with tangent sigmoid as the
transfer function. 70% out of the total engine test run data
chosen for training and the leftover 30% data employed to
test the performance. A back-propagation algorithm and the
Levenberg–Marquardt function opted to adjust the weights.
The predicted output parameters like BSFC, BTE, CO, HC,
and EGT resulted in the mean relative error of 2.24%, 2.97%,
3.64%, 6.66%, and 1.41%; root mean square error of 17.05g/
kWh, 0.51%, 0.04%, 1.41%, 19.66ppm, and 2.96°C; and a
correlation coefficient of 0.994, 0.992, 0.992, 0.996, 0.983
respectively.
Mehra etal. [68] carried out the ANN modeling of the
hydrogen-enriched CNG engine. `Hydrogen and CNG in a
separate tank were mixed manually. A model developed to
forecast the torque, BSFC, NOx, CO, HC, and CH4 using
engine load, excess air ratio, spark timing, and HCNG
ratio as inputs. The tansig and LM learning algorithms was
adopted in the ANN structure. The obtained Correlation
coefficient values were close to one.
3.1.3 Model Comparison andOptimization Studies
Researchers compared the engine characteristics predicted
by the different neural network models and other soft com-
puting methods based models. Some researchers carried
out multi-objective optimization studies on the ANN pre-
dicted outputs with different optimization methods. In this
context, Tasdemir etal. [69] investigated the engine char-
acteristics of the gasoline engine. The authors compared
the predicted results of the fuzzy expert system with ANN
models. Hydrocarbon emission, SFC, torque, and power
were selected as output parameters. Intake valve advance-
ment and speed selected as inputs. The back-propagation
algorithm with the tansig transfer function was adopted
in the 2-60-4 ANN network model. The gradient descent
was adopted as the learning function. The FES model was
designed using 48 fuzzy rules and the Mamdani approach
for fuzzy inference mechanism. Regression analysis in
Matlab and t-test in SPSS used to compare the results
of ANN and FES. Both the results are statistically very
close to each other and also to the experimental data. Both
methods are simple and fast in terms of the speed of com-
putation. In another study, Tosun etal. [70] predicted the
injection duration, SFC, and EGT at two different points
using ANN and linear and non-linear regression analysis.
The authors concluded the better accuracy of ANN than
other regression methods.
Zhang etal. [71] studied the method to detect and diag-
nose the misfire in an engine using the group of Luenberger
and the sliding mode technique. A Luenberger sliding mode
observer opted for the estimation of engine combustion
torque. The authors presented an ANN model based on the
experimental misfire events data and obtained engine com-
bustion torque. Three methods, including back-propagation
neural network, Elman neural network, and Support vector
machine, were applied and compared. The study showed
the designed ENN was able to predict the misfire data more
accurately in different test conditions than the other two
methods. The running time of SVM found significantly
lower than ENN and BPNN, whereas the running time of
BPNN was found higher than ENN.
Jo etal. [72] studied the turbocharged gasoline direct
engine, which involves the low-pressure cooled exhaust
gas recirculation to curtail NOx emissions. The long path
of LP-EGR leads to the transport delay, which leads to an
inaccurate estimation of LP-EGR flow. Further, non-linear
characteristics also make the estimation difficult. An accu-
rate estimation of LP-EGR is required to enhance the fuel
economy and for emission curtailments. An ANN model
involving 12 combustion inputs like maximum engine pres-
sure, IMEP, and 10 different mass fractions burnt in a cer-
tain range, used to precisely estimate the LP-EGR flow as
an output. All the inputs normalized in the range of 0 to 1.
The LM based back-propagation algorithm was adopted. The
three optimization algorithms viz HORD, TPE, and random
search were compared to get the optimized value of hyper-
parameters. The result showed that the efficacy of all algo-
rithms to search hyper-parameter was highly considerable.
The results for all three algorithms showed the R2 values of
more than 0.98, and RMSE values smaller than 0.76%.
Martinez etal. [73] developed a model to estimate the
engine NOx emission by the use of injection timing, torque,
intake pressure, speed, ignition point, and throttle data.
The 6-7-9-6-1 network architecture with logsig as transfer
function was selected. The model was modified with three
objective functions using the Ant colony optimization. The
optimization involves the sensitivity analysis using the
hyper-volume metrics and the VIKOR method for decision
making. The obtained correlation coefficient was close to
one.
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
A summary of the important studies are presented in
Table1.
3.2 Compression‑Ignition Engine
This section presents the neural network modeling studies on
the compression ignition engines using firstly, standard die-
sel fuel, secondly, alternative liquid and gaseous fuels, and
thirdly, the comparative modeling and optimization studies.
3.2.1 Models withStandard Fuel
The studies on diesel engines with various operating condi-
tions using standard diesel fuel are summarized. The few
studies adopted multi-input single-output models to pre-
dict each engine parameter by a separate neural network.
A performance map of the Mercedes Benz eight-cylinder
diesel engine was studied by Celik and Arcaklioglu [76].
The authors predicted the SFC curve, fuel–air equivalence
ratio, and EGT as separate ANN models. The engine power,
engine speed, and water temperature were the inputs. Four
neurons in a single hidden layer were adopted. Back-prop-
agation with SCG and LM algorithm showed R2 close to
0.99. RMSE was smaller than 0.03. In a similar study [77],
the fuel consumption of the mining truck was predicted by
ANN using loading time, idle time to load, empty travel
time, payload, idled empty time, loaded travel time as inputs.
A 6-9-9-1 ANN architecture comprising 9 neurons each in
two hidden layers was used and resulted in the MAPE of
around 10 percent. Bietresato etal. [78] studied the pre-
dictive capability of ANN to assess the performance of the
farm tractor indirectly. The authors predicted the BSFC and
torque using indirect measures like EGT and motor oil tem-
perature of four agriculture tractors. R2 values were close
to 0.996. The authors concluded the use of the Gaussian
function over the sigmoidal function.
Exhaust valve temperature and the heat transfer coeffi-
cient between the valve and its seat were determined with
two ANN models separately, by Goudarzi etal. [79]. Two
temperatures at different points of the seat, as obtained
from inverse heat problems, were selected as inputs to the
models. Ten learning algorithms were compared and LM
reported the best convergence characteristics. The model for
predicting the exhaust valve temperature contained 15 and
7 neurons in two hidden layers, whereas the heat transfer
coefficient model contained 10 and 5 neurons, respectively.
In a study on turbocharged diesel engine with the pre-
combustion chamber, the result of speed, throttle-position,
and injection pressure on the brake mean effective pressure,
power, engine torque, fuel flow, SFC, smoke, CO2, SO2, and
NOx were investigated. An ANN model with three inputs
viz; injection pressure, throttle position, and speed examined
to forecast the performance and emissions separately. The
test dataset fed to train and validate the model. A single
hidden layer having 5 to 15 neurons for performance param-
eters, and two hidden layers with combinations of 8-7, 9-5,
9-7, 10-7 neurons for emission parameters were used. The
back-propagation, along with the logsig transfer-function,
was adopted. Three different training functions viz; LM,
CGP, and SCG evaluated, and the authors reported the LM
function as fastest, whereas the CGP function reported the
highest number of errors. The result showed that R2, RMSE
values were close to 0.99, and 0.01and MAPE was less than
8.5. The study highly recommends the use of ANN only if
the experiments are producing steady-state results. Other-
wise, ANN may not be appropriate [80].
The studies based on multi-input multi-output models
are also carried out by some researchers. Roy etal. [42]
observed the consequences of varying EGR strategies on the
common rail direct injection engine and developed an ANN
model for EGR quantification and to enhance the onboard
diagnostics. The model was used to predict the effect of
load, diesel-injected, FIP, and EGR on the SFC, BTE, NOx,
CO2, and particulate matter. A 4-10-10-5 structure of four
inputs, five outputs with two hidden layers having ten neu-
rons in each was adopted. The logsig as the transfer function,
LM as the training algorithm, and minimum MSE as the
loss function criteria considered. The model gave the high
correlation coefficient value from 0.987 to 0.999; MAPE
values varied from 1.1 to 4.57%. Apart from the standard
correlation coefficients, the authors studied the mean square
relative error (MSRE), forecasting uncertainty (Theil uncer-
tainty U2), Nash Sutcliffe Coefficient of Efficiency (NSE),
and Kling–Gupta Efficiency (KGE). The values of MSRE
and Theil uncertainty U2 were relatively low. KGE and NSE
values were found highly in control, indicating the robust-
ness of the developed model.
Some of the authors used the computation fluid dynam-
ics generated data as the input of the neural network model.
Nikzadfar and Shamekhi [81] studied the individual roll of
10 engine inputs on the BSFC, torque, NOx, and soot of 1.5
L diesel engine for better controlling and calibration pur-
poses. Ten different engine input parameters include EGR
rate, temperature and pressure of air, speed, the weight of
primary and pilot fuel, exhaust and rail pressure, injection
crank angle, and pilot retard. The authors developed the
engine model using AVL boost simulation software using
engine geometric and phenomenal properties. AVL- mix-
ing control combustion model used for this purpose, which
was then validated by experimental test data. The simulated
engine model adopted for the development of a neural net-
work model. Four thousand datasets were generated out of
this engine model used for ANN modeling. Due to the non-
linear behavior of some of the output parameters, a two-stage
multi-layer perceptron model was adopted. Both the models
involve two hidden layers, the first contains ten neurons in
A.N.Bhatt, N.Shrivastava
1 3
Table 1 Summary of studies on the SI engine
Author Fuel/Blend Input Parameters Output parameters Algorithm /activation/
transfer function
Network configura-
tion
Maximum statistical
efficiency
Other prediction/opti-
mization method
Togun and Baysec
[55]
Gasoline Ignition advance,
throttling status, and
the engine speed
SFC and torque Back-propagation,
Levenberg–Mar-
quardt, Logsig
logsig
3-13-1
3-15-1
R = 0.99
Cay etal. [56] Gasoline Fuel flow, speed, inlet
manifold tem-
perature, torque, and
water temperature
BSFC, EGT, and
effective power
Back-propagation,
SCG, Logsig
5-7-3 R2 = 0.99
Sahin [57] Gasoline Ignition angle, engine
speed, peak ioniza-
tion current with
location
Excess air coefficient Back-propagation,
Scaled Conjugate
Gradient, Logsig
4-9-9-1 MAPE = 12.3,
R2 = 0.99
Cay etal. [61] Methanol-gasoline Blend percentage,
Engine speed,
torque, fuel type
SFC, air–fuel ratio,
CO, HC
Back-propagation,
SCG, BFGS, LM,
and RP, Logsig
4-7-1, 4-11-1, 4-14-1,
and 4-7-1
R2 = 0.99
Kapusuz etal. [62] Methanol and ethanol
blends with gasoline
Power, torque, fuel
consumption per
hour
BSFC Back-propagation,
LM, Tansig
3-10-1, 3-10-1, 3-7-1,
and 3-18-1
R = 0.99
Kiani etal. [63] Ethanol Blend percentage,
load, Engine speed
Torque, BP, NOx,
CO2, CO, and HC
Back-propagation,
LM, Tansig
3-25-25-6 R = 0.98
Najafi etal. [64] Ethanol Blend %, Engine
speed
Torque, power, BTE,
BSFC, CO, HC,
CO2, NOx, volumet-
ric efficiency
Back-propagation,
LM
2-23-9 R = 0.97
Danaiah etal. [66] Tert butyl alcohol Blend percentage,
Load, Engine speed
BSFC, BTE, CO,
CO2, HC, NOx, O2
Back-propagation,
LM
3-1-10 RMSE = 0.999%
Mehra etal. [68] HCNG Engine load, excess
air ratio, spark tim-
ing, and HCNG ratio
Torque, BSFC, NOx,
CO, HC, and CH4
Back-propagation
LM, Tansig
4-15-6 R = 0.999
Tasdemir etal. [69] Gasoline Intake valve advance-
ment, speed
Hydrocarbon emis-
sion, SFC, torque,
and power
Back-propagation,
Tansig,
2-60-4 0.999 FES
Martinez etal. [73] Gasoline Injection timing,
torque, intake pres-
sure, speed, ignition
point, and throttle
NOx Back-propagation,
Log-Sigmoid
6-7-9-6-1 R2 = 0.999 Ant colony optimiza-
tion
Uslu and Celik [74] Gasoline and i-amyl
alcohol
Speed, fuel blend,
compression ratio
BTE, BSFC, BMEP,
HC, CO, NOx
Back-propagation,
LM, Tansig,
3-10-6 R2 = 0.94 RSM
Liu etal. [75] Gasoline, Butanol,
propanol, ethanol,
methanol
Compression ratio,
speed, alcohol%,
alcohol type
Torque, fuel con-
sumption
HC, CO
GRNN, Gaussian and
Linear
4-185-2 R2 = 0.93 MLS-SVR, ANFIS
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
each layer whereas the second model contains ten and eight
neurons in each layer. Exhaust temperature, torque, and aspi-
rated air were the outputs of the first model. These outputs
were used for the second stage model to forecast the soot and
NOx. The study evaluated the individual role of these inputs
on the outputs using perturbation sensitivity analysis. The
authors concluded that the injected fuel mass significantly
affected engine performance and emissions parameters.
3.2.2 Models withAlternative Fuels
Many studies investigate the variety of alternative fuels by
adopting the neural network models to forecast the perfor-
mance emission characteristics of CI engines. The studies
presented are categorized into liquid and gaseous alternative
fuels.
3.2.2.1 Liquid Fuels The researchers investigated the use
of alternative liquid fuels like biodiesel of different origins,
alcohols, etc. by the neural network models. Some authors
initially present here, used the MISO models. Muralidharan
etal. [82] analyzed and developed a model to forecast the
emission, performance, and combustion parameters of the
engine using waste cooking oil biodiesel. Three different
ANN models with back-propagation algorithms utilized to
predict parameters separately. The input parameters consid-
ered were different blend percentages, load, compression
ratio, and crank angle. Trainbr, trainrp, traincgf, traingda,
traingdx, trainscg, trainbfg, and trainlm functions compared
with each other. The result showed the minimum conver-
gence time and MSE value with the trainlm function. After
testing the R2 for SFC, BTE, EGT, IMEP, BP, and mechani-
cal efficiency found as 0.9982, 0.998, 0.9994, 0.9972,
0.9995, and 0.9912, respectively.
Gurgen etal. [83] studied the cyclic variability of the
butanol-fuelled engine. The Coefficient-of- variance of
IMEP for engine running on different fuel mixtures and dif-
ferent speeds were experimentally measured. The observed
datasets were incorporated to design an ANN model to fore-
cast the cyclic variability as an output parameter with fuel
mixtures, and speed as inputs. SCG learning algorithm used
to train the 2–11-1 ANN network architecture. The R2 value
was in the range of 0.737 to 0.9677.
Arumugam etal. [84] used a 4-2-1 configured ANN
model to estimate the emission and performance characteris-
tics separately for the rapeseed biodiesel fuelled engine. Dif-
ferent percentages of biodiesel blend, BP, BSFC, and EGT
were the inputs to the neural network. The SFC is relatively
the same for all diesel blends except B20R (20% rapeseed
methyl ester + 80% Diesel), which get a slightly higher read-
ing than diesel. The obtained R-value was close to 1 and
MRE less than 5%.
The studies presented here onwards used MIMO models.
The diesel engine performance using preheated crude palm
oil was studied by Yusaf etal. [85]. The performance param-
eters were quite comparable to ordinary diesel, and break
power is slightly higher than diesel. The CO emission was
disappointing as its concentration in the exhaust was high,
which indicates the incomplete combustion of the CPO. The
power, BSFC, NOx, CO, CO2, and EGT predicted through
the neural network model, taking different percentage blend,
engine speed as inputs, and back-propagation algorithm to
teach the network. Initially, network weight and biases were
initiated arbitrarily. Gradient descent rule used to minimize
errors. The logsig function selected for the hidden layer and
a 2-25-6 network configuration was selected. The study
involves training the ANN by 80% out of the 40 sample
data. Nearly accurate results as compared to experimental
results observed with MSE of 0.0004.
Kshirsagar and Anand [86] modeled the varying injection
timing and pressure to predict engine performance using
Calophyllum inophyllum methyl ester. Two different ANN
model was developed, first with 4-16-3 architecture to fore-
cast the performance parameters and second with 4-14-14-6
architecture to forecast the emission characteristics. Load,
fuel blends, injection pressure, and timing were the inputs
to both models. R, NSE, forecasting uncertainty values were
in the range of 0.99879–0.99993, 0.990406–0.999802, and
0.011525–0.049462, respectively. These values indicate the
robustness and high effectiveness of the developed model.
Mega Motors Company, the largest car manufacturing
unit in Iran, studied the BSFC, torque, CO, and HC emission
of a two-cylinder engine running on waste cooking oil, and
developed a model to predict these parameters based on the
biodiesel blend and speed of the engine. Logsig activation
function with a single hidden layer containing 25 neurons
was selected. R values were near one for all outputs with
MSE was 0.0004 [50].
In a study, a 6–49-15 ANN network architecture used for
condition-based monitoring and fault detection of the vessel
diesel engine. The model used for predicting fuel consump-
tion and faulty conditions like faulty fuel injectors, clogged
air cooler, clogged filter, polluted turbine. The data from
propeller shaft torque, fuel oil flow meter, thermocouples,
and pressure sensors selected as inputs. The model approxi-
mates the results close to the actual one [87].
To investigate the thermal barrier coated engine, Kumar
etal. [88] used Lanthanum zirconate to coat the piston,
valves, and cylinder head. Pongamia Pinnata biodiesel was
tested in the study to analyze the BSFC, BTE, HC, NOx,
and CO. These parameters were predicted through the ANN
model having network configuration 3-6-5, whereas coating,
load, and fuel type took as inputs to ANN. For training, the
back-propagation algorithm was adopted. The transfer func-
tion for the hidden and output layer was Tangent sigmoid.
A.N.Bhatt, N.Shrivastava
1 3
MSE and MRE of the predicted results were 0.002 and 6.8%,
respectively.
In another study, Lapuerta etal. [89] analyzed the effect
of biodiesel on particulate matter emissions. A model was
developed to estimate the quantity of insoluble and soluble
particulates on different engine speeds and blend percent-
ages. The LM non-linear fitting approach opted for training
the network. The results showed that palmitic acid methyl
esters mainly affect the insoluble particulate present in
biodiesel. Oğuz etal. [90] used a 2-28-4 configured ANN
architecture to forecast the performance characteristics of
bioethanol-biodiesel fuelled engines. The torque, power, fuel
consumption, and BSFC predicted with the help of engine
revolutions and the fuel type as the inputs to the network.
The tansig transfer function was selected. The values of
training speed (β) and learning ratio (α) were chosen as 0.3.
The reliability significance value for the predicted results
was 99.94%.
Shivakumar etal. [45] studied the consequences of vary-
ing fuel injection timing and compression ratio on an engine
running on waste cooking. BSFC and NOX emission for bio-
diesel blended fuel found relatively higher than neat diesel,
whereas smoke and HC emissions were relatively less. These
results predicted through the developed ANN model. The
algorithm used to train the model was back-propagation. The
blend ratio, load, injection timing, and compression ratio
selected as inputs, whereas exhaust temperature, energy con-
sumption, and thermal efficiency as outputs parameters. The
MRE for emission and performance characteristics are 8%
and 5%, respectively.
Ilangkumaran etal. [91] prepared diethyl ether blended
biodiesel of fish oil. A model to forecast emission and per-
formance parameters was developed. The model had one
hidden layer with 20 neurons. Different load and percent-
age blend were inputs to the model. Correlation coefficients
of BTE and EGT were found as 0.997 and 0.999, whereas,
for exhaust emissions, it was 0.997 to 1. Gharehghani and
Pourrahmani [92] optimize the 36 combinations of water,
biodiesel, and ceO2 nano-particles to minimize exhaust
emissions. The optimum value determined by defining a new
parameter known as the performance-evaluation-of-diesel-
engine (PEDE). An ANN model was developed to get the
highest value of PEDE. The sensitivity analysis obtained the
optimum value of PEDE’s. Similarly, the effects of CeO2
nano-particle addition and consequent ANN model with
3-12-4 architecture with trainlm training rule showed a cor-
relation coefficient close to 1 [93].
Manieniyan etal. [94] carried out the wear analysis of
Mahua oil-fuelled diesel engine with hot and cold EGR. The
authors predicted the wear elements present in lubricating
oil, like Fe, Cu, Co, Zn, Pb, and Mg as output parameters
using the probabilistic neural network (PNN) and radial basis
function neural network. The hot EGR, Cetane number, Cold
EGR, and lube oil time considered as inputs. The smoothing
factor value of 0.8 and a hidden layer with 56 neurons gave
the least MSE for the PNN model. The optimum number
of centers for RBFNN was selected as 225. RBF width was
considered as 0.08. The authors concluded that both models
predicted the values close to the actual one. However, the
RBFNN model predicts better than the PNN model.
In a study, 12 input parameters considered to predict the
12 output parameters simultaneously in an ANN model. The
designed network model was to forecast the CO2, CO, HC,
NO, Root mean square vibrations (vertical, lateral, and lon-
gitudinal directions), Kurtosis (vibration parameter in verti-
cal, lateral, and longitudinal direction), torque, and power of
engine running on biodiesel with alumina nano-particle. The
12 input parameters were LHV, fuel viscosity, density, fuel
blend, speed, manifold pressure, fuel consumption, exhaust
temperature, oil temperature, ambience pressure, relative
humidity, O2 concentration. Architecture with 12-25-25-
12 structure having two hidden layers was used. R values
were close to 1 [95]. Shanmugam etal. [96] prepare hybrid
fuel by mixing ethanol and cottonseed biodiesel. A 2-31-6
configured ANN model was developed. Load and different
blends of hybrid fuel were used as inputs, whereas BTE,
CO2, CO, NOx, HC, and smoke took as the outputs of ANN.
The back-propagation algorithm and LM function adopted.
The correlation coefficient values of the predicted results
were in the range 0.975–0.999.
In another study, Sakthivel etal. [97] investigate the IC
engine fuelled with fish oil-based biodiesel. EGT, BTE, CO,
NOx, HC, CO2, and smoke predicted with the help of the
ANN model having network configuration 2-20-11. Back-
propagation algorithm with trainlm function used to train
the network. Load % and types of fuel blend took as input
parameters. The mean relative error and correlation coeffi-
cient of the predicted results were found as 0.02–3.97% and
0.957–0.999, respectively. Similar results were obtained by
other authors [98–100].
In another study, Canakci etal. [101] work on John
Deere 4276T model engine fuelled with waste frying oil-
based biodiesel to predict torque, load, cylinder pressure,
thermal efficiency, mass flow, fuel flow, injection pressure,
CO, NOx, CO2, EGT, smoke, and HC using five different
neural networks. Firstly different environmental conditions,
cetane numbers, and various fuel properties were consid-
ered as inputs, whereas fuel flow, air mass flow, injection
pressure as outputs to the network. Secondly, all previous
outputs, along with speed and fuel properties, were con-
sidered as inputs to predict various performance and emis-
sions parameters. The third and fourth networks in a similar
way considered to avoid statistical error values of injection
pressure. Finally, to predict the output coming out of the
first and second networks, a fifth network was used with the
input of the first network. Back-propagation, LM, and SCG
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
algorithms were used to train the network. Predicted results
have R2 values of 0.99, and the MAPE was less than five.
In a study to investigate the n heptanes-diesel combustion
in Ford 1.8L direct-injection diesel engine, a computation
fluid dynamics based model developed. RNG k-ε turbu-
lence model to simulate the turbulent flows in the combus-
tion chamber, Dukowicz model for heating and evaporation
of droplets, stochastic dispersion model to simulate the
interaction of turbulent eddies with particles, and finally
Extend Coherent Flame model for turbulent mixing were
adopted. An ANN model to predict the NOx, soot, and CO2
developed. The crank angle, pressure, equivalence ratio,
temperature, O2 concentration, and liquid mass evaporated
considered as input parameters. Levenberge Marquardt train-
ing algorithm with 6-18-3 architecture considered. The R2
values were 0.9951, 0.9995, and 0.9976, for NOx, soot, and
CO2, respectively [102]. In a similar study [103], the CFD
simulation model was used to forecast the wall heat flux
for an n-heptane fuelled diesel engine. In another study by
Salam and Verma [104], a CFD modeling software Diesel-
RK software was used to generate the microalgae fuelled
engine dataset. Diesel-RK employs a multi-zone combustion
model. A neural network model developed to forecast the
emission, performance, and combustion parameters of an
engine using the numerically simulated load, fuel blending,
and injection pressure as inputs to the model. A 3-10-17
network architecture adopted. R values were close to 0.98
for all the predicted outputs.
3.2.2.2 Gaseous Fuels The various researchers developed
the ANN models to study the alternative gaseous fuel in dual
fuel mode. The MISO model studies are initially presented.
Celebi etal. [105] analyze the acoustics and vibrations of
different biodiesel with natural gas addition in an unmodi-
fied engine. Speed, cetane number, CNG flow rate, and den-
sity as the input parameters used for predicting the vibration
and sound pressure level separately. The 4-4-1 and 4-5-1
network architectures were used to investigate the vibration
and SPL, respectively. Purelin and Logsig transfer functions
employed in the output and hidden layers, respectively. The
model outcomes were very close to the experimental values.
ANN modeling of micro combined heat unit operating
in CNG diesel dual fuel mode was carried out by Akkouche
etal. [106]. Three different models developed to predict the
airflow, pilot fuel flow, and exhaust temperature with 3-5-4-
1, 3-3-5-1, and 3-5-3-1 network architecture, respectively,
using biogas flow, methane contents, the power used as
inputs. Purelin and Logsig transfer functions employed in
the output and hidden layers, respectively. Gradient descent
with momentum function selected as learning function. R2
values were close to 1 for all three models.
Javed etal. [107] predicted the noise of the engine run-
ning on biodiesel blended with zinc oxide nano-particles
and H2 induction in dual fuel mode. Sound level in decibel
with a 4–8-1 network predicted using the load, amount of
nano-particles, biodiesel share, and hydrogen share as inputs
with a regression coefficient of 0.99.
The other researchers reported the use of the MIMO
model for engine characterization. In a study, Javed etal.
[108] examined an engine running on five different combi-
nations of biodiesel with H2 induction. The study aimed to
estimate the BTE, BSFC, EGT, NOx, HC, CO, CO2, and O2
at various loading conditions, biodiesel ratio, and H2 share.
A three-layered 3-16-8 network configuration was used,
where different load, biodiesel blend, and the amount of H2
selected as inputs. The data normalized between the ranges
of 0.1 to 0.9. The 63 experimental data (approximately 70%)
used to train the network, 13 datasets (approx 15%) used
to validate, and the remaining 13 data sets used to test the
whole model. A combination of seven different training
algorithms and five various training functions investigated.
LM learning algorithm with tangent sigmoid and logarith-
mic sigmoid transfer functions yields the best result. MSE,
MAPE, and overall regression coefficient of the model were
0.0011, 4.863%, and 0.9936, respectively, with a training
time of 78.135s.
A predictive model to investigate the capability of ANN
for the CNG dual-fuelled engine developed by Yusuf etal.
[109]. BSFC, BP, BTE, EGT, and emission parameters were
investigated as the outputs using engine speed and CNG
% as inputs. 22 neurons in a single hidden layer with LM
training function were selected. Correlation Coefficients for
output parameters were from 0.92 to 0.99.
The vibrations of an engine fuelled with hydroxyl (HHO)
gas-biodiesel examined by Uludamar etal. [110]. HHO gas
introduced via the intake manifold in varying amounts. An
ANN model designed to forecast the consequences of cetane
number, heating value, HHO %, and speed on the vibration
characteristics. Two neurons in the hidden layer gave sat-
isfactory results. The obtained Correlation coefficient was
close to one.
Syed etal. [111] in a similar study, compared the seven
learning algorithms in combination with three transfer func-
tions to evaluate an ANN model developed for hydrogen die-
sel dual-fuel mode engine. A comparative matrix with Seven
training algorithm viz; traingdx, trainlm, trainrp, traingda,
traincgf, trainbfg, and trainscg along with three transfer
functions viz; purelin, logsig, and tansig were prepared for
16 data sets. The matrix reveals that the quasi-Newton back-
propagation and tansig-tansig transfer function found the
best amongst other algorithms.
Taghavifar etal. [112] predicted the convective heat trans-
fer coefficient of the engine head, piston walls, and cylinder
liner of the hydrogen-fuelled diesel engine. In this study, a
CFD based model designed to generate the data. An ANN
model developed with liquid mass evaporated, equivalence
A.N.Bhatt, N.Shrivastava
1 3
ratio, and temperature as inputs. A 3–17-3 network configu-
ration was selected, which yield the RMSE value of 9.13.
3.2.3 Model Comparison andOptimization Studies
This section is divided into two kinds of studies. The first
part involves the comparative studies of the neural net-
work models and with other predictive models. The lat-
ter part includes the ANN modeling and optimization of
engine parameters by applying the different optimization
techniques.
Esonye etal. [113] evaluated the emissions and perfor-
mance of the Perkins diesel engine with different blends of
African pear seed biodiesel. Multi-input and multi output-
ANN model designed to forecast the BSFC, BTE, HC, CO,
and NOx emissions using the fuel blend, speed, and load
as the input parameters. A single hidden layer with seven
neurons along with Tan-sigmoid as transfer function was
selected. The correlation coefficients were between 0.9 and
0.99. The authors also applied the Nelder Mead downhill
optimization simplex method, which is generally used for
multidimensional optimization. The model predictive equa-
tion of Nelder Mead was used to obtain various solutions.
A comparison between ANN and Nelder mead predictive
models was presented. The R2 values for BTE, BSFC, CO,
NOx, and HC by the ANN model were 0.998, 0.811, 0.730,
0.601, and 0.842 respectively, Whereas Nelder mead predic-
tive model showed the values of 0.999, 0.974, 0.996, 0.985,
and 0.998 respectively. Hence, it demonstrates the suprem-
acy of the Nelder mead method over the ANN.
Shamshirband etal. [114] carried out a comparative
study of different models to predict the exegetic variables
of the engine using expanded polystyrene waste blended
with biodiesel. Seven parameters viz; exergy transfer rate to
coolant, fuel exergy rate, exergy rate of exhaust gas, exergy
efficiency, exergy transfer rate to ambient, exergy destruc-
tion rate, and sustainability index examined using different
engine speed, load, and biodiesel percent and polystyrene
contents as inputs. An ANN-based prediction model com-
pared with a support vector machine-based prediction model
coupled with quantum particle swarm optimization (SVM-
QPSO), radial basis function (SVM-RBF), firefly algorithm
(SVM-FFA), and discrete wavelet transform model (SVM-
WT). Results showed that the SVM-WT model showed
superiority over the other four approaches. R2 and R val-
ues were more than 0.9979 and 0.9989, proving the fidel-
ity of the model. However, in another study Yildirim etal.
[115] predicted the engine emissions, vibration, the noise of
engine blended with different biodiesels, and dual-fuelled
hydrogen induction through ANN and SVM. The authors
reported that the ANN outperforms the SVM.
Redel-Macías etal. [116] used a 1/3 octave band analyzer
to measure the combustion noise of the bio-fuelled engine.
A predictive model was developed for the biodiesel combus-
tion noise where power, speed, blend, diesel noise, and 1/3
octave frequency selected as inputs. An ANN model was
compared with the two polynomial regression models. Two
response surface models include the one with interaction
terms and another one without it. Neural network models
studied were product unit neural network and a hybrid of
PUNN with radial basis function (PUNN + RBFNN). ANN
models were found better than response surface models. It
observed that the noise predicted by the hybrid model and
PUNN were close to each other. The authors recommend the
use of pure PUNN due to its simplicity.
Aghbashlo etal. [117] developed the models to forecast
the exegetic performance of engine running on diesel bio-
diesel blend using ANN, genetic programming, extreme
learning machine, and extreme learning machine-with wave-
let transformation. Fuel exergy, exergy efficiency, cooling
water exergy, exhaust gas exergy, ambient exergy, exergy
destruction, and sustainability index were the seven output
parameters predicted by the models. Whereas biodiesel per-
cent expanded polystyrene wastes contents, load and speed
were the four input parameters. The authors reported the
yield of closest and accurate prediction by extreme learning
machine-with wavelet transformation method, followed by
extreme learning machine, ANN and genetic programming.
A similar study was carried out to compare the three differ-
ent models and to forecast the engine emissions and per-
formance using diesel-nano-particle blended fuel. Wavelet
neural network (WNN), back-propagation neural network
(BPNN), and non-linear autoregressive with exogenous
input (NARXNN) were the models compared. All the mod-
els predicted the result very close to actual results; however,
the authors concluded that the WNN with stochastic gradient
algorithm (WNN-SGA) performed better than the other two
models [118].
ANN and GEP based models were used to estimate
the engine emissions and performance opting EGR. GEP
is a population-based evolutionary algorithm. CO2, NOx,
PM, BSFC, and BTE were the outputs of the GEP model,
whereas the load, EGR, fuel-injected, FIP selected as inputs.
The large chromosomes were tested to minimize the error.
The ratio of correlation coefficients of GEP to the ANN
found more than one for all the predicted parameters, which
indicates that the GEP is more accurate than ANN [119].
Instantaneous exhaust emissions of NOx, CO2, particle
number in nucleation mode, particle number in accumula-
tion mode, and the geometric mean diameter of particles
were predicted and compared, using ANN and genetic
algorithm based symbolic regression method. Torque,
speed, acceleration, air temperature, airflow, fuel con-
sumption, and boost pressure were the inputs to models.
The authors concluded that both models predicted the
emissions accurately; however, NOx prediction by the
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
symbolic regression method and CO2 prediction by ANN
was found better than other methods [120].
Researchers optimized the outputs of the ANN model
by different optimization techniques. Diesel, blended
with castor oil biodiesel, was examined in a turbocharged
engine by Etghani etal. [121]. Based on the experimental
dataset, the authors designed an ANN model to forecast
power, BSFC, PM, NOx, CO, and CO2 using biodiesel per-
cent and engine speed data. A 2-15-6 network consisting
of fifteen neurons in a hidden layer selected. LM learning
algorithm, logistic sigmoid transfer functions with 20,000
epochs considered. The output of the model was optimized
considering the six objectives, i.e., maximize the BP and
minimize the BSFC, CO, CO2, NOx, and PM. The modi-
fied NSGA-II method, employing the ε-elimination algo-
rithm considered for optimization. The best trade-off solu-
tion of the six objective functions was determined by the
TOPSIS method.
Krishnamoorthi etal. [122] analyzed the straight veg-
etable oil blended with diethyl ether on different compres-
sion ratios, EGR rate, and engine speeds. Consequently, the
ANN model was developed and optimized. NOx, smoke,
CO, HC, CO2, BTE, and BSFC were the output parameters
forecasted by ANN using the load, compression ratio, and
speed as inputs. A 3-12-7 architecture with LM training
function was selected. R values range from 0.91 to 0.99.
The RSM based optimization method considered to maxi-
mize the engine performance parameters and minimize the
engine exhaust emission parameters. In another similar study
by the same authors [123], ANN modeling was adopted to
forecast the engine parameters. The particle swarm optimi-
zation, along with RSM was considered for investigating the
effect of compression ratio and EGR. Results concluded that
the ANN-PSO-RSM approach provided better results with
considerable accuracy.
Channapattana etal. [124] tested the Honne biodiesel
on various compression ratios, injection timings, and pres-
sure. An ANN model was designed to forecast the emissions
and performance parameters at optimum operating condi-
tions as obtained by the genetic algorithm. Weighted multi-
objective genetic algorithm-based optimization was carried
out to enhance the performance and emission curtailment
simultaneously. The optimum results obtained were adopted
to forecast performance and emissions. Four training func-
tions viz; traingdx, trainlm, trainrp, trainscg compared, and
trainlm function reported the least error.
Tertiary blends of diesel kerosene and ethanol were
prepared and examined by Bhowmik etal. [125]. ANN
model designed to forecast the BTE, energy consumption,
CO, HC, and NOx using the load, kerosene, and ethanol
blend. A 3-9-5 network topology was found to give the best
results. The multi-objective response surface methodology
(MORSM) was adopted to give the optimum result.
Roy etal. [126] examined the emission and performance
trade-off of the CNG dual-fuelled engine. An adaptive merit
function (AMF) designed for BSFC, PM, and NOx trade-off.
An ANN meta-model was designed to establish a correla-
tion between the control variable and the objective function.
Load, CNG energy share (CES), and FIP were selected as
the decision variable. The study involves the Latin Hyper-
cube sampling scheme as the design of experiment strategy.
A Particle swarm optimization technique was employed for
the optimization purpose. On different loading conditions,
the authors reported 2.7 to 9.9% higher AMF than the best
experimental operations. A similar study with hydrogen
dual-fuel in diesel engine under varying EGR strategies was
carried out by Banerjee etal. [127]. Multi-objective particle
swarm optimization on the ANN meta-model platform was
adopted to enhance the performance-emission trade-off of
the dual-fuel operation.
Deb etal. [128] developed an ANN model to predict the
performance and emission parameters of the diesel engine
using hydrogen in dual fuel mode. 2-15-15-6 network archi-
tecture, along with logsig-tansig activation function, was
used. The model was evaluated based on Theil U2, NSE,
and KGE values apart from traditional evaluation criteria,
which proved the robustness of the developed model. A
fuzzy approach implemented to achieve a multi-performance
characteristics index (MPCI) value for optimization. Nine
fuzzy rules were defined, and the highest value of MPCI was
selected to obtain optimum fitment in the transfer function
of the model.
Diesel–CNG dual-fuel engine investigated to prepare an
MLP-ANN model. CO and NOx emissions predicted using
CNG mass flow rate, diesel flow rate, intake temperature,
speed, power as inputs. Experimental and neural network
model uncertainties calculated. The CO and NOx were fore-
casted with the 5-13-1 and 5-14-1 architectures, respectively.
The NSGA-II approach obtained the optimal decision vari-
able based on concerning emissions. Crossover and Muta-
tion probability was selected as 0.7 and 0.4, respectively,
with a population size of 50. The Pareto optimal front was
obtained for CO and NOx trade-off [129].
A summary of the important studies are presented in
Table2.
3.3 HCCI Engine
Homogeneous charge compression ignition engines are
the modified form of compression ignition engines which
utilizes the concept of using homogeneous charge as used
in SI engines rather than heterogeneous charge used in CI
engine. It can be more beneficial in terms of performance
and emissions. Limitations of the HCCI engine using die-
sel-ethanol blends at higher loading conditions studied by
Bahri etal. [133]. The relation between the combustion
A.N.Bhatt, N.Shrivastava
1 3
noise level and cylinder pressure was studied, and after
that, an ANN noise level (ANL) model was developed to
estimate it. Experimental datasets were generated at dif-
ferent steady-state and cyclic loading conditions. The max
peak pressure and three pressures at a different crank angle
used as inputs of ANN and noise level as output. A 4-20-1
network architecture was selected, and the obtained R2 was
0.99. In an earlier similar study, the same authors [134]
detected the misfire event in ethanol fuelled HCCI engine
with five different pressure inputs. Three types of misfires
were considered in the study, namely, low temperature,
fuel cut off, and high dilution (ultra-lean air–fuel mixture).
Four hidden layers with ten neurons were selected. 7800
cyclic data used in the model to detect the misfire in the
engine with 100% accuracy. Bahri etal. [135], developed
a similar model for detecting the ringing operation of the
HCCI engine with ethanol and n-heptanes.
Rezaei etal. [136] analyze HCCI engine performance
with butanol and ethanol separately and developed ANN
models to predict various parameters like BTE, in-cylin-
der pressure, IMEP, HRR, NOx, CO, and HC. The fuel
blend and equivalence ratio were the input parameters of
each model. The authors compared the two neural net-
work models based on feed-forward and radial basis func-
tions. Both feed-forward and RBF model predicted the
performance with less than 4% MAPE. The feed-forward
network proved simple and less complicated, whereas a
radial basis function-based neural network took less train-
ing time.
Maurya and Saxena [137] predicted the ringing char-
acteristics of a hydrogen-fuelled HCCI engine on varying
operating and combustion conditions. The ringing intensity
predicted as the output of the two different ANN models.
The first model involves the combustion parameters as input,
whereas the second involves the operating parameters. The
combustion parameters studied were combustion phasing,
peak pressure, peak temperature, and combustion duration.
The speed, equivalence ratio, and temperature of the inlet
valve opted as operating parameters. The results were com-
pared with the kinetically simulated model. It was observed
that the ANN results were very accurate and the computa-
tion time was very less. Mean square error of 4.17 and 2.21
observed with a combustion-based model and operating
parameter based model, respectively.
Predicting the start of combustion is very crucial for
determining the engine characteristics of the HCCI engine.
The model to detect the start of combustion was developed
by Taghavi etal. [138]. Three models opting for MLP, RBF,
and NARX were compared. A genetic algorithm was used
in this study to optimize the network architecture. GA fairly
improved the regression coefficient of MLP and RBF models
from 0.8965 to 0.96166, and 0.7623 to 0.83991, respectively.
The NARX predicted result showed the best regression
coefficient and was very close to one. It also reduced the
computational time of the NARX model from 3.12 to 0.46s.
In a study, Anarghya etal. [139] investigated the HCCI
engine using methanol, different blends of isooctane with
n-heptanes (PRF), at a different speed, and air temperatures
in ANSYS fluent 3D simulation plate-form. The Zimont
turbulent combustion model, k-ω SST viscous model, and
emission model of fluent used. Consequently, a radial basis
neural network using the Gaussian activation function was
developed. The IMEP, heat release rate, pressure release
rate, CO, NOx, and HC were the predicted output param-
eters, whereas the air–fuel ratio, PRF percent, methanol,
crank angle were the input parameters. ANN-GA method
was adopted to optimize the weights and biases. The authors
compared RBFNN and ANN-GA models and observed that
ANN-GA models gave better results and close to experi-
mental values.
A summary of the important studies are presented in
Table3.
4 Conclusion
This paper presents a brief review of different studies on the
application of the artificial neural network and the methodol-
ogy adopted in solving the complex problems of the inter-
nal combustion engine. The various problems were divided
based on the type of engine, and the fuel used. It includes
the problems of SI, CI, HCCI engines using traditional, and
alternative fuels like biodiesel, alcohol, and gaseous fuels in
dual fuelling. Literature showed the use of different ANN
models like multi-input single-output and multi-input multi-
output approach. The paper profoundly covers the method-
ology for developing the different ANN models and their
statistical analysis for evaluation.
The various application studies involve the performance
and emission predictions, modeling for valve timing, EGR
rate estimation, knock intensity detection, noise prediction,
maintenance of ships, misfire detection, wall flux estimation,
heat transfer coefficient modeling, engine wear determina-
tion, and optimization problems. The various comparative
studies also discussed to compare the different prediction
and optimization models in terms of their accuracy, conveni-
ence, and computation time. In general, ANN was found to
be convenient over other methods due to its simplicity and
give considerable statistical efficiency.
The multi-input multi-output model was found more
convenient than the multi-input single-output model. The
Back-propagation with the Levenberg–Marquardt learn-
ing algorithm using a single hidden layer is commonly
employed for considerable accuracy and convergence speed
to determine most of the engine characteristics. The number
of neurons in the range of 10 to 20 in a hidden layer and
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
Table 2 Summary of studies on the CI engine
Author Fuel/Blend Input parameters Output parameters Algorithm /activation/
transfer function
Network configura-
tion
Maximum statistical
efficiency
Other prediction/opti-
mization method
Siami-Irdemoosa and
Dindarlo [77]
Diesel Loading time, idle
time to load, empty
travel time, payload,
idled empty time,
loaded travel time
Fuel consumption Back-propagation 6-9-9-1 MAPE = 10%
Bietresato etal. [78] Diesel EGT and motor oil
temperature
BSFC and torque Back-propagation,
Gaussian function
2-26-10-1 R2 close to 0.996
Roy etal. [42] Diesel load, diesel-injected,
FIP, and EGR
SFC, BTE, NOx,
CO2, and particulate
matter
Back-propagation,
LM, logsig
4-10-10-5 R = 0.999
Muralidharan etal.
[82]
WCO methyl ester Compression ratio,
blend percentage,
and load
SFC, BTE, BP,
IMEP, mechanical
efficiency, EGT
HC, CO2, CO, NOx
Ignition delay, HRR,
combustion pres-
sure, combustion
duration, mass frac-
tion burnt
Back-propagation,
LM
3-12-6
3-13-4
3-12-5
R2 = 0.998
Gurgen etal. [83] Diesel and butanol Fuel mixtures, and
speed
Cyclic variability Back-propagation,
SCG, Logsig, and
Purelin
2-11-1 R2 = 0.9677
Arumugam etal. [84] Rapeseed oil methyl
ester
Blend percentage,
BP,BSFC,EGT
BTE,NOx,CO2Back-propagation 4-2-1 MRE < 5%
Yusaf etal. [85] Crude palm oil Blend percentage,
Engine speed
BP, BSFC, CO, NOx,
EGT, O2
Back-propagation,
Logsig, and Purelin
2-25-6 MSE = 0.0004
Kshirsagar and Anand
[86]
Calophyllum inophyl-
lum methyl ester
Load, fuel blends,
injection timing, and
injection pressure
BTE, BSEC, EGT;
CO, CO2, HC, O2,
NO, soot
Back-propagation,
LM, Tansig
4-16-3, 4-14-14-6 R = 0.999
Kumar etal. [88] Pongamia Pinnata oil Load, fuel type BTE, BSFC, NOx,
HC, and CO
Back-propagation,
LM, Tansig
3-6-5 MSE = 0.002
Ghobadian etal. [50] Waste Cooking oil Blend percentage,
Engine speed
Torque, SFC, CO, HC Back-
propagation,LM,
Logsig and Linear
2-25-4
Oguz etal. [90] Bioethanol-biodiesel Engine revolutions
and the fuel type
Torque, power, fuel
consumption, and
BSFC
Back-
propagation,Tansig
2-28-4 reliability significance
value = 99.94%
Ilangkumaran etal.
[91]
Diethyl ether and
fish oil
Load and percentage
blend
BTE, HC, EGT, NOx,
CO, CO2, smoke
Back-propagation,
sigmoid and linear
2-20-7 R = 0.999
Manieniyan etal. [94] Mahua biodiesel hot EGR, Cetane
number, Cold EGR,
time
Wear value of Fe, Cu,
CO, Zn, Pb and Mg
PNN, RBFNN,
Gaussian function
R = 0.90-0.99
A.N.Bhatt, N.Shrivastava
1 3
Table 2 (continued)
Author Fuel/Blend Input parameters Output parameters Algorithm /activation/
transfer function
Network configura-
tion
Maximum statistical
efficiency
Other prediction/opti-
mization method
Sakthivel etal. [97] Fish oil Load % and fuel blend EGT, BTE, CO, HC,
NOx, CO2, and
smoke
Back-propagation,
sigmoid and linear
2-20-11 R = 0.999
Shivakumar etal. [45] Waste Cooking oil Load, CR, Blend
percentage, injection
timing
BTE,BSFC, EGT,
NOx
Back-propagation,
sigmoid and linear
4-24-3, 4-22-3 MRE < 8%
Shrivastava etal. [98] Karanja and Rosell oil Diesel, loading,
compression ratio,
Karanja ratio, Rosell
ratio
BTE, BSFC, EGT,
NOx, CO2, ITE,
MRPR, ID, smoke
Back-propagation,
LM, Tansig
5-7-9 R2 = 0.9915
Canakci etal. [101] Waste frying palm oil Lower heating value,
density, viscosity,
cetane number,
relative humidity,
pressure, dry bulb
temperature, Engine
speed
Torque, CO, NOx,
CO2, EGT, HC,
smoke, load,
cylinder pressure,
thermal efficiency,
mass flow, fuel flow,
injection pressure
Back-propagation,
LM, SCG, Logsig
8-8-11 R2 = 0.99
Babu etal. [130] Waste frying oil Pre, main and post
injection timing,
fuel
BTE, BSEC, CO, CD,
ID, CPP, CO2, HC,
NO, and Smoke
Back-propagation,
LM, Tansig
4-8-2
4-14-3
4-13-5
R = 0.999
Akkouche etal. [106]CNG Biogas flow, methane
contents, power
airflow, pilot fuel
flow, and exhaust
temperature
Back-propagation,
GDM, logsig and
purelin
3-5-4-1, 3-3-5-1,
3-5-3-1
R2 = 0.99
Celebi etal. [105] Diesel CNG Speed, cetane number,
CNG flow rate, and
density
vibration and sound
pressure level
Back-propagation,
LM, Logsig, and
Purelin
4-4-1, 4-5-1 R2 = 0.99
Javed etal. [108] Jatropha oil
Hydrogen
Load, biodiesel blend,
H2
BTE, BSFC, NOX,
HC, EGT, O2, CO,
CO2
Back-propagation,
LM, Logsig and
Tansig
3-16-8 MAPE = 0.0011
Uludamar etal. [110] Biodiesel and HHO Cetane number, heat-
ing value, HHO %,
and speed
Vibrations Back-propagation,
LM, Logsig, and
Purelin
4-2-1 R2 = 0.9986
Esonye etal. [113] African pear seed
biodiesel
Fuel blend, speed, and
load
BTE, BSFC, NOx,
HC, and CO
Back-propagation,
LM, Tan-sigmoid
3-7-5 R = 0.99 Nelder Mead downhill
optimization simplex
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
Table 2 (continued)
Author Fuel/Blend Input parameters Output parameters Algorithm /activation/
transfer function
Network configura-
tion
Maximum statistical
efficiency
Other prediction/opti-
mization method
Shamshirband etal.
[114]
Expanded polystyrene
waste
Exergy transfer rate
to coolant, fuel
exergy rate, exergy
rate of exhaust gas,
exergy efficiency,
exergy transfer rate
to ambient, exergy
destruction rate, and
sustainability index
Speed, load, biodiesel
blend, and polysty-
rene contents
Back-propagation,
Logsig
4-6-7 R2 = 0.99 SVM-QPSO, SVM-
RBF, SVM- FFA,
SVM-WT
Maccas etal. [116] Olive–pomace Bio-
diesel and palm oil
biodiesel
Biodiesel combustion
noise
1/3 octave band
frequency, speed,
power, biodiesel
percent, diesel noise
PUNN, RBFNN, and
Hybrid PUNN and
RBFNN, evolution-
ary algorithm
5-2-1
5-5-1
R2 = 90.89–97.8% RSM
Krishnamoorthi etal.
[122]
Straight vegetable oil Compression ratios,
EGR rate, and
engine speeds
NOx, smoke, CO,
HC, CO2, BTE, and
BSFC
Back-propagation,
LM
3-12-7 R = 0.99 RSM
Aydin etal. [131] Biodiesel Injection pressure,
biodiesel blend, load
EGT, BTE BSFC,
NOx, HC, CO,
smoke
Back-propagation,
LM, logistic sig-
moid
3-10-7 R2 = 0.985 RSM
Krishnamoorthi etal.
[123]
Chaulmoogra oil Compression ratio and
EGR
CO, BTE, SFC, NOx,
Smoke, HC
Back-propagation,
LM
3-12-6 R = 0.99 ANN-PSO-RSM
Etghani etal. [121] Castor oil Blend percentage,
Engine speed
BP, BSFC, NOx,
CO2, CO, PM
Back-propagation,
LM logistic sigmoid
2-15-6 R = 0.999 NSGA-II and TOPSIS
ε-elimination,
Hariharan etal. [132] Hydrogen and Lemon-
grass oil
Load, H2, LGO BSEC, BTE, NOx,
HC, CO, and smoke
Back-propagation,
LM, Logsig
3-12-6 R = 0.9945 RSM
A.N.Bhatt, N.Shrivastava
1 3
Table 3 Summary of studies on the HCCI engine
Author Fuel/Blend Input parameters Output parameters Algorithm /activation/
transfer function
Network configuration Maximum
statistical
efficiency
Other prediction/
optimization
method
Bahri etal. [133] Diesel-ethanol Peak pressure and three
pressures at a different
crank angle
Noise level Back-propagation, LM 4-20-1 R2 = 0.99
Bahri etal. [134] Diesel-ethanol Five different pressure Three misfires (low tem-
perature, fuel cut off,
and high dilution)
Back-propagation, LM,
Logsig
5-10-10-10-10-1 MSE = 0.01
Rezaei etal. [136] Butanol and ethanol Fuel blend and equiva-
lence ratio
BTE, in-cylinder pres-
sure, IMEP, HRR,
NOx, CO, and HC
Back-propagation, and
RBF
LM, Tansig
2-15-7 MSE < 4%
Maurya and Saxena [137] Hydrogen Pressure, temperature,
combustion dura-
tion, and combustion
phasing, equivalence
ratio, speed, inlet valve
temperature
Ringing intensity at dif-
ferent conditions
Back-propagation, LM 3-16-1
4-20-1
MSE = 2.21%
Taghavi etal. [138] Diesel Speed, Equivalence
Ratio, EGR, pressure,
Temperature, Octane
number
Start of combustion Back-propagation,
RBFNN, NARX
6-25-20-20-1
6-10-5-1
R = 0.93 GA
Anarghya etal. [139] Methanol Air–fuel ratio, PRF per-
cent, methanol, crank
angle
IMEP, heat release rate,
pressure release rate,
CO, NOx, and HC
RBFNN, Gaussian acti-
vation function
R2 < 0.9927 GA
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
logarithmic sigmoid and tangent sigmoid transfer function
found to give optimum results. This review shows that ANN
can be efficiently used to predict complex engine perfor-
mance, combustion, and emission characteristics and helps
in a cost-effective search for sustainable alternative fuel with
enhanced engine characteristics.
5 Future Scope
A wide variety of literature available on the prediction of
engine characteristics with different test conditions using
the artificial neural network. However, prediction studies
incorporating most of the fuel properties along with test
parameters can be future scope. The combustion chamber
modification, particulate matters particle size estimation, is
one of the critical areas missing in the literature.
Training the neural network is time-consuming, further
comparative studies on the different networks, Deep neural
network involving a higher number of hidden layers, recur-
rent neural network, convolutional neural network are there-
fore recommended.
Declarations
Conflict of interest The authors declare that they have no conflict of
interest.
References
1. Kalghatgi G (2019) Development of fuel/engine systems—the
way forward to sustainable transport. Engineering 5:510–518.
https:// doi. org/ 10. 1016/j. eng. 2019. 01. 009
2. Tsita KG, Kiartzis SJ, Ntavos NK, Pilavachi PA (2019) Next
generation biofuels derived from thermal and chemical conver-
sion of the Greek transport sector. Therm Sci Eng Prog. https://
doi. org/ 10. 1016/j. tsep. 2019. 100387
3. Vandepaer L, Cloutier J, Amor B (2017) Environmental impacts
of Lithium Metal Polymer and Lithium-ion stationary batteries.
Renew Sustain Energy Rev 78:46–60. https:// doi. org/ 10. 1016/j.
rser. 2017. 04. 057
4. Huang B (2018) Recycling of lithium-ion batteries: recent
advances and perspectives. J Power Sources 399:274–286.
https:// doi. org/ 10. 1016/j. jpows our. 2018. 07. 116
5. Zhen X, Wang Y, Liu D (2020) Bio-butanol as a new generation
of clean alternative fuel for SI (spark ignition) and CI (compres-
sion ignition) engines. Renew Energy 147:2494–2521. https://
doi. org/ 10. 1016/j. renene. 2019. 10. 119
6. Alonso Raposo M, Ciuffo B, Ardente F etal (2019) The future
of road transport—implications of automated, connected, low-
carbon and shared mobility
7. Kalghatgi GT (2015) Developments in internal combustion
engines and implications for combustion science and future
transport fuels. Proc Combust Inst 35:101–115. https:// doi. org/
10. 1016/j. proci. 2014. 10. 002
8. Daraei M, Thorin E, Avelin A, Dotzauer E (2019) Potential
biofuel production in a fossil fuel free transportation system:
a scenario for the County of Västmanland in Sweden. Energy
Procedia 158:1330–1336. https:// doi. org/ 10. 1016/j. egypro.
2019. 01. 327
9. Quiroz-Pérez E, Gutiérrez-Antonio C, Vázquez-Román R (2019)
Modelling of production processes for liquid biofuels through
CFD: a review of conventional and intensified technologies.
Chem Eng Process Process Intensif 143:107629. https:// doi. org/
10. 1016/j. cep. 2019. 107629
10. Karthickeyan V, Thiyagarajan S, Geo VE etal (2019) Simulta-
neous reduction of NOx and smoke emissions with low viscous
biofuel in low heat rejection engine using selective catalytic
reduction technique. Fuel 255:115854. https:// doi. org/ 10. 1016/j.
fuel. 2019. 115854
11. Veza I, Said MFM, Latiff ZA (2019) Progress of acetone-
butanol-ethanol (ABE) as biofuel in gasoline and diesel engine:
a review. Fuel Process Technol 196:106179. https:// doi. org/ 10.
1016/j. fuproc. 2019. 106179
12. Shrivastava N (2017) Experimental investigation of performance,
emission, and noise parameters of water-emulsified Karanja
biodiesel: a prospective Indian fuel. J Braz Soc Mech Sci Eng
39:1009–1017. https:// doi. org/ 10. 1007/ s40430- 016- 0581-z
13. Shrivastava N, Shrivastava D, Shrivastava V (2018) Experimen-
tal investigation of performance and emission characteristics of
diesel engine running on jatropha biodiesel with nanoparticels
as additives. Int J Green Energy 15:136–143. https:// doi. org/ 10.
1080/ 15435 075. 2018. 14288 07
14. Shrivastava N, Varma SN, Pandey M (2013) Experimental inves-
tigation of diesel engine using EGR and fuelled with Karanja
oil methyl ester. Int J Sustain Eng 6:307–315. https:// doi. org/ 10.
1080/ 19397 038. 2012. 749310
15. Sarıdemir S, Etem Gürel A, Ağbulut Ü, Bakan F (2020) Investi-
gating the role of fuel injection pressure change on performance
characteristics of a DI-CI engine fuelled with methyl ester. Fuel.
https:// doi. org/ 10. 1016/j. fuel. 2020. 117634
16. Ağbulut Ü, Sarıdemir S (2018) A general view to converting
fossil fuels to cleaner energy source by adding nanoparticles. Int
J Ambient Energy. https:// doi. org/ 10. 1080/ 01430 750. 2018. 15638
22
17. Sarıdemir S, Ağbulut Ü (2019) Combustion, performance, vibra-
tion and noise characteristics of cottonseed methyl ester–diesel
blends fuelled engine. Biofuels. https:// doi. org/ 10. 1080/ 17597
269. 2019. 16676 58
18. Dwivedi G, Jain S, Sharma MP (2011) Impact analysis of bio-
diesel on engine performance—a review. Renew Sustain Energy
Rev 15:4633–4641. https:// doi. org/ 10. 1016/j. rser. 2011. 07. 089
19. Verma P, Sharma MP, Dwivedi G (2016) Impact of alcohol on
biodiesel production and properties. Renew Sustain Energy Rev
56:319–333. https:// doi. org/ 10. 1016/j. rser. 2015. 11. 048
20. Verma P, Sharma MP, Dwivedi G (2016) Evaluation and
enhancement of cold flow properties of palm oil and its biodiesel.
Energy Rep 2:8–13. https:// doi. org/ 10. 1016/j. egyr. 2015. 12. 001
21. Dwivedi G, Sharma MP (2014) Prospects of biodiesel from Pon-
gamia in India. Renew Sustain Energy Rev 32:114–122. https://
doi. org/ 10. 1016/j. rser. 2014. 01. 009
22. Dwivedi G, Sharma MP (2014) Impact of cold flow properties
of biodiesel on engine performance. Renew Sustain Energy Rev
31:650–656. https:// doi. org/ 10. 1016/j. rser. 2013. 12. 035
23. Shrivastava N, Khan MZ (2017) Application of soft computing in
the field of internal combustion engines: a review. Arch Comput
Methods Eng. https:// doi. org/ 10. 1007/ s11831- 017- 9212-9
24. Wong PK, Tam LM, Li K, Vong CM (2010) Engine idle-speed
system modelling and control optimization using artificial intel-
ligence. J Automob Eng. https:// doi. org/ 10. 1017/ CBO97 81107
415324. 004
A.N.Bhatt, N.Shrivastava
1 3
25. Kannan GR, Balasubramanian KR, Anand R (2013) Artificial
neural network approach to study the effect of injection pressure
and timing on diesel engine performance fueled with biodiesel.
Int J Automot Technol 14:507–519. https:// doi. org/ 10. 1007/
s12239
26. Haleem A, Javaid M, Vaishya R (2020) Effects of COVID-19
pandemic in daily life. Curr Med Res Pract. https:// doi. org/ 10.
1016/j. cmrp. 2020. 03. 011
27. Madurai Elavarasan R, Pugazhendhi R (2020) Restructured soci-
ety and environment: a review on potential technological strat-
egies to control the COVID-19 pandemic. Sci Total Environ.
https:// doi. org/ 10. 1016/j. scito tenv. 2020. 138858
28. Nicola M, Alsafi Z, Sohrabi C etal (2020) The socio-economic
implications of the coronavirus and COVID-19 pandemic: a
review. IJS Publishing Group Ltd
29. Karonis D, Lois E, Zannikos F etal (2003) A neural network
approach for the correlation of exhaust emissions from a diesel
engine with diesel fuel properties. Energy Fuels 17(5):1259–1265
30. Traver ML, Atkinson RJ, Atkinson CM (1999) Neural network-
based diesel engine emissions prediction using in-cylinder com-
bustion pressure. SAE Pap 1
31. Hutter M (2005) Universal artificial intelligence
32. Agwu OE, Akpabio JU, Alabi SB, Dosunmu A (2018) Artificial
intelligence techniques and their applications in drilling fluid
engineering: a review. J Pet Sci Eng 167:300–315. https:// doi.
org/ 10. 1016/j. petrol. 2018. 04. 019
33. Haleem A, Javaid M, Khan IH (2019) Current status and applica-
tions of Artificial Intelligence (AI) in medical field: an overview.
Curr Med Res Pract. https:// doi. org/ 10. 1016/j. cmrp. 2019. 11. 005
34. Wang H, Chaffart D, Ricardez-Sandoval LA (2019) Modelling
and optimization of a pilot-scale entrained-flow gasifier using
artificial neural networks. Energy 188:116076. https:// doi. org/
10. 1016/j. energy. 2019. 116076
35. Kartheeswaran S, Christopher Durairaj DD (2017) A data-paral-
lelism approach for PSO-ANN based medical image reconstruc-
tion on a multi-core system. Informatics Med Unlocked. https://
doi. org/ 10. 1016/j. imu. 2017. 05. 001
36. Ferreira LB, da Cunha FF, de Oliveira RA, Fernandes Filho
EI (2019) Estimation of reference evapotranspiration in Brazil
with limited meteorological data using ANN and SVM—a new
approach. J Hydrol. https:// doi. org/ 10. 1016/j. jhydr ol. 2019. 03.
028
37. Falat L, Pancikova L (2015) Quantitative modelling in economics
with advanced artificial neural networks. Procedia Econ Financ.
https:// doi. org/ 10. 1016/ s2212- 5671(15) 01619-6
38. Dwivedi YK, Hughes L, Ismagilova E etal (2019) Artificial
Intelligence (AI): multidisciplinary perspectives on emerging
challenges, opportunities, and agenda for research, practice and
policy. Int J Inf Manage. https:// doi. org/ 10. 1016/j. ijinf omgt. 2019.
08. 002
39. Kalogirou SA (2001) Artificial neural networks in renewable
energy systems applications: a review. Renew Sustain Energy
Rev 5:373–401
40. Haykin S (1994) Neural networks: a comprehensive foundation.
Macmillan. https:// doi. org/ 10. 1017/ S0269 88899 82140 44
41. Huang A, Zhang X, Li R, Chi Y (2018) Memristor neural net-
work design. Memristor Memristive Neural Netw. https:// doi. org/
10. 5772/ intec hopen. 69929
42. Roy S, Banerjee R, Bose PK (2014) Performance and exhaust
emissions prediction of a CRDI assisted single cylinder diesel
engine coupled with EGR using artificial neural network. Appl
Energy 119:330–340. https:// doi. org/ 10. 1016/j. apene rgy. 2014.
01. 044
43. Meireles MR, Almeida PE, Simões MG (2003) A comprehen-
sive review for industrial applicability of ANNs. IEEE Trans Ind
Electron 50:585–601
44. Uzun A (2012) A parametric study for specific fuel consump-
tion of an intercooled diesel engine using a neural network. Fuel
93:189–199. https:// doi. org/ 10. 1016/j. fuel. 2011. 11. 004
45. Shivakumar, Srinivasa Pai P, Shrinivasa Rao BR etal (2011)
Artificial Neural Network based prediction of performance
and emission characteristics of a variable compression ratio CI
engine using WCO as a biodiesel at different injection timings.
Appl Energy 88:2344–2354. https:// doi. org/ 10. 1016/j. apene rgy.
2010. 12. 030
46. Choi Y, Chen JY (2005) Fast prediction of start-of-combustion
in HCCI with combined artificial neural networks and ignition
delay model. Proc Combust Inst 30(2):2711–2718. https:// doi.
org/ 10. 1016/j. proci. 2004. 08. 143
47. Parlak A (2006) Application of artificial neural network to pre-
dict specific fuel consumption and exhaust temperature for a
Diesel engine. Appl Therm Eng 26:824–828. https:// doi. org/ 10.
1016/j. applt herma leng. 2005. 10. 006
48. Arcakliog E (2005) Performance maps of a diesel engine. Appl
Energy 81:247–259. https:// doi. org/ 10. 1016/j. apene rgy. 2004. 08.
003
49. Zweiri YH, Seneviratne LD (2007) Diesel engine indicated
torque estimation based on artificial neural networks, pp 791–798
50. Ghobadian B, Rahimi H, Nikbakht AM etal (2009) Diesel engine
performance and exhaust emission analysis using waste cooking
biodiesel fuel with an artificial neural network. Renew Energy
34:976–982. https:// doi. org/ 10. 1016/j. renene. 2008. 08. 008
51. Yusaf TFTFF, Yousif BFF, Elawad MM etal (2015) Predic-
tion and reduction of diesel engine emissions using a combined
ANN-ACO method. Appl Energy 38:139–150. https:// doi. org/
10. 1016/j. asoc. 2015. 04. 059
52. Hemeida AM, Hassan SA, Mohamed AAA etal (2020) Nature-
inspired algorithms for feed-forward neural network classifiers:
a survey of one decade of research. Ain Shams Eng J. https:// doi.
org/ 10. 1016/j. asej. 2020. 01. 007
53. Kaviani S, Sohn I (2020) Influence of random topology in artifi-
cial neural networks: A survey. ICT Express. https:// doi. org/ 10.
1016/j. icte. 2020. 01. 002
54. Kalogirou SA (2003) Artificial intelligence for the modeling
and control of combustion processes : a review. Progress Energy
Combust Sci 29(6):515–566
55. Kara Togun N, Baysec S (2010) Prediction of torque and specific
fuel consumption of a gasoline engine by using artificial neural
networks. Appl Energy 87:349–355. https:// doi. org/ 10. 1016/j.
apene rgy. 2009. 08. 016
56. Cay Y (2013) Prediction of a gasoline engine performance with
artificial neural network. Fuel 111:324–331. https:// doi. org/ 10.
1016/j. fuel. 2012. 12. 040
57. Şahin F (2015) Effects of engine parameters on ionization cur-
rent and modeling of excess air coefficient by artificial neural
network. Appl Therm Eng 90:94–101. https:// doi. org/ 10. 1016/j.
applt herma leng. 2015. 06. 100
58. Gölc̈u M, Sekmen Y, Erduranli P, Salman MS (2005) Artificial
neural-network based modeling of variable valve-timing in a
spark-ignition engine. Appl Energy 81:187–197. https:// doi. org/
10. 1016/j. apene rgy. 2004. 07. 008
59. Hazar H, Gul H (2016) Modeling analysis of chrome carbide
(Cr3C2) coating on parts of combustion chamber of a SI engine.
Energy 115:76–87. https:// doi. org/ 10. 1016/j. energy. 2016. 08. 083
60. Çay Y, Çiçek A, Kara F, Saǧiroǧlu S (2012) Prediction of engine
performance for an alternative fuel using artificial neural net-
work. Appl Therm Eng 37:217–225. https:// doi. org/ 10. 1016/j.
applt herma leng. 2011. 11. 019
61. Çay Y, Korkmaz I, Çiçek A, Kara F (2013) Prediction of engine
performance and exhaust emissions for gasoline and methanol
using artificial neural network. Energy 50:177–186. https:// doi.
org/ 10. 1016/j. energy. 2012. 10. 052
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
62. Kapusuz M, Ozcan H, Ahmad J (2015) Research of performance
on a spark ignition engine fueled by alcohol e gasoline blends
using arti fi cial neural networks. Appl Therm Eng 91:525–534.
https:// doi. org/ 10. 1016/j. applt herma leng. 2015. 08. 058
63. Deh Kiani MK, Ghobadian B, Tavakoli T etal (2010) Applica-
tion of artificial neural networks for the prediction of perfor-
mance and exhaust emissions in SI engine using ethanol- gaso-
line blends. Energy 35:65–69. https:// doi. org/ 10. 1016/j. energy.
2009. 08. 034
64. Najafi G, Ghobadian B, Tavakoli T etal (2009) Performance
and exhaust emissions of a gasoline engine with ethanol blended
gasoline fuels using artificial neural network. Appl Energy
86:630–639. https:// doi. org/ 10. 1016/j. apene rgy. 2008. 09. 017
65. Yu HS, Arcakliog E (2007) Comparative study of mathematical
and experimental analysis of spark ignition engine performance
used ethanol–gasoline blend fuel. Appl Therm Eng 27:358–368.
https:// doi. org/ 10. 1016/j. applt herma leng. 2006. 07. 027
66. Danaiah P, Kumar PR, Rao YVH (2013) Performance and emis-
sion prediction of a tert butyl alcohol gasoline blended spark-
ignition engine using artificial neural networks. Int J Ambient
Energy 36:37–41. https:// doi. org/ 10. 1080/ 01430 750. 2013.
820147
67. Sayin C, Ertunc HM, Hosoz M etal (2007) Performance and
exhaust emissions of a gasoline engine using artificial neural
network. Appl Therm Eng 27:46–54. https:// doi. org/ 10. 1016/j.
applt herma leng. 2006. 05. 016
68. Mehra RK, Duan H, Luo S etal (2018) Experimental and arti-
ficial neural network (ANN) study of hydrogen enriched com-
pressed natural gas (HCNG) engine under various ignition tim-
ings and excess air ratios. Appl Energy 228:736–754. https:// doi.
org/ 10. 1016/j. apene rgy. 2018. 06. 085
69. Tasdemir S, Saritas I, Ciniviz M, Allahverdi N (2011) Artificial
neural network and fuzzy expert system comparison for pre-
diction of performance and emission parameters on a gasoline
engine. Expert Syst Appl 38:13912–13923. https:// doi. org/ 10.
1016/j. eswa. 2011. 04. 198
70. Tosun E, Aydin K, Merola SS, Irimescu A (2017) Estimation of
operational parameters for a direct injection turbocharged spark
ignition engine by using regression analysis and artificial neural
network. Therm Sci. https:// doi. org/ 10. 2298/ TSCI1 60302 151T
71. Zheng T, Zhang Y, Li Y, Shi L (2019) Real-time combustion
torque estimation and dynamic misfire fault diagnosis in gasoline
engine. Mech Syst Signal Process 126:521–535. https:// doi. org/
10. 1016/j. ymssp. 2019. 02. 048
72. Jo Y, Min K, Jung D etal (2019) Comparative study of the
artificial neural network with three hyper-parameter optimiza-
tion methods for the precise LP-EGR estimation using in-cyl-
inder pressure in a turbocharged GDI engine. Appl Therm Eng
149:1324–1334. https:// doi. org/ 10. 1016/j. applt herma leng. 2018.
12. 139
73. Martínez-Morales J, Quej-Cosgaya H, Lagunas-Jiménez J etal
(2019) Design optimization of multilayer perceptron neural net-
work by ant colony optimization applied to engine emissions
data. Sci China Technol Sci 62:1055–1064. https:// doi. org/ 10.
1007/ s11431- 017- 9235-y
74. Uslu S, Celik MB (2020) Performance and exhaust emission
prediction of a SI engine fueled with I-amyl alcohol–gaso-
line blends: an ANN coupled RSM based optimization. Fuel
265:116922. https:// doi. org/ 10. 1016/j. fuel. 2019. 116922
75. Liu W, Safdari Shadloo M, Tlili I etal (2020) The effect of
alcohol–gasoline fuel blends on the engines’ performances and
emissions. Fuel 276:117977. https:// doi. org/ 10. 1016/j. fuel. 2020.
117977
76. Çelik V, Arcaklioğlu E (2005) Performance maps of a diesel
engine. Appl Energy 81:247–259. https:// doi. org/ 10. 1016/j. apene
rgy. 2004. 08. 003
77. Siami-Irdemoosa E, Dindarloo SR (2015) Prediction of fuel con-
sumption of mining dump trucks: a neural networks approach.
Appl Energy 151:77–84. https:// doi. org/ 10. 1016/j. apene rgy.
2015. 04. 064
78. Bietresato M, Calcante A, Mazzetto F (2015) A neural network
approach for indirectly estimating farm tractors engine perfor-
mances. Fuel 143:144–154. https:// doi. org/ 10. 1016/j. fuel. 2014.
11. 019
79. Goudarzi K, Moosaei A, Gharaati M (2015) Applying artificial
neural networks (ANN) to the estimation of thermal contact con-
ductance in the exhaust valve of internal combustion engine.
Appl Therm Eng 87:688–697. https:// doi. org/ 10. 1016/j. applt
herma leng. 2015. 05. 060
80. Arcaklioğlu E, Çelıkten İ (2005) A diesel engine’s performance
and exhaust emissions. Appl Energy 80:11–22. https:// doi. org/
10. 1016/j. apene rgy. 2004. 03. 004
81. Nikzadfar K, Shamekhi AH (2014) Investigating the relative con-
tribution of operational parameters on performance and emis-
sions of a common-rail diesel engine using neural network. Fuel
125:116–128. https:// doi. org/ 10. 1016/j. fuel. 2014. 02. 021
82. Muralidharan K, Vasudevan D (2014) Applications of artificial
neural networks in prediction of performance, emission and
combustion characteristics of variable compression ratio engine
fuelled with waste cooking oil biodiesel. J Braz Soc Mech Sci
Eng 37:915–928. https:// doi. org/ 10. 1007/ s40430- 014- 0213-4
83. Gürgen S, Ünver B, Altın İ (2018) Prediction of cyclic variability
in a diesel engine fueled with n-butanol and diesel fuel blends
using artificial neural network. Renew Energy 117:538–544.
https:// doi. org/ 10. 1016/j. renene. 2017. 10. 101
84. Arumugam S, Sriram G, Shankara Subramanian PR (2012)
Application of artificial intelligence to predict the performance
and exhaust emissions of diesel engine using rapeseed oil methyl
ester. Procedia Eng 38:853–860. https:// doi. org/ 10. 1016/j. pro-
eng. 2012. 06. 107
85. Yusaf TF, Yousif BF, Elawad MM (2011) Crude palm oil fuel for
diesel-engines: experimental and ANN simulation approaches.
Energy 36:4871–4878. https:// doi. org/ 10. 1016/j. energy. 2011. 05.
032
86. Kshirsagar CM, Anand R (2017) Artificial neural network
applied forecast on a parametric study of Calophyllum ino-
phyllum methyl ester-diesel engine out responses. Appl Energy
189:555–567. https:// doi. org/ 10. 1016/j. apene rgy. 2016. 12. 045
87. Basurko OC, Uriondo Z (2015) Condition-based maintenance for
medium speed diesel engines used in vessels in operation. Appl
Therm Eng 80:404–412. https:// doi. org/ 10. 1016/j. applt herma
leng. 2015. 01. 075
88. Vinay Kumar D, Ravi Kumar P, Kumari MS (2013) Prediction
of performance and emissions of a biodiesel fueled lanthanum
zirconate coated direct injection diesel engine using artificial
neural networks. Procedia Eng 64:993–1002. https:// doi. org/ 10.
1016/j. proeng. 2013. 09. 176
89. Lapuerta M, Rodrı J (2005) Neural networks estimation of diesel
particulate matter composition from transesterified waste oils
blends. Fuel 84:2080–2085. https:// doi. org/ 10. 1016/j. fuel. 2005.
04. 029
90. Oğuz H, Sarıtas I, Baydan HE (2010) Prediction of diesel engine
performance using biofuels with artificial neural network. Expert
Syst Appl 37:6579–6586. https:// doi. org/ 10. 1016/j. eswa. 2010.
02. 128
91. Ilangkumaran M, Sakthivel G, Nagarajan G (2016) Artificial neu-
ral network approach to predict the engine performance of fish
oil biodiesel with diethyl ether using back propagation algorithm.
Int J Ambient Energy 37:446–455. https:// doi. org/ 10. 1080/ 01430
750. 2014. 984082
92. Gharehghani A, Pourrahmani H (2019) Performance evalua-
tion of diesel engines (PEDE) for a diesel-biodiesel fueled CI
A.N.Bhatt, N.Shrivastava
1 3
engine using nano-particles additive. Energy Convers Manag
198:111921. https:// doi. org/ 10. 1016/j. encon man. 2019. 111921
93. Soukht Saraee H, Taghavifar H, Jafarmadar S (2017) Experi-
mental and numerical consideration of the effect of CeO2
nanoparticles on diesel engine performance and exhaust emis-
sion with the aid of artificial neural network. Appl Therm Eng
113:663–672. https:// doi. org/ 10. 1016/j. applt herma leng. 2016.
11. 044
94. Manieniyan V, Vinodhini G, Senthilkumar R, Sivaprakasam S
(2016) Wear element analysis using neural networks of a DI die-
sel engine using biodiesel with exhaust gas recirculation. Energy
114:603–612. https:// doi. org/ 10. 1016/j. energy. 2016. 08. 040
95. Hosseini SH, Taghizadeh-Alisaraei A, Ghobadian B, Abbasza-
deh-Mayvan A (2020) Artificial neural network modeling of per-
formance, emission, and vibration of a CI engine using alumina
nano-catalyst added to diesel-biodiesel blends. Renew Energy
149:951–961. https:// doi. org/ 10. 1016/j. renene. 2019. 10. 080
96. Shanmugam P, Sivakumar V, Murugesan A, Ilangkumaran M
(2011) Performance and exhaust emissions of a diesel engine
using hybrid fuel with an artificial neural network. Energy
Sources Part A Recover Util Environ Eff 33:1440–1450. https://
doi. org/ 10. 1080/ 15567 036. 2010. 539085
97. Sakthivel G, Ilangkumaran M, Nagarajan G (2012) Predicting
the engine performance using ethyl ester of fish oil with the aid
of artificial neural network. Int J Ambient Energy 34:145–158.
https:// doi. org/ 10. 1080/ 01430 750. 2012. 740429
98. Shrivastava P, Salam S, Verma TN, Samuel OD (2020) Experi-
mental and empirical analysis of an IC engine operating with
ternary blends of diesel, karanja and roselle biodiesel. Fuel
262:116608. https:// doi. org/ 10. 1016/j. fuel. 2019. 116608
99. Mohamed Ismail H, Ng HK, Queck CW etal (2012) Artificial
neural networks modelling of engine-out responses for a light-
duty diesel engine fuelled with biodiesel blends. Appl Energy
92:769–777. https:// doi. org/ 10. 1016/j. apene rgy. 2011. 08. 027
100. Canakci M, Erdil A, Arcaklioğlu E (2006) Performance and
exhaust emissions of a biodiesel engine. Appl Energy 83:594–
605. https:// doi. org/ 10. 1016/j. apene rgy. 2005. 05. 003
101. Canakci M, Necati A, Arcaklioglu E, Erdil A (2009) Expert sys-
tems with applications prediction of performance and exhaust
emissions of a diesel engine fueled with biodiesel produced from
waste frying palm oil. Expert Syst Appl 36:9268–9280. https://
doi. org/ 10. 1016/j. eswa. 2008. 12. 005
102. Taghavifar H, Taghavifar H, Mardani A etal (2014) Appraisal of
artificial neural networks to the emission analysis and prediction
of CO2, soot, and NOx of n-heptane fueled engine. J Clean Prod.
https:// doi. org/ 10. 1016/j. jclep ro. 2015. 03. 035
103. Taghavifar H, Taghavifar H, Mardani A etal (2015) A numerical
investigation on the wall heat flux in a di diesel engine fueled
with n-heptane using a coupled CFD and ANN approach. Fuel
140:227–236. https:// doi. org/ 10. 1016/j. fuel. 2014. 09. 092
104. Salam S, Verma TN (2019) Appending empirical modelling to
numerical solution for behaviour characterisation of microalgae
biodiesel. Energy Convers Manag 180:496–510. https:// doi. org/
10. 1016/j. encon man. 2018. 11. 014
105. Çelebi K, Uludamar E, Tosun E etal (2017) Experimental and
artificial neural network approach of noise and vibration char-
acteristic of an unmodified diesel engine fuelled with conven-
tional diesel, and biodiesel blends with natural gas addition. Fuel
197:159–173. https:// doi. org/ 10. 1016/j. fuel. 2017. 01. 113
106. Akkouche N, Loubar K, Nepveu F etal (2020) Micro-combined
heat and power using dual fuel engine and biogas from discon-
tinuous anaerobic digestion. Energy Convers Manag 205:112407.
https:// doi. org/ 10. 1016/j. encon man. 2019. 112407
107. Javed S, Baig RU, Murthy YVVS (2018) Study on noise in a
hydrogen dual-fuelled zinc-oxide nanoparticle blended bio-
diesel engine and the development of an artificial neural network
model. Energy 160:774–782. https:// doi. org/ 10. 1016/j. energy.
2018. 07. 041
108. Javed S, Murthy YVVS, Ulla R, Rao DP (2015) Development of
ANN model for prediction of performance and emission char-
acteristics of hydrogen dual fueled diesel engine with Jatropha
Methyl Ester biodiesel blends. J Nat Gas Sci Eng 26:549–557.
https:// doi. org/ 10. 1016/j. jngse. 2015. 06. 041
109. Yusaf TF, Buttsworth DR, Saleh KH, Yousif BF (2010) CNG-
diesel engine performance and exhaust emission analysis with
the aid of artificial neural network. Appl Energy 87:1661–1669.
https:// doi. org/ 10. 1016/j. apene rgy. 2009. 10. 009
110. Uludamar E, Tosun E, Tüccar G etal (2017) Evaluation of vibra-
tion characteristics of a hydroxyl (HHO) gas generator installed
diesel engine fuelled with different diesel–biodiesel blends. Int
J Hydrogen Energy 42:23352–23360. https:// doi. org/ 10. 1016/j.
ijhyd ene. 2017. 01. 192
111. Syed J, Baig RU, Algarni S etal (2017) Artificial Neural Network
modeling of a hydrogen dual fueled diesel engine characteris-
tics: an experiment approach. Int J Hydrogen Energy 42:14750–
14774. https:// doi. org/ 10. 1016/j. ijhyd ene. 2017. 04. 096
112. Taghavifar H, Taghavifar H, Mardani A etal (2015) On the mod-
eling of convective heat transfer coefficient of hydrogen fueled
diesel engine as affected by combustion parameters using a cou-
pled numerical-artificial neural network approach. Int J Hydro-
gen Energy 40:4370–4381. https:// doi. org/ 10. 1016/j. ijhyd ene.
2015. 01. 140
113. Esonye C, Onukwuli OD, Ofoefule AU, Ogah EO (2019) Multi-
input multi-output (MIMO) ANN and Nelder-Mead’s simplex
based modeling of engine performance and combustion emission
characteristics of biodiesel-diesel blend in CI diesel engine. Appl
Therm Eng. https:// doi. org/ 10. 1016/j. applt herma leng. 2019. 01.
101
114. Shamshirband S, Tabatabaei M, Aghbashlo M etal (2016) Sup-
port vector machine-based exergetic modelling of a di diesel
engine running on biodiesel-diesel blends containing expanded
polystyrene. Appl Therm Eng 94:727–747. https:// doi. org/ 10.
1016/j. applt herma leng. 2015. 10. 140
115. Yıldırım S, Tosun E, Çalık A etal (2019) Artificial intelligence
techniques for the vibration, noise, and emission characteristics
of a hydrogen-enriched diesel engine. Energy Sources Part A
Recover Util Environ Eff. https:// doi. org/ 10. 1080/ 15567 036.
2018. 15505 40
116. Redel-Macías MD, Hervás-Martínez C, Gutiérrez PA etal (2018)
Computational models to predict noise emissions of a diesel
engine fueled with saturated and monounsaturated fatty acid
methyl esters. Energy 144:110–119. https:// doi. org/ 10. 1016/j.
energy. 2017. 11. 143
117. Aghbashlo M, Shamshirband S, Tabatabaei M etal (2016) The
use of ELM-WT (extreme learning machine with wavelet trans-
form algorithm) to predict exergetic performance of a DI diesel
engine running on diesel/biodiesel blends containing polymer
waste. Energy 94:443–456. https:// doi. org/ 10. 1016/j. energy.
2015. 11. 008
118. Rahimi molkdaragh R, Jafarmadar S, Khalilaria S, Soukht Saraee
H, (2018) Prediction of the performance and exhaust emissions
of a compression ignition engine using a wavelet neural network
with a stochastic gradient algorithm. Energy 142:1128–1138.
https:// doi. org/ 10. 1016/j. energy. 2017. 09. 006
119. Roy S, Ghosh A, Das AK, Banerjee R (2015) Development
and validation of a GEP model to predict the performance and
exhaust emission parameters of a CRDI assisted single cylin-
der diesel engine coupled with EGR. Appl Energy 140:52–64.
https:// doi. org/ 10. 1016/j. apene rgy. 2014. 11. 065
120. Domínguez-Sáez A, Rattá GA, Barrios CC (2018) Prediction
of exhaust emission in transient conditions of a diesel engine
fueled with animal fat using Artificial Neural Network and
Application ofArtificial Neural Network forInternal Combustion Engines: AState oftheArt…
1 3
Symbolic Regression. Energy 149:675–683. https:// doi. org/ 10.
1016/j. energy. 2018. 02. 080
121. Etghani MM, Shojaeefard MH, Khalkhali A, Akbari M (2013)
A hybrid method of modified NSGA-II and TOPSIS to opti-
mize performance and emissions of a diesel engine using bio-
diesel. Appl Therm Eng 59:309–315. https:// doi. org/ 10. 1016/j.
applt herma leng. 2013. 05. 041
122. Krishnamoorthi M, Malayalamurthi R, Sakthivel R (2019)
Optimization of compression ignition engine fueled with die-
sel—chaulmoogra oil—diethyl ether blend with engine param-
eters and exhaust gas recirculation. Renew Energy 134:579–
602. https:// doi. org/ 10. 1016/j. renene. 2018. 11. 062
123. Krishnamoorthi M, Malayalamurthi R (2018) Engine char-
acteristics analysis of chaulmoogra oil blends and corrosion
analysis of injector nozzle using scanning electron micros-
copy/energy dispersive spectroscopy. Energy 165:1292–1319.
https:// doi. org/ 10. 1016/j. energy. 2018. 10. 112
124. Channapattana SV, Pawar AA, Kamble PG (2017) Optimisation
of operating parameters of DI-CI engine fueled with second
generation Bio-fuel and development of ANN based predic-
tion model. Appl Energy 187:84–95. https:// doi. org/ 10. 1016/j.
apene rgy. 2016. 11. 030
125. Bhowmik S, Paul A, Panua R etal (2018) Performance-exhaust
emission prediction of diesosenol fueled diesel engine: An
ANN coupled MORSM based optimization. Elsevier
126. Roy S, Das AK, Bose PK, Banerjee R (2014) ANN metamodel
assisted Particle Swarm Optimization of the performance-
emission trade-off characteristics of a single cylinder CRDI
engine under CNG dual-fuel operation. J Nat Gas Sci Eng
21:1156–1162. https:// doi. org/ 10. 1016/j. jngse. 2014. 11. 013
127. Banerjee R, Mikulski M, Chakraborty A etal (2017) ANN
meta-model assisted MOPSO application in an EPA-Tier
4 constrained emission-performance trade-off calibration
problem of a hydrogen-diesel-EGR dual fuel operation. Fuel
208:746–778. https:// doi. org/ 10. 1016/j. fuel. 2017. 07. 037
128. Deb M, Majumder P, Majumder A etal (2016) Application
of artificial intelligence (AI) in characterization of the per-
formance–emission profile of a single cylinder CI engine
operating with hydrogen in dual fuel mode: an ANN approach
with fuzzy-logic based topology optimization. Int J Hydrogen
Energy 41:14330–14350. https:// doi. org/ 10. 1016/j. ijhyd ene.
2016. 07. 016
129. Lotfan S, Ghiasi RA, Fallah M, Sadeghi MH (2016) ANN-based
modeling and reducing dual-fuel engine’s challenging emis-
sions by multi-objective evolutionary algorithm NSGA-II. Appl
Energy 175:91–99. https:// doi. org/ 10. 1016/j. apene rgy. 2016. 04.
099
130. Babu D, Thangarasu V, Ramanathan A (2020) Artificial neural
network approach on forecasting diesel engine characteristics
fuelled with waste frying oil biodiesel. Appl Energy. https:// doi.
org/ 10. 1016/j. apene rgy. 2020. 114612
131. Aydın M, Uslu S, Bahattin Çelik M (2020) Performance and
emission prediction of a compression ignition engine fueled
with biodiesel-diesel blends: a combined application of ANN
and RSM based optimization. Fuel. https:// doi. org/ 10. 1016/j. fuel.
2020. 117472
132. Hariharan N, Senthil V, Krishnamoorthi M, Karthic SV (2020)
Application of artificial neural network and response surface
methodology for predicting and optimizing dual-fuel CI engine
characteristics using hydrogen and bio fuel with water injection.
Fuel 270:117576. https:// doi. org/ 10. 1016/j. fuel. 2020. 117576
133. Bahri B, Shahbakhti M, Aziz AA (2017) Real-time modeling of
ringing in HCCI engines using artificial neural networks. Energy
125:509–518. https:// doi. org/ 10. 1016/j. energy. 2017. 02. 137
134. Bahri B, Aziz AA, Shahbakhti M, Muhamad Said MF (2013)
Understanding and detecting misfire in an HCCI engine fuelled
with ethanol. Appl Energy 108:24–33. https:// doi. org/ 10. 1016/j.
apene rgy. 2013. 03. 004
135. Bahri B, Shahbakhti M, Kannan K, Aziz AA (2016) Identifi-
cation of ringing operation for low temperature combustion
engines. Appl Energy 171:142–152. https:// doi. org/ 10. 1016/j.
apene rgy. 2016. 03. 033
136. Rezaei J, Shahbakhti M, Bahri B, Aziz AA (2015) Performance
prediction of HCCI engines with oxygenated fuels using artificial
neural networks. Appl Energy 138:460–473. https:// doi. org/ 10.
1016/j. apene rgy. 2014. 10. 088
137. Maurya RK, Saxena MR (2018) Characterization of ringing
intensity in a hydrogen-fueled HCCI engine. Int J Hydrogen
Energy 43:9423–9437. https:// doi. org/ 10. 1016/j. ijhyd ene. 2018.
03. 194
138. Taghavi M, Gharehghani A, Nejad FB, Mirsalim M (2019)
Developing a model to predict the start of combustion in HCCI
engine using ANN-GA approach. Energy Convers Manag
195:57–69. https:// doi. org/ 10. 1016/j. encon man. 2019. 05. 015
139. Anarghya A, Rao N, Nayak N etal (2018) Optimized ANN-GA
and experimental analysis of the performance and combustion
characteristics of HCCI engine. Appl Therm Eng 132:841–868.
https:// doi. org/ 10. 1016/j. applt herma leng. 2017. 12. 129
Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
