Fault Detection - Science topic

In 2024, AI is revolutionising fault detection in fuel cell (FC) systems by employing advanced machine learning (ML) algorithms, such as neural networks and support vector machines, to analyse sensor data for early anomaly detection. These AI-driven methods facilitate real-time monitoring, enabling predictive maintenance and fault diagnosis.

Researchers utilise AI to process large datasets from FC operations, employing techniques like data fusion and pattern recognition to identify faults. This approach enhances system reliability and efficiency by reducing downtime, optimising performance, and lowering maintenance costs. The integration of AI results in prolonged FC lifespan, improved energy efficiency, and overall cost reduction, advancing the viability and sustainability of fuel cell technology.

Dear Doctor

Go To

"ABSTRACT:

Uninterruptible power supply is the main motive of power utility companies that motivate them for identifying and locating the different types of faults as quickly as possible to protect the power system prevent complete power black outs using intelligent techniques. Thus, the present research work presents a novel method for detection of fault disturbances based on Wavelet Transform (WT) and Independent Component Analysis (ICA). The voltage signal is taken offline under fault conditions and is being processed through wavelet and ICA for detection. The time-frequency resolution from WT transform detects the fault initiation instant in the signal. Again, a performance index is calculated from independent component analysis under fault condition which is used to detect the fault disturbance in the voltage signal. The proposed approach is tested to be robust enough under various operating scenarios like without noise, with 20-dB noise and variation in frequency. Further, the detection study is carried out using a performance index, energy content, by applying the existing Fourier transform (FT), short time Fourier transform (STFT) and the proposed wavelet transform. Fault disturbances are detected if the energy calculated in each scenario is greater than the corresponding threshold value. The fault detection study is simulated in MATLAB/Simulink for a typical power system."

Hi,

You could find it in "NASA's Open Data Portal"

Best

Thank you for the response Rim Hamdaoui. I would definitely look into the resources suggested by you.

@amoghShukla thank you

Any idea about how to write code for Fault detection in WSN using Deep learning??

Richard Fenner and Steve Mounce have done work using Kalman filters for leak detection in water distribution networks Seyed.

For comparison, you may have a look at https://www.sciencedirect.com/science/article/abs/pii/S0888327004000111 and

CNN and RNN are deep learning algorithms which are also examples of ANN while ANN is a type of machine learning.

ANN has different architectures such as FNN, CNN, RNN, GAN, auto-encoders, SOM, and so on. These different atchitectures are suitable for different kinds of tasks as well as nature and volume of data.

I think that in your question, you referred to FNN ( with F referring to feedforward) as ANN. FNN works as a supervised learning algorithm for mapping numerical input vectors to output vectors (either continuous or categorical outputs). CNN is suitable for learning a model for at least 2D data such as image, RNN is for sequence modelling.

It all depends on the task. I have not worked on fault detection before but I suspect it could involve classifying variables which are likely to be numerical values. The volume of the data too can also contribute to why FNN is used because it is less data-hungry. Even in small data problems involving 2D or more, FNN can be used on the flattened data because it works better on small dataset than CNN.

If there is enough train data for all fault type cases to clamp to input and output, there will not be a need for a fixed threshold.

Fourier Transform only provides information about the frequency components in a signal. In order to detect faults both frequency and time information is needed -'Which frequency component was detected in the signal and at what time?'

There is no obligation to use only wavelet transform. Start with short-time fourier transform (STFT) and then try other transforms. What matters is joint time-frequency analysis!

I strongly recommend this article: https://web.iitd.ac.in/~sumeet/WaveletTutorial.pdf

Siddharth Kamila The primary distinction is that wavelets are localized in both time and frequency, whereas ordinary Fourier transforms are solely concentrated in frequency.

While the Fourier transform generates a frequency domain representation of the signal, the wavelet transform creates a time and frequency domain representation of the signal, giving quick access to localized information about the signal.

The wavelet transform (WT) employs short windows at high frequencies and long windows at low frequencies, as opposed to the typical STFT, which has a single-window size. Wavelets rely on the employment of a mother wavelet function that may be scaled and altered to correspond with signal abnormalities or occurrences.

Gear meshing frequency is a kinematic (rotation speed-related) parameter. If you have another maximum in the spectrum under loading conditions, this effect can probably be related to another source (gear coupling or bearing). Of course, if you have reduced rotation speed under load (motor power drop), you will have shifted the frequency of gear meshing.

Dear Siddharth Kamila:

You can benefit from these valuable Link about your topic:

@https://de.mathworks.com/help/predmaint/ug/Fault-Detection-Using-Data-Based-Models.html.

@https://github.com/AgHarsh/Fault-Detection-in-Power-Microgrid

@https://github.com/matlab-deep-learning/Fault-Detection-Using-Deep-Learning-Classification

##################

Also you can take a look at:

"Fault Detection of the Power System Based on the Chaotic Neural Network and Wavelet Transform"

Abstract:

The safety and stability of the power supply system are affected by some faults that often occur in power system. To solve this problem, a criterion algorithm based on the chaotic neural network (CNN) and a fault detection algorithm based on discrete wavelet transform (DWT) are proposed in this paper. MATLAB/Simulink is used to establish the system model to output fault signals and travelling wave signals. Db4 wavelet decomposes the travelling wave signals into detail signals and approximate signals, and these signals are combined with the two-terminal travelling wave location method to achieve fault location. And the wavelet detail coefficients are extracted to input to the proposed chaotic neural network. The results show that the criterion algorithm can effectively determine whether there are faults in the power system, the fault detection algorithm has the capabilities of locating the system faults accurately, and both algorithms are not affected by fault type, fault location, fault initial angle, and transition resistance.

@https://www.hindawi.com/journals/complexity/2020/8884786/

I hope it will be helpful...

Best wishes...

Siddharth Kamila This tutorial might help https://www.youtube.com/watch?v=FvuDdfO649A

Hi, sir, did you find the fault dataset?

Would it be practical to train a neural network (NN) to diagnose faults based on inputs from the sensor (S).

{ sensor performance inputs } -> NN -> output specific fault F1 or F2 or F3

The training data set would define (discover) unique sensor properties when each specific fault was purposely made.

Because there could be many combinations of faults, for example for F1, F2, F3 single faults

no faults

F1 fault

F2 fault

F3 fault

there are 4 possibilities. But the number of fault conditions increases rapidly for multiple faults with many devices, Fi, i = 1, n. So the AI approach may not be practical.

Dear Prof. Messaoudi,

I am interested to work with you since Renewable energy is one of my main and favorite research field. I have also completed a research in past based on renewable energy sources and their effectiveness.

Hi Amashisha David

Yes. The LabVIEW simulation tool can provide it.

The following paper may help you: "MOTOR CURRENT SIGNATURE ANALYSIS BASED FAULT DIAGNOSIS OF INDUCTION MOTOR" http://ijcns.com/pdf/ijpcscvol6no2-1.pdf

If you are interested in assessing the performance of modelling approaches, please let me recommend these works:

If your topic assumes the comparison of alternative approaches, the FEW-L1 workflow presented in the article above can be used.

Eliseo Galli The bearings in big plants.

!(...) could be defined a border across which one cannot permutate. E.g. <alpha> * !(<beta>) could permit all variants of <alpha> and all variants of <beta>, but no recombinations of parts of alpha with parts of beta..

Regards,

Joachim

two points

Dear Mr. Widagdo Purbowaskito,

We have done a few experiments in past years on BLDC motor. Based on MCSA, 3^rd harmonic components of motor currents are analyzed to localize and isolate different faults. We did not do "fault thresholding" since it will be very specific for a system and it will differ based on the loads connected to the motor. However, for a constant load and operating environment, the magnitude of the 3^rd harmonic components increases as the fault severity rises.

To my knowledge, there is no specific study that establishes a "severity index" based on MCSA. I suggest you look for different frequency domain features i.e. harmonic magnitudes, root variance frequency, peak frequency, SNR, entropy, THD, etc. Not all features will be suitable for all systems, you can find the most suitable features based on your problem definition. I can suggest you the following papers and you may look into the references cited in these papers for similar studies.

(1) T. A. Shifat and J. W. Hur, "An Effective Stator Fault Diagnosis Framework of BLDC Motor Based on Vibration and Current Signals," in IEEE Access, vol. 8, pp. 106968-106981, 2020, doi: 10.1109/ACCESS.2020.3000856.

(2) T. A. Shifat and H. Jang-Wook, "Remaining Useful Life Estimation of BLDC Motor Considering Voltage Degradation and Attention-Based Neural Network," in IEEE Access, vol. 8, pp. 168414-168428, 2020, doi: 10.1109/ACCESS.2020.3023335.

Yes dear Loai Gomaa this what I said before.

But I would like to expect that the short circuit may have less inductive and capacitive parasitic than the electrical loads. Consequently, if this is the case, the rate of rise of the current at shortcircuit would be larger.

But this still needs verification by more investigations.

Best wishes

Dear Vivek Kumar Sharma,

Look at these papers, maybe useful:

Hardware problem, or sometimes the software

This is a trial an error process in a multi.machine system, first of all you need to know a clearing time for a stable operating condition and another time for an unstable condition, then you can try another time from such an interval. Should the resulting condition is unstable you have to reduce the clearing time and run the simulation again, in this way you are reducing the interval until you reach the CCT

Dear Farzin,

Nice sharing. Thank you.

Ashish

Dear syed taha Ahmed

you can see the attach file for more details

It depends on the type of Cyber Attack. Some can, and some can be difficult to detect (like the Replay Attack).

Nice Contribution Sauvik Biswas

Mohammed Deeb

Thanks for the answer, but I want to implement this using libsvm toolbox. The SVDD function here differs from the one in the aforementioned toolbox. There are some variables like positive cost or negative cost which I do not know how to select.

Rahul Kumar I don't go along with that idea, there is a broad range of papers on this regard. Intensive research looking at recent developments and classification, however, is missing.

BTW, I appreciate your response and the recommended book.

Dear

i suggest K-mean

Hi Syed Ali,

You may follow: https://www.timeshighereducation.com/ which for all PostDoc jobs and my be helpful for you.

Good Luck

Ali

you can do patient matching (like propensity score based methods) or kNN, and look at patients who do not match.

I am engaged in modeling of computer central processing papeline. Problems of pape in other spheres are not my area.

Hello Raghuraj , do you know something about Security Requirements Engineering? There are many problems associated to that topic, and I think, is very important. Please, read this paper

Distance relay is used to detect the fault by line impedance.

A technique to detect the location of the different faults on a transmission lines for quick and reliable operation of protection schemes. The simulation is developed in MATLAB to generate the fundamental component of the transient voltage and current simultaneously both in time and frequency domain. One cycle of waveform, covering pre-fault and post-fault information is abstracted for analysis. The discrete wavelet transform (DWT) is used for data preprocessing. It is applied for decomposition of fault transients, because of its ability to extract information from the transient signal, simultaneously both in time and frequency domain. MATLAB software is used to simulate different operating and fault conditions on high voltage transmission line, namely single phase to ground fault, line to line fault, double line to ground and three phase short circuit. The application of the wavelet transform to estimate the fault location on transmission line has been investigated. The ability of wavelets to decompose the signal into frequency bands in both time and frequency allows accurate fault detection. The most suitable wavelet family has been made to identify for use in estimating the fault location on transmission line. Simulation of single line to ground fault (S-L-G) for 735kv, 300km transmission line was performed using SIMULINK MATLAB SOFTWARE. The waveforms obtained from SIMULINK have been converted as a MATLAB file for feature extraction. DWT has been used to analyze the signal to obtain the coefficients for estimating the fault location

Hi, have a look here,

Hello. I assume you mean synthetic aperture radar (SAR) images taken from some altitude above the earth. If this is your data source, you can get plenty of images (many thousands). Assuming you want to use a supervised machine learning technique, you will also need ground truth, or annotated images (masks) that say where the fault zone is in each appropriate image. Once you have both image data {X(i)} and the corresponding masks {Y(i)}, you have a supervised learning problem to which many techniques are applicable. The main contenders are regression trees and forests, support vector machines and convolutional neural networks (CNN). CNNs can usually outperform the other approaches if you have enough training data, but they can be difficult to optimise and generally require a GPU rather than a CPU. If you don't use CNNs, you will need some kind of statistical description of what a fault zone looks like, that is, you need to characterise the fault zone in terms of its statistical features (edges, etc). Many feature descriptors are available for this purpose.

Please recommend if you find this answer useful.

I am not sure that understand the whole idea, without single line diagram, but let me try to give you few thoughts. If this neutral busbar common, you should have to use highest current transformer ratio, 3000/x A. Perhaps is better to use secondary values of 1A. With this value, it is easiest to handle many issues, which brings 5A nominal secondary current values. Today is also common to have 1A even on generator side. Utility company mostly new solution have with 1A.

Did you find the dataset? I am also interested in getting such a dataset.

Hi

i thin you must attention to connection and interaction type between bodies

suggest test different connection type and select better of that

Dear Roger,

unfortunately, I don't know any benchmark specially related to Oil and Gas industry. However, there are some industrial process benchmarks that could be very interesting and may provide challenges very similar to that faced in some Oil and Gas processes.

1) Tennessee Eastman process

Download: https://depts.washington.edu/control/LARRY/TE/download.html

Paper: https://www.sciencedirect.com/science/article/pii/S2405896315010666

2) DAMADICS

Download: https://cs2ac.upc.edu/en/research-projects/ue-projects/Damadics

Paper: https://doi.org/10.1016/j.conengprac.2005.06.015

3) Sim3Tanks

Download: https://github.com/e-controls/Sim3Tanks

Paper: https://doi.org/10.1109/ACCESS.2018.2874752

In particular, Sim3Tanks was developed by the research group that I'm engaged, and we can provide you more informations if you need.

Best regards,

Iury Bessa

Dear Guru,

I'm never implement a system in timed Petri net, your idea is an interesting fashion. Now I'm applying the techniques of other modeling languages with Colored Petri net for partially variable observation of a process design. such as:

May you get the solutions as soon.

You can read the literature yourself and come up with your own conclusions. It would do you good, you would sound less like a non-expert rambling about something you have very little clue about. I am done here.

Hi Richard,

Thank you for the suggestion. I will visit the lab and see what is possible.

Regards,

Madhurjya

Modern numerical distance protection tracking all the time all values for the purpose of correct operation. Therefore, the any angle is also track. In numerical distance protection system, there are several parameters for defining characteristic of quadrilateral distance zones. With those parameters, it can be influence on relay protection operations. Slope of distance protection characteristic can be fine-tune for various goals. For example, high loaded lines, phase to phase faults, phase to ground fault, power swing, pole slip, etc. For you can be useful to read manuals for relay protection from several vendors, GE, SEL, Toshiba, ABB, Siemens, etc.

Dear Sdi,

Please refer to the followings:

1. http://research.cs.wisc.edu/dmi/svm/psvm/

2.

3. https://www.researchgate.net/topic/MATLAB

It will be not be easy. Designing, developing and implementing a practical working Kalman filter has a number of pitfalls. Here are a few tips derived from a good deal of experience with extended Kalman filters in all sorts of aerospace applications:

1. Make sure that the partials with respect to state variables are right.

2. Check that the observation errors are close to Gaussian and independent from one measurement time to the next

3. Pay much attention to the differential equations of motion and their numerical solution method. Best to use a self-starting Runge-Kutta method.

4. Approximate the state transition matrix by differentiating the integration formula----an old, very practical aerospace engineering trick.

5. Devote a good deal of attention to Process Noise in the state transition---a good practical method is to numerically integrate the equations of motion with the operational RK method along the baseline trajectory along with a higher order method, and difference resultant state vectors at each integration point and average the first and second order statistics of the vector differences. The first average of the first difference will give you some notion of bias while the averaged second order statistics will be your process noise matrix.

6. Avoid using the noxious notion of covariance error analysis in which your run the filter along the baseline trajectory without processing observations, just updating the state error covariance matrix. This can be very misleading.

7. Construct the entire filter first and run the observations through it----that's the sine qua non.

8. Be sure to compute chi-square of the prediction-observation. This statistic will tell you whether your extended KF is working---It should average to the dimension of the measurement.

I hope you find these tips helpful. Let me know how things work out , if you need help, and best of luck.

Roy Danchick

Fault detection in power system means not prediction even before it is initiated. Strictly speaking its about methods to predict and detect faults at an early stage in power systems. PMU measurements are used to predict the conditions on power systems. There is very little literature available in this area of research. Most of the cases, Artificial Neural Networks or Fuzzy logics are used for this task.

Some of the papers are as follows.

Thanks a lot. Have a great day.

Regards,

Hehong

Thanks Karima

Can you please share with the SVM code with me?

Dear Paulin,

I suggest you to see links in topic.

-Safety Mechanisms for the Reliable Operation of 3D Vehicles

https://repositorio.iscte-iul.pt/.../Sem%20indice%20biblio%20e%20...

-ء إ تث ءؤؤ حئءخ تثءج

-Mass threshold for 'harmless' drones - Anders la Cour-Harbo, 2017

-hard science - Design me a Mars drone - Worldbuilding Stack Exchange

https://worldbuilding.stackexchange.com/.../design-me-a-mars-dro...

-Aalborg Universitet Mass threshold for 'harmless' drones La ... - VBN

Best regards

Dear Faeze Sdi,

Why proximal SVM are you using? You may try other algorithms for fault detection.

Dear Madhurjya Dev,

According to recent survey, rotating components in wind turbine can experience a number of failures during operation, where around 50% of failures are related to bearings, 10% are relevant to rotors, and 12% are related to other faults including unbalance. During my long study and research in this arena, I find bearing is the single most competent, which is account for maximum failure.

So, you may start with bearing fault detection and diagnosis (FDD). If you talk about approach, data-driven approach is the best alternative of model-based approach, specifically for large scale industries, due to the rapid development of data mining, data acquisition technique, and machine learning technique since last several decades.

I may share you recent articles relevant to your request.

Thanks,

M M Manjurul Islam

University of Ulsan, South Korea

Dear Jiri Kovar,

I don't know exactly about GETFRAME. Would you please explain more?

I want to create motional time window in order to fault detection in each window.

Dear Aisha,

We have already developed an induction motor model mainly developed to extract faults signals (i.e. stator current). details bout this mode could be found in Dr. Elhoussin Elbouchikhi PhD thesis:

Regards,

Mohamed

Hello,

SVM evaluation can be improved. But, before that the problem needs to be clearly defined. Feature selection based on your domain may increase the SVM performance also. In addition, parameter tuning is also a fair choice, indeed.

Thanks,

Sobhan

I have use CCA algorithm for fault detection. The paper entitled "Fault Detection for Non-Gaussian Processes Using Generalized Canonical Correlation Analysis and Randomized Algorithms" may help you.

A review of process fault detection and diagnosis

Part III: Process history based methods

Dear Hehong,

Regarding the first point of your questions, you can have a look on our recent works on fault detection in an electric grid:

1.

2.

3.

Regards,

Mohamed

Dear Eric,

welcome,

This is a project to specify, concept, design and implement a time domain reflectometer .

The steps to develop such an instrument is to put requirements and specifications of such instrument such as:

- The test object, the test method. the length of the transmission line, the speed of the pulse transmission on the transmission line.

From these data you can obtain the ranges of the time of flight of pulses on the transmission line.

- The measuring concept is to transmit a narrow pulse along the transmission line. Part of this pulse will be reflected back to the transmitting side where it will be received. By measuring the time delay between the transmitted pulse and the received pulses one can determine the time of flight of the signal T. Knowing the speed of propagation v of the waves on the transmission line one can calculate the distance to the reflection point based on the relation L = vT/2

- So you need a pulse generator with adjustable pulse width and pulse repetition rate, You can get such parameters from reference equipment.

You need a receiver of the reflected pulses where you they may need to be amplified as they may be attenuated on the transmission line.

Then you need to sample the two pulses or pulse streams, the transmitted and the received with high sampling rate since the time measurement accuracy is limited by this sampling time.

After sampling you convert them into digital form and store them in a memory. You need to repeat this for few consequent pulses or any suitable time window. Then by correlation of the two wave forms you can get the delay time. So you need to build all these building blocks using on FPGA.

There other methods that can be used to calculate the delay time but this method may be one of the suitable ones.

The rest of the calculations can be done also using FPGA building blocks.

- After conception using matlab for example you can develop your vhdl code and then synthesize this code on FPGA.

- Then you can down load your design on an fpga to test it.

This may be a short outline about the tasks that one carry out in the project.

wish you sucess

Recently I came accross this reference:

S. Hanninen, Single phase earth faults in high impedance grounded networks, Espoo: Technical Research Centre of Finland, VTT Publications , 2001

It reports on measurement data collected during the two years period. Hope this helps...

Hi Anjana, you can use this reference to eliminate the fault. Best Regards Dino

I wrote this topic "Software Enterprise Development for the Organization" for HOAQ Journal. Please access it through the link below.

Dear samir,

Thanks for your help.

Regards

Farzin

Reservoir Computing methods are well suited to modelling the deterministic components of the signal. I have seen, in passing, a number of papers that have used Echo State Networks to classify faults in mechanical engineering.

@Roy Sir. May be the combination of Lb and the DC-Link capacitor achieve resonant situation. This circumstances can be taken care by choosing a different value of inductor. However the DC-Link voltage is not still stabilized. The difficulties is with the duty cycle. During the charging and discharging the duty cycle increases. May be it is due to the dynamics initiated by two controller. Kindly check the attachment for duty pulses of the buck-boost converter . Fault is initiated at 3 sec.