Challenges and Opportunities of Computational Intelligence in Industrial Control System (ICS)

Artificial intelligence (AI) is not a fancy term anymore, or not limited to only researchers and academia. AI is currently becoming a part and parcel of our daily life, we are using AI/ intelligent systems by knowing or without knowing. Event product manufacturers are also trying to incorporate AI with their products to make it more preferable to consumers and trying to get the full benefit of using AI in their production and control units even for business decisions. Therefore, In our paper, we give a comprehensive survey of recent advances in Computational intelligence in industrial Control Systems and cover many usages of how industrial Control Systems are getting benefits from using Computational intelligence. We covered multiple domains like Manufacturing, Energy Management, Transportation, Food and Beverage Industry, and Pharmaceutical Industry, how these industries are utilizing multiple CI-based control systems like Programmable Logic Controllers, Distributed Control Systems, Supervisory Control and Data Acquisition, Industrial Automation, and Control Systems, Intelligent Electronic Devices and found benefits in their operations and manufacturing which helping them to focus more in innovation and improvement of their products. We believe that this survey shall be valuable to researchers across academia and industry.


I. INTRODUCTION
Artificial intelligence (AI) is transforming many industries, and its integration with industrial control systems (ICS).It has the potential to alter how manufacturing and industrial processes are handled and optimized.The term "Industrial Control Systems" refers to the mix of hardware and software that is used to monitor and control the operation of industrial processes and machinery [20].Traditionally, these systems relied on pre-programmed logic and human intervention for decision-making.However, the introduction of AI technologies opens up new possibilities for automation, predictive maintenance, real-time data analysis, and overall process improvement [23].
Industrial Control Systems, which include technologies such as SCADA, DCS, and PLC, have formed the backbone of industrial automation, allowing complicated processes to be monitored and controlled [4].Incorporating AI into these systems opens up a plethora of previously unreachable possibilities and benefits.The change from traditional rulebased, deterministic control to data-driven, adaptive decisionmaking is the primary driving reason behind the use of AI in ICS.Real-time sensor data, historical records, and other contextual information can be used by AI algorithms to optimize operations, detect possible faults, and enable autonomous decision-making [2].This revolutionary synergy improves the operational agility and competitiveness of enterprises in a variety of industries [3].AI can predict possible equipment failures by evaluating sensor data and equipment performance, allowing for proactive maintenance, minimizing downtime, and enhancing asset reliability.Furthermore, AIdriven anomaly detection enables real-time monitoring, allowing for early detection of deviations and rapid corrective actions to avoid major failures.Furthermore, AI augments human expertise through cognitive automation, which handles difficult tasks such as natural language processing, data analysis, and decision-making with unparalleled speed and precision [14].This human-machine collaboration provides vital insights to industrial operators, engineers, and decisionmakers, allowing them to make better-educated decisions.
These developments, however, bring with them new obstacles.To protect critical infrastructures from potential risks and attacks, the integration of AI in ICS demands sophisticated cybersecurity measures.Addressing ethical concerns, maintaining openness, and encouraging responsible AI use are all critical to sustaining public trust and confidence in the technology.

II. LITERATURE REVIEW
This paper [12] describes an improved intrusion detection approach for IoT-driven smart building control systems.Combining an ensemble AE-OCSVM model yields a 99.6% F1 score, resulting in an excellent intrusion detection system.However, false alarms in the OCSVM model and sporadic anomaly identification in the AE model necessitate more investigation.Future studies will try to improve model accuracy, investigate novel methodologies, and integrate ensemble methods with other unsupervised ML models to improve system performance.
This work [8] uses machine learning to predict cardiac issues using BRFSS survey data from 400k US adults.Xgboost achieves the highest accuracy of 91.30% when tested against six models: Bagging, Random Forest, Decision Tree, K-Nearest Neighbor, and Naive Bayes.The paper, however, recognizes the limitations of relying simply on accuracy for model evaluation.Future studies will try to enhance prediction accuracy by using larger datasets and relevant features, as well as investigating additional categorization techniques such as deep learning.
This article [6] presents a blockchain-based smart grid energy storage unit (ESU) charge coordination system that enhances transparency, reliability, and privacy while ensuring efficient power delivery to ESUs.It advocates for a decentralized charge coordination system with smart contracts, emphasizing the importance of a cryptocurrency-independent blockchain for widespread adoption.Future implementation will focus on utilizing blockchain technology suitable for larger smart grid applications.
The paper [2] presents a voice-activated, push-buttoncontrolled intelligent wheelchair for elderly, disabled, and patients.It increases user independence and mobility with forward, backward, left, right, and stop capabilities.The proposed model includes Arduino Mega, battery, motor driver, Bluetooth module, button switch, and car chassis.Future work involves testing, improvement, implementation, production research, and real-world deployment to enhance user mobility and freedom.
The study [15] introduces FedeX, a federated anomaly detection system for Industrial Control Systems (ICS) in Smart Manufacturing (SM) based on Federated Learning (FL), achieving high performance and lightweight deployment on edge devices.FedeX outperforms existing algorithms on various criteria using liquid storage and SWAT datasets, incorporating Explainable Artificial Intelligence (XAI) for interpretable reasons.
The research [3] presents an AI-powered IIoT framework for optimizing Caterpillar generator set (genset) operations, using real-time data from Modbus RTU and ML.Net.The platform aims to improve fuel efficiency, predict maintenance, and promote sustainability.Future work will explore advanced optimization algorithms for enhanced industrial efficiency.The study showcases the potential of IIoT and AI to revolutionize genset operations and boost economic rewards.
The study [5] proposes a signaling game-based defense using Moving Target Defense (MTD) to combat Stealthy Link Flooding Attacks (SLFAs).The technique dynamically modifies network settings to effectively protect with minimal overhead.Testing in a Mininet-based network shows comparable SLFA protection to complex MTD systems but with less overhead.Future work will focus on extending the architecture to protect against new DDoS-based threats.
This research [17] introduces a novel deep learning-based technique for detecting smart meter fraud and identifying possibly suspect Advanced Metering Infrastructure (AMI) relay nodes.By analyzing a large dataset of 1 million smart meter recordings, the suggested technique achieves an outstanding accuracy of 85.7% while maintaining a remarkably low false positive rate of only 5.7%.Future efforts will focus on improving the method's security capabilities and addressing dataset imbalance to improve the accuracy of detecting abnormalities and rogue nodes inside the AMI network.
This publication [16] proposes a unique automated approach for detecting paddy leaf illness in the context of Advanced Metering Infrastructure (AMI).Surprisingly, the Inception-ResNet-V2 model obtains a great accuracy of 92.68%, demonstrating its use in this application.However, it is worth noting that the model's performance may be influenced by dataset imbalance, demanding changes to maintain robustness in complicated settings.In the future, the study intends to investigate a broader range of disease types, finetune convolutional neural network (CNN) models, assess the efficacy of detection criteria, and conduct comparative analyses pitting Inception-ResNet-V2 against other CNN models designed for detecting plant leaf diseases.
This study [18] proposes a unique multi-agent reinforcement learning technique for optimal UAV work distribution.The technique models the path planning problem as a multiagent economic game, allowing for cooperative and competitive resource allocation (POIs).UAVs make judgments on movement and POI trading using Q-learning and a greedy strategy, resulting in successful work allocation.REPLAN-NER outperforms commonly used RL-based trajectory search techniques.Future work will concentrate on improving the training process, resolving problematic tactics, and suggesting new enhancements.
This study [11] proposes an adaptive load-shedding solution for under-frequency load shedding (UFLS) in power grids, taking frequency, voltage, and frequency derivative into account without the use of relay telecommunication.When compared to the typical Iranian grid technique, the DigSilent simulation outperforms it.The algorithm's simplicity and ease of implementation utilizing digital relay setup tools are stressed, and future studies may involve testing it in other settings and experimenting with adjustments to increase performance even more.
This research [20] providesa comprehensive security architecture for Smart Healthcare Systems (SHS) based on IoT, ubiquitous computing, and machine learning.HealthGuard identifies harmful activities such as interference, fraudulent data insertion, and device manipulation.The dataset includes vital sign data from eight medical devices for both healthy and diseased people.HealthGuard incorporates four machine learning approaches (ANN, Decision Tree, Random Forest, and k-Nearest Neighbor).The results demonstrate 91% accuracy and a 90% F1 score for detecting harmful activities in the Smart Healthcare System.

III. BASIC OF ICS
An Industrial Control System (ICS) is a complex network that includes control loops, human interfaces, and remote diagnostics and maintenance tools.Control loops regulate a controlled process by utilizing sensors, actuators, and controllers (e.g., PLCs).Sensors measure physical qualities and send this information to the controller as controlled variables, which interprets the signals and generates manipulated variables for the actuators to operate on [24].Operators and engineers use human interfaces to monitor, configure set points, control algorithms, and alter parameters in the controller.These interfaces also provide real-time process status as well as historical data.Diagnostics and maintenance utilities are also important in preventing, detecting, and recovering from anomalous operations or failures.Control loops can be organized in nested or cascade structures, with the set point of one loop dependent on the process variable of another.Throughout the operation, many control loops operate continuously, with cycle times ranging from milliseconds to minutes.Figure 1 depicts the basic operation of an ICS.In an industry control system (ICS), sensors gather realtime data on physical parameters, serving as a control input, while actuators carry out control actions in response to commands.Using control algorithms, the core controller processes sensor data to produce the desired results.Realtime data, visual representation, and user control are all provided by the HMI.Through a communication network, components are linked together.If necessary, safety systems will launch emergency measures after data analysis stores past data for monitoring.Redundancy and security measures are also included in ICS to assure ongoing and protected operations.This section explains the basic components of ICS utilized in control Some of these components are applicable to SCADA systems, DCS, and PLCs in general, while others are specialized to a specific type of system.The following is a representation of the fundamental control equation for a feedback control system: where: e(t) is the error at time t, SP is the setpoint (desired value), P V (t) is the process variable (actual value) at time t.
The controller algorithm processes the error signal to calculate the controller output CO(t): where f (e(t)) is the control algorithm function that determines the appropriate control action based on the error.
The feedback loop is finished when the actuators receive the controller output CO(t) and change the process variable to bring it closer to the setpoint.For more sophisticated control techniques and implementations employed in several industrial applications, this fundamental feedback control system serves as the basis.

IV. TYPES OF ICS
Industrial Control Systems (ICS) encompass a variety of types, each designed to serve specific functionalities and address varying complexities in control operations.The following section provides an overview of the most commonly utilized control systems in industrial settings: 1) Programmable Logic Controllers (PLCs): PLCs are highly advanced solid-state control systems [23] equipped with user-programmable memory.Within these memory modules, specific instructions are stored to facilitate a wide range of functions, including I/O control, logic operations, precise timing, counting, PID (Proportional-Integral-Derivative) control, communication protocols, arithmetic operations, as well as data and file processing.Figure 2 depicts the PLC Control System.2) Distributed Control Systems (DCS): DCS represents a sophisticated industrial control system specifically designed for distributed operations [14].In this approach, multiple control systems or processes operate autonomously, offering a departure from traditional centralized units.DCS relies on decentralized intelligence, allowing for efficient control and coordination of industrial processes.Improve fault isolation and communication resilience, and broaden SCADA with realtime analytics for improved distribution network performance.[2] Voice-activated intelligent wheelchair for the aged, disabled, and patients using Arduino Mega, Bluetooth module, and other components.
Improves user movement and independence.testing, improvement, implementation, production research, and real-world deployment.
Machine Learning.Net (ML.Net) Future research will investigate advanced optimization algorithms.[5] Effective defense against Stealthy Link Flooding Attacks (SLFAs) with low overhead employing Moving Target Defense (MTD).

Unspecified
Future emphasis on extending architecture to protect against future DDoS-based threats.
[6] a smart grid energy storage unit charge coordination system based on blockchain.
Improves openness, dependability, and privacy.[7] Biotechnological fermentation is vital in food and beverage production, yet intelligent control faces challenges from living organisms and changing conditions.
Fuzzy logic and neural fuzzy hybrids Enhance precision and reliability.
[9] Pipeline leakage concerns need immediate response; real-time DCS and CAO-SCADA monitoring successfully mitigates losses.models using z-test, verifying assumptions on pipeline failure, system variations, and industrial control.
Focusing on compliance, ICT, and SCADA policies will optimize operations and reduce losses in the future.[10] The hybrid intelligent-classic control technique in nonlinear cyber-physical and IoT systems.

Neural network (NN)
Improving the flexibility of the intelligent estimator and evaluating performance and real-world applications.[11] Adaptive load-shedding solution for underfrequency load shedding in energy systems.
Iranian grid method.try in other conditions, experimenting with modifications for improved performance.[12] Improved intrusion detection for IoT-driven smart buildings AE-OCSVM model achieving 99.6% F1 score.
False alarms in OCSVM and intermittent anomaly identification.[13] IEDs with distributed intrusion detection improve cybersecurity in IEC 61850 networks Proposed IEDs efficiently monitor anomalies in systems and communication (GOOSE, SV), accurately mitigating cyber threats.
Research should focus on improving simultaneous assault detection, expanding protocol coverage, and testing IEDs against sophisticated incursions on numerous devices.[15] Anomaly detection solution for Industrial Control Systems (ICS) Federated Learning (FL) methods.Will employ XAI for interpretable reasons.
explore more illness kinds, fine-tune models, and compare to other CNN models.[17] Deep learning-based technique for detecting smart meter fraud and identifying questionable AMI relay nodes.
Achieves 85.7% accuracy with a low false positive rate.
increase security capabilities, as well as address dataset imbalance.
[18] Multi-agent reinforcement learning technique for optimal labor allocation among UAVs using Q-learning.

REPLANNER.
improve training, address difficult techniques, and suggest improvements.
[19] Manufacturing embraces extensive network integration for control, diagnostics, safety, and e-manufacturing over the internet.
Network performance in control, diagnostics, and safety is evaluated and shown on a reconfigurable industrial testbed.
wireless network usage, integration in control and safety, changing industrial protocols, with Ethernet and wireless becoming prominent for cost-effectiveness and flexibility.[20] Comprehensive security architecture for Smart Healthcare Systems (SHS) based on IoT, ubiquitous computing, and machine learning.
In detecting dangerous activities, Health-Guard obtains 91% accuracy and a 90% F1 score.
Hydro power plant test to validate honeypot assess attack data.[22] Evolving IACS and SCADA systems face cybersecurity difficulties as a result of OT/IT network convergence, growing attack surfaces, and data analysis requirements.
(IADS) framework is proposed, which combines strategies to improve SCADA security monitoring.
address platform administration issues, automate machine learning, evaluate various anomaly detection techniques, and implement privacy-preserving mechanisms for data confidentiality.[23] Highlights complicated industrial automation by emphasizing safety, reliability, determinism, distributed structures, explicit timing, event-triggered computation, and enhanced security.
PLC approaches support deterministic, distributed models with an emphasis on innovation look into new programming paradigms, promote safety, validate proposed models, and encourage practical implementation.IACS solutions are characterized by their secure infrastructure, facilitating seamless information transfers and communications [22].These systems also employ smart devices, such as sensors installed on machinery, to effectively collect data.The integration of hardware, software, and communication alternatives ensures the transformation of sensor data into actionable information for optimized decision-making.

6) Programmable Automation Controllers (PACs):
PACs represent versatile automation controllers that encompass higher-level instructions for control and automation processes.Their application spans diverse sectors, including critical infrastructure and various industrial control systems (ICS) applications [25], making them valuable assets in enhancing efficiency and performance.7) Intelligent Electronic Devices (IEDs): IEDs are electronic components equipped with integrated microprocessors, enabling digital communication through various protocols like Fieldbus and real-time Ethernet.These intelligent devices play a pivotal role [13] in enabling seamless and efficient data exchange within complex industrial control systems.
V. APPLICATION OF ICS 1) Manufacturing Automation: The integration of Industrial Control Systems (ICS) has revolutionized manufacturing automation [19].This paper explores various critical aspects, including distributed control, diagnostics, and safety processes, while evaluating network performance characteristics and strategic implementation of cutting-edge network technologies within a reconfigurable factory testbed.The study also discusses emerging networking trends and challenges for future advancements.2) Energy Management: Industrial Control Systems (ICS) are vital in monitoring, controlling, and optimizing energy processes in various sectors, including power plants, renewables, and smart grids.SCADA systems play a crucial role in collecting real-time data, identifying inefficiencies, and ensuring reliable power supply, while smart grids integrated with ICS enable dynamic balancing and effective load management.The seamless integration empowers operators to optimize energy utilization and promote sustainability.3) Oil and Gas Industry: In the oil and gas industry, ICS (Industrial Control System) plays a crucial role, with SCADA (Supervisory Control and Data Acquisition) being instrumental in detecting pipeline leakages through real-time monitoring, which includes DCS and CAO-ICS Technology, minimizing losses effectively.The study conducted in Nigeria implemented [9] SCADA technology, along with DCS and CAO-ICS systems, in a specific oil and gas terminal, obtaining data through validated questionnaires and conducting z-test hypotheses.This integration of SCADA, DCS, and CAO-ICS ensures efficient pipeline monitoring, reducing leakage risks, and optimizing oil transportation, thereby enhancing overall system effectiveness and safety.4) Water Treatment and Distribution: Water treatment and distribution systems rely on efficient and reliable operation, facilitated by Industrial Control Systems (ICS) integrating components like Supervisory Control and Data Acquisition (SCADA), Distributed Control Systems (DCS), Programmable Logic Controllers (PLC), and Remote Terminal Units (RTU).These ICS components provide various advantages for water treatment and distribution: • Real-Time Monitoring: SCADA systems collect real-time data from sensors, enabling informed decisions, and optimizing efficiency, and response times.

VI. AI BASED ICS
AI in industrial control systems combines AI concepts and technology to improve and automate industrial operations, hence improving efficiency, productivity, and safety in a variety of industries.It uses artificial intelligence (AI) algorithms to evaluate data, make choices, and automate processes, resulting in enhanced performance and streamlined operations [10].
Artificial intelligence (AI) is able to forecast equipment failures by examining sensor data and maintenance logs, enabling proactive maintenance to cut downtime and increase overall equipment effectiveness.Real-time AI algorithms optimize resource allocation and control parameters in complicated industrial processes to boost productivity and lower energy use.AI-based control systems can detect abnormalities and departures from typical behavior, enabling prompt reactions to avert system failures or mishaps.By examining sensor data to satisfy predetermined standards, AI keeps an eye on and maintains product quality.Artificial intelligence (AI) permits the creation of autonomous industrial systems that can carry out operations and make decisions without the involvement of a direct human.
By anticipating demand, controlling inventory levels, and streamlining logistics and transportation routes, AI improves supply chain operations [6].AI reduces costs and encourages sustainability by optimizing energy use in industrial buildings.AI continuously scans industrial processes for safety issues and assists in evaluating and reducing risks.It improves human-machine collaboration by letting operators concentrate on making complicated decisions while AI takes care of the tedious work.To gather insights and make data-driven decisions, artificial intelligence (AI) techniques like machine learning and deep learning analyze enormous amounts of data created in industrial processes.

VII. CHALLENGES AND OPPORTUNITIES OF ICS
Industrial Control Systems (ICS) are crucial in modern industries.However, ensuring their secure and efficient operation poses challenges due to insufficient security in current infrastructures.Common countermeasures are ineffective, leaving ICS facilities vulnerable to attacks.The complex nature of ICS further complicates the detection of known and unknown threats.Despite challenges, opportunities exist to improve ICS security.Architectures show repetitive communication patterns, aiding anomaly detection and pattern recognition.Analyzing application logs using pattern mining helps detect potential anomalies.Integrating ICS with public networks introduces new cybersecurity challenges, increasing vulnerability to attacks.Researchers propose advancements in architecture, policies, system scanning, authentication, access control, encryption, and intrusion detection.ICS integration with cutting-edge technologies like IoT offers production process enhancements.Balancing opportunities with cybersecurity challenges remains critical.Continued research and development in ICS security are essential to stay ahead of evolving threats and ensure system safety and reliability.Indepth data analysis, suitable detection systems, and industry collaboration enhance ICS security.

VIII. CONCLUSION AND FUTURE WORK
The incorporation of AI in Industrial Control Systems (ICS) has opened unparalleled possibilities for industrialization, machine learning-based maintenance, and simultaneous data analysis, intensifying functioning resilience and competitiveness across industries.Predictive maintenance, AIdriven outlier identification, and AI-driven automation are among the crucial applications propelling this transformation.However, to effectively utilize the advantages of AI in ICS, executing cybersecurity measures, moral considerations, and accountable AI utilization is fundamental to ensure public faith and assurance in this technology.The limitation of the paper is not an applied AI Model.Future research could concentrate on creating different interpretable AI models and techniques based on ICS, contributing to further advancements in the field and enhancing the potential of AI in industrial control systems.Additional research could also explore the integration of AI with adaptive control strategies to enhance the adaptability and efficiency of ICS applications, making them more robust and effective in dynamic industrial environments.

Fig. 3 .
Fig. 3. SCADA System Genaral Layout 4) Remote Terminal Units (RTUs): RTUs are essential microprocessor-based electronic devices integrated into Industrial Control Systems (ICS).They play a crucial role in establishing seamless connections between various hardware components and Distributed Control Systems (DCS) or SCADA [1].Functioning as remote telemetry units, RTUs are responsible for gathering sensor data from control loops and transmitting this information to the central command of the ICS.5) Industrial Automation and Control Systems (IACS):IACS solutions are characterized by their secure infrastructure, facilitating seamless information transfers and communications[22].These systems also employ smart devices, such as sensors installed on machinery, to effectively collect data.The integration of hardware, • Process Optimization: DCS and PLCs allow precise control over processes like chemical dosing, enhancing water quality and treatment efficiency.• Remote Monitoring and Maintenance: RTUs facilitate remote monitoring and quick issue resolution, especially useful for large-scale systems.• Data Analytics and Predictive Maintenance: