Artificial Intelligence and Machine Learning for Sustainable Manufacturing: Current Trends and Future Prospects

The integration of AI in sustainable manufacturing makes it possible to achieve reductions in environmental impacts, optimize resource utilization, and reinforce circular economy principles. Through the integration of AI in sustainable manufacturing practices, new opportunities have emerged to reduce the environmental impact, maximize resource utilization, and support a circular economy [15,16]. Research has also shown potential impacts for sustainable areas like biomedical [17,18,19,20,21], water [22,23,24], nanomaterials [25,26,27,28,29,30], Composites [31,32,33], safety [34,35], etc. This section examines some of the key areas where AI is making a significant difference in sustainable manufacturing. Because of the Fourth Industrial Revolution, which is characterized by the fusion of digital, biological, and physical advancements, AI has been identified as a central technology lever in the modernization of manufacturing processes [36]. The manufacturing sector is under increasing pressure to adopt more sustainable practices due to global concerns about climate change, resource depletion, and environmental degradation. The toolkit provided by AI, which has the ability to handle large amounts of data, identify complicated patterns, and respond to input intelligently and quickly, is powerful in addressing these challenges [37]. Sustainable manufacturing, as defined by the U.S. Department of Commerce, involves “the creation of manufactured products that use processes that minimize negative environmental impacts, conserve energy and natural resources, are safe for employees, communities, and consumers and are economically sound” [38]. AI technologies are uniquely positioned to support each aspect of this definition, from optimizing energy use and resource consumption to enhancing worker safety and improving economic outcomes. The applications of AI in sustainable manufacturing span the entire product lifecycle, from design and production to use and end-of-life management [39]. Applications of AI in sustainable manufacturing include energy optimization through consumption monitoring and efficiency enhancement, equipment longevity via predictive analytics, supply chain network refinement, waste minimization protocols, and lifecycle-oriented product engineering methodologies. Each area represents a critical leverage point for improving the sustainability of manufacturing operations. By harnessing AI technologies, manufacturers can not only reduce their environmental footprint but also drive innovation, improve product quality, and enhance their competitive position in an increasingly sustainability-conscious market. Moreover, the integration of AI with other emerging technologies, such as the Internet of Things (IoT), blockchain, and advanced robotics, creates new synergies and possibilities for sustainable manufacturing. For instance, the combination of AI and IoT enables the creation of “digital twins”—virtual replicas of physical assets or processes that can be used for real-time monitoring, predictive maintenance, and optimization of resource use [40]. As we delve into each application area, it is important to note that while AI offers tremendous potential, its implementation in sustainable manufacturing is challenging. Issues such as data quality, integration with legacy systems, workforce adaptation, and ethical considerations must be carefully addressed to fully realize the benefits of AI in this domain [41]. In the following subsections, we explore each key application area in detail, examine current implementations, quantify benefits where possible, and discuss emerging trends and future prospects. By providing a comprehensive overview of AI applications in sustainable manufacturing, this section aims to offer valuable insights for researchers, practitioners, and policymakers working towards a more sustainable and efficient manufacturing sector. Table 1 presents a comprehensive analysis of the AI applications in sustainable manufacturing processes and operations. The table systematically categorizes various research initiatives that address critical challenges in manufacturing sustainability through AI methodologies. This systematic compilation provides valuable insights into the current state of AI integration into sustainable manufacturing practices and identifies promising directions for future research and implementation. 2.1. Applying AI and ML for Energy Management and Efficiency AI applications, particularly the incorporation of ML algorithms and IoT sensors, have transformed energy management into sustainable manufacturing. These advanced systems provide real-time monitoring and analysis of energy usage patterns across various manufacturing processes, equipment, and facilities. Sophisticated AI models, including deep neural networks (DNN) and RL algorithms, can forecast energy requirements, detect inefficiencies, and autonomously modify operational parameters to enhance energy efficiency. AI-powered energy management systems have demonstrated significant potential for reducing energy consumption in manufacturing plants, with research indicating reductions of greater than 10–20% [61]. According to a comprehensive review, the implementation of AI has consistently led to energy reductions exceeding 25% across various applications [62]. This substantial improvement in energy efficiency can be attributed to the AI’s ability to optimize processes, predict energy demand, and enable more dynamic and responsive energy management systems. For instance, in Chinese manufacturing, incorporating an additional unit of industrial robots per hundred workers leads to an approximately 2.5% reduction in enterprise energy intensity [63]. Labor-intensive and technology-intensive sectors show more significant reductions in energy intensity than capital-intensive sectors [64]. Digital twins allow manufacturers to model and optimize energy-consumption scenarios without interrupting ongoing operations [40]. These technologies are especially potent when combined with smart grid integration, enabling manufacturing facilities to adjust their energy usage flexibly based on grid demand and pricing fluctuations [41]. Moreover, AI algorithms can enhance production scheduling to capitalize on off-peak energy rates and renewable energy availability, thereby contributing to both cost reduction and environmental sustainability objectives [65,66]. Recent case studies from the automotive and chemical manufacturing industries have shown that AI-enabled energy management systems can significantly reduce carbon emissions while maintaining or improving production output. Figure 3 depicts a comprehensive framework for energy management in smart systems, integrating anomaly detection and diagnosis using unsupervised learning, energy demand prediction through time-series forecasting and historical data analysis, and real-time optimization leveraging sensor networks, AI algorithms, and actuators to enhance efficiency metrics. Time-series forecasting models have demonstrated significant utility in predicting the energy demand within manufacturing facilities with enhanced precision [67]. These models analyze historical energy consumption data, production schedules, and environmental variables such as meteorological conditions to forecast future energy requirements. This capability enables manufacturers to optimize production scheduling, negotiate more advantageous energy contracts, and effectively plan renewable energy integration. Research has compared six ML algorithms, including ANN, Gaussian regression (GR), and SVM, to predict the power demand and supply in smart grids. Their analysis revealed that these algorithms performed similarly in both quantitative and qualitative measures when optimized, particularly for predictable patterns such as solar PV generation and daily power consumption [68]. A recent study was conducted over a three-year period from January 2020 to January 2023, using energy consumption data collected from three institutional buildings that had implemented smart building ecosystems. The experimental methodology involved comprehensive data preprocessing, including handling of missing data and feature importance analysis, with statistical evaluation of data normality through skewness and kurtosis measurements. For clarity, we have revised the reporting of our metrics to use more appropriate precision levels: the Boosting Regressor (GBR) model achieved mean absolute percentage error (MAPE) values of 9.3%, 12.3%, and 4.0% Center of Law and Society (CLAS), Nursing and Health Innovation (NHAI), and Cronkite, respectively, while mean absolute error (MAE) values were 71.0 and 53.8 for CLAS and Cronkite. All three buildings were analyzed using three supervised machine learning algorithms: GBR, Long Short-Term Memory (LSTM), and Random Forest (RF), with consistent evaluation metrics applied across all cases to ensure a fair comparison of model performance [69]. A comprehensive analysis of DL techniques for building energy consumption prediction revealed that memory-based architectures (LSTMs) demonstrate superior performance compared to stateless networks (MLPs) in time-series analysis, even without data aggregation through CNNs. Research conducted on an operational office building, specifically the ICPE facility, indicated that Recurrent Neural Networks (RNNs) achieved a 50% reduction in Normalized Mean Absolute Error (NMAE), whereas CNNs exhibited optimal performance in datasets characterized by missing values and consistent trends [70]. A study presented a novel DL-based approach for energy prediction in machine tools, demonstrating significant performance improvements with accuracy increasing from 19.14% to 74.13% for grinding machines and from 64.89% to 85.61% for milling machines, compared to conventional handcrafted-feature methods. The innovative methodology, as shown in Figure 4, employs unsupervised learning for extracting sensitive energy consumption features from raw machinery data, coupled with supervised learning for prediction model development, thereby establishing a more generalized and efficient approach to machine tool energy modelling that requires minimal expert intervention. It is important to note that these predictions are achieved without requiring additional hardware measurements or sensor installations. The model utilizes existing building data infrastructure to generate accurate energy consumption forecasts, making it a cost-effective solution for building energy management systems [71]. 2.2. Predictive Maintenance and Equipment Optimization Using AI & ML Predictive maintenance powered by AI is revolutionizing equipment management in manufacturing, contributing significantly to sustainability efforts by reducing downtime, extending the equipment lifespan, and optimizing resource use [72]. A comprehensive study of AI-driven predictive maintenance in U.S. manufacturing revealed that increased AI integration leads to significant reductions in downtime and improved operational efficiency while highlighting the critical role of organizational culture and employee training in successful AI implementation [73]. Figure 5 provides a comprehensive process flow for AI-assisted predictive maintenance, comprising six key stages. The sequence begins with the identification of maintenance requirements, progresses through the implementation of AI tools and data analysis, advances to maintenance prediction and scheduling, and culminates in execution. This systematic approach elucidates how AI transforms conventional maintenance practices into a proactive data-driven process that enables organizations to anticipate and address equipment maintenance needs prior to the occurrence of failures. Recent reviews indicate that ANN, particularly LSTM and Bi-LSTM architectures, along with random forests, have emerged as predominant methods for industrial equipment performance monitoring and failure prediction, with applications spanning diverse sectors, including the oil and gas, automotive, and coal industries [74]. A comprehensive review of AI methods for condition monitoring and fault detection of induction motor bearings highlights their critical role in industrial environments, with particular emphasis on various AI techniques, including signal processing, data fusion, and expert systems, for early fault diagnosis and the prevention of catastrophic failures [75]. A study reported that Deep Forest and Gradient Boosting algorithms achieved an average accuracy exceeding 90% in fault-detection models [76]. A laser welder predictive maintenance model based on the LSTM-AE algorithm achieved an accuracy rate of 97.3% and detected equipment abnormalities 27 h before failure, reducing downtime by 18 h compared with existing processes [77]. Random Forest was found to be the most effective in predicting the machine conditions and failure modes for a milling machine [78]. These AI-driven solutions enable organizations to move beyond rigid calendar-based maintenance routines to more intelligent predictive approaches by continuously learning from historical data and real-time equipment performance metrics. This transformation allows manufacturers to strategically minimize unnecessary maintenance interventions while ensuring that critical equipment receives timely attention, leading to reduced production disruptions and extended equipment lifespans. A novel ML-based approach using RL algorithms has shown significant improvements in reducing the downtime and maintenance costs of parallel working machines [79]. In the aviation industry, a Deep Q-learning algorithm has been successfully employed to optimize long-term maintenance scheduling for aircraft fleets, resulting in fewer scheduled checks and increased fleet availability [80]. The knowledge-enhanced reinforcement learning (KERL) method was developed for multi-machine systems, achieving higher overall business rewards and learning to avoid failures in stochastic environments [81]. A study on a real-world use case of semiconductor front-end wafer fabrication showed that RL policies can learn competitive joint scheduling strategies for dispatching and maintenance, considering both internal and external opportunistic opportunities [82]. Some studies reported even higher improvements in overall equipment effectiveness (OEE). For instance, a case study in the automotive industry demonstrated an OEE increase of over 5% by implementing an AI-enhanced methodology for digital shift-books [83]. Another study in a manufacturing company showed an improvement in OEE from 54.09% to 60.15% (a 6.06 point increase) by implementing Total Productive Maintenance (TPM) and Single Minute Exchange of Dies (SMED) [84]. A multinational semiconductor company in the Philippines reported an increase in OEE from 68% to 87%, indicating a 19-point increase [85]. 2.3. Waste Reduction and Recycling Using AI & ML AI technologies play a significant role in promoting circular economy principles and minimizing the environmental impact of manufacturing through waste reduction and recycling efforts [22,86]. In the manufacturing sector, digital technologies, including AI, IoT, and blockchain, enhance visibility, traceability, and collaboration in supply chains, thereby contributing to the implementation of circular economy principles [87]. AI-driven innovations in waste management include optimizing waste collection routes, automating sorting processes, and enhancing recycling efficiency. ML algorithms analyze historical data to predict waste generation patterns, thereby enabling more effective resource allocation [88]. Notably, the application of AI technologies in waste management varies between developed and developing regions. In the United States, AI is utilized to optimize advanced waste management infrastructure, whereas in Africa, AI applications focus on scalable and adaptable solutions, such as crowd-sourced waste reporting and smart bins for real-time collection route optimization [88]. AI technologies are crucial enablers for transitioning to a circular economy model. They facilitate the design of sustainable products, enable new circular business models, and support the broader infrastructure necessary to scale circularity [89]. Multilayer CNNs have been the subject of research to evaluate the effectiveness of innovative approaches in smart city waste transportation [90]. Contemporary investigations have explored the application of robotic systems for categorizing recyclable waste from demolition and reconstruction activities, a process that demands precise segregation of various materials before demolition commences [91]. The adoption of robotic technology has been shown to improve the efficacy of recyclable waste management for resource utilization [92]. Within this framework, researchers have developed DL models to estimate the volume of recoverable waste materials obtainable from structures prior to their demolition [93]. The outcomes of these studies indicate the viability of accurately predicting the quantities of waste extracted from a building post-demolition based on its essential characteristics. Figure 6 presents a comprehensive roadmap for integrating AI into sustainable manufacturing and recycling processes, which is depicted as a winding road with key milestones. The foundational level illustrates the Material Flow in Manufacturing, which focuses on visualizing the movement of materials during the production process. The progression then advances to Predictive Modeling for Waste Generation, wherein AI systems analyze data to forecast waste patterns [94]. The roadmap encompasses the Design for Recyclability, utilizing AI tools to enhance product recyclability from the conception phase. The trajectory continued to include CV & Waste Sorting, demonstrating how AI systems can identify and classify waste materials. Ultimately, this culminates in the Optimization of Recycling Processes, where AI is employed to maximize the recovery efficiency. The experimental analysis of garbage classification demonstrated performance variations based on image quality has been studied. High-quality images achieved detection confidence levels approaching 1.0 for common items, whereas complex backgrounds reduced detection reliability. The implementation of GAN-based image enhancement improved the classification accuracy from 89.26% to 91.98%, with kitchen waste detection increasing from 81.82% to 87.88%. These results quantitatively validated the critical role of image quality optimization in waste classification systems. DL models, such as MobileNetV2, EfficientNet, and YOLO variants, have been successfully applied to garbage detection and classification tasks [95,96,97]. The development of region-specific waste databases has proven crucial for improving model performance. For instance, the Taiwan Recycled Waste Database (TRWD) outperformed the commonly used TrashNet dataset when training a YOLO-v3 model for detecting domestic waste specific to Taiwan [97]. This highlights the importance of tailoring waste sorting systems to the local context. Intelligent waste sorting systems that combine computer vision, DL, and robotics show great promise in automating and improving recycling processes. These systems can achieve high accuracy rates and real-time performance even on embedded devices [98]. Figure 7 illustrates a circular waste management approach that uses AI and ML. It depicts a continuous cycle of waste generation through detection using IoT technology, device interconnection, collection, data visualization, segregation, and recycling. This advanced approach facilitates efficient monitoring, automated sorting, and sustainable waste-handling practices. The literature has documented the development of an IoT-based digital bin capable of waste disposal management [99] and air pollutant forecasting in its immediate surroundings [100]. Concurrently, researchers have explored biogas prediction methodologies using DL models [101]. Several investigations have focused on predicting industrial waste generation quantities, employing digital research designs that incorporate deep data augmentation techniques [102] and DL algorithms [103]. ML algorithms have demonstrated considerable potential in enhancing the efficiency of renewable energy generation from waste recycling. Research indicates that the application of these techniques can reduce energy consumption by up to 30%, increase recycling output, and decrease greenhouse gas emissions. The 30% reduction was calculated based on the aggregate annual energy consumption data, taking into account seasonal variations and normalized for processing volume [104]. Evolutionary computation, which simulates natural selection to solve optimization problems, has proven effective in identifying high-performance reinforcement-learning policies. These approaches can automatically discover optimal representations, manage continuous action spaces, and address partial observability, in some instances surpassing the performance of traditional temporal difference methods [105]. Recent advancements have led to the development of adaptive evolutionary algorithms that are specifically tailored to multitask reinforcement-learning environments. For example, the A-MFEA-RL algorithm incorporates crossover and inheritance mechanisms to refine the exchange of genetic material, leveraging the multilayered structure of modern DL-based reinforcement learning models [106]. This approach has shown competitive success rates across simultaneously addressed tasks and promotes knowledge exchange among synergistic tasks. The integration of AI, IoT, and blockchain enhances visibility, traceability, and collaboration in manufacturing supply chains, optimizes resource utilization, and reduces waste generation [87]. These technologies play a crucial role in facilitating the transition towards a circular economy by improving the efficiency and effectiveness of recycling and remanufacturing processes [107]. The Chemical Production and Waste Material Recovery Framework (CP&WMRF) used an innovative system that outperforms traditional AI-ML approaches in sustainable chemical production and waste management, achieving superior results across multiple metrics, including resource efficiency (97.5%) and waste recovery (98.7%), surpassing AI-ML’s 91.5%. The framework integrates circular economy principles with advanced technologies to transform industrial practices, showcasing how environmentally conscious solutions can be both economically viable and significantly more effective than conventional methods [108]. AI technologies are making significant contributions to waste reduction and recycling efforts in manufacturing, promoting circular economy principles and minimizing environmental impacts. AI has transformed product design and lifecycle management through four principal innovations. Generative design tools driven by AI optimize sustainability criteria, resulting in designs that are 25% lighter and 30% more recyclable, as evidenced in automotive components. The implementation of smart product platforming (SPP) powered by AI and generative AI, in conjunction with big data analytics and machine learning, has been demonstrated to facilitate personalized product design and manufacturing suitable for environmentally sustainable products in the circular economy [109]. This research indicates that SPP and big data analytics are fundamental determinants for manufacturing environmentally sustainable products, ultimately promoting circular economy applications. The integration of ML with Life Cycle Assessment (LCA) has demonstrated its effectiveness in predicting environmental impacts and improving design iterations [110]. The application of ML in material selection has led to substantial reductions in the environmental footprint, with some studies reporting more than a 20% reduction in almost all environmental impact indicators [111]. ML algorithms have proven to be valuable tools for enhancing material selection processes, particularly for evaluating environmental impact and recyclability. Their integration with LCA methodologies has led to significant reductions in the environmental footprint, often exceeding 20% [111]. Figure 8 depicts an AI-enhanced LCA framework for additive manufacturing, illustrating the progression from design requirements through CAD, STL, and G-code generation to the final 3D-printed components. This framework integrates design and process parameters into LCA/LCIA models, employing ML to analyze environmental impacts (climate change, toxicity, ozone depletion, and land use) and to elucidate the correlations between manufacturing parameters and environmental outcomes [110]. AI also plays a crucial role in low-carbon manufacturing by optimizing production and industrial structures, helping achieve carbon emission reduction targets, and sustainable resource development [112]. The integration of AI and IoT, forming Artificial Intelligence of Things (AIoT), offers numerous benefits to the manufacturing industry, including improved efficiency, waste reduction, and enhanced safety measures [113]. The integration of AI across these areas enables manufacturers to create more sustainable products while optimizing their environmental performance throughout their lifespans. In the near future, Artificial superintelligence (ASI) may potentially surpass human cognitive capabilities across all domains, although it remains a theoretical concept, as strong AI has not yet been achieved for a more sustainable future.