ArticlePDF Available

Abstract and Figures

This paper presents the development and evaluation of a digital method for multi-criteria optimized production planning and control of production equipment in a case-study of an Austrian metal casting manufacturer. Increased energy efficiency is a major requirement for production enterprises, especially for energy intensive production sectors such as casting. Despite the significant energy-efficiency potential through optimized planning and the acknowledged application potential for sophisticated simulation-based methods, digital tools for practical planning applications are still lacking. The authors develop a planning method featuring a hybrid (discrete-continuous) simulation-based multi-criteria optimization (a multi-stage hybrid heuristic and metaheuristic method) for a metal casting manufacturer and apply it to a heat treatment process, that requires order batching and sequencing/scheduling on parallel machines, considering complex restrictions. The results show a ~10% global goal optimization potential, including traditional business goals and energy efficiency, with a ~6% energy optimization. A basic feasibility demonstration of applying the method to synchronize energy demand with fluctuating supply by considering flexible energy prices is conducted. The method is designed to be included in the planning loop of metal casting companies: receiving orders, machine availability, temperature data and (optional) current energy market price-data as input and returning an optimized plan to the production-IT systems for implementation.
Content may be subject to copyright.
Manufacturing and
Materials Processing
Journal of
Article
Simulation-Based Multi-Criteria Optimization of
Parallel Heat Treatment Furnaces at a
Casting Manufacturer
Thomas Sobottka 1, 2, *, Felix Kamhuber 1and Bernhard Heinzl 3
1Fraunhofer Austria Research GmbH, 1040 Vienna, Austria; felix.kamhuber@fraunhofer.at
2Department of Industrial and Systems Engineering, Institute of Management Science,
Vienna University of Technology, 1040 Vienna, Austria
3Research Division of Automation Systems, Institute of Computer Engineering,
Vienna University of Technology, 1040 Vienna, Austria; bernhard.heinzl@tuwien.ac.at
*Correspondence: thomas.sobottka@fraunhofer.at; Tel.: +43-676-888-61626
Received: 10 August 2020; Accepted: 14 September 2020; Published: 17 September 2020


Abstract:
This paper presents the development and evaluation of a digital method for multi-criteria
optimized production planning and control of production equipment in a case-study of an Austrian
metal casting manufacturer. Increased energy eciency is a major requirement for production
enterprises, especially for energy intensive production sectors such as casting. Despite the significant
energy-eciency potential through optimized planning and the acknowledged application potential
for sophisticated simulation-based methods, digital tools for practical planning applications are
still lacking. The authors develop a planning method featuring a hybrid (discrete-continuous)
simulation-based multi-criteria optimization (a multi-stage hybrid heuristic and metaheuristic
method) for a metal casting manufacturer and apply it to a heat treatment process, that requires
order batching and sequencing/scheduling on parallel machines, considering complex restrictions.
The results show a ~10% global goal optimization potential, including traditional business goals and
energy eciency, with a ~6% energy optimization. A basic feasibility demonstration of applying the
method to synchronize energy demand with fluctuating supply by considering flexible energy prices
is conducted. The method is designed to be included in the planning loop of metal casting companies:
receiving orders, machine availability, temperature data and (optional) current energy market
price-data as input and returning an optimized plan to the production-IT systems for implementation.
Keywords:
production planning and control; simulation; optimization; genetic algorithm; heuristics;
energy eciency; heat treatment; scheduling; batching; case study
1. Introduction
Heat treatment processes represent one of the most energy-intensive manufacturing processes
of cast steel production and usually accounts for ~70% of the total process energy consumption [
1
].
Thus, increased energy eciency in steel production is a key factor for competitiveness in global
markets, due to long-term rising costs of energy. The manufacturing industry accounts for 36% of
CO
2
emissions and is responsible for 31% of the corresponding primary energy consumption [
2
]
and digital planning methods are a major potential contributor toward increasing energy eciency
for production companies [
3
]. A higher level of energy-eciency can be achieved by either thermal
process (parameter) improvements [
4
6
], optimized demand-side management [
7
], or production
planning [
8
12
] improvements, e.g., ecient production scheduling. The casting manufacturing
system comprises of casting, heat treatment including cooling, and subsequent grinding. According to
Wang [
1
], significant variables and uncertainties exist in the operation of heat treatment equipment,
J. Manuf. Mater. Process. 2020,4, 94; doi:10.3390/jmmp4030094 www.mdpi.com/journal/jmmp
J. Manuf. Mater. Process. 2020,4, 94 2 of 21
which have to be considered within scheduling, such as gas heating values, processing times, due dates,
and setup-times. Since the availability of ready-to-use digital planning methods for optimized energy
eciency are both crucial for industrial implementations of this planning objective and unavailable
for most applications, this paper aims at contributing a case study of a successful development and
evaluation of an optimized integrated industrial planning including a presentation of the underlying
method and its development.
The contribution at hand focuses on the optimization of parallel heat treatment furnace processes
using the upstream casting process as input data. The method is based on a digital model of
the real-world system, i.e., a simulation, parametrized with measurements and data from the real
system and additional real-time data from the enterprise resource planning (ERP) system (e.g., orders,
due-dates, routing tables) as well as variable energy prices for electricity (optional, in some scenarios).
The digital model is utilized by an optimization method as the evaluation function. Once the optimized
plan is obtained, the plan is fed back into the ERP/MES (manufacturing execution system) for execution,
thus closing the loop from real-world system to digital system/twin and back. The planning is meant to
operate on a regular basis, e.g., updated planning every day with additional updates in case of changed
circumstances (e.g., machine breakdowns or unavailable material), as a rolling horizon planning with
a planning horizon of ~1 week.
The simulation itself follows a hybrid discrete/continuous modelling approach that combines
discrete material flow simulation with continuous modelling of the energy flows. This makes it possible
to accurately model transient energetic processes, such as the heating process of furnaces. Especially in
the context of cyber-physical systems (CPS), hybrid simulation is well suited as a tool to model both
the cyber part (discrete) and the physical part (continuous) and provide an integrated view for analysis.
This integrated view enables to incorporate dynamic eects that lie in the interactions between the
physical and the cyber domains and thereby uncover additional potentials for dynamic optimizations.
The use case simulation builds on previous works, in particular a hybrid simulation engine together
with a library of hybrid modelling components that can be instantiated, connected, and configured
for dierent use cases. This library and some of the components have been extended to incorporate
additional requirements imposed by the use case. This includes a crane for loading and unloading
the heat treatment furnaces as well adapting the furnace temperature control to time-dependent
temperature profiles.
This paper is structured as follows: The following Section 2provides a review of relevant literature.
This is followed by an introduction into the case-study and its requirements in Section 3. In Section 4,
the base method developed by the authors in previous work is briefly presented. This is followed by
the development of the major components of the case-study specific method presented in Section 5
(simulation) and Section 6(optimization). Section 7features the case-study evaluation and discussion
of experimental results, followed by a conclusion and outlook in Section 8.
2. Background and Related Work
2.1. Literature on Optimized Planning Approaches
The paper at hand focusses on the heat treatment process in a casting plant. Thus, the optimized
planning considers a batch process and order scheduling/sequencing on parallel machines, with several
technological restrictions—e.g., not all products are eligible for processing on all machines—and a
complex system behavior, i.e., limited crane handling capacity for loading/unloading of heat treatment
furnaces and a complex energy system with heat exchange between considerable material masses,
the furnaces and the production hall.
There are several approaches for an optimized batching and scheduling of orders, and they are
based on heuristics/rules, metaheuristics, or a combination of both. Yu employed a genetic algorithm
(GA) to optimize hybrid flow-shop scheduling and considered machine eligibility [
13
]. Cheng developed
an ant colony optimization for batching in a considerably simpler system of a single machine process [
14
].
J. Manuf. Mater. Process. 2020,4, 94 3 of 21
Baykaso˘glu covered the cases of parallel heat treatment furnaces as a parallel machine scheduling
problem (PMSP), features technological constraints (i.e., not all products can be processed on all
furnaces), and introduces a greedy randomized adaptive search [
15
]. This search algorithm combines a
rule-based approach with limited stochastic elements and a local search component. This is meant
to increase optimization speed but potentially leaves some global optimization potential through
a stochastic search component untapped. Lin develops an iterated hybrid metaheuristic based on
the population-based metaheuristic called electromagnetism-like algorithm for a parallel machine
scheduling problem [
16
]. Yilmaz addresses the PMSP with a GA, featuring random key chromosomes,
enabling a simultaneously optimized machine assignment and scheduling of orders [
17
]. Jiang’s
optimization for the entire steelmaking-continuous casting process chain in a steel production plant
is based on dierential evolution in a multi-stage optimization with dispatching rules to handle the
problem complexity [
12
]. The heat treatment process and the associated batching is not considered in
detail. The approaches listed above solve the order batching and/or the scheduling/sequencing, usually
for standardized datasets to compare algorithm performance. They feature simplified system behavior
and do not include simulation to consider the dynamic behavior of resources and the energy system.
Approaches featuring simulation can be found in Radim’s heuristic and simulation-based
scheduling and batching method for parallel heat treatment furnaces [
18
]. The rule-based batching
follows Goldratt’s theory of constraints, is applied to a real-world production facility but it does not
make use of stochastic optimization methods such as metaheuristics. All approaches mentioned thus
far feature objective functions for the optimization with one or two part-goals and do not explicitly
consider energy consumption.
Approaches considering energy consumption are the following: Huang presents an energy
saving scheduling for a flexible flow-shop production, based on a GA [
19
], without simulation and
without batching. Cheng used the GA for a two criteria batch optimization, pursuing CO2 emission
reduction [
14
]. Wang considered an objective function containing tardiness of orders and energy (gas)
consumption, using a GA for batch scheduling of orders in a heat-treatment application [
1
]. In these
approaches, the energy consumption was considered via functions and simplified models without the
simulation of complex system behavior.
The following approaches consider the system behavior of the production system (material flow
and production process) and the associated energy behavior (e.g., energy consumption and emission
of heat) through simulation: Thiede used co-simulation to model the material flow, production
equipment, and components of the production and its energy system [
20
]. The optimization is
conducted manually and scenario-based without an automatic optimization module. Rager’s approach
utilizes a discrete event simulation (DES) in combination with a GA to model the material flow, while the
energy consumption is considered with deterministic values [
21
]. The simulation-based optimization
features sequencing and scheduling of orders and does not include the control of equipment; e.g.,
switching machines on and o. The authors of the publication at hand were co-developers of a
simulation-based optimization for a multi-criteria optimization, including energy eciency, with a
comprehensive simultaneous simulation of both energy and production system behavior in a hybrid
discrete-continuous simulation method [
22
]. This approach is the basis for the method developed for
the heat treatment use-case, which is introduced in detail in Sections 5and 6. The base-method is
explained in Section 4.
The approaches listed thus far, as representatives of relevant reference approaches in literature,
have used heuristics and metaheuristics to solve the NP-hard problem of batching and scheduling of
orders on parallel machines. The GA and other evolution-based metaheuristics are the most common
successfully applied algorithms. The system behavior is either modelled mathematically within the
optimization with simplified features, or simulated, to consider more detailed real-world systems,
with the optimization using the simulation as its evaluation function. Concerning the simulation
of complex real-life systems there are fewer available approaches and those that are available do
not cover batching processes representing an important feature of the heat treatment in the use-case
J. Manuf. Mater. Process. 2020,4, 94 4 of 21
to be solved. There are examples in the literature for data-based and machine-learning approaches
to modelling energy demand [
23
], however these are currently not suitable for and adapted to the
requirements of optimized production planning with the necessary detailed level of system behavior.
Nonetheless, the integration of data-based modelling may oer future benefits, e.g., in reducing the
computational and modelling eort associated with complex simulations (there have been preliminary
steps in this direction for the base method of this paper [
24
]). Following the literature analysis,
a simulation-based optimization with a customized GA was the method chosen for the requirements in
this publication—this is also consistent with the result of method comparisons the authors conducted
for similar planning problems in the development of the base method presented in Section 4.
2.2. Literature Concerning Simulation
There already exists a substantial body of work regarding simulation of production systems
for energy eciency [
25
27
]. This is driven by an increasing demand from companies for software
supporting simulation and integration of energy eciency performance [
28
]. Dias [
29
] provides a
comprehensive overview about the most common DES environments. The challenge is to model both
the material-flow/production-process behavior and the energy system behavior and their interactions.
A simplified way, also found in approaches listed in Section 2.1, is to consider the energy consumption
deterministically in a DES. In the use-case at hand, the thermal-physical behavior is important, thus both
a discrete and continuous simulation behavior is required. A more comprehensive review of simulation
methods and applications used in the manufacturing domain is provided in [
30
]. The authors conclude
an increasing interest in hybrid simulation techniques for handling complex real-world simulation tasks.
However, this topic is still only covered by a minority of available publications [
26
]. Typically, hybrid
simulation takes the form of co-simulation [
27
,
28
] that couples dierent simulation environments in
a way that the respective sub-models exchange data at given times during a simulation run using a
suitable coupling middleware. Co-simulation, however, limits the level of integration between discrete
and continuous simulation, the communication across simulation environments is computationally
expensive, and the models and their elements are dicult to reuse for other applications since the
system boundaries and basic model structure diers in both sub-models [
29
]. In contrast, our hybrid
simulation approach incorporates discrete-event simulation (DES) and continuous simulation models
at the building block level of an object-oriented simulation.
The authors of [
31
] employ a simulation-based approach for system analysis and propose
an integrated simulation of interdisciplinary systems, although the models implement a simplified
hybridization without differential equations, thereby excluding transient dynamics. In [
32
], a co-simulation
is proposed for a similar application. Other applications for hybrid simulation of continuous and
discrete systems can be found in smart grids [33], distributed control systems [34], or healthcare [35].
A dierent hybrid process modelling approach [
6
] based on numerical simulation and analytical
equations proposes to calculate the radiation, convection, and conduction heat transfer in heat
treatment processes to optimize the workpiece loading and the thermal schedule. The furnace and
atmosphere temperatures are simplified in this approach to be uniform and equal and disregards the
three-dimensional shape of the individual workpieces. The contribution is also focused on the process
modelling view only and neglects the production planning and scheduling processes, which is central
to the case-study at hand.
3. Case Study Introduction
In this section the casting manufacturer case study is introduced as it represents the requirement
for the specific planning method to be developed. Figure 1gives a schematic overview of the main
components. The case study is based on an Austrian casting manufacturer and comprises five parallel
heat treatment furnaces/ovens of dierent sizes, capacity, and energy requirements.
J. Manuf. Mater. Process. 2020,4, 94 5 of 21
Figure 1.
Schematic overview of the production layout comprising five heat treatment furnaces with
associated cooling stations and first in–first out (FIFO) loading buers as well as a crane for loading
and unloading.
Four of the furnaces are powered by natural gas, while the remaining one uses electric energy
for heating. They can perform a total of nine dierent heat treatment processes (clearing, annealing,
warm-up, relaxation, glowing, normalizing, air/water/oil-based steel tempering) diering in their
individual temperature profiles and heat treatment durations for each individual article type.
These combinations result in roughly 150 dierent heat treatment programs. Downstream of the
furnaces are stations for cooling the workpieces after heating. Depending on the treatment, this takes
place in the air or in a water or oil bath. The five furnaces are loaded and unloaded by a single crane
that can only operate on one station at a time. Each furnace is loaded from a holding area in front
of it that acts as a FIFO (first in–first out) buer. There, the workpieces are batched and packed onto
furnace grates. After treatment, the furnace grates are removed individually to be carried to the
subsequent station (cooling). Depending on the temperature requirements of the treatment process,
the furnace may still be kept switched on to maintain the temperature. The opening of the furnace
door for removing a grate however causes a drop in temperature, which must be compensated again.
The resulting optimization problem is a combination of a batching task and the scheduling and
sequencing of orders on parallel machines. The batches, consisting of furnace grates filled with cast
iron workpieces, must be created by the planning method in order to be processed. The resulting
process consists of the following steps:
1.
Combining jobs to batches, which are sequenced and scheduled for processing on the heat
treatment furnaces;
2. Transfer of batches to the furnaces via crane;
3. Processing in the furnaces;
4. Release and transfer to cooling stations via crane.
The arrival times of jobs for heat treatment, i.e., the system considered in this paper, are determined
by the schedule of the upstream casting process. This casting schedule is provided by the ERP system,
which also factors in a considerable amount of rework, since a considerable share of the products have
to undergo one or multiple additional heat treatment process iterations, because of quality issues that
are only identifiable after the heat treatment process. The actuating variables of the optimization are:
order batching, assigning batches to heat treatment furnaces, scheduling/sequencing the processing of
J. Manuf. Mater. Process. 2020,4, 94 6 of 21
the batches in the furnaces, and controlling the machines (times for switching the machines on and o
and thus control the heating process).
4. Basic Planning Method
The presented approach is based on the planning method developed in preceding research
by the authors [
22
,
36
], utilizing hybrid simulation as an evaluation function for an iterative
metaheuristic optimization module. The basic optimization procedure and interaction between
simulation, optimization, and real-world production planning IT (ERP) are depicted in Figure 2
Figure 2. Overview of the optimization process.
The ERP/MES provide the main input data for the method: current orders and due dates.
This paper is focused on the interaction with ERP/MES systems, since these are the IT systems
responsible for production planning and control in a typical industrial automation system landscape [
37
].
These ERP/MES systems are complemented by SCADA (Supervisory Control and Data Acquisition)
software [
38
] for handling the acquisition and storage of production feedback data. These data are
then aggregated and sent to the ERP/MES system, from where they can be used for initialization and
alignment of the simulation with the real world.
Together with (optional) variable energy prices (e.g., electricity spot market prices) and weather
data (in the current development state this is restricted to outside temperature trend), this is the starting
point for the optimization. The optimization initializes simulation evaluations of all intermediate
solutions—the simulation feedback in turn is evaluated by the optimization and its objective function,
thus informing the next optimization step. This cycle is repeated until a good solution is found and the
resulting optimized plan is sent to the planners, ready to be updated into the ERP/MES for execution.
The following two subsections summarize the two main components of the planning method: the
simulation and the optimization modules. The batching module is part of the case study specific
additions to the method that will be presented in Section 6.
4.1. Hybrid Simulation
The aim of the simulation is to provide evaluations of dierent production scenarios in terms
of energy usage, production flow, and delivery tardiness. The simulation thus forms the basis for a
simulation-supported optimization procedure to optimize production plans for the heat treatment
stations while also considering energy eciency.
In the context of interdisciplinary assessment of energy eciency in production, it is important to
be able to model aspects from dierent engineering domains with high accuracy. While the material
J. Manuf. Mater. Process. 2020,4, 94 7 of 21
flow is typically modelled as discrete entities within a discrete-event simulation, energy flow is more
accurately described using continuous dierential equations. Continuous representation of energy
flow, as opposed to discrete energy profiles, enables to accurately incorporate transient dynamics, like
for example the heat-up process of a furnace, which have an impact on the overall energy consumption.
To accurately capture the dynamic interactions between these two domains, a hybrid
discrete/continuous simulation approach with a tight integration between the discrete and the
continuous models is required. Realizing this integration at the component level improves modularity
and reusability of hybrid component-based models, which is crucial in practice for managing the
complexity of real-world applications [36].
For these reasons, we employ a formal model description, called hyPDEVS [
39
] for hybrid system
modelling [
22
]. Compared to using typical co-simulation methods, hyPDEVS oers tighter hybrid
integration and improved modularity [
36
] by following a strict component-based paradigm that defines
atomic and coupled components, which can be combined to create new application models. This enables
modular hierarchical model development of hybrid simulation models that are easily extensible.
From a software perspective, these components can be implemented in an object-oriented manner,
in which component classes can be instantiated and reused in dierent contexts. This enables to
implement libraries of model components for various engineering domains. Application engineers
then use these pre-defined component models to create application models for new use cases [40].
In the context of the use case in Figure 1, the furnace is an example of such a class (called oven),
with dierent parameter values for each instance to represent the various characteristics (size, capacity,
heat transfer, etc.) of the dierent stations. Another example of a class is the handling resource
presented in Section 5.
Based on the hyPDEVS formalism, a standalone hybrid simulator was developed together with
a library of components for modelling industrial production systems [
22
,
41
]. The components are
designed for reusability and take into account entity exchange as well as energy balance equations.
More details on the hybrid simulation are described in [39,42].
4.2. Metaheuristic Optimization
Production planning optimization problems are NP-hardand thus require approximation methods
that provide near optimal solutions in good time. Since the search space in these optimization scenarios
typically feature multiple local optima, metaheuristics with their ability to seek global optima with
stochastic principles for a wide range of problems have been chosen for the optimization method.
Multiple metaheuristics have been compared concerning their optimization performance. A GA, with a
set of tuning and customization measures for optimal optimization performance, was chosen as the
best fit for the method [43]. The customizations are:
A guided search by adapted operators in the GA (this prevents the GA from mixing parts of the
chromosome that determine dierent variables and for some variables focusses the mutations in a
certain direction, thus reducing the number of practically infeasible solutions);
A memory function from the Tabu Search algorithm (this prevents the algorithm from evaluating
solutions that have already been evaluated before);
A mixed integer optimization (setting an optimal step size for the optimization significantly
reduces the search space without compromising accuracy in practical terms);
Hybridization by combining the GA with Pattern Search and determining the optimal
population size.
The developments and adaptations described in Section 6adapt the approach to the requirements
of the casting manufacturer—mainly comprising new models for complex heat treatment furnaces,
optimization strategies to support complex batch-forming processes and an optimization method
supporting longer planning horizons in this industry sector.
J. Manuf. Mater. Process. 2020,4, 94 8 of 21
5. Hybrid Simulation for the Case Study
Using the hyPDEVS simulator and component libraries described in Section 4.1, the use case
under study (see Figure 1) was implemented in C++. The library includes the model of an industrial
furnace for thermal treatment of entities (work pieces, etc.). The model considers the thermal mass of
the furnace as well as a product-specific entity mass, including the furnace grate. Given that, the heat
transfer of waste heat to the outside as well as heat input into the workpieces are calculated, consisting
mainly of heat conduction with a correction factor for radiation and convection.
Compared to other previous use cases, some of which are presented in [
41
,
44
], the furnace model
had to be extended to incorporate specific requirements: on the one hand, the furnace must be able to
follow a given time-dependent temperature profile (instead of just maintaining a constant temperature),
which is dierent for each treatment process. A selection of these temperature profiles is illustrated in
Figure 3.
Figure 3. Dierent temperature profiles for dierent heat treatment processes.
On the other hand, a sudden drop in furnace temperature from opening the furnace door must be
considered in the model. This was modelled by a time-dependent heat transfer coecient, which can
be specified over a time series. The furnace station is modelled as a hyPDEVS-coupled component,
meaning that it itself is comprised of other, smaller components, representing e.g., the entity queue,
thermal cell, or the temperature controller. There are also dedicated (atomic) components for reading
and handling the time series values.
The crane for loading and unloading the heating furnaces has been implemented as a new class,
see Figure 4. The crane represents a bottleneck in the use case and must consider intervals for loading
and unloading as well as during the cooling period. An internal state machine defines the process logic
of the class, see Figure 5. From the default state standby, an entity – which in this case constitutes a
furnace grate—can be picked up at its input EIN (state incoming), the duration of which is modelled as
T1. Subsequently, the crane moves the entity to its destination within a time interval T2 (state holding)
before it exits at the output EOUT (state output). When the receiving station has acknowledged the
arrival of the entity (signal EOUTcom), the crane is blocked for an additional interval T3 (state waiting),
during which the crane moves back to the starting point.
Figure 4. Crane component which is implemented as a handling resource class.
J. Manuf. Mater. Process. 2020,4, 94 9 of 21
Figure 5. State machine for the handling resource.
The state machine is kept simple in order to not unnecessarily complicate the entity flow. It does
use a setup duration matrix but assumes uniform durations in the scenarios presented hereafter.
It also does not consider any human operator needed to operate the crane and instead assumes ideal
availability. Breakdown behavior of equipment can be added later, the main reason to not consider it in
this case-study was the lack of data to describe the behavior; a safety margin in the available capacity
accounts for production disturbances for now. Furthermore, the model only accounts for the material
flow, and is therefore purely discrete. Any energy required for the crane operation is assumed to be
negligible and has thus not been modelled.
This abstract modelling on the other hand enables to reuse the handling resource class for other
logistical purposes, other than a crane. A handling resource may for example also represent a forklift or an
elevator. This makes these components versatile and reusable in dierent contexts. After implementing
the overall use case model, component parameters were calibrated based on production feedback data
as well as energy consumption data for a given week in 2018. This ensures that the overall energy
demand is depicted correctly, which is relevant for our application, while the modelling simplifications
only result in minor deviations in the temporal aspect of the heat transfer.
The model validation was conducted as follows: for each of the five heat treatment furnaces,
the simulation model was parametrized with the actual energy consumption (gas and electricity)
recorded/measured in multiple reference weeks in 2018, considering the production program executed
in those weeks. The parameter values (in particular for the thermal mass
Mth
and heat transmission
coecient UA) were selected in such a way that, for each furnace
i
, the Root Mean Square (RMS) value
(1) of the error between calculated and measured consumption, is minimized for all production jobs
j
.
RMSi=r1
niXjJob siWj,means Wj,calc2, (1)
The resulting parametrization equals the machine-specific physical characteristics, which was
implemented in the simulation models for the production equipment (i.e., dierential equation systems).
The measured/recorded energy consumption values were obtained from the installed data acquisition
system in the company and, for gas consumption, were additionally checked against raw meter
readings recorded in a dierent system. Following that, the actual production program that was
produced in the reference weeks of our optimization scenario and which is dierent from the programs
in the parametrization reference weeks, was simulated (the actual and not the optimized program was
simulated). The simulation results—mainly energy consumption and basic discrete behavior—were
then compared with the actual recorded/measured data for the validation period. The results, presented
J. Manuf. Mater. Process. 2020,4, 94 10 of 21
in Table 1show that the RMS errors during the validation period are comparable to those during the
original parametrization.
Table 1. RMS and relative error for parametrization and validation.
UA [W/K] Mth [J/K] RMSpar RMSvalid erelvalid
Oven 1 284.27 1.643 ×10732.95 44.87 14.16%
Oven 2 520.37 1.923 ×10736.76 55.05 5.14%
Oven 3 45.06 2.487 ×10645.81 80.79 7.06%
Oven 4 385.21 1.015 ×10718.07 72.91 5.72%
Oven 5 464.00 3.075 ×10765.78 70.81 9.72%
For most major equipment, we have found deviations of simulated vs. actual/measured energy
consumption of below 10%, and no relevant faults in the discrete system behavior. We followed up
several of the slightly higher than average deviations (e.g., heat treatment runs for certain product
batches and associated heat treatment programs) and could explain the deviations largely with
spontaneous deviations from the standard process, e.g., running a heat treatment process for longer
than the planned/technologically required duration. The planning and simulation are not meant to
simulate these deviations since they are not the approved process and are meant to be minimized
in the future. Thus, the validation of the simulation behavior could be confirmed, and the model is
ready for use in the optimization. Existing and new heat treatment programs can be reliably simulated,
based on their parameters (mainly temperature level and duration) with sucient accuracy and in
flexible scenarios, e.g., with dierent outside temperatures or with dierent furnace filling degrees and
corresponding thermal masses. This increases accuracy for flexible scenarios, compared with fixed and
recorded values, and it also makes future maintenance of the system more ecient, since not every
new heat treatment program has to be measured individually before it can be simulated.
6. Optimization Method for the Case Study
The major use-case specific requirement for the optimization is the batching of orders and
assigning the batches to parallel furnaces under technological restrictions. Orders requiring dierent
temperature profiles, and sometimes additional processes such as hardening, have to be batched
into furnace grates, which in turn have to be batched to a furnace processing at one of the available
furnaces, each with dierent capabilities, concerning temperature profiles and additional hardening
processes. For this task, a batching heuristic module was developed, creating a valid and initial
solution and optimized batches from the input data from the enterprise resource planning (ERP) system.
Furthermore, the optimization module is extended by a tailor-made customized problem-specific
heuristic optimization module forming a multi-stage optimization together with a GA and a weighted
sum [
45
] based model formalization. The resulting modules of the optimization, which are described
in the following subsections, following the model formalization, are:
A rule-based deterministic batching heuristic;
A deterministic exchange-based heuristic optimization;
A GA optimization.
6.1. Model Formalization
The multi-stage optimization model consists of a knowledge-based enumerative exchange heuristic
and a superimposing GA. It pursues three part-goals that aim at optimizing dierent planning measures
to minimize the manufacturing costs; the first part-goal evaluates the additional costs of shortage
and storage of heat treatment output (representing orders
i
aggregated to oven batches) against their
corresponding due dates
ti
via a cost-function (in
). The second part-goal, being represented by the
J. Manuf. Mater. Process. 2020,4, 94 11 of 21
total energy consumption (oil, gas) and its corresponding (constant) energy prices, is calculated and
evaluated as cost-term with the feedback of the simulation. The third part-goal evaluates the total load
factor of the oven batches
cj
versus the maximum possible load factor
ck
, evaluating the opportunity
costs (i.e., business not realized through unused furnace space/capacity) as a linear function of unused
capacity. The objective Equation (2), applied in the GA, scalarizes the problem by calculating a weighted
and scaled fitness value. The scaling is applied using the part-goals of the best solution from the
preceding batching heuristic optimization phase.
Minimize f ti, e, g, cj/ck=
ω1
s1
m
P
i=1f1(ti)+ω2
s2
n
P
j=1ejk1+gjk2+ω3
s3(1
o
P
k=1
n
P
j=1
cj
ck)
ω1ω3,s1s3. . . part goal weightings and scaling factors
m,n,o. . . total number of orders, oven batches and furnaces
f1. . . cost function evaluating shortage and storage per order
e,g,k1,k2. . . energy consumption and price for electricity and gas
cj,ck. . . batch load factor and maximum possible load factor
(2)
The part-goal function
f1
is calculated based on the completion time on order level (
ti
) versus
its corresponding due date. Early completion times are penalized by storage costs through a flatter
partial function, while late completion times are rated by tardiness costs, described by a steeper
partial function.
The objective function is accompanied by two constraints that have to be considered during
the optimization process. The first constraint does not allow batch orders to start (
t_startj)
during
weekends, when no stais available. This means oven batches cannot be started after Saturday
morning 05:00 a.m. (represented by
tconst =
119
h simulation time
), reflecting the shift plan of the plant.
This condition is repeated every period length
p
(
p=
168
h simulation time)
, resulting in the following
mathematical formulation:
t_startj(mod p)tconst,j{1, . . . ,n}(3)
The second constraint requires that the combination
sj,k
of each batch
j
being selected for the
processing on a certain oven
k
always must result in a valid service combination
sj,k=1
for this batch,
see Equation (4).
sj,k=1, j{1, . . . ,n},k{1, . . . ,o}(4)
Each batch
j
has a common heat treatment program. Table 2describes the relationship between
heat treatment and furnace resulting in valid
(s=1)
and invalid combinations. An invalid combination
(s=0)cannot be selected as a possible part of a new solution by the GA.
Table 2. Heat treatment to furnace service combinations matrix.
Heat Treatment 12345
100110
200110
301101
411110
511111
611111
711111
811111
911111
J. Manuf. Mater. Process. 2020,4, 94 12 of 21
The dataset used in the following presentation contains two planning weeks from 2018
(46/2018–47/2018). The prices for electricity and gas represent constant market prices (34
/MWh for
natural gas and 90
/MWh for electricity). While four oven units are operated with gas, one unit is
operated with electricity.
6.2. Batching Heuristic
The first step of the heuristic, input-data preprocessing, comprises the cleaning of invalid
(furnace overcrowding,
. . .
) or missing (e.g., material master) input data from the raw ERP data set.
The accumulation of order lines into batches to reduce conversions and adjustment of furnaces is
based on a similar principle as Lenort’s TOC approach (see Section 2) for heat treatment furnaces.
The heuristic approach for batching was chosen because of the eciency and speed, compared to
metaheuristic/stochastic approaches, especially since it works without the need for computationally
expensive simulation evaluations. Another advantage is that the optimization procedure can be
formalized into rules quite well, thus reducing the advantage of a random global search that a GA
could provide.
First, the heuristic sorts all orders based on their ERP completion times (derived from planned
delivery dates) and order additional priority variable based on information from sales. A furnace
occupancy matrix is initialized and for each order the necessary number of grates is calculated. The key
principle is to find “batch crystallization” points to initialize batches and to subsequently fill the created
batches by iterating over the available orders with increasing order due dates. The batching creation
process prioritizes orders requiring specific furnaces and fills the batch with orders requiring the
same process (same program number) within an adjustable batch time span (orders in this range of
completion dates are eligible for batching in that planning run), set to 72 h for most orders in the case
study. Thereby the heuristic tries to utilize the most suitable oven (i.e., reserving scarce capacity for
orders requiring high oven capabilities), in terms of size, providing earliest possible free capacity for the
to be created batch. The heuristic then iteratively fills up the furnaces oering the remaining capacity to
batches that can be processed on all furnaces. Finally, the initial solution for the following optimization
steps, consisting of all created batches assigned to specific furnaces, including an estimation of the lead
time, is evaluated by the hybrid simulation once.
Between each oven batch, an adjustable interval of one hour is kept for (a) considering the
material handling delay of the crane and (b) reheating the furnace to another program-specific
temperature—both are considered in detail during the simulation evaluations, but considering them
already in the optimization keeps the optimization away from practically infeasible solutions. These two
conditions are dependent on the lot-sequence in terms of (heat treatment-specific) reheating and the
number of grates on each oven batch. The heuristic, requiring the feedback from the hybrid simulation
to be executed, pursues the overall objective of minimizing the intervals between each batch on the
corresponding furnace. The visualization of the initial solution created by the batching heuristic is
shown in Figure 6.
Figure 6. Total furnace occupancy.
J. Manuf. Mater. Process. 2020,4, 94 13 of 21
The results indicate that all five furnaces (also referred to as ovens) are loaded equally in terms of
lead-times and total furnace occupancy times. The batch heuristic results are deterministic. In both
planning weeks Saturdays feature an uneven load because of the constraint that an oven-batch is not
allowed to start after 05:00 a.m. on Saturday.
6.3. Hybrid Optimization Approach
A hybrid optimization for the assignment of batches to furnaces and scheduling of furnace
processing runs takes over after the batching heuristic, which provides the optimized batches.
It consists of a deterministic exchange-based heuristic optimization followed by a customized
GA. The exchanged-based heuristic performs a pairwise-exchange of batches between furnaces
enumeratively. This procedure is controlled by two parameters: the value of the first parameter defines
the maximum amount of time, that exchanged oven batches are allowed to dier in their estimated
oven occupancy times (30 min in the scenarios presented herein), while the second parameter is
used for the maximum allowed deviation of the estimated starting times between the exchanged
batches. The exchange operation is only executed if the respective constraints—oven capacities and
heat treatment process restrictions—are met. Furthermore, each solution from the iterative exchange
procedure is only kept, if the fitness value, calculated by the feedback of the hybrid simulation, is better
than the previous best solution. For the data set at hand and the chosen parameters, this greedy
heuristic strategy results in ~900 iterative solution evaluations. The exact value depends on the scope
of the respective data set and the chosen values for the two input parameters.
6.4. Genetic Algorithm Customization
The GA is implemented in MATLAB
®
(R2018b) and adapts a GA from the Global Optimization
Toolbox. The GA, as the second optimization phase, receives the best solution from the previous
deterministic heuristic as the starting point. The implemented stochastic uniform selection operator,
ensuring that every individual within a line corresponds to a section of this line length proportional to its
scaled value, remains untouched. The algorithm, moving along the line of individuals in steps of equal
size, selects the individuals for the next generation stochastically. Because of its stochastic character,
this selection operator is suitable for small GA populations. The diversity and exploratory character of
the algorithm in this case is preferable to e.g., ranking based or tournament selection operators.
The crossover operator is adapted to create suitable modifications to the sequencing and assignment
of the furnace batches. It uses either the pairwise-exchange operator or the single-point-mutation
operator to change the time or sequence of an individual. The composition of the batches themselves
remain unchanged during the GA optimization.
The parameterization of the GA features, aside from very small population sizes (max. 9 individuals
per generation), one elite individual and the disabled utilization of the GA mutation operator.
This operator does not add any functionality, as the mutation functionality is already included
within the customized crossover operator. The crossover customization features a pairwise-exchange
mechanism for oven batches, and an alteration of the start time of batches, thus mimicking the
functionality of the GA single-point mutation operator.
Dierent population sizes are tested within the case study, in combination with a Divide and
Conquer (DaC) GA compared with a global search GA: For the DaC GA, a global-search GA with a
larger population size and only order/batch swapping allowed, looks for an optimized sequencing,
while the second local search GA takes the best individual from the first GA as input, using a smaller
population size. The objective of the approach is to first search from a more global perspective to
provide an optimized sequence and then, in a second phase, to search for the optimal starting points
of each sequenced oven batch. The Divide and Conquer strategy is compared with the results of a
global-search GA strategy, in which both sequencing and scheduling are optimized simultaneously
(see Section 7.1).
J. Manuf. Mater. Process. 2020,4, 94 14 of 21
6.5. Constraint Handling and Validation
The three model-specific constraints described in the model formalization (see Section 6.1) are
implemented as follows: The shift plan constraint (1) and the constraint regarding the furnace service
combination matrix (2) are implemented within the optimization routine, while the model-specific
crane-unit constraint (3) is covered by the simulation itself, by not allowing the simultaneous handling
of more than one batch. The handling-unit specific constraint is modelled in the optimization algorithm
by using a minimal distance between each oven batch of half an hour.
The optimization results of test scenarios were validated with the planners of the casting
manufacturer. The actual production plans and simulated outcomes were compared with the
optimized plan and simulated outcomes. No relevant errors were found, it was confirmed that the
optimized plans are practically executable, and that the optimization results are plausible. Together
with the already validated simulation behavior, this means that the digital planning method is ready for
a proof of concept analysis documented in the next chapter. It is necessary to note that the optimization
validation has been conducted with historical data as well as the following experiments. This is a
practical limitation for a development in this stage. An actual real-life test for a future planning horizon
(i.e., the following week from now) would require implemented live-interfaces between company IT
(ERP/MES/control SW), as well as the external data input (e.g., outside temperature forecast and actual
temperature as well as optional energy market data). This is not an option mainly due to economic
restrictions for all parties involved.
7. Experimental Results for the Case Study
The results from the case study include two runs for each scenario of the customized GA.
An important part of this research addresses the search behavior of the two GA versions:
A Global Search GA with dierent population sizes (small populations with 3, 6 and 9 individuals);
A Divide and Conquer based, two-phase hierarchical GA, using the optimal population size from
the Global Search GA and utilizing only one modification operator in the global and local search
phase respectively.
Furthermore, the results cover the influence and potential of variable energy day-ahead price
markets. Because of computationally expensive evaluations of the hybrid simulation, only small
population sizes are covered by the experimental results section. This is necessary because the algorithm
is meant to be used as an operative planning tool, which must provide optimization results within
a limited time, e.g., in overnight runs or even shorter update runs in the event of sudden changes
(e.g., machine breakdowns or external influences aecting the capacity and material availability).
The manual optimization currently implemented in the company is equivalent to the results of the
batching heuristic, which is the starting point for the optimization trends for the hybrid heuristic
and GA; thus the optimization potential at the end of the simulation runs is roughly equivalent to
the potential gains through the planning method over the current manual planning process for the
heat treatment.
7.1. Discussion of the Case Study Results
The first part of the case study contains the comparison between the GA running with or without
the enumerative heuristic. The comparison, illustrated in Figure 7, shows the superiority of the GA in
combination with the prior heuristic optimization phase. The heuristic oers the distinct advantage of
systematically optimizing the load factor. This results in a 3.5% better performance after 250 evaluation
steps when the heuristic and the GA are combined.
J. Manuf. Mater. Process. 2020,4, 94 15 of 21
Figure 7.
Comparison of genetic algorithm (GA) in combination with the heuristic and GA without
the heuristic.
The first part of the case study results clearly shows the superiority of very small population-sizes,
similar to small population sizes in grey wolf optimizers (GWO). Within the standard GWO the omega
wolves of a pack are updated, next to a stochastic random change, in relation to the best three wolves
(alpha, beta, and gamma) [
46
]. The explorative character in the GWO algorithm is expressed by the
wolves leaving the original search path to a certain extent and searching into a new direction, while the
exploitation is based on the trajectory of the best three wolves. The GA developed herein is based on a
very small population (similar to the size of a wolf pack) with the population size varying between 3
and 9 individuals. In the context of long execution times of the hybrid-simulation resulting in a limited
number of possible evaluations/optimization steps, the first result (Figure 8) shows the superiority
and eciency of a GA (in runs with 9.000 evaluations each) with a population size of 3 individuals
over a GA with a larger population size of 6 or 9 individuals. The global goal improvement of the best
parameterization of the GA ensures at least 10% global goal optimization potential.
Figure 8. Results with dierent population sizes.
Figure 9shows the individual part goal trends within the optimization, using the optimized
parameterization introduced before. Shortages, carrying a very high penalty, are reduced by 70%
within the GA, while the energy costs can be reduced by approximately 6%, and the load factor of the
J. Manuf. Mater. Process. 2020,4, 94 16 of 21
batches is improved in total by 11%. In almost none of the observed optimization runs can the load
factor be further improved by the GA.
Figure 9. Part goal trends.
The total direct CO
2
emission potential is derived from the reductions in energy consumption.
According to a national electricity-supplier [
47
] the environmental impact of electricity during the
production of the electricity mix is about 170 g/kWh, while natural gas accounts for about 198 g/kWh
(combustion calorific value). This value is calculated from the stoichiometric combustion equation for
CH
4
using the molar mass for CO
2
. The total direct CO
2
emission reduction potential is about 6 tons
for the given dataset. This is equivalent to around 200 tons emission reduction per year, depending on
the actual furnace utilization.
Figure 10 compares the best parameterization of the global search GA (with population size set
to 3) with the sequential hierarchical (Divide and Conquer) GA. The latter utilizes 3 individuals in
the first swapping phase and 2 individuals in the following single-point-mutation phase. The figure
clearly shows that the sequential GA suers in its eciency because of its restricted fitness landscape
and opportunities in the corresponding optimization phases.
Figure 10. Sequential GA and best global GA.
J. Manuf. Mater. Process. 2020,4, 94 17 of 21
7.2. Consideration of Dynamic Energy Prices
In addition to increasing the energy eciency in the multi-criteria optimization, the planning
approach also oers the potential for a better synchronization of energy demand from production sites
with the fluctuating energy supply. The energy supply fluctuates significantly over time—especially
for electricity and reinforced by the necessary energy transformation toward renewable energy sources,
the supply fluctuates significantly every day. If the production sector can better adjust its energy
demand through optimized planning, i.e., by considering the availability of energy as input data
for an optimized planning, the entire energy system becomes more resource and potentially CO
2
ecient. Currently, there are significant hurdles for companies to access the flexible energy trade,
most of which are based in the inability for smaller industrial energy consumers (“small” in this case
includes industrial production plants as the casting manufacturer in the case study) to predict and plan
their demand. This results in fixed energy taris for most production plants in Europe—sometimes
with a degree of flexibility but never with direct access to long- and short-term energy markets and
stock-exchanges. The presented simulation-based approach is both able to better predict and plan
the energy demand for the next days and it can consider flexible energy prices as input data for
the optimization.
One scenario within the experimental case study covers the consideration of dynamic day-ahead
prices [
48
,
49
] for electricity and gas, instead of static energy prices. The objective of this scenario
is to evaluate the potential of the utilization of a dynamic pricing model versus the current static
pricing (flat tari). Figure 11 shows the price trends for electricity and gas for the time-period of
the used dataset. This visualization shows that prices for electricity vary more widely than those
for gas and the current fixed tarifor electricity per MWh is above the spot market price for the
entire two reference weeks. Table 3compares the influence of constant and variable energy prices on
the optimization outcome. The comparison shows that while there is a cost advantage of using the
variable prices, it is relatively small compared to the absolute energy cost and even more so compared
to the total cost of all cost categories considered in the objective function. It also shows that for this
specific scenario the optimization in its final result (optimized) seems to have opted to sacrifice lower
energy costs for the minimization of other cost categories—thus, the relative energy cost advantage of
the optimized solution is lower than the advantage in the initial solution. There are several possible
explanations: since in this case and for the given datasets the furnaces must run most of the time,
the optimization cannot utilize variable prices extensively, e.g., by shifting energy-intensive orders to
times of cheaper energy.
Figure 11. Dynamic energy costs (Week 46/2018).
J. Manuf. Mater. Process. 2020,4, 94 18 of 21
Table 3. Influence of variable and constant energy prices (in ).
Variable Price
Energy Cost
Contant Price
Energy Cost
Advantage of Variable
Price for Energy Cost
Initial 18.185 18.411 226
Optimized 17.286 17.373 88
relative improvement 4.94% 5.64%
The minor variations in natural gas prices do not lead to significant dierences in optimization
potential. Within this use-case 4 out of 5 heat treatment furnaces are operated with gas. This indicates
only limited total optimization potential for flexible energy procurement in this specific case, with
the limited considered system-boundaries. For the entire production plant, the total consumption of
electricity (two electric arc furnaces for the casting process, machine tools and auxiliary electricity
consumers such as hall lighting, etc.) is much higher, which would also increase the potential
for advantages due to flexible energy prices. It is also possible to combine multiple industrial
energy consumers to larger virtual consumers and there are already service providers that oer
this coordination. This clustering of industrial energy consumers could also increase the impact for
optimized planning in combination with flexible energy prices.
However, there are additional factors that must be considered when assessing the use of optimized
planning with flexible energy prices: there are costs associated with trading on energy markets and
engaging in complex energy portfolio management introduces additional risks for the company.
A major risk factor is that there can be times of (e.g., externally induced) of high energy prices that
can be significantly higher than fixed taris and hard to predict. If the planned energy consumption
associated with an optimized production planning cannot be fulfilled, e.g., because of internal or
external complications, there is a risk of penalties or high surcharges. Also, although tools like the
one presented in this paper significantly improve the predictability of energy demand, operating of
flexible energy markets would require companies to engage in a complex field requiring expertise in a
field usually far removed from the core competences of production enterprises. Thus, the potential
for optimized planning in combination with flexible energy management should be invested in more
detail to provide actionable advice for companies.
8. Conclusions and Outlook
Our contribution shows that the basic concept of a multi-stage hybrid optimization featuring
hybrid simulation for the planning task of batching and scheduling, plus assigning orders—batches
of orders in this case—to parallel machines is practically feasible and provides considerably better
performance than the limited manual planning currently implemented in the company. This is
especially important since the automated optimization would be equivalent to hours of manual
planning, which is often not practically feasible and had previously not been conducted in comparable
detail at the casting manufacturer.
Considering the optimization options, the results show that a combination of an enumerative
heuristic and a specific GA is more ecient and robust than a GA implementation only. The global
search GA, with a pairwise-exchange operator and the single-point mutation operator combined
within the search mechanism, is superior compared to a Divide and Conquer optimization approach.
The latter GA approach leaves parts of the search space unexplored in each optimization phase,
leading to an inferior solution. The case study at hand shows that GA implementations with small
population sizes are more ecient than larger population sizes for the given problem complexity
for short optimization runs. This is due to the structure of the large search space, requiring many
iterations—since the number of possible simulation-evaluation iterations is limited in practice, smaller
population sizes ensure the available iterations are spent on searching for good solutions rather than
trying to ensure a more robust optimization behavior.
J. Manuf. Mater. Process. 2020,4, 94 19 of 21
The hybrid simulation for the energetic evaluation of the production plans does not just rely on
fixed energy profiles but performs a physical simulation that considers time-dependent interactions
between material and energy flows. For example, loading the furnaces can only be started after
pre-heating is completed, which in turn depends on the starting temperature and the environment.
Only this integrated approach really does justice to the complexity of modern CPS, in which physical
and cyber aspects interact with each other, thereby resulting in emergent behavior.
The method can produce results in time frames suitable for MES/APS and thus can be included in the
planning loop of industrial companies. A major challenge for the application is the required quality and
detail of data: process information has to be up to date and the initial eort for modelling—especially
modelling and parametrization of the physical (thermodynamic) simulation for the production
equipment—is demanding for companies, especially given the typical current data maturity level
found in the metal casting industry. Future development in the field of digitization and data integration
could therefore play a key role in enabling advanced planning tools. The immediate next step would
be widening the system boundaries to include the upstream casting process and its large electric arc
furnaces, which will increase the optimization potential but also the optimization complexity. Another
next step would be validating the method for operative planning, by implementing interfaces for input
and output data exchange on a regular basis, which thus far has only been simulated for historical data
for a reference period.
The potential of using planning tools like the one presented herein for synchronizing industrial
energy demand with fluctuating energy supply has been briefly considered in this case study. A basic
feasibility test of the concept has been accomplished and potential savings for the company could be
shown. However, this was only a preliminary test and a more in-depth consideration of the practical
implementation potential and implications for companies is necessary. This is one of the future
directions of this research.
Another future research trajectory is the integration with longer planning horizons, which would
enable the method to provide planning support for mid- to long-term production smoothing on the
one hand and considering long-term energy markets on the other, completing the support for complex
energy portfolio management. Challenges will be data availability and planning with risk, as well
as finding suitable modelling simplifications with reduced detail that are derived from the detailed
short-term model.
Author Contributions:
Conceptualization, T.S.; methodology, T.S., F.K., and B.H.; software, F.K. and B.H.;
validation, T.S., F.K., and B.H.; formal analysis, T.S., F.K., and B.H.; investigation, T.S., F.K., and B.H.; resources;
data curation, T.S., F.K., and B.H.; writing—original draft preparation, T.S., F.K., and B.H.; writing—review and
editing, T.S.; visualization, T.S., F.K., and B.H.; supervision, T.S.; project administration, T.S.; funding acquisition,
T.S., F.K., and B.H. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by the Austrian Research Promotion Agency (FFG), grant number 858655.
Acknowledgments:
The authors would like to thank all case study partners for their contribution, especially the
staat the casting manufacturer in the data acquisition phase.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Wang, J.; Qiao, F.; Zhao, F.; Sutherland, J.W. Batch scheduling for minimal energy consumption and tardiness
under uncertainties: A heat treatment application. CIRP Ann. 2016,65, 17–20. [CrossRef]
2.
He, Y.; Li, Y.; Wu, T.; Sutherland, J.W. An energy-responsive optimization method for machine tool selection
and operation sequence in flexible machining job shops. J. Clean. Prod. 2015,87, 245–254. [CrossRef]
3.
Matta, A. Automation technologies for sustainable production [TC Spotlight]. IEEE Robot. Autom. Manag.
2019,26, 98–102. [CrossRef]
4.
Zhang, S.; Xing, Y.; Wang, D.; Wang, N.; Jimenez, G. Optimal heat treatment furnace based on new robust
point approximation strategy. J. Clean. Prod. 2020,254, 120097. [CrossRef]
J. Manuf. Mater. Process. 2020,4, 94 20 of 21
5.
Rao, R.V.; Rai, D.P. Optimization of selected casting processes using Jaya algorithm. Mater. Today Proc.
2017
,
4, 11056–11067. [CrossRef]
6.
Kang, J.; Rong, Y. Modeling and simulation of load heating in heat treatment furnaces. J. Mater. Process. Technol.
2006,174, 109–114. [CrossRef]
7.
Ramin, D.; Spinelli, S.; Brusaferri, A. Demand-side management via optimal production scheduling in
power-intensive industries: The case of metal casting process. Appl. Energy 2018,225, 622–636. [CrossRef]
8.
Tiwari, M.K.; Mukhopadhyay, A.; Sanyal, D. Parameter optimization through performance analysis of model
based control of a batch heat treatment furnace with low NOx radiant tube burner. Energy Convers. Manag.
2005,46, 2114–2133. [CrossRef]
9.
Tang, H.; Chen, R.; Li, Y.; Peng, Z.; Guo, S.; Du, Y. Flexible job-shop scheduling with tolerated time interval
and limited starting time interval based on hybrid discrete PSO-SA: An application from a casting workshop.
Appl. Soft Comput. 2019,78, 176–194. [CrossRef]
10.
Rahal, S.; Li, Z.; Papageorgiou, D.J. Proactive and reactive scheduling of the steelmaking and continuous
casting process through adaptive robust optimization. Comput. Chem. Eng. 2020,133, 106658. [CrossRef]
11.
Qin, H.; Fan, P.; Tang, H.; Huang, P.; Fang, B.; Pan, S. An eective hybrid discrete grey wolf optimizer for the
casting production scheduling problem with multi-objective and multi-constraint. Comput. Ind. Eng.
2019
,
128, 458–476. [CrossRef]
12.
Jiang, S.-L.; Zheng, Z.; Liu, M. A multi-stage dynamic soft scheduling algorithm for the uncertain
steelmaking-continuous casting scheduling problem. Appl. Soft Comput. 2017,60, 722–736. [CrossRef]
13.
Yu, C.; Semeraro, Q.; Matta, A. A genetic algorithm for the hybrid flow shop scheduling with unrelated
machines and machine eligibility. Comput. Oper. Res. 2018,100, 211–229. [CrossRef]
14.
Cheng, B.; Li, K.; Chen, B. Scheduling a single batch-processing machine with non-identical job sizes in fuzzy
environment using an improved ant colony optimization. J. Manuf. Syst. 2010,29, 29–34. [CrossRef]
15.
Baykaso˘glu, A.; Ozsoydan, F.B. Dynamic scheduling of parallel heat treatment furnaces: A case study at a
manufacturing system. J. Manuf. Syst. 2018,46, 152–162. [CrossRef]
16.
Lin, Y.-K.; Hsieh, F.-Y. Unrelated parallel machine scheduling with setup times and ready times. Int. J.
Prod. Res. 2013,52, 1200–1214. [CrossRef]
17.
Eroglu, D.Y.; Özmutlu, H.C.; Özmutlu, S. Genetic algorithm with local search for the unrelated parallel
machine scheduling problem with sequence-dependent set-up times. Int. J. Prod. Res.
2014
,52, 5841–5856.
[CrossRef]
18.
Lenort, R.; Roman, K.; Andrea, S. Heuristic algorithm for planning and scheduling of forged pieces heat
treatment. 2012. Available online: http://dspace.vsb.cz/handle/10084/89844 (accessed on 16 September 2020).
19.
Huang, R.-H.; Yu, S.-C.; Chen, P.-H. Energy-saving scheduling in a flexible flow shop using a hybrid genetic
algorithm. J. Environ. Prot. 2017,8, 1037–1056. [CrossRef]
20.
Thiede, S.; Schönemann, M.; Kurle, D.; Herrmann, C. Multi-level simulation in manufacturing companies:
The water-energy nexus case. J. Clean. Prod. 2016,139, 1118–1127. [CrossRef]
21.
Rager, M. Energieorientierte produktionsplanung: Analyse, konzeption und umsetzung; Gabler Verlagg: Wiesbaden,
Germany, 2008.
22.
Sihn, W.; Sobottka, T.; Heinzl, B.; Kamhuber, F. Interdisciplinary multi-criteria optimization using hybrid
simulation to pursue energy eciency through production planning. CIRP Ann.
2018
,67, 447–450. [CrossRef]
23.
Reinhardt, H.; Bergmann, J.-P.; Münnich, M.; Rein, D.; Putz, M. A survey on modeling and forecasting the
energy consumption in discrete manufacturing. Procedia CIRP 2020,90, 443–448. [CrossRef]
24.
Sobottka, T.; Kamhuber, F.; Faezirad, M.; Sihn, W. Potential for machine learning in optimized production
planning with hybrid simulation. Procedia Manuf. 2019,39, 1844–1853. [CrossRef]
25.
Herrmann, C.; Thiede, S. Process chain simulation to foster energy eciency in manufacturing. CIRP J.
Manuf. Sci. Technol. 2009,1, 221–229. [CrossRef]
26.
Schönemann, M.; Schmidt, C.; Herrmann, C.; Thiede, S. Multi-level modeling and simulation of manufacturing
systems for lightweight automotive components. Procedia CIRP 2016,41, 1049–1054. [CrossRef]
27.
Garwood, T.L.; Hughes, B.R.; Oates, M.R.; O’Connor, D.; Hughes, R. A review of energy simulation tools for
the manufacturing sector. Renew. Sustain. Energy Rev. 2018,81, 895–911. [CrossRef]
28. Bunse, K.; Vodicka, M.; Schönsleben, P.; Brülhart, M.; Ernst, F.O. Integrating energy eciency performance
in production management – gap analysis between industrial needs and scientific literature. J. Clean. Prod.
2011,19, 667–679. [CrossRef]
J. Manuf. Mater. Process. 2020,4, 94 21 of 21
29.
Dias, L.M.S.; Vieira, A.A.C.; Pereira, G.A.B.; Oliveira, J.A. Discrete simulation software ranking—A top list of
the worldwide most popular and used tools. In Simulating Complex Service Systems; IEEE: Washington, DC,
USA, 2016; pp. 1060–1071.
30.
Jahangirian, M.; Eldabi, T.; Naseer, A.; Stergioulas, L.K.; Young, T. Simulation in manufacturing and business:
A review. Eur. J. Oper. Res. 2010,203, 1–13. [CrossRef]
31.
Thiede, S. Energy Eciency in Manufacturing Systems; Springer Science & Business Media: Berlin, Germany,
2012.
32.
Junge, M. Simulationsgestützte Entwicklung und Optimierung einer energieezienten Produktionssteuerung.
Ph.D. Thesis, Kassel University, Kassel, Germany, 2007.
33.
Lin,H.; Sambamoorthy, S.; Shukla, S.K.; Thorp,J.; Mili,L. Powersystem and communicationnetworkco-simulation
for smart grid applications. In Proceedings of the ISGT 2011, Anaheim, CA, USA,
17–19 January 2011
; pp. 1–6.
[CrossRef]
34.
Carloni, L.P.; Passerone, R.; Pinto, A.; Angiovanni-Vincentelli, A.L.; Sangiovanni-Vincentelli, A.
Languages and tools for hybrid systems design. Found. Trends Electron. Des. Autom.
2006
,1, 1–193.
[CrossRef]
35.
Chahal, K.; Eldabi, T. Applicability of hybrid simulation to different modes of governance in UK healthcare.
In Proceedings of the 2008 Winter Simulation Conference, Miami, FL, USA, 7–10 December 2008;
pp. 1469–1477.
36. Heinzl, B.; Raich, P.; Preyser, F.; Kastner, W. Simulation-based Assessment of Energy Eciency in Industry:
Comparison of hybrid simulation approaches. IFAC PapersOnLine 2018,51, 689–694. [CrossRef]
37. Nof, S.Y. Springer Handbook of Automation; Springer: Berlin/Heidelberg, Germany, 2009.
38. Love, J. Process Automation Handbook; Springer Science and Business Media LLC: London, UK, 2007.
39.
Raich, P.; Heinzl, B.; Preyser, F.; Kastner, W. Modeling techniques for integrated simulation of industrial
systems based on hybrid PDEVS. In Proceedings of the 2016 Workshop on Modeling and Simulation of
Cyber-Physical Energy Systems (MSCPES), Vienna, Austria, 11 April 2016; pp. 1–6.
40.
Sobottka, T.; Kamhuber, F.; Rössler, M.; Sihn, W. Hybrid simulation-based optimization of discrete parts
manufacturing to increase energy eciency and productivity. Procedia Manuf.
2018
,21, 413–420. [CrossRef]
41.
Heinzl, B.; Raich, P.; Preyser, F.; Kastner, W.; Smolek, P.; Leobner, I. Modular Hybrid Modeling based on DEVS
for interdisciplinary simulation of production systems. In Proceedings of the 31st European Simulation and
Modelling Conference—ESM’2017, Lisbon, Portugal, 25–27 October 2017; pp. 157–161.
42.
Heinzl, B. Hybrid Modeling of Production Systems: Co-Simulation and DEVS-Based Approach.
Diploma Thesis, TU Wien, Vienna, Austria, 2016.
43.
Sobottka, T.; Kamhuber, F.; Sihn, W. Increasing energy eciency in production environments through an
optimized, hybrid simulation-based planning of production and its periphery. Procedia CIRP
2017
,61,
440–445. [CrossRef]
44.
Sobottka, T. Eine anwendungsorientierte simulationsbasierte Methode, unter Berücksichtigung von
Energieezienz, in der optimierenden Planung von Produktion und Logistik. Ph.D. Thesis, Vienna University
of Technology, Vienna, Austria, 2018.
45.
Freitas, A.A. A critical review of multi-objective optimization in data mining. ACM SIGKDD Explor. Newsl.
2004,6, 77–86. [CrossRef]
46.
Wang, J.S.; Li, S.X. An improved grey wolf optimizer based on dierential evolution and elimination
mechanism. Sci. Rep. 2019,9, 7181. [CrossRef] [PubMed]
47.
EVN Energievertrieb GmbH & Co KG. EVN Produktionformation für Ihren Stromtarif. 2020. Available online:
https://www.evn.at/Downloads/Businesskunden/Strom/Produktinformation_BUS_konventionell.aspx
(accessed on 23 July 2020).
48.
Central European Gas Hub AG. Day Ahead Market. 2020. Available online: https://www.cegh.at/(accessed on
23 July 2020).
49.
Energy Exchange Austria. Spotmarkt Strom/Österreich. 2020. Available online: https://www.exaa.at/de
(accessed on 23 July 2020).
©
2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
... Areas/sectors of use [26,[66][67][68][69][70][71][72][73][74][75] Level of machine [66,76] Level of plant [66,[69][70][71][72][73][74][75]77,78] Level of plant with supporting technology [26,66,68,79] Integration of parameters [26,[66][67][68][69][70][71][72][73][74][75]77,78,80] Analysis, optimization, strategy validation, and other benefits [2,26,[66][67][68][69][70][71][72][73][74][75]77,78,81,82] ...
... Areas/sectors of use [26,[66][67][68][69][70][71][72][73][74][75] Level of machine [66,76] Level of plant [66,[69][70][71][72][73][74][75]77,78] Level of plant with supporting technology [26,66,68,79] Integration of parameters [26,[66][67][68][69][70][71][72][73][74][75]77,78,80] Analysis, optimization, strategy validation, and other benefits [2,26,[66][67][68][69][70][71][72][73][74][75]77,78,81,82] ...
... Areas/sectors of use [26,[66][67][68][69][70][71][72][73][74][75] Level of machine [66,76] Level of plant [66,[69][70][71][72][73][74][75]77,78] Level of plant with supporting technology [26,66,68,79] Integration of parameters [26,[66][67][68][69][70][71][72][73][74][75]77,78,80] Analysis, optimization, strategy validation, and other benefits [2,26,[66][67][68][69][70][71][72][73][74][75]77,78,81,82] ...
Article
Full-text available
This paper addresses electricity consumption management in manufacturing enterprises. The research aims to provide manufacturing enterprises with an effective tool to control electricity costs. Recently, some factors have been observed to affect the rapid changes in the operating conditions of enterprises. These include the transformation of the power sector toward renewable energy, the disruption of supply chains resulting from a coronavirus pandemic, political crises, and process automation. A method for the analysis and management of electricity consumption in enterprises based on simulation modeling is proposed. The simulation model contains predefined objects representing physical system elements and the data processing algorithm. The production order execution time, energy consumption, employee overtime, and machine load are included in the model. The results show that it is possible to determine the level of power available for the process completion and its influence on the production volume and realization time. In the studied case, when the available power was reduced by half, there was an increase in order execution time of nearly 25 percent and an increase in energy consumption of nearly 15 percent. The method can be used in the operational activities of enterprises as well as extended to different types of production processes.
... While many previous studies have proposed methods for forming a batch under the constraint of the weight capacity of a given furnace, they ignore the spatial constraints imposed by the size of the furnace in relation to the shapes and dimensions of the workpieces in the batch [1][2][3][4][5]. Previous studies [1,2] have dealt with the constraint on the temperature of heat treatment; [1] assumed that the target temperature is explicitly given for each workpiece, whereas [2] formed a batch with the workpieces whose heat treatment processes to go through are the same with no reference to their target temperatures. ...
... While many previous studies have proposed methods for forming a batch under the constraint of the weight capacity of a given furnace, they ignore the spatial constraints imposed by the size of the furnace in relation to the shapes and dimensions of the workpieces in the batch [1][2][3][4][5]. Previous studies [1,2] have dealt with the constraint on the temperature of heat treatment; [1] assumed that the target temperature is explicitly given for each workpiece, whereas [2] formed a batch with the workpieces whose heat treatment processes to go through are the same with no reference to their target temperatures. The time and energy costs were either given or estimated in most previous studies. ...
... While many previous studies have proposed methods for forming a batch under the constraint of the weight capacity of a given furnace, they ignore the spatial constraints imposed by the size of the furnace in relation to the shapes and dimensions of the workpieces in the batch [1][2][3][4][5]. Previous studies [1,2] have dealt with the constraint on the temperature of heat treatment; [1] assumed that the target temperature is explicitly given for each workpiece, whereas [2] formed a batch with the workpieces whose heat treatment processes to go through are the same with no reference to their target temperatures. The time and energy costs were either given or estimated in most previous studies. ...
Article
Full-text available
Scheduling heat treatment jobs in a hot press forging factory involves forming batches of multiple workpieces for the given furnaces, determining the start time of heating each batch, and sorting out the order of cooling the heated workpieces. Among these, forming batches is particularly difficult because of the various constraints that must be satisfied. This paper proposes an optimization method based on an evolutionary algorithm to search for a heat treatment schedule of maximum productivity with minimum energy cost, satisfying various constraints imposed on the batches. Our method encodes a candidate solution as a permutation of heat treatment jobs and decodes it such that the jobs are grouped into batches satisfying all constraints. Each candidate schedule is evaluated by simulating the heating and cooling processes using cost models for processing time and energy consumption, which are learned from historical process data. Simulation experiments reveal that the schedules built using the proposed method achieve higher productivity with lower energy costs than those built by human experts.
... Due to this, the development of rescheduling has become essential, either repairing the original schedule or generating again an optimal schedule [21]. Thus, algorithms have been developed that allow approaching the problem of scheduling with different interruptions [22], algorithms that allow dealing with the machine's setup times [23], jobs with specific routes that avoid some steps [24,25], multi-criteria algorithm methods [26,27] and even algorithms that consider several sources of interruption simultaneously (job variation, job cancellation, job time variation, machine breakdown and new jobs) [28][29][30][31][32][33]. Industry 4.0 and smart manufacturing [13] are also capable of creating a virtual copy of the physical world These are three types of semi-products, and each one is obtained in a different machine. ...
... Due to this, the development of rescheduling has become essential, either repairing the original schedule or generating again an optimal schedule [21]. Thus, algorithms have been developed that allow approaching the problem of scheduling with different interruptions [22], algorithms that allow dealing with the machine's setup times [23], jobs with specific routes that avoid some steps [24,25], multi-criteria algorithm methods [26,27] and even algorithms that consider several sources of interruption simultaneously (job variation, job cancellation, job time variation, machine breakdown and new jobs) [28][29][30][31][32][33]. Industry 4.0 and smart manufacturing [13] are also capable of creating a virtual copy of the physical world and make decentralized decisions [34], so new approaches can be seen based on decision-making schema [20] or adaptative scheduling [35]. ...
... When looking for optimal planning for improved steel plant production, the type of methodologies that should be considered, as they are the most advantageous, are those mixed methodologies, in which algorithms and programming systems are used under human supervision [26,49]. Normally, the decision on the method to apply is not carried out by the scheduling algorithm, but by the coordinator or plant manager [11,[50][51][52]. ...
Article
Full-text available
During the steelmaking and continuous casting process in the steel plant, it is common to encounter delays that affect initial planning. Furthermore, continuous casting machines themselves can lose much of their performance in the event of closure of one or more of their casting strands. The situation that is generated, far from being a planning problem, forces consideration of a vision of cost analysis when deciding changes in the planned sequences. This study presents a detailed analysis of the different circumstances that can cause strands closures or sequence breaks, their consequences and the different options available to minimize losses. Finally, an algorithm capable of analyzing the workshop situation and making the most favorable decision to optimize production is proposed, analyzed and compared with the efficiency of the original scheduling method in a real steel plant. The new algorithm proves its efficiency in all situations, with a time-saving average of 26.41 min per decision taken.
... Angel et al. [46] introduced a randomized approximation algorithm for the problem of energy-consumption scheduling for unrelated parallel machines with the average weighted completion time. Sobottka et al. [47] demonstrated the development and evaluation of a digital method for multiobjective optimization problems considering traditional business aims and energy efficiency in a metal-casting manufactory. The improved method, including hybrid simulation-based multi-criterion optimization as a multi-stage hybrid heuristic and metaheuristic method, was employed in a heat-treatment process, which requires order batching and sequencing on parallel machines under complex restrictions. ...
Article
Full-text available
Environmental concerns and rising energy prices put great pressure on the manufacturing industry to reduce pollution and save energy. Electricity is one of the main machinery energy sources in a plant; thus, reducing energy consumption both saves energy costs and protects our planet. This paper proposes the novel method called variable neighborhood strategy adaptive search (VaNSAS) in order to minimize energy consumption while also considering job priority and makespan control for parallel-machine scheduling problems. The newly presented neighborhood strategies of (1) solution destroy and repair (SDR), (2) track-transition method (TTM), and (3) multiplier factor (MF) were proposed and tested against the original differential evaluation (DE), current practice procedure (CU), SDR, TTM, and MF for three groups of test instances, namely small, medium, and large. Experimental results revealed that VaNSAS outperformed DE, CU, SDR, TTM, and MF, as it could find the optimal solution and the mathematical model in the small test instance, while the DE could only find 25%, and the others could not. In the remaining test instances, VaNSAS performed 16.35–19.55% better than the best solution obtained from Lingo, followed by DE, CU, SDR, TTM, and MF, which performed 7.89–14.59% better. Unfortunately, the CU failed to improve the solution and had worse performance than that of Lingo, including all proposed methods.
... Angel et al. [40] obtained feasible solutions within the bounded expected value by employing a randomized approximation algorithm for energy consumption scheduling on unrelated parallel machines with the average weighted completion time. Sobottka et al. [41] presented the development and evaluation of a digital method for multi-criteria optimization, considering traditional business goals and energy efficiency, for production scheduling in metal casting manufacturing. The developed method, featuring hybrid (discrete and continuous) simulation-based multi-objective optimization (a multistage hybrid heuristic and metaheuristic), was employed in a heat treatment process which required batch ordering and sequencing/scheduling on parallel machines and considering complex restrictions. ...
Article
Full-text available
Environmental and economic considerations create a challenge for manufacturers. The main priorities for production planning in environmentally friendly manufacturing industries are reducing energy consumption and improving productivity by balancing machine load. This paper focuses on parallel machine scheduling to minimize energy consumption (PMS_ENER), which is an indicator of environmental sustainability when considering machine-load balance problems. A mathematical model was formulated to solve the proposed problem and tested using a set of problem groups. The findings indicated that the mathematical model could find an optimal solution within a limited calculation time for small problems. For medium and large problems, the mathematical model could also find the optimal solution within a limited calculation time, but worse than all metaheuristics. However, finding an optimal solution for a larger problem is time-consuming. Thus, a novel method, a hybrid differential evolution algorithm with adaptive large neighborhood search (HyDE-ALNS), is presented to solve large-scale PMS_ENER. The new mutation and recombination formula for the differential evolution (DE) algorithm proposed in this article obtained promising results. By using the HyDE-ALNS, we improved the solution quality by 0.22%, 7.21%, and 12.01% compared with a modified DE (MDE-3) for small, medium, and large problems respectively. In addition, five new removal methods were designed to implement in ALNS and achieve optimal solution quality.
... A multi-agent simulation with predictive capabilities was used to support the planning and potentially also the operation of these setups. • Sobottka et al. [7] presented a simulation-based approach that allows for an energy aware scheduling of heat treatment furnaces in a casting company. While using data from the company's enterprise resource planning (ERP) and manufacturing execution system (MES), an optimized planning of operations was enabled, which led to improved environmental and economic performance. ...
Article
Full-text available
Digitalization is a major change driver in manufacturing and is nowadays typically linked to terms like Industry 4 [...]
Article
Full-text available
The manufacturing industry faces environmental challenges due to its high electrical energy demand. Measures for improving the energy efficiency of manufacturing processes are applied to reduce the total energy consumption while eco-friendly renewable energy sources are harnessed increasingly. However, the volatile nature of renewables drives a change in energy allocation and the electrical grid. Both energy efficiency and energy allocation may benefit from accurate energy consumption modeling and forecasting. In this paper we constitute the state of the art of existing modeling techniques in the domain of discrete manufacturing and derive future research activities.
Article
Full-text available
Advanced production planning and scheduling approaches increasingly rely on simulation-based optimization methods. This entails the problem of a high computational effort due to complex models, resulting in limitations for the practical application of otherwise powerful methods. While machine-learning methods offer a potential for performance improvement, approaches for real-life applications with a high complexity are still lacking. This paper explores the potential for machine learning, especially artificial neural networks, used as surrogate models, to improve the performance of a recently developed planning method for real life production planning applications. The simulation considered in this paper is a complex hybrid discrete-continuous model, enabling the method to pursue energy efficiency simultaneously with economic goals, in a complex multi-criteria goal system. The artificial neural network is trained via offline learning and is meant to provide a computationally cheap evaluation of intermediate planning solutions, compiled by an optimization algorithm during an optimization run. The approach is developed and evaluated in a case-study on the food industry, indicating a basic feasibility of the approach but also pointing out necessary future challenges to be solved towards practical applicability.
Article
Full-text available
The grey wolf optimizer (GWO) is a novel type of swarm intelligence optimization algorithm. An improved grey wolf optimizer (IGWO) with evolution and elimination mechanism was proposed so as to achieve the proper compromise between exploration and exploitation, further accelerate the convergence and increase the optimization accuracy of GWO. The biological evolution and the “survival of the fittest” (SOF) principle of biological updating of nature are added to the basic wolf algorithm. The differential evolution (DE) is adopted as the evolutionary pattern of wolves. The wolf pack is updated according to the SOF principle so as to make the algorithm not fall into the local optimum. That is, after each iteration of the algorithm sort the fitness value that corresponds to each wolf by ascending order, and then eliminate R wolves with worst fitness value, meanwhile randomly generate wolves equal to the number of eliminated wolves. Finally, 12 typical benchmark functions are used to carry out simulation experiments with GWO with differential evolution (DGWO), GWO algorithm with SOF mechanism (SGWO), IGWO, DE algorithm, particle swarm algorithm (PSO), artificial bee colony (ABC) algorithm and cuckoo search (CS) algorithm. Experimental results show that IGWO obtains the better convergence velocity and optimization accuracy.
Thesis
This thesis documents the development of a planning method that simultaneously increases the energy efficiency as well as the economic performance of complex production systems. The method utilizes a multi-criteria optimization based on a hybrid simulation that is able to exhibit both continuous and discrete behaviour. Energy efficiency has become an important goal for manufacturing enterprises, largely due to a combination of rising societal and political pressure towards a more sustainable way of conducting business and growing energy prices in the long-term. There is considerable potential in this area to be found in the optimized production planning and control (PPC), as well as the synchronized control of equipment in the periphery, together with a goal system that features both classical economic goals and energy efficiency in the objective function of an integrated planning approach. However, despite the recognized demand, the necessary planning methods are missing for manufacturing companies, resulting in wasted potential. Even the scientific approaches publicized thus far are lacking crucial elements: Especially the integrated planning of the discrete material flow and order processing together with the thermal-physical behaviour of equipment – e.g. industrial ovens or chillers – and its interdependencies is not sufficiently provided by available approaches. As a result, the associated optimization potential is left unutilized. This research aims at developing a planning method that integrates energy efficiency into the goal system of production planning optimization. To achieve this, a modelling concept supporting the integrated consideration of production systems is being developed. Next, a corresponding novel hybrid simulation method that supports both the discrete and continuous behaviour of production and energy systems is evolved. Following that, an optimization module utilizing the simulation as an evaluation function and enabling the simultaneous optimization of the short-term production plan and optimized control of relevant equipment in the periphery is engineered. The resulting overall method is then evaluated in a case study featuring a food production facility, revealing an overall optimization potential of up to 50%, amid energy savings of up to 30%.
Article
In this work, a new model of air to air heat pump (AAHP) using industrial continuous heat treatment furnace (ICHTF) that runs by wind generation and solar Stirling engine are presented. The suggested approach for making fire includes an internal and an external airflow fans, extension valve, compressor, condenser, and evaporator to provide energy with temperature higher than 1000 K for thermal demands. While the combustion reactions is happening, AAHP does not provide thermal energy; it is moved from the environment to furnace section. Accordingly, application of wind, solar and AAHP not only decreases the emission and ozone depleting substances, but also significantly rises the saving rate in electricity and petroleum products consumptions. The paper presents a mathematical approach, that utilizes energy and exergy, which is used for modeling the suggested ICHTF in presence of green energy, e.g. solar, wind and air. The proposed method is compared with different optimization algorithms. In addition, a robust periodic optimization technique for economic issues is presented for minimization of the overall energy supplement costs utilizing the predicted wind velocity as the problem input and considering the higher and lower bounds of wind energy for simulating wind volatility. Obtained results proof the validity of proposed model.
Article
The schedule of a steelmaking and continuous casting (SCC) process defines the unit assignments of casts and ladles, and the processing sequence and timing in each unit. These decisions have to meet strict production requirements such as limited storage duration and continuous operation under various disruptions. Uncertain processing time is a common factor that may invalidate the initial schedule. Traditional deterministic reactive scheduling implements corrective measures for every disruption event by either repairing the deterministic schedule or rescheduling from scratch. In this work, we present a novel adaptive robust optimization method for SCC scheduling under uncertainty. Compared to the deterministic scheduling method, adaptive robust proactive scheduling generates an initial schedule immune to most major disruptions eliminating the need for most changes. Adaptive robust hybrid scheduling presents an attractive trade-off between solution conservatism and excessive scheduling modifications.
Article
Sustainability has become a major component of competition among manufacturers worldwide. Approximately one-third of the energy consumed in the United States is used for manufacturing. Sustainable production is the top concern for manufacturers and consumers because generating greenhouse gases and harmful emissions and disposing of waste generated during the manufacturing processes could pollute the environment. Therefore, production system design, analysis, and control play key roles in providing cleaner energy and reducing emissions, waste, and pollution, as well as minimizing the impact of climate change and conserving natural resources and energy during the product’s entire life cycle. Therefore, sustainable production automation (SPA), which involves designing, structuring, and engineering operations and products, is a significant part of manufacturing. The technical committee (TC) on SPA promotes the idea of developing innovative algorithms, models, heuristics, hardware, and software in broad areas.
Article
In this paper, we addressed two significant characteristics in practical casting production, namely tolerated time interval (TTI) and limited starting time interval (LimSTI). With the consideration of TTI and LimSTI, a multi-objective flexible job-shop scheduling model is constructed to minimize total overtime of TTI, total tardiness and maximum completion time. To solve this model, we present a hybrid discrete particle swarm optimization integrated with simulated annealing (HDPSO-SA) algorithm which is decomposed into global and local search phases. The global search engine based on discrete particle swarm optimization includes two enhancements: a new initialization method to improve the quality of initial population and a novel gBest selection approach based on extreme difference to speed up the convergence of algorithm. The local search engine is based on simulated annealing algorithm, where four neighborhood structures are designed under two different local search strategies to help the proposed algorithm jump over the trap of local optimal solution. Finally, computational results of a real-world case and simulation data expanded from benchmark problems indicate that our proposed algorithm is significant in terms of the quality of non-dominated solutions compared to other algorithms.
Article
Since there are some special constraints in a real foundry enterprise, including the limitation of starting time in some casting operations and the transportation time between two adjacent operations, processing interval constraint (PIC) and job transportation time (JTT) are introduced in this paper. With the consideration of PIC and JTT, a multi-objective casting production scheduling model is constructed to minimize makespan, the total production cost and the total delivery delay time. A hybrid discrete multi-objective grey wolf optimizer (HDMGWO) is developed to solve this model. An initialization strategy based on reducing job transportation time and processing time (RTP) are designed to improve the quality of initial population. A improved tabu search (ITS) algorithm is embedded into grey wolf optimizer (GWO) to overcome the premature convergence of the GWO. A modified search operator of GWO is designed to tackle discrete combinatorial optimization. A case example of the real foundry enterprise is illustrated to evaluate the effectiveness of proposed HDMGWO. Experimental results demonstrate that the proposed HDMGWO is superior in terms of the quality of solutions compared to five multi-objective algorithms. Real running in a casting ERP system verifies the applicability of the proposed scheduling model and the HDMGWO.