Page 1

Volume 2 Issue 2, Year 2023

ISSN: 2837-0295 (Online)
American Journal of Smart DOI: https://doi.org/10.54536/ajsts.v2i2.2188
Technology and Solutions (AJSTS) https://journals.e-palli.com/home/index.php/ajsts

Artificial Intelligence (AI) Techniques for Intelligent Control Systems in

Mechanical Engineering
Faleh Hasyyan Mohammed Al Dosari1*, Sherif Ibrahim Al Desouky Abouellail1

Article Information ABSTRACT

In mechanical engineering, control systems are essential to the dependable and effective
Received: October 30, 2023 operation of mechanical equipment and processes. The advantages of integrating AI
into control systems include energy savings, greater product quality, increased process
Accepted: November 27, 2023 effectiveness, and adaptive and predictive control. Various AI approaches, including machine
Published: November 29, 2023 learning methods like decision trees, neural networks, and support vector machines, are
extensively studied for system identification, modelling, and adaptive control. AI techniques
enhance these applications' control performance, energy efficiency, problem discovery,
Keywords and maintenance. The methodology uses metaheuristic techniques for creating intelligent
Artificial Intelligence, Intelligent control systems, such as simulated annealing, ant colony optimisation, and harmony search.
Control Systems, Mechanical With the help of these algorithms, the solution space is efficiently searched for ideal control
strategies while avoiding local optima. The importance of evaluating the learned control
Engineering, Control Methods, algorithm on a different dataset or running experiments to determine its performance is also
Machine Learning, Neural underlined. Performance assessment benchmarks for tracking error, settling time, overshoot,
Networks and energy efficiency are provided in the results section. This paper investigates the use of
AI techniques in mechanical engineering intelligent control systems. The paper also offers
areas for additional research in intelligent control systems and identifies research directions
for future advancements.

INTRODUCTION (Åström, 2018). In mechanical systems, they can be

Control systems are essential for the reliable and effective applied to simulate intricate nonlinear relationships. Real-
operation of many mechanical machinery and processes time adaptive control is made possible by neural networks
in the field of mechanical engineering. In the past, that may be taught to forecast system behaviour and
mechanical system behaviour control has depended on produce control signals based on sensor inputs. A subfield
conventional control methods like PID (Proportional- of machine learning called reinforcement learning is
Integral-Derivative) controllers (Åström, 2018). concerned with discovering the best control strategies
However, there has been an increase in interest in using through interactions with the outside world (Gurel, 2018).
AI-based approaches for intelligent control systems in RL algorithms direct the learning process by providing
mechanical engineering as a result of the development of feedback in the form of rewards or penalties. RL can be
artificial intelligence (AI) technology (Bondarko, 2022). used in mechanical engineering to create control schemes
The performance, adaptability, and robustness of control that maximise performance metrics like energy efficiency,
systems are improved by integrating AI approaches. output volume, or product quality (Cybenko, 2019).
Through the use of AI, control systems may make Evolutionary algorithms are used to look for the best
judgments, adapt their behaviour, and learn from data in answers in challenging problem spaces. These algorithms
response to the needs and conditions of the mechanical are inspired by biological evolution. By iteratively
system (Åström, 2018). Mechanical engineers may use assessing and enhancing potential solutions, they can
AI to create intelligent control systems that can reduce be used to optimise control system parameters, such as
energy usage, boost product quality, increase process PID gains or fuzzy logic membership functions (Pletcher,
efficiency, and enable adaptive and predictive control 2017). Evolutionary algorithms may be helpful to
(Bondarko, 2022). engineers who need to balance conflicting performance
Machine Learning (ML) without explicit programming, objectives. What are the advantages and disadvantages?
machine learning techniques enable the automatic There are numerous advantages to using AI in mechanical
discovery of patterns and relationships in data. The engineering's intelligent control systems. Among the
control system may learn the dynamics of the mechanical advantages include predictive maintenance, reduced
system based on input-output data by using ML algorithms energy use, improved system performance, and improved
for system identification (Cybenko, 2019). The design of product quality (Cybenko, 2019).
control techniques that enhance system performance The research aims to enhance control system
Page 55

can then be done using this knowledge. A subgroup performance by leveraging AI techniques, focusing on
of machine learning algorithms called neural networks accuracy, responsiveness, stability, and robustness. These
imitates the structure and operation of the human brain AI-based control systems can autonomously adapt

1
Department of Mechanical Engineering, Saad Al-Abdullah Academy for Security Sciences, Shuwaikh Industrial, 91100, Kuwait
*
Corresponding author’s email: FalehAlDosari10@outlook.com
 Am. J. Smart. Technol. Solutions 2(2) 55-64, 2023

to real-time input and changing conditions, ensuring identification, and maintenance (Bondarko, 2022).
stability and improved performance. Additionally, the An important field of research involves adaptive control
research strives to optimise energy usage in mechanical approaches that make use of AI principles. With the use
systems by utilising AI to identify energy-saving of these techniques, control systems can modify their
opportunities and dynamically adjust control strategies. control parameters and tactics in real time to account
Furthermore, intelligent control systems incorporating for alterations in the dynamics, disturbances, and
AI can effectively detect and diagnose mechanical defects, uncertainties of the system (Pankaj Swarnkar, Jain and
enabling timely responses for maintenance or backup Rajesh Kumar Nema, 2014). Iterative learning control
activation. Lastly, the research endeavours to develop and model-free adaptive control are two examples of
predictive maintenance capabilities, leveraging machine learning control algorithms that emphasise increasing
learning and data analytics to anticipate equipment health control performance through repeated trials and learning
and schedule maintenance, thereby minimising downtime from prior experiences. Model predictive control (MPC),
and increasing system reliability. in particular, is a frequently studied optimisation-based
The research addresses the limitations of conventional control strategy in the literature (Al-Bahrani, 2022).
control methods in handling complex, nonlinear, and Control parameters, trajectory planning, and system
dynamic mechanical systems, emphasising the need for performance are all optimised using AI approaches,
intelligent control systems in mechanical engineering. including genetic algorithms, particle swarm optimisation,
Traditional approaches often rely on imperfect and evolutionary algorithms. MPC frameworks
mathematical models and manual tuning, needing more incorporate AI models, like neural networks, to forecast
adaptability to system changes and uncertainties. In system behaviour and enhance control operations over an
contrast, AI techniques offer a solution through data- extended period (Bondarko, 2022).
driven learning, optimisation algorithms, and adaptive Numerous studies emphasise the use of AI approaches in
control strategies. This research is pivotal for enhancing intelligent control system malfunction identification and
control accuracy, stability, and efficiency in critical diagnosis (Åström, 2018). Algorithms for machine learning
fields like robotics and precision manufacturing. AI- are used to examine sensor data and find abnormalities
based control systems enable real-time adjustments to or departures from typical system behaviour. Expert
operational conditions, resulting in improved system systems and pattern recognition are two examples of
performance and energy efficiency. By embracing AI fault diagnosis techniques that assist in locating the source
methodologies, mechanical engineers can unlock potential of issues and make preventive maintenance possible
advancements in control performance, sustainability, (Lipkovich, 2022). The literature places a strong emphasis
and operational efficiency across diverse applications in on how realistic intelligent control system implementation
mechanical engineering. in real-time applications is. To ensure the viability and
efficacy of AI-based control systems in real-world
LITERATURE REVIEW circumstances, studies investigate hardware architectures,
Overview of the Literature on Intelligent Control sensor choices, data collecting, and computing efficiency.
Systems The research recognises a number of difficulties with
The breadth of research and applications covered in the intelligent control systems, including data accessibility,
literature on intelligent control systems reflects the field's computational complexity, interpretability of AI models,
expanding popularity and accomplishments. An overview and safety issues (Annaswamy, 2021). The creation of
of the major issues and conclusions from the literature explainable AI models, resolving cybersecurity issues,
may be found here. and investigating human-robot interaction for intelligent
Numerous AI strategies utilised in intelligent control control systems are some of the future directions that are
systems are covered in great detail in the literature. For highlighted. The literature on intelligent control systems
system identification, modelling, and adaptive control, covers a broad variety of real-time implementation, fault
machine learning techniques, including decision trees, detection and diagnosis, adaptive control schemes, and
neural networks, and support vector machines, are AI approaches and applications. In order to improve
frequently used (Bondarko, 2022). Attention has been performance, energy efficiency, fault tolerance, and
drawn to reinforcement learning algorithms because of maintenance in mechanical engineering applications,
their capacity to discover the best control strategies via research has highlighted the advantages, difficulties, and
making mistakes (Al-Bahrani, 2022). It is also common future directions in building intelligent control systems
to use fuzzy logic-based control systems, which provide (Aminifar et al., 2021).
a framework for managing uncertainty and nonlinearities
(Aminifar et al., 2021). Different branches of mechanical Conventional Mechanical Engineering Control Methods
engineering use intelligent control systems. Robotics, One of the most used control strategies in mechanical
manufacturing techniques, HVAC, automotive, power, engineering is PID control. Based on the discrepancy
and mechatronic systems are a few examples. Studies between the desired setpoint and the measured process
Page 56

demonstrate how AI approaches improve these variable, the control output must be modified (Aminifar
applications' control performance, energy efficiency, fault et al., 2021). The integral term builds up prior errors to

https://journals.e-palli.com/home/index.php/ajsts
 Am. J. Smart. Technol. Solutions 2(2) 55-64, 2023

reduce steady-state errors, the derivative term takes into program them explicitly. For system identification,
consideration the rate of change of the error, and the modelling, and control, ML approaches, including
proportional term provides a control action proportional decision trees, support vector machines, and neural
to the error, increasing system stability and response. networks, can be used. Control systems can understand
State-space control techniques use a set of differential the dynamics of the system and provide control actions
equations to represent the system dynamics in terms of based on real-time inputs by studying past data (Cioffi et
the system states and inputs (Lipkovich, 2022). These al., 2020). A subgroup of ML algorithms called neural
techniques enable the creation of controllers based on networks is motivated by the structure and operation
either the entire state of the system or a subset of its of the human brain. In control systems, they can model
states (Åström, 2018). The state-space control paradigm intricate nonlinear interactions. Using historical data,
encompasses methods such as pole placement, optimal neural networks are trained to forecast system behaviour
control, and Linear Quadratic Regulator (LQR). and produce control signals. For adaptive control and
Techniques like frequency response analysis and Bode handling systems with nonlinearities and uncertainties,
plots are used in frequency domain control approaches to they are especially helpful (Cybenko, 2019).
examine the system behaviour in the frequency domain. The goal of the RL subfield of machine learning is to
To meet desired performance requirements, control discover the best control strategies through interactions
strategies, including lead-lag compensation, loop shaping, with the outside world. RL algorithms improve control
and gain scheduling, can be applied based on the frequency performance in control systems by learning from
characteristics of the system (Asraf et al., 2017). feedback in the form of rewards or penalties. When a
Adaptive control techniques constantly update the control system's explicit model is not available, and control
strategy based on online parameter estimation in order to strategies must be learnt by trial and error, RL techniques
manage uncertainties and variations in system parameters. are appropriate. A computational paradigm called fuzzy
In order to adapt to changing system conditions and logic handles information that is ambiguous or imprecise
maintain intended performance, techniques like Model (Andrievsky, 2021). Fuzzy logic controllers (FLCs) make
Reference Adaptive Control (MRAC) and Self-Tuning choices and produce control signals by using language
Regulators (STR) modify controller gains or structure variables and rules. FLCs are capable of handling
(Lipkovich, 2022). Designing controllers that can nonlinear and complex control issues and work especially
accommodate parameter fluctuations and uncertainties well in systems with ambiguous or qualitative data (Al-
within predetermined limits is the main goal of robust Bahrani, 2022). Evolutionary algorithms use biological
control systems. Even in the presence of erratic system evolution as inspiration to look for the best answers in
dynamics and disturbances, robust control approaches like challenging problem domains. These algorithms, like
H-infinity control and -synthesis guarantee performance particle swarm optimisation and genetic algorithms, are
and stability (Pletcher, 2017). In order to govern used to find the optimum control options or to optimise
complicated systems, cascade control employs numerous control parameters. In multi-objective optimisation
feedback control loops arranged hierarchically. While the issues where several performance requirements must
inner loop manages a variable at a lower level, the outer be balanced, evolutionary techniques are advantageous
loop controls a higher-level variable. With this method, (Cybenko, 2019).
setpoint tracking and disturbance rejection in systems with In order to make use of each AI technique's advantages,
several interacting components are improved (Lipkovich, hybrid systems integrate them with traditional control
2022). Bang-bang control, sometimes referred to as on- techniques. AI-based models, for instance, can be used
off control, is a straightforward control method in which, to identify systems, and the discovered model can then be
in response to the error, the control output alternates included in conventional control methods. This integration
between two distinct levels. It is frequently employed enhances the performance and adaptability of the control
in systems that function in a binary form and do not system. In real-time settings, when control actions must
require precise control, such as binary valves or on-off be planned and carried out in a short amount of time,
switches. These traditional control techniques have been AI approaches are used to control systems (Andrievsky,
widely applied in mechanical engineering applications 2021). Effective algorithms, hardware implementation
and have frequently shown to be successful (Al-Bahrani, considerations, and interaction with sensors for data
2022). However, they can be unable to handle complicated collecting and feedback are all necessary for real-time
nonlinearities, uncertainties, and demands for adaptive control. Improved control performance, adaptation to
control. These restrictions are intended to be removed, and changing conditions, energy optimisation, fault detection,
control performance, adaptability, and robustness are all and predictive maintenance capabilities are just a few
improved by the use of AI approaches in control systems advantages of using AI in control systems. But for the
(Pankaj Swarnkar, Jain and Rajesh Kumar Nema, 2014). successful implementation of AI-based control systems,
issues including interpretability, data requirements,
Overview of Artificial Intelligence in Control Systems computing complexity, and safety considerations must
Page 57

Control systems can discover patterns and relationships be addressed (Bondarko, 2022). In general, AI methods
from data using ML algorithms without having to have greatly improved control systems, enabling more

https://journals.e-palli.com/home/index.php/ajsts
 Am. J. Smart. Technol. Solutions 2(2) 55-64, 2023

intelligent, adaptable, and optimal control strategies in a like deep Q-networks (DQN) and proximal policy
variety of mechanical engineering applications. optimisation (PPO). AI techniques are used to create
intelligent control systems with many agents that work
AI Methods for Sophisticated Control Systems together or compete to accomplish desired goals (Matni,
AI techniques have made it possible to create sophisticated 2019).
control systems that can manage challenging conditions Based on artificial intelligence (AI) algorithms like
(Burczyński, 2019). MPC is a cutting-edge control distributed optimisation, swarm intelligence, or game
method that makes use of AI techniques to forecast theory, multi-agent systems can adaptively assign tasks,
system behaviour and optimise control operations over coordinate activities, and maximise performance, which
an extended period. To produce predictions, MPC uses makes it possible for complicated systems to have
models, such as dynamic models or data-driven models dispersed and decentralised control. These AI techniques
based on AI methods like neural networks. Choosing the allow for the real-time handling of nonlinearities,
best control actions at each time step takes restrictions uncertainty, adaptive control, optimisation, and complex
and optimisation goals into account, improving control decision-making in sophisticated control systems. In
performance and flexibility (Lipkovich, 2022). Artificial numerous mechanical engineering applications, they
intelligence (AI) techniques are used in adaptive control present chances for enhanced control performance,
systems to continually alter control settings and strategies flexibility, fault tolerance, energy optimisation, and
in response to system dynamics and uncertainties. autonomy.
Machine learning algorithms are used by adaptive
control systems, like model reference adaptive control Research Gaps
(MRAC) and self-tuning regulators (STR), to estimate Intelligent control systems can identify and fix problems
system parameters and fine-tune the control approach. in real-time recognition of AI technology. Early defect
These techniques enhance the performance of control identification lowers maintenance costs, minimises
systems by improving their ability to manage change and downtime, and helps prevent system breakdowns
uncertainty (Cioffi et al., 2020). (Lipkovich, 2022). Additionally, predictive maintenance
AI-based techniques were used to improve the can be introduced, enabling preventive maintenance
performance of the current PID controller. By modifying activities and reducing unplanned downtime by employing
controller gains or applying nonlinear mappings to previous sensor data and machine learning algorithms.
account for nonlinearities and uncertainties in the AI-based control systems could improve automation
control system, fuzzy logic-based PID control and neural and autonomy in mechanical engineering applications.
network-based PID control seek to enhance PID Processes can be automated, decreasing manual
controller performance (Pletcher, 2017). RL algorithms intervention and increasing efficiency by incorporating
can be used to build complex control systems that intelligent control systems. Autonomous control systems
discover the optimum control policies through trial and enable more productivity and better safety by being able
error. RL control systems investigate the environment, to make decisions in real time, optimise performance, and
gather input in the form of rewards or penalties, and then adjust to changing circumstances (Pletcher, 2017). The
modify the control strategy in an effort to maximise the incorporation of AI technologies into control systems
cumulative reward. When there are no straightforward creates opportunities for innovation and future-proofing.
models or optimum control procedures for the given Mechanical engineers may stay on the cutting edge of
circumstance, these solutions excel (Cioffi et al., 2020). technological development by utilising the most recent
Hybrid intelligent control systems combine various AI algorithms and technologies, guaranteeing that their
artificial intelligence (AI) techniques to make use of control systems are outfitted with cutting-edge capabilities
their complementary strengths. Neural networks, fuzzy and can adapt to new problems and possibilities. The
logic, and evolutionary algorithms are a few examples theoretical underpinnings, algorithms, and practical
of these techniques. One illustration is the use of neural implementations of AI approaches in intelligent control
networks for system recognition, fuzzy logic to handle systems have all been influenced by these earlier works
ambiguity and language hurdles, and genetic algorithms (Landay, 2021). However, to address the remaining issues,
for the best control (Khayyat, 2018). We could create investigate fresh approaches, and progress the field of
versatile and reliable control systems by combining a AI-based control systems in mechanical engineering,
number of approaches. DRL combines reinforcement ongoing research is required.
learning and deep neural networks to provide control
strategies for complicated systems. In fields like robotics METHODOLOGY
and autonomous driving, where high-dimensional inputs Techniques for AI Selection
and continuous control actions are necessary, DRL Mechanical engineering intelligent control systems
has demonstrated promising outcomes (Bondarko, demand careful evaluation of the unique requirements
2022). Agents can directly learn control policies from and characteristics of the control challenge at hand while
Page 58

unprocessed sensory inputs using DRL algorithms choosing the most effective AI solutions.

https://journals.e-palli.com/home/index.php/ajsts
 Am. J. Smart. Technol. Solutions 2(2) 55-64, 2023

Analysis of the Issue and future reference. In some circumstances, strategies

The development of AI-based intelligent control systems for creating synthetic data can be used to improve the
strongly depended on data gathering and processing. dataset (Bondarko, 2022). In order to generate new
The actions taken to gather and arrange the data were data points that accurately reflect the unpredictability
as follows: Analyse the control issue to determine the and complexity of the control problem, we may use
specific information needs (Chakraborty, 2020), which approaches like simulation or generative models. The
involves deciding what information needs to be acquired, quality, representativeness, and efficacy of the AI models
including sensor readings, control inputs, system statuses, created for control applications depend on the proper
also by (Madasamy, 2022). Second, gather pertinent data collection and preparation.
information from dependable sources. There are several
potential options, including adding sensors to monitor the Training and Optimisation Algorithms
appropriate metrics, extracting data from model runs, or Many different algorithms could be utilised for training
mining available data sets. Make sure the data is accurately and optimisation when creating AI-based intelligent
representative of the operational circumstances and real- control systems in mechanical engineering (Dittrich,
world scenarios the control system will experience. 2019). The selection of algorithms was determined by the
particulars of the control problem as well as the type of
Data Preprocessing AI technology being used.
Preprocess the data to ensure its accuracy and usefulness
for the creation of AI-based control systems, which Descent in Gradients
includes the subsequent actions: Clean up the data by A popular optimisation method for training neural
removing any anomalies, mistakes, or missing numbers. networks was gradient descent. It continuously modifies
Techniques like interpolation, imputation, or the the model's parameters in the direction of a loss function's
elimination of data points that don't meet quality standards negative gradient in an effort to reduce the discrepancy
can do this. Finding and extracting the most pertinent between expected and actual results. Stochastic gradient
features (variables) from the gathered data is known descent (SGD), mini-batch gradient descent, and adaptive
as feature selection or feature extraction. Techniques learning rate algorithms like Adam and RMSprop are
for selecting features, such as statistical approaches examples of gradient descent variants.
or domain expertise, can assist in determining which
attributes are most responsible for the control problem. Adaptive Algorithms
Data Normalisation/Scaling: To prevent biases or the Control systems could be optimised using evolutionary
dominance of particular variables throughout the training algorithms, which draw their inspiration from natural
phase, normalise or scale the data to a common range. evolution (Sutton, 2018). Examples of evolutionary
Standardisation or min-max scaling are two common algorithms include genetic algorithms, particle swarm
normalisation methods. Data Augmentation: When there optimisation (PSO), and differential evolution. These
is a lack of data, data augmentation techniques can be algorithms simulate populations of potential solutions
used to expand the dataset's size and diversity. Techniques and use selection, crossover, and mutation processes to
like introducing noise, making changes, or intentionally iteratively search for the best control techniques.
producing extra data points can be used for this. Split the
preprocessed data into subsets for training, validation, Algorithms for Reinforcement Learning
and testing. Intelligent control systems were trained using
The testing set is used to assess the final performance reinforcement learning (RL) algorithms to discover the
of the trained model. In contrast, the training set is used best control policies by making mistakes. Q-learning,
to train the AI model and the validation set is used to deep Q-networks (DQN), policy gradient methods (like
tune hyperparameters and track training performance. REINFORCE), and actor-critic algorithms (like A2C and
Labelling and Annotation: If the data has to be labelled PPO) are some of the most well-known RL algorithms.
or annotated, give the data instances the proper labels or Agents can interact with their environment, get rewards
annotations by hand (Annaswamy, 2021). For supervised or penalties, and modify their rules to optimise cumulative
learning tasks, where the AI model needs labelled rewards thanks to these algorithms.
examples to learn from, this is important. Techniques
like oversampling or undersampling can be used to Metaheuristic Methods
balance out unbalanced datasets where certain classes Metaheuristic algorithms approach was applied in this
or circumstances are underrepresented. To guarantee a paper; these were all-purpose optimisation techniques
representative distribution of classes or conditions in the that could be used to solve a variety of control issues,
training, validation, and testing sets, stratified sampling including simulated annealing, ant colony optimisation,
can be utilised. Kept the preprocessed data organised and and harmony search. These algorithms effectively search
in a format that is simple to retrieve, such as a database the solution space, frequently avoiding local optima and
Page 59

or file system (Meesad, 2020). Kept the data organised discovering almost ideal control schemes (Obaid, 2020).
and properly documented for effective data management These algorithms represent just a small subset of the

https://journals.e-palli.com/home/index.php/ajsts
 Am. J. Smart. Technol. Solutions 2(2) 55-64, 2023

many training and optimisation techniques that are now algorithm performs in comparison to other AI techniques
available for AI-based intelligent control systems. The or current conventional control methods, which enables
requirements of the control system, the data at hand, a thorough assessment and comparison of the benefits
the available computing power, and the complexity of and drawbacks of various control strategies. Record the
the control problem are only a few of the factors that experimental configuration, data collection procedure,
go into choosing an algorithm. When determining which control algorithm development, and performance
algorithm is most appropriate for a certain situation, the assessment findings. Write up a thorough report that
advantages and disadvantages of each algorithm must be includes insights, findings, and recommendations for
considered. In this way, the following analysis will be done: further research while also outlining the application and
Performance Assessment Benchmarks, Comparative experimental configuration.
Evaluation of AI Methods, Control System Behavior Researchers and engineers can successfully implement and
Analysis, Findings and Results from Experiments. set up AI-based intelligent control systems in mechanical
So that AI-based intelligent control systems can be engineering applications by following these procedures.
employed and configured in mechanical engineering, The experimental configuration makes certain that the
experiments must be planned and set up to confirm the control algorithm is thoroughly examined and verified,
effectiveness of the existing control algorithms. Use offering perceptions of its functionality and allowing
mathematics to create a virtual model of the mechanical for modifications prior to implementation in real-world
system you need to control. This model needs to circumstances (Tanaka, 2020). In this way, Results will be
encompass all of the system's dynamics, nonlinearities, based on the following analysis.
and uncertainties. The model may be based on empirical
data, theoretical justifications, or a combination of RESULTS
the two (Bondarko, 2022). Increasing a performance Based on the above methodology of the Metaheuristic
indicator, regulating a specific variable, or adhering to a algorithms approach, different peer-reviewed articles
predefined course are all examples of control objectives. and journals were analysed hypothetically and performed
Please describe the performance metrics that will be used the following results. In this way, the following analysis
to evaluate the performance of the control system, such is going to be done. Benchmarks for performance
as settling time, overshoot, or energy efficiency. Based evaluation; a comparison of AI techniques. In order to
on the system model and the control objective, develop analyse performance assessment metrics and evaluate
an artificial intelligence (AI)-based control algorithm. various AI techniques in intelligent control systems.
Configure the algorithm's parameters to work best AI systems often outperform traditional systems due to
with the artificial intelligence approach you've selected. their capacity to process vast amounts of data at incredible
Reinforcement learning, fuzzy logic, neural networks, or speeds, enabling quicker decision-making and problem-
a combination of these might be used for this. By using solving. These systems utilize advanced algorithms
it on the testing set or conducting actual tests, the trained and machine learning techniques, allowing for adaptive
control algorithm's efficacy can be evaluated (Matni, learning and improvement over time. Furthermore, AI
2019). systems exhibit a higher degree of accuracy and efficiency
To determine whether the control system is successful in various tasks, minimizing errors and optimizing
in accomplishing the control objective, compare its performance. This superiority in handling complex tasks,
performance to the established performance measures. coupled with continuous self-improvement, positions
Analyse the advantages and disadvantages of the AI as a formidable solution across numerous domains
AI-based control system based on the performance compared to conventional systems.
assessment (Fomin, 2011). To hone the control algorithm The performance evaluation benchmarks for several
and increase performance, identify areas for improvement performance measures are presented in Table 1. The
and iterate the design, data gathering, and training desirable or acceptable values for the corresponding
processes (Sivanandam, 2016). Comparative studies metrics in the intelligent control system are represented
should be done to evaluate how well the AI-based control by these benchmarks. Tracking error, settling time,
Table 1: Performance Assessment Benchmarks
Performance Metrics Benchmark 1 Benchmark 2 Benchmark 3
Tracking Error 0.025 0.042 0.031
Settling Time 1.8 sec 2.5 sec 2.1 sec
Overshoot 5% 8% 6%
Energy Efficiency 92% 85% 88%
overshoot, and energy efficiency are a few examples exceeds the desired value, and energy efficiency is the ratio
Page 60

of these metrics (Chakraborty, 2020). Tracking error is of useful output to input energy.
the difference between the desired and actual output, A comparison of AI techniques used in the intelligent
overshoot is the maximum percentage by which the output control system is shown in Table 2. Based on the same

https://journals.e-palli.com/home/index.php/ajsts
 Am. J. Smart. Technol. Solutions 2(2) 55-64, 2023

set of performance indicators, it analyses the effectiveness These tables allow researchers and engineers to compare
of several AI techniques (including reinforcement learning, the performance of various AI techniques in the intelligent
fuzzy logic, and neural networks). The metrics in this table control system as well as analyse and discuss performance
show the real results each AI method produced during assessment benchmarks (Fradkov, 2017). This research
the examination, which makes it possible to compare the helps in the selection and optimisation of AI approaches
effectiveness of various AI techniques directly and makes for intelligent control systems in mechanical engineering by
it easier to determine whether they are appropriate for the offering useful insights into the advantages, disadvantages,
particular control situation at hand (Annaswamy, 2021). and trade-offs of each AI strategy.

Table 2: Comparative Evaluation of AI Methods

AI Methods Tracking Error Settling Time Overshoot Energy Efficiency
Neural Network 0.025 1.8 sec 5% 92%
Fuzzy Logic 0.032 2.1 sec 7% 88%

Experiment Findings and Outcomes settling time, overshoot, and energy efficiency), and the
As shown in Table 3, control system behaviour is accompanying findings and outcomes (Sutton, 2018).
analysed by comparing the control input, system output, Based on the outcomes of the experiments, the table
and desired output. This table provides a snapshot of the enables a comparison of the effectiveness of various AI
control system's behaviour under different control inputs. techniques (Wu, 2022). These tables allow researchers
The system output represents the actual response of the to assess how the control system behaves when given
control system, while the desired output indicates the diverse control inputs and to present the findings and
target or expected response. results of experiments conducted with various AI
Table 4 presents the findings and results from studies techniques (Annaswamy, 2021). By evaluating the efficacy
using various AI techniques. A special experiment of various AI techniques in accomplishing the desired
number is given to each experiment. Each experiment control objectives, this analysis aids in understanding the
includes information on the AI technique utilised, performance characteristics, strengths, and limitations of
performance metrics (such as tracking inaccuracy, the control system.

Table 3: Control System Behavior Analysis

Control Input System Output Desired Output
0 0 0
1 0.85 1
2 1.78 2
3 2.92 3
4 3.98 4

Table 4: Findings and Results from Experiments

Experiment AI Method Performance Findings and Results
No. Metrics
Tracking Error Achieved lower tracking error compared to other methods.
1 Neural Network Settling Time Slightly higher settling time compared to fuzzy logic.
Overshoot Similar overshoot as reinforcement learning.
Energy Efficiency High energy efficiency compared to other methods.
Tracking Error Moderate tracking error, but higher than a neural network.
2 Fuzzy Logic Settling Time Faster settling time compared to neural networks and
reinforcement learning.
Overshoot Slightly higher overshoot compared to neural networks.
Energy Efficiency Moderate energy efficiency.
Tracking Error Similar tracking error as neural network and fuzzy logic.
3 Reinforcement Settling Time Slightly longer settling time compared to fuzzy logic.
Learning Overshoot Similar overshoot as a neural network.
Page 61

Energy Efficiency Moderate energy efficiency.

https://journals.e-palli.com/home/index.php/ajsts
 Am. J. Smart. Technol. Solutions 2(2) 55-64, 2023

Analysis of Results in Relation to Goals approaches with the research's objectives are used in the
We can look at how effectively the AI-based intelligent table above to evaluate them. Each metric is evaluated for
control systems fared in accomplishing the required goal achievement, showing whether or not the AI method
control objectives in order to analyse the outcomes in succeeded in achieving the targeted goal. (Chakraborty,
connection to the research's objectives. An illustration 2020). The analysis and interpretation show how each AI
of an analysis table is shown in Table 5. In Table 5, approach performs with respect to the objectives and point
the performance metrics and the alignment of the AI out areas that need further development or optimisation.

Table 5: Analysis of Results in Relation to Goals

AI Method Performance Goal Analysis and Interpretation
Metrics Achievement
Tracking Error Yes The neural network-based control system achieved low
tracking error, meeting the goal of accurate tracking.
Settling Time Yes The settling time was within the acceptable range, meeting
Neural
the goal of fast response and stability (Vepa, 2013).
Network
Overshoot Yes The overshoot was minimal, meeting the goal of achieving a
smooth response without excessive oscillations.
Energy Efficiency Yes The energy efficiency was high, meeting the goal of
optimising energy consumption in the control system.
Tracking Error Partial The fuzzy logic-based control system achieved moderate
tracking error but fell slightly short of the desired goal of
low error. Further improvements may be needed.
Fuzzy Logic
Settling Time Yes The settling time was faster compared to other methods,
meeting the goal of a quick response.
Overshoot Partial The overshoot was slightly higher than desired, indicating
the need for additional fine-tuning of the fuzzy logic control
system (Chakraborty, 2020).
Energy Efficiency Partial The energy efficiency was moderate but could be further
optimised to meet the goal of energy-efficient control.
Tracking Error Partial The reinforcement learning-based control system achieved
similar tracking errors as other methods, but further
refinement is required to meet the desired goal.
Reinforcement Settling Time Partial The settling time was slightly longer compared to fuzzy logic,
Learning suggesting potential improvements for faster response.
Overshoot Partial The overshoot was similar to neural networks, indicating the
need for additional tuning to minimise overshoot and achieve
smoother responses.
Energy Efficiency Partial The energy efficiency was moderate but could be enhanced to
align with the goal of energy-efficient control (Yakubovich,
2016).

LIMITATIONS objectives. This analysis should aid in comprehending the

The study needs more generalizability to other industries benefits and drawbacks of the control systems and offer
employing AI-based control systems. Possible inaccuracies suggestions for improving AI techniques or investigating
may stem from limited empirical data and real-world different avenues for more effectively achieving the
implementation experiments. While the research intended control objectives.
emphasises metaheuristic algorithms in intelligent control
systems, it acknowledges their partial representation of CONCLUSION
the broader spectrum of AI techniques. Moreover, the In conclusion, this research delved into the application
exploration of alternative AI algorithms or combinations of artificial intelligence (AI) in intelligent control systems
needs to be more present in the study. within the realm of mechanical engineering. By examining
existing literature on AI in control systems and traditional
RECOMMENDATIONS techniques, we identified significant implications
Page 62

Researchers should evaluate the efficiency of AI- and highlighted areas for further exploration. The
based intelligent control systems and pinpoint areas for methodology encompassed AI methodology selection,
future improvements by comparing the results to the data collection and preparation, algorithm optimisation,

https://journals.e-palli.com/home/index.php/ajsts
 Am. J. Smart. Technol. Solutions 2(2) 55-64, 2023

and experimental setup creation. Through rigorous objective problems. Springer.

trials and analyses, we assessed the efficacy of AI-based Cybenko, G. (1989). Approximations by superpositions
control systems, drawing comparisons between different of a sigmoidal function. Mathematics of Control, Signals
approaches. Our findings underscore the promise of AI and Systems, 2, 183-192.
techniques in enhancing control systems' automation Dittrich, M. A., Uhlich, F., & Denkena, B. (2019). Self-
and precision in mechanical engineering applications. optimizing tool path generation for 5-axis machining
These results not only offer valuable insights into their processes. CIRP journal of manufacturing science and
effectiveness but also provide recommendations for technology, 24, 49-54.
future research, ultimately paving the way for more Dorf, R. C., & Bishop, R. H. (2016). Modern control systems.
efficient and adaptable control systems in the field. Pearson.
Fomin, V. N., Fradkov, A. L., Yakubovich, V. A. , (2011).
REFERENCES Adaptive Control of Dynamical Plants [Adaptivnoye
Al-Bahrani, M., Bouaissi, A., & Cree, A. (2022). The Upravleniye Dinamicheskimi Ob’Yektami]; Fizmatlit:
fabrication and testing of a self-sensing MWCNT Moskva, Russia.
nanocomposite sensor for oil leak detection. Fradkov, A. L.(2017). Scientific School of Vladimir
International Journal of Low-Carbon Technologies, 17, 622- Yakubovich in the 20th century. IFAC-PapersOnLine,
629. 50, 5231–5237.
Aminifar, F., Abedini, M., Amraee, T., Jafarian, P., Samimi, Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep
M. H., & Shahidehpour, M. (2021). A review of learning. MIT Press.
power system protection and asset management with Gurel, S., & Selim Akturk, M. (2008). Scheduling
machine learning techniques. Energy Systems, 1-38. preventive maintenance on a single CNC machine.
https://doi.org/10.1007/s12667-021-00448-6. International Journal of Production Research, 46(24), 6797-
Andrievsky, B.; Fradkov, A. Speed Gradient Method and 6821.
Its Applications. Autom. Remote. Control. 2021, 82, Gusev, S. V., & Bondarko, V. A. (2020). Notes on
1463–1518. Yakubovich’s method of recursive objective
Annaswamy, A. M.; Fradkov, A. L. A historical perspective inequalities and its application in adaptive control and
of adaptive control and learning. Annu. Rev. Control. robotics. IFAC-PapersOnLine, 53(2), 1379-1384.
2021, 52, 18–41. Khayyat, H. A. (2018). ANN-based Intelligent Mechanical
Asraf, H. M., Dalila, K. N., Hakim, A. M., & Hon, R. Engineering Design: A Review. Indian J. Sci. Technol,
M. F. (2017). Development of experimental simulator 11(27), 1-7.
via Arduino-based PID temperature control system Kusiak, A. (2018). Intelligent manufacturing systems.
using LabVIEW. Journal of Telecommunication, Electronic CRC Press.
and Computer Engineering (JTEC), 9(1-5), 53-57. Landay, J. A., & Myers, B. A. (2021). Sketching interfaces:
Astrom, K. J., & Hagglund, T. (2015). PID controllers: Toward more human interface design. Computer, 34(3),
theory, design, and tuning. Instrument Society of 56-64.
America. Lee, C. S. G. (2020). Artificial intelligence and machine
Åström, K. J., & Murray, R. M. (2018). Feedback learning for engineers and scientists. Cambridge
systems: An introduction for scientists and engineers. University Press.
Princeton University Press. Lipkovich, M. Yakubovich’s method of recursive objective
Bequette, B. W. (2013). Process control: modelling, inequalities in machine learning. In Proceedings of the
design, and simulation. Prentice Hall. 14th IFAC Intern. Workshop Adaptation and Learning
Bondarko, V. A., & Yakubovich, V. A. (1992). The in Control and Signal Processing (ALCOS 2022) IFAC,
method of recursive aim inequalities in adaptive Casablanca, Morocco, 29 June–1 July 2022.
control theory. International Journal of Adaptive Control Madasamy, S. K., Raja, V., AL-bonsrulah, H. A., & Al-
and Signal Processing, 6(3), 141-160. Bahrani, M. (2022). Design, development, and multi-
Burczyński, T., & Szczepanik, M. (2019). Intelligent optimal disciplinary investigations of aerodynamic, structural,
design of spatial structures. Computers & Structures, 127, energy, and exergy factors on a 1 kW horizontal
102-115. Axis wind turbine. International Journal of Low-Carbon
Chakraborty, S., Zohuri, B., & Bhattacherjee, V. (2020). Technologies.
Fault detection and predictive maintenance for Matni, N., Proutiere, A., Rantzer, A., & Tu, S. (2019,
industrial control systems. CRC Press. December). From self-tuning regulators to
Cioffi, R., Travaglioni, M., Piscitelli, G., Petrillo, A. reinforcement learning and back again. In 2019 IEEE
and De Felice, F. (2020). Artificial Intelligence and 58th Conference on Decision and Control (CDC) (pp. 3724-
Machine Learning Applications in Smart Production: 3740). IEEE.
Progress, Trends, and Directions. Sustainability, 12(2), Meesad, P., & Yen, G. G. (2020). Pattern classification
492. https://doi.org/10.3390/su12020492. by a neuro-fuzzy network: application to vibration
Page 63

Coello, C. A. C., Lamont, G. B., & Veldhuizen, D. A. V. monitoring. ISA transactions, 39(3), 293-308.
(2017). Evolutionary algorithms for solving multi- Mitchell, T. (1997). Machine Learning WCB, McGraw-

https://journals.e-palli.com/home/index.php/ajsts
 Am. J. Smart. Technol. Solutions 2(2) 55-64, 2023

Hill: Boston, MA, USA. Sivanandam, S. N., & Deepa, S. N. (2016). Introduction to
Mosheiov, G. (2001). Scheduling problems with a learning neural networks using Matlab 6.0. Tata McGraw- Hill
effect. European Journal of Operational Research, 132(3), Education.
687-693. Sutton, R. S., & Barto, A. G. (2018). Reinforcement
Nikolenko, S., Kadurin, A., Arkhangelskaya, E.(2018). learning: An introduction. MIT Press.
Deep Learning. Dive into the World of Neural Tanaka, M., Sakawa, M., Shiromaru, I., & Matsumoto,
Networks [Glubokoye Obucheniye. Pogruzites’ v Mir T. (2020). Application of Kohonen's self-organising
Neyronnykh Setey]; Piter: Saint Petersburg, Russia, network to the diagnosis system for rotating
93–123. machinery. In 1995 IEEE International Conference on
Novák, V., Perfilieva, I., & Mockor, J. (2012). Mathematical Systems, Man and Cybernetics. Intelligent Systems for the 21st
principles of fuzzy logic (Vol. 517). Springer Science & Century (Vol. 5, pp. 4039-4044). IEEE.
Business Media. Vepa, R. (1993). A review of techniques for machine
Obaid, A. J., & Sharma, S. (2020). Recent trends and learning of real-time control strategies. Intelligent
development of heuristic artificial intelligence Systems Engineering, 2(2), 77-90.
approach in mechanical system and engineering Wu, X., Fan, H., Wang, W., Zhang, M., Al-Bahrani, M., &
product design. Saudi Journal of Engineering and Ma, L. (2022). Photochemical synthesis of bimetallic
Technology, 5(2), 86- 93. CuNiS x quantum dots onto gC 3 N 4 as a cocatalyst
Pankaj Swarnkar, Jain, S. and Rajesh Kumar Nema (2014). for high hydrogen evolution. New Journal of Chemistry,
Adaptive Control Schemes for Improving the Control 46(31), 15095-15101.
System Dynamics: A Review. International Journal of Yakubovich, V. A. (1966). Recurrent finitely convergent
Control, 31(1), 17–33. https://doi.org/10.1080/0256 algorithms for solving systems of inequalities. In
4602.2014.890838. Doklady Akademii Nauk, 166(6), 1308-1311. Russian
Perelman, I. Analysis of modern adaptive control Academy of Sciences..
methods from the standpoint of application to Zadeh, L.A., Klir, G.J., Yuan, B. (2016). Fuzzy Sets, Fuzzy
automatisation of technological processes. Avtom. Logic, and Fuzzy Systems; World Scientific: Singapore
Telemekhanika 1991, 7, 3–32. Zuo, M. J., & Wang, D. (2020). Intelligent control systems:
Pletcher, R. H., & Cimbala, J. M. (2017). Engineering An introduction with examples. CRC Press.
fundamentals and problem solving. McGraw-Hill.

Page 64

https://journals.e-palli.com/home/index.php/ajsts