Autonomous control systems using artificial intelligence, machine learning, and digital twins in Industry 4.0

The fourth industrial revolution, also referred to as Industry 4.0, is defined by the incorporation of intelligent digital technology into industrial production processes [1]. Therefore, smart production facilities with autonomous optimization capabilities are provided in order to increase the performance of part production [2]. Generally, traditional industrial control systems are based on rule-based logical control principles, including deterministic and classical feedback control approaches [3]. Therefore, these control principles are effective in the conditions of linear and stable control systems. However, in the conditions of nonlinear and uncertain control systems, traditional control principles are not effective and are associated with challenges [4]. Autonomous control systems based on artificial intelligence (AI), machine learning (ML), and digital twin (DT) simulation are the main control systems associated with the fourth industrial revolution [5]. The incorporation of digital technology into traditional control systems results in the creation of intelligent control systems. DT technology offers a dynamic environment for real-time monitoring and scenario simulation, while AI and ML offer optimization and decision-making capabilities [6]. AI and ML enhance these systems by learning from real-time operational data and autonomously adapting control strategies for improving process performance. The autonomous control systems used in Industry 4.0 are based on continuous sensing, data-driven models, and the use of feedback loops [7]. The real-time feedback from the system updates the twin model, which in turn enables the continuous learning process of the autonomous controller system. This process enables the convergence of the learning process to be greatly accelerated [8].

The paradigm of the Internet of Things has been further reinforced with the development of DT simulation tools that can create virtual models of physical assets to monitor, predict, and control them in real time [9]. In addition, AI-based DT simulation tools can be used to enable autonomous decision-making, prediction, and optimization in smart manufacturing systems [10]. As a result, the developed autonomous control systems based on AI-driven DT simulation are able to perceive the environment, learn from operational data, predict the future, and adapt controlling processes without human intervention in order to ensure accurate controls during the production process [11].

AI and ML are at the forefront of making autonomous industrial control possible through learning-based modeling, predictive analytics, adaptive control strategies, anomaly detection, and real-time decision intelligence [12]. The paradigm for modeling and identifying systems changes from offline to continuous, intelligent, and adaptive processes through the integration of AI, ML, and DT simulation [13]. The use of these technologies can lead to improved prediction accuracy through autonomous decision-making, continuous model adaption, and real-time estimates of system parameters [14]. ML algorithms can offer a learning-driven approach, which can remove the need to have explicit models of the systems. Deep learning techniques further enhance adaptive control by enabling the approximation of high-dimensional control policies and value functions [15]. Moreover, deep reinforcement learning supports real-time decision-making in large-scale and multi-variable industrial processes where conventional control design becomes intractable. DTs facilitate continuous validation and refinement of predictive models in terms of improving accuracy and robustness of prediction algorithms within the decision-making process [16].

AI and ML can provide data-driven modeling, prediction, and decision-making capabilities during the smart production process of Industry 4.0 [17]. AI- and ML-powered autonomous controllers can optimize production processes, manage distributed manufacturing plants, and change machine variables in order to enhance the productivity of part manufacturing [18]. Additionally, ML can help robots recognize patterns, adapt to changes, and instantly acquire the best robot handling techniques. Industrial processes could be autonomously controlled with increased efficiency and dependability when AI and ML are coupled with DT and cyber-physical systems. This provides the technological basis for the intelligent manufacturing systems that Industry 4.0 envisions. Some of the applications of AI- and ML-based autonomous industrial control systems include fault detection and predictive maintenance, adaptive scheduling and resource allocation, process optimization, and quality control [19]. Therefore, autonomous control systems can proactively adjust machine settings, production schedules, and maintenance activities before faults occur or become critical failures. As a result, the integration of AI and ML models within the DT framework enables adaptive updating of diagnostic models, increasing the precision and resilience of real-time defect diagnosis during the autonomous control system of smart manufacturing [20].

The integration of AI, ML, and DT technologies enables and facilitates a paradigm shift from reactive to proactive and prescriptive industrial control. The advanced algorithm of AI, ML, and DTs in autonomous control systems has the capability to learn from virtual interactions, solve multi-objective functions, and enhance control strategies before being used in real-world systems [6-21]. The integration of these technologies can be considered as complete package which can enhance efficiency, quality, added value and safety in industrial environments. Therefore, predictive analytics and self-learning control systems can provide real-time process optimization, energy-efficient processing, and dynamic quality control in order to identify anomalies and errors and increase system stability for different conditions. Therefore, autonomous industrial control systems can provide opportunities for highly informed, multi-objective decision-making which can successfully resolve challenges related to productivity, quality, energy, and operational safety using data-driven decision-making and virtual control integration [22]. This is a key requirement for realizing the vision for smart, flexible, and self-optimizing factories that capture the very essence of Industry 4.0.

The concept of autonomous control systems in the context of Industry 4.0 needs to be regarded as a multilayer cyber-physical system that comprises several components such as physical processes, sensing and communication network systems, cloud computing systems, data analytics systems, intelligent decision-making systems, and DT systems [23]. Each layer interacts with another one in real time, resulting in complex dynamics and nonlinear interactions among machines, controllers, software, and human operators [24]. Such systems' complexity results not only from physical production processes but also from the integration of diverse sources of data, distribution of control systems, and coordination of multi-agent decision-making processes. AI, ML, and DT technologies play a crucial part in achieving synchronization, adaptiveness, prediction, and coordination between various layers [25].

The originality of the proposed research lies in its comprehensive and integrative analysis of autonomous industrial control systems, achieved through the exploration of the synergistic potential of AI-, ML-, and DT-based approaches within the framework of Industry 4.0. Unlike previously published studies that examine these technologies in isolation, the originality of the present research lies in establishing the interrelationships among intelligent system modeling, predictive analytics, adaptive self-learning control, real-time anomaly detection, multi-objective optimization, quality control, and energy management within an integrated autonomous control framework. Furthermore, the research provides a comprehensive perspective on the potential of DT-based approaches for the development of fully autonomous, self-optimizing, and resilient industrial processes. While this provides a clear understanding of the current state of the art, it also highlights the major research gaps and identifies future directions for developing scalable, reliable, and sustainable autonomous industrial ecosystems. This review presents the role that AI, ML, and DT simulation can play together in the development of the concept of autonomous control systems, as envisioned by the Industry 4.0 paradigm. All the sections presented together point towards the fact that the technological basis for the development of the next-generation smart and self-optimizing industrial systems lies with the convergence of data-driven intelligence, learning, and DT simulation.

Data-driven intelligence, adaptive learning, and high-fidelity virtual modeling form the technological foundation for next-generation smart and self-optimizing industrial operations. Moreover, this study also discusses challenges related to interoperability, scalability, data fidelity, and security, which remain open research issues. As a result, a novel study in the analysis of autonomous control systems enabled by AI, ML and DT simulation in Industry 4 is presented in order to enhance productivity in the process of part production.

System modeling and identification are critical prerequisites for the achievement of a trustworthy and autonomous industrial control system [26]. It is challenging to accurately predict and model industrial processes due to the nonlinearity and complexity of production procedures. However, traditional methods of model-based control are dependent on physical modeling and linearization, which may be unsuitable for modeling the dynamics of contemporary industry. The application of AI, ML, and DT simulation allows for intelligent modeling and identification of industrial production through data-driven and adaptive modeling of cyber-physical systems [27]. ML methods such as neural network and Gaussian process modeling allow for learning system dynamics through the use of sensor measurements [28]. The ability to learn system dynamics without requiring physical modeling equations, to capture non-linear and time-varying behaviors, and to adaptively manage production settings are key advantages of ML algorithms in industrial system modeling [29]. Moreover, DTs enable virtual testing and scenario simulations without interfering with real-world operations by using sensing data to maintain consistent synchronization with the real systems [30]. This characteristic permits proactive modification of controls and avoidance of risks when modeling the process of production procedures. This data-driven modeling is essential for autonomous controllers that must operate reliably despite process uncertainties during the modeling procedure [31]. Several key approaches have been developed to achieve accurate and adaptive modeling as follows:

(i) Data-driven modeling in intelligent industrial systems: The application of AI and ML techniques is useful in the implementation of data-driven modeling, in which the model learns the dynamics of the system [32]. This model has the potential to learn the nonlinear relationships, disturbances, and uncertainties of the process, which are difficult to model [33].

(ii) Reinforcement learning for system identification and adaptive control: Reinforcement learning can be viewed as an extension of system identification, in which continuous interaction between the agent and the process is maintained [34]. This interaction as exploration enables the agent to simultaneously learn the system dynamics and develop an appropriate control policy. This approach is particularly suitable when an accurate mathematical model of the process is unavailable or difficult to derive [35-36].

(iii) DT simulation in a high-fidelity modeling process: DT simulation offers a virtual synchronized model of physical industrial assets, acting as a sophisticated environment for intelligent system identification and modeling. The DT model combines physical models and data-driven intelligent models like AI/ML to attain accurate hybrid models [37].

(iv) Hybrid physics-informed AI models for industrial identification: The technology models intricate industrial processes by fusing the great predictive capacity of AI with the dependability and interpretability of physical principles. Because the system combines the flexible pattern recognition of data-driven ML algorithms with the high-level reasoning of first-principles mechanistic models, it can be utilized to accurately model and simulate complex industrial processes. These physics-informed AI models have found a lot of applications during production simulation as they integrate first-principles understanding with ML algorithms. Industrial systems may perform reliable system identification even with sparse, noisy, or restricted sensor data due to this synergy [38-39].

The application of AI, ML and DTs for intelligent system modeling of industrial production procedures is shown in Figure 1.

Consequently, intelligent system modeling and identification form the cognitive foundation of autonomous industrial control systems in Industry 4.0 [11]. AI, ML, and DT technologies enable controllers to operate autonomously, perform optimally, and maintain resilience throughout autonomous control processes by guaranteeing that such industrial processes are represented by self-evolving, high-fidelity models [40].

The predictive analytics are an essential foundational component for the development of the proactive and autonomous decision-making process for the control systems based on Industry 4.0 [41]. Predictive analytics could make it easier to make proactive, self-optimizing decisions about increasing the industrial control systems' part manufacturing process' productivity [42]. By using AI, ML, and DT modeling technologies, predictive analytics would be able to change the conventional reactive decision-making process into a proactive approach for forecasting future system behavior during the process of part production [43]. The creation of intelligent, self-adaptive, and self-sustaining industrial control systems could be significantly facilitated by these predictive analytics. Therefore, based on the past data production systems, predictive analytics are able to use ML to forecast system states, failures, and performance indicators [44]. To enhance the process of reducing product defects and predictive maintenance during manufacturing operations, predictive analytics might be used for quality prediction of component fabrication [45]. The integration of predictive analytics into autonomous industrial control offers several benefits, including improved reliability, enhanced operational efficiency, reduced downtime, and optimized resource utilization [46]. Applications of predictive analytics for proactive decision-making processes can be presented as follows:

(i) Role of predictive analytics in autonomous industrial control: In conventional industrial automation systems, control actions are normally initiated after system deviations or faults have occurred, thus leading to delays, inefficiencies, and increased system risks [47]. This limitation has been overcome by the predictive analytics approach, whose main advantage lies in its ability to forecast system performance in the future based on past and current system data [48]. Through continuous monitoring and learning, AI and ML systems can recognize trends, forecast system deviations, and suggest appropriate system control actions before system failures occur. Therefore, this approach enhances autonomous systems with increased system uptimes, quality, and reliability [49]. Autonomous systems can now function proactively rather than reacting in response due to the application of predictive analytics as a decision support mechanism.

(ii) ML techniques for predictive modeling: Predictive models for industrial processes could be constructed using a variety of ML methods. Regression models, artificial neural networks, and gradient boosting machines are examples of supervised learning methods that can be used to create prediction models [50]. They can be applied to predict the values of the process variables, equipment condition parameters, and manufacturing outputs. The time series analysis algorithms like recurrent neural networks and long short-term memory networks can be applied to make efficient predictions of sensor data streams.

(iii) Integration with DTs for scenario-based prediction: The predictive analysis tool is improved with the assistance of the DT simulation. It is feasible to establish a virtual environment in order to predict and analyze future condition of production procedures [16]. The use of ML-based prediction models in conjunction with DT algorithms ensures real-time synchronization between virtual and physical assets [37-51]. Therefore, without the actual effects of variables in real environment, the system can virtually predict the consequences of different potential situations using the DT simulation techniques [52].

Figure 2 shows the predictive analytics for proactive decision-making processes within the part production process using AI, ML and DT algorithms.

Predictive models facilitate prompt and informed decision-making that is in line with the goals of intelligent and self-optimizing factories by allowing early identification of abnormalities and potential disruptions. However, several challenges remain that can affect the effectiveness and portability of predictive models. Predictive analytics requires three essential ingredients: the availability of good quality data, good feature extraction capabilities, and the constant evolution of models according to changing circumstances [53]. Since edge-cloud systems combine these three components by considering computational limitations and other operational needs in industrial contexts, they provide an attractive option for predictive analytics. Predictive analytics facilitates proactive and adaptive decision-making, which is a crucial component of Industry 4.0 autonomous control systems.

Adaptive and self-learning control policies are one of the most vital components of autonomous industrial control systems within the Industry 4.0 era [54]. Adaptive control systems continuously modify and update their control strategies based on real-time data, the environment, and the objectives, whereas traditional industrial control systems with set parameters can only depend on the model and tuning control parameters [55]. It is feasible to create intelligent control policies that can continuously learn, optimize, and fine-tune themselves for the optimal outcomes while the conditions in the industrial environment change continuously by merging the ideas of AI, ML, and DT simulations [11-56].

The integration of AI and ML into DTs is particularly valuable for simulating interactions among multiple agents before implementing any changes in the real system. This allows industrial control systems to continuously improve performance without human intervention [57]. One of the most significant contributions of AI/ML is the ability to automatically learn optimal control strategies from interaction with the environment [58]. Online policy adaptation to changing operating conditions and autonomous tuning of process parameters to obtain optimal control of nonlinear and multi-variable systems are some advantages of the adaptive and self-learning control systems using AI, ML and DTs [59]. As a result, adaptive and self-learning control policies offer several advantages in autonomous industrial systems, including improved adaptability, enhanced robustness, reduced need for manual tuning, and optimized system performance [60]. These benefits collectively contribute to increased productivity, higher product quality, and improved energy efficiency in industrial processes. These benefits are achieved through several key mechanisms and concepts, which are outlined as follows:

(i) Concept of adaptive and self-learning control: Adaptive control, as a concept, is concerned with a control system that has the ability to fine-tune its own control parameters in response to changes in the underlying dynamics or the appearance of disturbances [61]. This leads to the concept of self-learning control, in which the control system learns optimal control policies directly through its own experiences and interactions with the environment [62]. These ideas have important implications for the dynamic behavior of modern industries, characterized by nonlinearity, varying parameters, and interactions between the sub-systems. As a result, self-organizing and adaptive control systems have significant implications for autonomous industrial processes that enable self-healing, self-stabilization, and performance enhancement of controlling processes [63].

(ii) ML approaches for adaptive control: To create adaptive self-learning control strategies, ML offers a wide range of methods which can obtain the optimal condition from nonlinear effects, uncertainties, and disruptions during control processes [64]. Supervised learning can be applied to adjust controller parameters according to predicted responses during control procedures [65]. In addition, reinforcement learning has now become the preferred method for learning optimal control policies in complicated industrial systems. In order to optimize cumulative conditions such as productivity, efficiency, and quality, reinforcement learning controllers interact with the environment, monitor the system's status, and refine their actions [66].

(iii) Role of DT simulation in policy learning and adaptation: DT simulation is required to be implanted in order to provide safe and effective creation of adaptive and self-learning control rules [67]. This is due to the reason that DT simulation offers a very realistic virtual representation of the actual system, allowing controllers to experiment with different learning-based strategies in a risk-free setting [68]. Therefore, the controllers can digitally connect with DTs of the real-world system to explore various tactics and choose the optimum policies without the need for real-world system experiments and expenditures [67].

(iv) Multi-agent and distributed self-learning control: New smart factories constructed from a web of connected machines, robots, and production modules can work together as a system [69]. Therefore, adaptive, self-learning control can be extended to multi-agent systems, where a number of intelligent agents make decisions to achieve broad production objectives [70]. Distributed units can learn cooperative strategies and exchange knowledge in order to analyze and modify their decisions in real time by utilizing ML coordination techniques. Distributed self-learning control is well suited to flexible manufacturing systems with frequent configuration changes, enabling distributed decision-making while maintaining system-wide coordination [71].

The application of AI, ML and DT algorithms for adaptive and self-learning control systems within the part production process is presented in Figure 3.

However, several challenges should be addressed in order to fully realize the benefits of this approach. In particular, online learning must be guaranteed to remain stable and safe, especially in safety-critical industrial processes. At the same time, real-time learning algorithms can be computationally expensive, and in many cases, large amounts of data are required, which can be a limiting factor in real-world deployment [72]. Thus, understanding the learned control policies and verifying procedures can be considered as an advanced topic of research work in order to provide integrated adaptive control systems. As a result, these approaches serve as key enablers of self-optimizing and resilient cyber-physical production systems, driving the transition from conventional automation to fully autonomous and intelligent control envisioned in Industry 4.0.

In order to provide autonomous and accurate control systems in Industry 4.0, real-time anomaly detection and problem localization are essential [73-74]. Real-time detection of unexpected and aberrant patterns is considered important due to control challenges of sophisticated, connected, and data-intensive systems [75]. The combination of AI, ML, and DT simulations can provide advanced autonomous control systems in order to deliver robust capabilities in the areas of real-time monitoring, fault detection, and smart correction during the controlling process [20]. ML-based anomaly detection systems can work by continuously sweeping through data streams in real time in order to obtain the pattern of failures [76]. There are prominent ML approaches which can be applied in real-time fault detection systems such as deep learning autoencoders, statistical learning and clustering, and Bayesian fault detection models [77]. These methods have the potential to provide early fault detection, which opens the door to independent corrective action. Some advantages of the utilization of the AI system and the DT system in the real-time anomaly detection and fault diagnosis include the reduction of unexpected downtime, improved reliability, safety, efficiency, and the maintenance of product quality [78]. AI-driven DTs can also support predictive maintenance and adaptive scheduling for the fault diagnosis procedures. Real-time anomaly detection and fault diagnosis within autonomous control systems can be presented as follows:

(i) Importance of real-time fault awareness in autonomous control: Maintaining steady and seamless operation is crucial for autonomous industrial control systems, as is making sure that plant behavior is sufficiently understood [79]. Faults can occur in sensors, actuators, or cyber-physical components, among other parts of the system. If they are not quickly identified and corrected, they can cause catastrophic failure or performance deterioration [80]. Simple threshold-based alerts are frequently insufficient in dynamic and nonlinear industrial contexts as defects may manifest as abnormal patterns or subtle deviations rather than sudden shifts [81]. Anomaly detection can address this limitation by providing continuous monitoring using real-time sensor and data analysis in order to identify abnormal system behavior at an early stage. By integrating these detectors into closed-loop control systems, corrective actions can be immediately implemented in order to prevent faults from escalating or propagating in terms of maintaining the stability of the autonomous system [82].

(ii) ML techniques for anomaly detection: This domain offers an effective approach for detecting anomalies and recognizing faults occurring during the controlling process of systems [83]. Using historical data, supervised learning provides the foundation for identifying recognized defect types and predicting their recurrence rate [84]. However, due to the lack of historical error data, fault information is typically insufficient. By identifying patterns of typical behavior, unsupervised and semi-supervised learning approaches are helpful in this situation for defect identification [85]. Clustering, principal component analysis, and autoencoders are efficient techniques in modeling normal behaviors for anomaly detection procedures [86]. Moreover, deep learning algorithms, like autoencoders and recurrent neural networks, are able to capture temporal dependencies and recognize anomalies in data. Thus, the process can provide a means of immediately detecting abnormal situations in the system functioning process. Probabilistic models such as Bayesian methods are quite popular in the field of fault detection due to their effectiveness in identifying the underlying cause of problems by examining the correlations between system variables [87]. As a result, they are used for detecting, localizing and classifying the fault in terms of anomaly detection of systems using ML algorithms [88].

(iii) DT-assisted fault diagnosis and prognostics: DT simulation can improve real-time anomaly detection and fault diagnosis through the application of a synchronized virtual model [89]. As a result, areas of defect can be revealed by comparing the physical system's actual condition with the DT model's predicted condition. This provides a better fault diagnosis compared to the usual analysis of the data [90]. Additionally, to more accurately estimate the equipment's remaining usable life, DT simulations can be applied to predict the fault's evolution in terms of providing reliable fault diagnosis and prognostics procedures [91].

(iv) Closed-loop fault-tolerant autonomous control: Real-time anomaly detection and fault diagnosis are the essential elements of fault-tolerant control, allowing AI-powered decision-making modules to respond effectively when a system situation turns incorrect [92]. This guarantees that the systems can continue to function with the appropriate degree of fault tolerance without requiring human involvement. Anomaly detection can also be utilized for cooperative fault management in distributed systems, including multi-agent systems, which can be used to lessen the impact of errors which can cause major disruption in the system [93].

The real-time anomaly detection and fault diagnosis process in the part production process using DT simulation is shown in Figure 4.

However, in order to effectively implement the system in real-world industrial settings, several challenges must still be addressed. The industrial data streams usually have a high number of dimensions and may contain noisy or incomplete data or even non-stationary data, which might affect the accuracy of the models developed. Secondly, real-time fault detection requires highly efficient processing systems that must be based on edge computing and/or cloud computing [74-94].

In summary, real-time anomaly detection and fault diagnosis can significantly enhance the autonomy and reliability of industrial control systems by providing continuous insight into system state and enabling intelligent fault handling [95]. Anomalies could be detected through the use of AI, ML, and dynamic simulation, while the root cause could also be established [96]. As a result, appropriate corrections could be implemented in order to provide smart control systems using real-time anomaly detection and fault diagnosis. This technology is important in the development of resilient and reliable autonomous control systems, which are fundamental concepts of Industry 4.0.

Industrial control systems should be able to simultaneously optimize across several critical metrics in order to enhance quality and minimize cycle times, energy consumption, waste and carbon emissions [97]. The autonomous control systems have to simultaneously manage multiple objectives in order to remain productively and sustainably effective during part production [98]. High-speed algorithms combined with DTs enable systems to evaluate and implement control actions in order to adapt to real-time production circumstances. Multi-objective real-time optimization of control systems is essential, particularly for autonomous industrial applications operating within the complex and dynamic environment of Industry 4.0 [99].

When AI and ML are utilized to solve multi-objective optimization problems, it is possible to simultaneously consider multiple performance criteria and constraints [100]. Autonomous control systems can choose Pareto-optimal solutions, which are the best compromises between competing objectives, by defining control issues as multi-objective decision problems [101]. As a result, the combination of AI, ML, and DT simulations could be highly efficient optimization systems which can assess and choose the best action in real time by using virtual simulation [102]. The key components and operational mechanisms of this integrated optimization framework are as follows:

(i) Role of virtual environments and DTs: The idea of a DT allows the real world to be imported into a kind of sandbox environment. By creating an accurate virtual replica of the real-world process, optimal control actions can be evaluated without directly affecting the physical system [103]. The DT continuously monitors the current system state and predicts its response to various control actions in real time [104]. In order to identify the best method of action for achieving the desired performance objectives, the procedure enables autonomous controllers to simulate and assess several tactics inside the virtual model [105]. As a result, the optimal action can be applied to the real world with much fewer uncertainties and much more confidence in order to enhance the performance of industrial control systems [106].

(ii) AI and ML techniques for multi-objective optimization: The AI and ML paradigms are essential components of real-time multi-objective optimization for industrial control systems. The genetic algorithm and particle swarm optimization methods have been widely used in order to determine the Pareto-optimal control policies in nonlinear and constrained industrial settings [107]. In addition, the reinforcement learning technique can be utilized to design the control policies that optimize the cumulative rewards, where multiple goals are incorporated via the weighted reward functions [108]. Thus, it could be advantageous to utilize the deep reinforcement learning approach, which might be very successful in a high-dimensional, nonlinear industrial domain where constructing the optimization model is quite difficult [109]. The reinforcement learning algorithms can further improve the control rules and pursue near-optimal, real-time control inside the DT framework [4-110]. Moreover, the surrogate modeling method, such as neural network-based models and Gaussian processes, can be effectively utilized to capture the complicated behavior of the industrial process [111]. As a result, computational costs associated with the online optimization procedure can be substantially minimized.

(iii) Closed-loop optimization in DT-based control frameworks: Closed-loop control can be effectively implemented by integrating multi-objective optimization methods within a DT-based framework [112]. Real-time data is sent into the DT before being processed by the model. The output from the model goes through various control strategies that are compared against several objectives. Next, the optimization tool using AI and ML algorithms is used in order to obtain the most optimal solution [113]. Thus, control strategies can be analyzed and refined based on feedback regarding the effectiveness of the control model during industrial processes [114].

(iv) Applications in smart manufacturing and industrial systems: The scope of multi-objective real-time optimization within a virtual setup includes a broad range of Industry 4.0 applications. For example, within the context of intelligent manufacturing systems, real-time optimization can be used to optimize the machine variables to enhance productivity, surface finish, and tool life [16]. Real-time optimization could be used to optimize processes to reduce energy consumption and waste materials without lowering productivity in sectors with high energy consumption [115]. In the context of autonomous robotics, real-time optimization could be employed to optimize the robots' movements in order to achieve a balance between safety requirements, speed, and accuracy during working conditions [116].

Multi-objective real-time optimization of control actions in virtual environments using DT simulation procedures is presented in Figure 5.

However, despite several advantages, some challenges accompany the implementation of multi-objective real-time optimization in a virtual setting [117]. The reaction time could be delayed by the computational load, which is quite high in big systems. The DT models must be very accurate, as they have a direct influence on the outcome of the optimization process, which relies on the reliability of the simulations. The formulation of objective functions and the assignment of their corresponding weights constitute a complex process.

In summary, real-time multi-objective optimization within DT environments is driving the development of more autonomous and intelligent industrial control systems. Using AI, ML, and DT simulations, an autonomous system would be able to make the necessary compromises between conflicting goals, predict outcomes of possible decisions, and take measures to avoid any problems even before they occur. This self-tuning efficiency and robustness are at the heart of the concept of autonomous cyber-physical systems in Industry 4.0.

To monitor production in real time, identifying defects and adjusting process parameters autonomously to maintain high standards of production, quality control and adaptive process tuning are implemented in the Industry 4.0 era [16-118]. The processes are managed by autonomous control systems that integrate AI, ML, and real-time sensor data in order to monitor and analyze a variety of process parameters, including temperature, pressure, vibrations, feed rates, and material properties [79]. The primary objective is to ensure consistent quality and optimal performance despite changing or uncertain operating conditions. These systems leverage AI, ML, DT simulations and the Internet of Things to monitor production in real time, identifying defects and adjusting process parameters autonomously in terms of quality control and enhancement of part production [119]. The integrated system has enabled intelligent quality control systems that continuously monitor production processes, predict potential quality degradation, and adjust process parameters in real time [120].

By enabling systems to self-correct in response to material or environmental variability, adaptive process tuning ensures process stability without the need for human intervention to improve defect detection accuracy over time [121]. AI and DT simulation can predict where errors are likely to develop before the failures occur for enabling preventive fixes by examining process data patterns [122]. Therefore, a proactive self-optimizing control loop has replaced a reactive quality control strategy based on a post-production quality control mindset in the control process [123]. This closed-loop system allows for ongoing learning and modification of control policies, which can consider new materials, production requirements, and constantly shifting operational restrictions during production procedures [124]. As a result, smart and continuous quality control and adaptive process tuning by autonomous control systems integrated by AI, ML and DT simulation can be achieved within the Industry 4.0 era. The realization of these capabilities relies on several key technological components and functional mechanisms, which are discussed as follows:

(i) Intelligent quality monitoring in smart manufacturing: Autonomous control systems use AI and ML to search through large, complicated datasets for patterns related to product quality [118-125]. Therefore, the quality control systems become sensitive to even the slightest changes that might indicate a problem as they understand the intricate correlations between process variables and the quality of the outcomes [126]. In contrast to statistical process control, which necessitates the establishment of limitations, quality monitoring systems make use of AI's capacity to modify the process in response to changes.

(ii) Predictive quality estimation using ML: ML plays a vital role in predictive quality control, which forecasts the quality of the product before the batch is complete [127]. Supervised learning algorithms, which include neural networks, support vector machines, and ensemble regressors, can be used for predictive quality control [128]. They can be trained on the real-time sensor data, along with the historical data, to forecast the quality indicators [129]. This helps the control systems detect the possible quality problems early, allowing for adjusting the process settings accordingly [4]. Time series algorithms, which include recurrent networks and temporal convolutional networks, can be used for predictive quality control, as they can easily monitor the changes in the process variables and the impact on the quality [130].

(iii) DT-enabled virtual quality assessment: The DT is a powerful tool for quality control because it provides a virtual space for experimenting with how changes in the process can impact the quality of the product [16]. The DT can continuously monitor the real process in real time, accurately reflecting how changes in control parameter settings influence the resulting quality attributes [131]. The DT provides a virtual space for autonomous systems to experiment with a wide range of tuning options without ever stopping or changing the real process. Continuous updates enable discrepancies between predicted and actual process behaviors to serve as early indicators of potential issues [132]. Based on these insights, the DT can recommend optimal parameter adjustments to maintain process quality [133].

(iv) Adaptive process tuning through AI-driven control policies: Adaptive process tuning can be applied in order to provide automatic adjustments for controlling processes and ensure that performance and quality are consistently high when facing changes to the environment and materials [134]. AI-based autonomous controllers use predictive models and reinforcement learning and optimization techniques to select the optimal actions in order to optimize quality, productivity, and efficiency [135]. For example, in a three-dimensional printing operation, the controller can dynamically adjust the deposition of the three-dimensional printer, depending on predicted outcomes of the quality of the finish or the accuracy of the final piece. Similarly, in chemical plants, the controllers adjust temperatures, flow rates, and reaction times to maintain a consistent mix of products and output rates. To develop a self-optimizing factory floor that can maintain high quality and performance during production operations, these modifications are performed via adaptive process tuning using AI-driven control [136].

(v) Closed-loop integration of quality control and process optimization: The integration of quality control and adaptive process tuning forms a closed-loop autonomous control framework [137]. Thus, real-time sensor data are analyzed using AI and ML models in order to estimate current and future quality levels. Then, the DT simulates alternative control actions and predicts their effects on quality metrics. Based on these evaluations, the autonomous controller selects optimal parameter adjustments that ensure compliance with quality requirements while optimizing overall process performance [16-138].

Thus, the adoption of autonomous quality control and adaptive process tuning promises many advantages, including product consistency, reduced defect rates, improved resource efficiency, and greater flexibility in operation. Autonomous systems help ensure customer satisfaction through the elimination of quality problems before they occur. Quality control and adaptive process tuning by autonomous control systems using AI, ML and DT algorithms is shown in Figure 6.

However, there are several challenges that have to be overcome for successful implementation of the method [139]. The real-time performance evaluation of the method should be implemented by handling large quantities of data from sensors which need a large volume of computational work [140]. Moreover, the quality prediction is dependent on the availability of large quantities of high-quality data for training the dataset. The interpretability of the method is another important factor that needs to be considered [141]. In addition, the decisions made by the machine for the quality prediction should be accurate and reliable in order to be considered during performance evaluation of processes [142].

In summary, the quality control and adaptive parameterization of the processes, with the power of AI, ML, and DT simulations, can considerably increase the intelligence and autonomy level of the control systems. With the quality control and adaptive parameterization, the autonomous control systems can sustain quality, adapt to changes in the processes, and optimize to improve quality. These are important to the development of self-optimizing, robust, and resilient manufacturing, which is the essence of the Industry 4.0 intelligent production environments.

Industrial sectors are among the largest consumers of global energy, with many of their processes operating under varying load conditions and multiple operational constraints [143]. Thus, energy efficiency has emerged as a major goal for Industry 4.0, considering factors such as increased cost of energy, environmental issues, and maximization of value extracted from resources [144]. Traditional energy management approaches, which are largely static and rule-based methods, are often insufficient for effectively monitoring and responding to the dynamic nature of real-time system changes such as fluctuations in load, machine status, and production scheduling. This has resulted in inefficient utilization of resources, unexpected peak demands, and increased operational costs [145]. Smart and autonomous control systems, enabled by AI, ML, and DTs, are transforming industrial operations and advancing the vision of an energy-aware, more sustainable manufacturing environment [146]. Smart control systems track the dynamic distribution of electrical energy across machines, production lines, and support systems, and perform real-time adjustments to maintain an optimized and efficient consumption profile [147]. Through advanced analysis of energy consumption data, these control systems detect inefficiencies, forecast future demand, and implement optimal strategies to reduce energy usage [148]. The realization of these functions relies on four key technological components as follows:

(i) AI and ML for energy consumption modeling and prediction: The technologies of AI and ML are significant in providing prediction and modeling tools for energy consumption in complex industrial processes [149]. For instance, supervised learning can be used to make predictions and models of how certain variables, schedules, and energy consumption are related in a process [150]. Additionally, unsupervised learning can be used to identify any patterns in energy consumption, which may demand some action to correct the situation [151]. Prediction tools enable the estimation of future energy consumption under different scenarios, supporting proactive decision-making for energy management [152]. The time series prediction tools, such as recurrent neural networks and long short-term memory networks, are very efficient in modeling energy consumption over a given period of time. Through constant refreshment of their predictions using new data streams, ML-based controllers can always predict energy surges and adjust the process variables to ensure maximum energy efficiency [153].

(ii) DT-based simulation for energy optimization: A DT provides a real-time virtual playground to test and fine-tune the way energy is used. This is accomplished by precisely simulating the actual industrial setup, allowing the assessment of various control rules, production tactics, and machine configurations without interfering with actual operations [154]. This means that autonomous control systems can perform numerous scenario-based energy optimizations and determine the best energy-saving actions before anything is implemented in the real world [155]. For example, the DT can balance the speed at which production is pushed versus the energy used by the production process. Thus, it can provide capabilities for the control system to find the optimal balance between energy savings and production rates [154-156]. As a result, continuous synchronization between the digital and physical domains can provide accurate simulation of real-world systems in order to optimize energy usage.

(iii) Adaptive energy management through autonomous control policies: Adaptive, self-learning techniques are used by autonomous control systems to dynamically optimize energy use at every stage of operation, from individual machines to the production line and the entire factory [157]. Control systems that employ reinforcement learning, for example, try out the industrial environment, utilizing feedback from real-world savings and improved performance to find the best strategies for energy saving. Through iterative learning and continuous data integration, these systems progressively develop strategies that balance energy efficiency and productivity, while adhering to all operational constraints [158]. In order to prevent the demand for energy from exceeding the available resources, load balancing and peak shaving which refer to scheduling energy-intensive procedures during off-peak hours are also included in adaptive energy management [159]. In addition, the overall energy savings in a smart factory are mostly dependent on the regulation of auxiliary systems like cooling, heating, ventilation, and compressed air [160]. The systems can be managed through autonomous control of energy usage in order to enhance the productivity of systems.

(iv) Integration with multi-objective optimization frameworks: Industrial operations aiming to meet energy-efficiency targets must balance multiple priorities, including productivity, product quality, equipment reliability, and environmental performance [158]. The autonomous control systems, with the ability to consult with DTs and the assistance of AI and ML in a multi-objective optimization framework, can evaluate these conflicting objectives in real time and determine the Pareto optimal set of options [154-161]. The autonomous control system can select the appropriate control actions to meet the desired productivity targets, as well as reduce the use of energy and the impact on the environment, in an Industry 4.0 scenario [162].

The advantages of using AI, ML, and DT systems for energy-efficient systems include reduced operating costs and carbon footprint, longer equipment life, and improved productivity and sustainability of production [163]. However, the hurdles include the challenge of modeling the complex and untidy energy flows among different industrial assets. Moreover, key challenges include limited computing power for model optimization, inconsistent data across sources, difficulties integrating new technologies with existing energy management systems, and a lack of standardized metrics for evaluating energy performance [164]. The challenges can be addressed with robust data acquisition infrastructure, computing power, and system architecture.

Thus, the approach towards attaining energy efficiency within industries is currently receiving an accelerated boost through the integration of AI, ML, and DT technology. In such technological developments, the bar for sustainability and intelligence in autonomous control systems within Industry 4.0 is raised since such systems are capable of forecasting and optimizing their energy usage. As a result of this development, there have been some notable improvements on the side of energy-efficient autonomous control systems.

AI, ML, and DT-based autonomous control systems can be regarded as sophisticated cyber-physical engineering systems consisting of several interdependent levels of operations [165]. This type of system combines sensory devices, communication systems, intelligent data processing, DTs, autonomous control, and feedback into an industrial control system. Interactions between physical entities and virtual worlds facilitate predictive analysis, adaptive learning, and intelligent decision-making in smart control systems [166]. This kind of system architecture allows continuous monitoring, optimization, and autonomous coordination within large industrial facilities. Besides, decentralized decision-making and multi-agent collaboration make the operation of the system more complex due to nonlinear relationships between subsystems. Thus, it becomes important to understand these multilevel systems in order to develop the Industry 4.0 control systems [167]. The AI–ML–DT-enabled autonomous control systems can be presented as follows:

(i) Multi-layer cyber-physical architecture: The autonomous control system can be defined as a multi-layered cyber-physical architecture that comprises the sensing layer, the communication layer, the computing layer, the decision-making layer, and the actuation layer [168]. The interaction between these different layers helps in real-time monitoring, learning, and autonomous functioning in the context of Industry 4.0.

(ii) Integration of AI, ML, and DT frameworks: The interaction of AI and ML algorithms with DTs can be applied for predictive analysis, experimentation, and optimization of adaptive control systems [67]. It is necessary that both be continuously synchronized for autonomous decision-making in order to increase the performance of autonomous control systems.

(iii) Distributed and multi-agent coordination mechanisms: Large industrial processes are constructed from distributed machines, robots, edge devices, and intelligent agents working together within the manufacturing network [169]. Hence, the architecture must consider issues such as coordination methods, decentralized decision-making, latency in communications, and optimization through cooperation among autonomous agents.

(iv) Closed-loop feedback and real-time data processing: The architecture must incorporate closed-loop feedback loops where sensor data is collected, analyzed, and acted upon in real time [170]. Such an approach allows the system to detect any anomalies, learn from itself, optimize its operations, and provide resilient autonomous control.

In conclusion, AI–ML–DT-based autonomous control systems constitute a new generation of intelligent cyber-physical industrial structures for Industry 4.0. The interplay between the sensing, analytics, virtual modeling, and adaptive control layers allows for optimizing the operations of the manufacturing environment in real time and making them autonomous [12]. However, some issues with regard to integration, scalability, interoperability, and coordination of the agents distributed across the network still pose challenges that require further investigation.

Despite major achievements in the field of autonomous control systems based on the application of AI, ML, and DT simulations, several challenges remain regarding the full implementation of these systems [171-172]. These challenges are associated with incorporating intelligent data models into industrial control systems during the modeling process of the production process as follows:

(i) Data availability, quality, and heterogeneity: There exists a requirement for collecting large amounts of data from different sources using sensors, machines, and manufacturing processes, enabled by the capabilities of AI and ML algorithms for intelligent and autonomous control systems [11-118]. However, collected data can be incomplete, inconsistent, and noisy owing to faulty sensors, delayed communication, or outdated systems in several industries [173]. Moreover, different data representations on different machines make it difficult to integrate them into one model. This also applies to insufficiently labeled data for identifying unique faults and anomalies, which prevents the creation of automated control systems [174].

(ii) Model accuracy, generalization, and interpretability: It is challenging to develop AI and ML solutions that are always accurate and widely applicable across different operational processes [175]. It is also difficult due to the fact that industrial systems generally exhibit nonlinearity and dynamics, along with other disturbances [176]. Data-driven models may not be able to retain their high-performance levels, mainly because the data they work on may change over time. This means that model drift, which is generally influenced by changing production needs, aging of equipment, and changing environmental factors, may cause a decline in the performance levels of the models [177-178]. In addition, advanced ML techniques, including deep learning, have very poor interpretability characteristics [179]. This is another issue, as trust, validation, and regulatory issues can be challenging, especially in safety-critical applications.

(iii) Real-time computational and scalability constraints: The essence of autonomous industrial control lies in real-time data analysis, prediction, and control adaptation [180]. However, the application of AI and ML methods, especially deep learning and optimization algorithms, requires substantial computational capabilities [181]. Therefore, low-latency inference becomes one of the primary challenges since using such approaches in practice for industrial processes often relies on edge-computing infrastructure, including memory and computing power restrictions [182]. Additionally, scalability represents another critical issue due to the high number of connected devices within a smart industry. The vast quantity of information provided by various devices and sensors necessitates a well-coordinated interaction between edge and cloud computing infrastructures [183].

(iv) Integration with legacy systems and interoperability issues: A significant proportion of industrial plants continue to depend on outdated control infrastructures and siloed communication architectures. The systems were not designed to support contemporary digital solutions such as real-time analytics, advanced automation, and seamless data exchange [184]. This mismatch creates substantial barriers to interoperability, limiting the effective integration of intelligent, connected technologies [185]. Integrating advanced AI and ML technologies along with DT solutions into existing industrial control infrastructures such as programmable logic controllers and distributed control systems introduces substantial technical and organizational challenges [186]. Data exchange and decision-making are hampered by the incompatibility of various hardware, software, and communication platforms [187]. In addition, the lack of common standards in DT solutions and AI-based industrial control systems is the main challenge in the widespread adoption of DT solutions [185].

(v) Model fidelity and DT synchronization: The efficiency of autonomous control using DT technology relies on the performance of DT models [188]. Since it is challenging to develop accurate models that accurately depict the physics involved in such complex physical systems, developing a decent DT model can be a challenging issue [189]. When discrepancies exist between the physical system and its DT representation, both predictive accuracy and control decisions can become unreliable [190]. Achieving real-time alignment between the two requires robust communication infrastructure, precise sensor calibration, and advanced data fusion methods capable of integrating heterogeneous, high-frequency data streams [189].

(vi) Safety, reliability, and robustness concerns: The autonomy control systems should be dependable and accurate in spite of all the curve balls that get thrown around on the factory floor, including the unknowns, the hazards, and the critical issues [191]. However, the main challenge lies in ensuring that decisions made by autonomous AI systems are consistently safe, reliable, and verifiable, particularly when deployed in high-risk environments such as heavy machinery operations, hazardous chemical processing, and mission-critical manufacturing workflows [192]. Autonomous system safety is indeed an arduous endeavor since erroneous predictions, disturbances, and unreliable data can negatively impact system performance. In order to make sure that the autonomy control system is safe, there should be an assurance of formal verification and validation [192-193].

(vii) Cybersecurity and data privacy risks: The advent of Industry 4.0 will lead to an increase in connected autonomous control systems. This will also lead to an increase in cybersecurity risks such as data breaches, harmful intrusions, and tampering of control signals [194]. The efficient flow of data is an essential part of ML and AI algorithms. Therefore, any breach of data quality or incorporation of old data can make the predictions inaccurate, leading to incorrect or dangerous outcomes [195]. The secure management of critical data and the secure interconnectivity of systems are major challenges. The challenges can be addressed by ensuring proper encryption and authentication [196].

(viii) System complexity as nonlinear interactions among subsystems: Another major challenge is the increasing complexity of the system due to non-linear interactions between subsystems, heterogeneous data environments in industry, distributed decision-making systems, and collaboration between various intelligent agents working within Industry 4.0 [197]. The modern Industry 4.0 environment is characterized by the presence of interconnected machines, robots, edge-cloud systems, the Industrial Internet of Things, and DTs that work simultaneously in the uncertain and time-varying environment [198]. Such an environment produces heterogeneous, high-dimensional, and multi-source data flows that need robust mechanisms for data fusion and synchronization. The complexity of autonomous control systems for industry should be addressed not only from the standpoint of non-linear dynamics of processes but also from the point of view of interaction between heterogeneous cyber-physical subsystems, distributed decision-making systems and cooperation between intelligent agents [199]. Furthermore, new issues with communication delays, consensus, scalability, and cooperative optimization among autonomous agents are brought up by distributed control systems and multi-agent cooperation [200]. Thus, future autonomous control systems should incorporate more advanced AI, ML, and DT techniques.

Besides technical barriers, implementation costs, lack of qualified workforce, and technological averseness are also organizational barriers to the implementation of AI- and DT-based autonomous control systems [201]. The design, implementation, and maintenance of intelligent control systems demand multidisciplinary expertise from the fields of control engineering, data science, and industrial informatics, which is not easily available in many manufacturing enterprises [202]. In addition, clearly defined performance criteria and demonstrated long-term operational benefits are essential to justify high-level industrial transformation enabled by autonomous control systems.

The advancement of fully autonomous and intelligent industrial control systems through the aid of AI, ML, and DT simulations is faced with various challenges due to their interrelation process [12-60]. Data, model, and computational scalability, system integration, system safety, and cybersecurity are among the key challenges. These are several challenges that need to be overcome by standardizing the process, ensuring the explainability and trustworthiness of AI, enabling scalability in edge-to-cloud platforms and synchronizing the DTs [203]. All these hurdles need to be overcome in order to achieve an effective autonomous Industry 4.0 control system.

The research process began by modeling and identifying intelligent systems; therefore, the demonstration of the significance of AI and ML ensures a precise and clear understanding of a given industrial nonlinear process, as depicted by data. Since these models are constantly evolving when a DT is involved, a clear and reliable understanding of a process, which is a crucial factor in informing autonomous decision-making processes, should be guaranteed. Therefore, predictive analytics enables the prediction of future events and process conditions within industrial systems. The adaptive and self-learning control policies are essential components of ensuring autonomy in an industrial process. The control policies are essential in ensuring robust control despite uncertainties and disturbances in the process. Real-time anomaly detection and fault diagnosis techniques are also essential in ensuring health awareness. The review also highlights the optimization of control actions in real time within the DT environment, with the potential of using DTs to support the safe testing of conflicting objectives such as productivity, energy, and quality. This is important as it helps decision-makers make rational decisions in complex industrial processes.

The review further highlights the capability of autonomous control systems to support quality control and continuously adjust operating parameters to maintain product quality and optimal process performance. The ability to provide energy-efficient performance within an industrial setup is another advantage associated with the use of autonomous control with the help of AI, ML, and DTs. The review reveals that with the understanding of energy use, the prediction of the demand for energy, as well as the enhancement of the operation of things, it is possible to create an environmentally friendly process of production. Other barriers for the use of AI, ML, and DTs include data quality issues, lack of knowledge of the behavior of models, computational power issues, difficulties in integration with existing technologies, cybersecurity challenges, and the need to establish trust in AI-driven decision-making.

As a consequence, a new paradigm must emerge in control systems that shifts away from the current trend in automated systems to intelligent, autonomous, and adaptable systems. This paradigm can be made possible through the integration of capabilities provided by AI, ML, and DT simulations. These three technologies can provide support in implementing concepts associated with Industry 4.0, such as predictive, robust, energy-efficient, and quality-focused control systems. In this regard, efforts should focus on developing new learning paradigms, standardizing DTs, and creating secure and low-latency edge-cloud infrastructures to bridge the gap between simulations and reality.

Future research directions for autonomous control systems based on the use of AI, ML, and DT simulations in Industry 4.0 include robustness and scalability of those systems for practical application. Autonomous control systems could be developed based on explainable and robust AI models by overcoming the existing challenges. In order to improve the transparency of multi-objective decision-making processes, future research is anticipated to concentrate on the creation of scalable architectures for edge-cloud optimization, federated learning-based distributed optimization techniques, and transparent explanation techniques for AI-based decision-making processes. In addition, the use of hybrid optimization techniques that combine the use of physical insights with ML techniques is also expected to enhance the robustness and generalization capabilities associated with the decision-making processes.

As the development of Industry 4.0 continues towards fully connected and interoperable cyber-physical ecosystems, future architectures for autonomous control could facilitate smooth and synchronous communication and coordination among various types of machines, assembly lines, robots, cloud-edge systems, and DTs functioning in several factories and within complex supply chains. In this regard, distributed intelligence, cooperative multi-agent systems, and interoperability standards are crucial for providing consistent and synchronous decision-making in large-scale industrial networks. Additionally, future research could explore the dynamic coupling issues of interdependent subsystems, in which case disturbances, delays, and local control strategies could be transferred across the whole production ecosystem. Thus, advanced knowledge and research will be required in decentralized control, federated learning, edge-cloud cooperation, and resilient coordination.

Key research directions include:

• Hybrid physics–ML modeling: Integrate first-principles models with data-driven learning to improve generalization, interpretability, and robustness in varying industrial regimes.

• Simulation-to-reality transfer and domain adaptation: Develop reliable methods to bridge the gap between DT simulations and real plant behavior under distribution shifts.

• Trustworthy and explainable autonomous control: Advance interpretable AI and formal verification methods to build confidence in autonomous decision-making, especially for safety-critical industrial processes.

• Edge-native and resource-aware learning: Design lightweight, latency-aware ML models that operate efficiently on edge devices for real-time closed-loop control.

• Multi-objective optimization under uncertainty: Formulate scalable controllers that balance productivity, quality, energy, and safety while accounting for stochastic disturbances.

• Federated and privacy-preserving industrial learning: Enable collaborative model training across distributed plants without exposing sensitive operational data.

• Cybersecure AI-enabled control architectures: Investigate resilient control strategies that detect, withstand, and recover from adversarial attacks and data poisoning.

• Continual and lifelong learning for evolving processes: Create adaptive policies that update online without catastrophic forgetting as equipment ages or processes drift.

• Standardized interoperability for DTs: Establish common data models, application programming interfaces, and ontologies to ensure seamless integration across heterogeneous Industry 4.0 platforms.

• Energy-aware and sustainable control strategies: Creating AI-enabled controllers that explicitly consider carbon footprint, energy efficiency, and sustainability metrics alongside traditional performance objectives.

In summary, advancing autonomous control systems within Industry 4.0 requires recognizing the future potential of synergistically integrating intelligent modeling, real-time analytics, adaptive learning, and DT infrastructures. Once the challenges are addressed and the benefits of the new advancements in AI, ML, and simulation-based techniques are realized, next-generation autonomous control systems will have the potential for self-optimization and sustainability.