A Data-Centric Framework for Implementing Artificial Intelligence in Smart Manufacturing

Abstract

The manufacturing segment is undergoing a rapid transformation as manufacturers integrate artificial intelligence (AI) and machine learning (ML). These technologies increasingly rely on data-driven architectures, which enable manufacturers to manage large volumes of data from machines, sensors, and other sources. As a result, they optimize operations, increase productivity, and reduce costs. This paper examines the role of AI in manufacturing through the lens of data-driven architecture. It focuses on the key components, challenges, and opportunities involved in implementing these systems. The paper explores various data types and architecture models that support AI-driven manufacturing, with an emphasis on real-time analytics. It highlights key use cases in manufacturing, including predictive maintenance, quality control, and supply chain optimization, and identifies the essential components required to implement AI successfully in smart manufacturing. The paper emphasizes the critical importance of data governance, security, and scalability in developing resilient and future-proof AI systems. Finally, it reviews a data-centric framework with essential components for manufacturers aiming to leverage these technologies to drive sustained growth and innovation.

1. Introduction

Over the past several decades, the manufacturing sector has led technological adoption, evolving from basic automation and industrial robotics to advanced digital production systems, such as IoT-enabled machinery and AI-driven quality control [1,2]. These innovations have transformed production processes, resulting in greater efficiency, enhanced product quality, and significant cost savings [3,4,5]. However, the next significant leap forward is now being driven by the integration of AI, ML, and big data technologies [6]. With the rise of the Internet of Things (IoT), manufacturers are collecting unprecedented amounts of real-time operational data through connected systems, sensors, and devices [7,8]. This data, often referred to as industrial big data, has the potential to fundamentally change manufacturing operations by enabling data-driven decision-making, optimizing workflows, and improving overall efficiency [9,10]. However, despite the promise, many manufacturers still struggle to fully harness the value of this data [11].

Traditional methods of data analytics often fall short in making sense of the vast and complex information generated across modern production lines. As a result, manufacturers often find themselves with underutilized data that could drive better decisions and process improvements [12,13]. In this context, AI and ML technologies offer a promising solution [14,15]. By applying advanced analytics and machine learning algorithms, manufacturers have demonstrated improved fault detection, predictive maintenance, and real-time process optimization, leading to measurable gains in efficiency and product quality [5]. For example, predictive maintenance models reduce unplanned downtime by anticipating equipment failures before they occur [16]. Similarly, AI-driven quality control systems have been shown to identify defects more accurately and faster than traditional inspection methods [17].

At the heart of AI-driven manufacturing systems lies data-driven architecture [18,19,20]. These advanced infrastructures facilitate the collection, storage, processing, and analysis of large-scale data streams from diverse sources such as IoT devices, production metrics, and enterprise systems [21,22,23]. Data-driven architectures provide a robust foundation for integrating different data sources, enabling the seamless flow of information across the manufacturing ecosystem [24]. By leveraging AI within these frameworks, manufacturers can achieve a range of benefits, including predictive maintenance, optimized production schedules, enhanced quality control, and improved supply chain management [25,26,27].

However, there are significant challenges associated with leveraging diverse data types in manufacturing environments, one of the most notable being data heterogeneity [28,29,30]. Manufacturing data is often varied in nature, with different formats, structures, and levels of granularity [31,32]. For instance, sensor data tends to be structured and numerical, while image-based data from machine vision systems is unstructured. Each type of data requires distinct methods for preprocessing and analysis. For example, visual data used for quality control can have significant variability due to factors such as lighting, resolution, and orientation, making it difficult for AI models to generalize across environments [33,34,35]. Additionally, time-series data in manufacturing environments introduces its own set of challenges, including the handling of noisy signals, missing data, and anomalies that may signal impending failures [10].

The growing volume of data further complicates these challenges. As the number of connected devices in smart factories increases, so does the need for high-speed, scalable data infrastructure [36,37,38]. Traditional data storage solutions may be inadequate for managing the influx of data generated by IoT sensors operating at high frequencies [39]. Real-time analytics and AI models demand low-latency processing and advanced edge computing capabilities to deliver timely insights that drive operational improvements [40,41]. Ensuring the quality of data is another critical issue. Inconsistent or incomplete data due to faulty sensors, communication failures, or integration problems can undermine the effectiveness of AI models [42,43,44]. Poor-quality data leads to inaccurate predictions, making robust data governance and preprocessing protocols essential [45].

Additionally, many manufacturers still rely on older equipment or proprietary systems that were not designed to communicate with newer data platforms. Integrating these systems into data-driven infrastructures requires considerable effort in data mapping, standardization, and ensuring interoperability between old and new technologies [46].

On top of all the listed challenges, data security and privacy are major concerns in the digital transformation of manufacturing [47]. With the increasing reliance on cloud services and third-party analytics platforms, manufacturers must ensure that sensitive data such as intellectual property, production secrets, and proprietary designs remains protected. Data breaches could lead to severe consequences, including theft of intellectual property, operational disruptions, and reputational damage. Ensuring robust cybersecurity measures, including encryption, secure communication channels, and access control systems, is crucial to safeguarding AI-driven manufacturing environments [48].

Given these evolving opportunities and challenges, this paper presents a review on the role of data-driven architectures in enabling AI integration within manufacturing. This study describes the crucial role of data-driven architectures in enabling the successful integration of AI into manufacturing environments. It presents a detailed analysis of the hypothesis of massive data generation in factories. The paper also explores the challenges and opportunities of managing diverse data modalities, such as time-series, visual, and sensor data, within AI-powered systems. Additionally, it digs deeper into the technical aspects of various architectures, computing framework, and AI learning methods, emphasizing their importance in supporting the collection, storage, and analysis of large-scale industrial data. By addressing issues related to data quality, heterogeneity, and system integration, the paper outlines strategies to overcome the barriers to effectively applying AI in manufacturing.

2. Sectors in Manufacturing

The manufacturing industry spans diverse sectors, including automotive, electronics, pharmaceuticals, food and beverage, textiles, aerospace, heavy equipment, plastics, steel, chemicals, furniture, 3D printing, energy equipment, and consumer goods as shown in Figure 1. Each sector already leverages or has the scope to leverage AI to enhance production, improve quality, streamline supply chains, and boost efficiency [49].

In automotive manufacturing, AI is used for autonomous driving, predictive maintenance, and supply chain optimization, while electronics manufacturing uses AI for quality control, precision in semiconductor fabrication, and the automation of assembly lines [50,51]. In pharmaceuticals, AI accelerates drug discovery, optimizes production processes, and ensures high-quality standards [52,53]. The food and beverage sector benefits from AI-driven predictive maintenance, supply chain optimization, and process automation [54,55].

AI also plays a key role in textiles and apparel, improving production efficiency, design, and labor automation [56,57]. In aerospace and defense, AI enhances predictive maintenance, production optimization, and design processes. The heavy equipment industry uses AI for performance improvement, predictive maintenance, and autonomous machinery [50]. Similarly, AI optimizes processes in plastics, rubber, steel, and metal manufacturing, improving efficiency, quality, and production schedules [58].

In chemicals manufacturing, AI models chemical reactions and optimizes processes to enhance efficiency [59]. The furniture industry applies AI for design, demand prediction, and production automation [60]. Also, 3D printing and energy equipment manufacturing incorporate AI for design optimization, real-time monitoring, and predictive maintenance [61].

Finally, in the consumer goods and warehousing/logistics sectors, AI plays a key role in demand forecasting, supply chain optimization, and task automation, all of which lead to increased efficiency and cost reduction [62]. These are just a few examples of how AI is transforming these industries, providing businesses with a competitive advantage through improved decision-making, automation, and predictive maintenance. However, there are numerous other use cases that AI can enable.

3. The Big Challenge of Massive Data in Manufacturing

In the manufacturing sector, the rapid advancement of technologies such as IoT, automation, and smart sensors has led to an explosion of data [63,64]. This “overabundant data” is generated at every stage of production, from machine performance metrics to supply chain logistics and quality control [65]. While this data holds immense potential for improving efficiency, reducing costs, and optimizing operations, it often becomes overwhelming to manage and analyze [66]. Manufacturers face the challenge of not only storing and processing massive amounts of data but also extracting actionable insights from it [67]. Without the right tools and strategies, this vast amount of data can lead to information overload, which can hinder decision-making instead of improving it [68,69]. However, when leveraged properly through advanced analytics, machine learning, and AI, this data can provide a competitive edge, helping companies improve production quality, prevent downtime, and streamline operations [70,71].

We propose a hypothesis regarding the amount of data generated in a car manufacturing factory. An example of this is shown in Figure 2. Following the approach of Mudgal et al. [72], we extend their specific example where two high-definition cameras are used to monitor the welding process for each vehicle. Considering the camera to be capturing 30 frames per second (FPS) on 1080p resolution, each camera generates approximately 5 MB of data per second. Assuming each welding operation on a car may last about 10 min (600 s), each camera will produce 3 GB of data per car. With 100 cars manufactured daily, each camera generates 300 GB of data per day. With two cameras, the total data produced is approximately 600 GB daily. Over 30 operational days in a month, this amounts to 18,000 GB (or 18 TB). Annually, this results in around 219,000 GB (or 219 TB) of data generated by just these two cameras in the factory [73,74].

However, this is only a small part of the process. Other operations, such as riveting, painting, and assembly, also require monitoring and produce significant amounts of data [74]. These operations involve additional cameras, sensors, and equipment that continuously record data, often in high-definition or even 3D formats, to ensure quality control and safety [75]. As more advanced technologies like robotics and AI are integrated into the production line, data from sensors monitoring pressure, temperature, and movement will further increase. This highlights how data production in a manufacturing environment can easily scale to several terabytes per day, posing significant challenges in managing, analyzing, and extracting actionable insights from such vast volumes of information [73].

In a modern car manufacturing factory, the volume of data generated is staggering, not only from cameras monitoring welding and other operations but also from various sensors embedded throughout the production line [76,77]. For example, welding guns, which are crucial for the welding process, produce about 1.5 MB of data per second during operation. With 10 welding guns in use and 100 cars produced daily, this adds up to approximately 900 GB of data per day just from the welding guns [74].

In addition, there are sensors on robotic arms that handle tasks such as lifting, assembling parts, and performing precision movements. These robots generate data about their positions, speeds, and operational status, typically around 0.5 MB per second per robot. Assuming 20 robots are active in the factory, each working 10 h a day, the total data generated by these robots is significant. For instance, if each robot generates 0.5 MB per second and operates for 36,000 s (10 h), it produces 18 GB of data daily. With 20 robots working in parallel, this brings the total to 360 GB per day from robotic arms alone [78].

Additionally, environmental sensors, such as those monitoring temperature, humidity, and air quality, are essential for maintaining optimal production conditions and worker safety. While each environmental sensor may generate a smaller amount of data—around 0.1 MB per second—the cumulative data from numerous sensors across the factory can still be substantial. For example, with 50 environmental sensors continuously collecting data (0.1 MB per second over 24 h), the total data produced by these sensors would amount to 432 GB per day [79].

Considering the described data volume and to sum up, the data generated by just 2 cameras, 10 welding guns, 20 robots, and 50 environmental sensors, the factory produces about 2.3 TB of data daily. Over the course of a month, this amounts to approximately 68.76 TB, and annually, the factory generates around 836.58 TB of data as shown in Figure 3. However, this is a simplified estimate. In reality, factories are much larger, with hundreds or even thousands of cameras, welding guns, robots, and sensors. As a result, the actual data production could be at least 100 times higher than our hypothesis. This highlights the significant challenge that manufacturing plants face in storing, managing, and processing the enormous amounts of data they generate challenges that will only grow as technologies continue to evolve [80].

The overwhelming volume of data produced in a manufacturing environment, such as from cameras monitoring welding, riveting, painting, and other operations, presents several significant challenges. First, data storage becomes a critical issue. With terabytes of data generated daily, ensuring there is enough storage capacity to safely house this information without slowing down operations can be costly and complex. This leads to the second challenge, data management. Organizing and categorizing such large volumes of data in a way that makes it easy to retrieve and analyze is no small feat. Without proper systems, data can become siloed or disorganized, making it difficult to track and extract meaningful insights [81].

Another major hurdle is data processing. Traditional data processing methods often cannot keep up with the scale and complexity of this information. Manufacturing systems generate data not just from cameras but from a variety of sensors and machines, and analyzing this data in real time or near real time is essential for maintaining optimal production and quality standards. However, the sheer speed at which data is generated makes it hard to process and act upon quickly enough [82].

Additionally, data analysis itself presents challenges. With so much data coming from different sources and in various formats, identifying relevant patterns, trends, and anomalies can be overwhelming. Employing advanced machine learning or AI solutions is often necessary, but these require specialized expertise and computational resources [83]. Moreover, data quality is another concern ensuring that the data collected is accurate and reliable is crucial for making correct decisions [69,84]. Poor-quality or inconsistent data can lead to erroneous conclusions, affecting product quality or operational efficiency.

Finally, security and privacy become more pressing. Protecting sensitive information and ensuring that data is not vulnerable to breaches are critical, particularly in manufacturing where intellectual property and trade secrets are at stake. The larger the data footprint, the greater the risk of cyberattacks, and the more stringent the measures needed to secure the data.

These challenges, including storage, management, processing, and security, highlight the complexities manufacturers face in effectively utilizing the vast amounts of data generated daily. To address these issues, there is a need for a unified framework that leverages AI-powered analytics platforms, enabling manufacturers to streamline operations and unlock the full value of their data.

4. Data Modalities in Manufacturing

Manufacturing processes generate a wide variety of data types, each offering valuable insights that can be harnessed by AI and ML algorithms [21,85]. In data-driven manufacturing, various types of data, known as data modalities, are collected from different sources across the manufacturing ecosystem. Each modality presents unique challenges and opportunities in terms of collection, processing, and analysis. As manufacturing environments become more digitized, the diversity of data types, ranging from structured sensor data to unstructured image data, requires advanced methods of integration and analysis to extract meaningful insights. Below, we explore some of the primary data modalities in manufacturing, the challenges they pose, and how AI can leverage them to optimize operations.

4.1. Sensor Data (Time-Series Data)

Sensor data, particularly in the form of time-series data, plays a vital role in manufacturing by providing continuous, high-frequency information about machine health and production processes [86]. Embedded sensors monitor critical parameters such as temperature, pressure, vibration, speed, humidity, and fluid levels, offering real-time insights into equipment performance. However, working with sensor data comes with several challenges. One of the primary issues is noise and signal interference, caused by environmental factors, sensor calibration problems, or electrical interference. This noise can obscure meaningful patterns, making it difficult to extract accurate insights. Another challenge is missing or incomplete data, which can occur when sensors malfunction or data transmission is interrupted. AI systems need to be capable of handling these gaps, either by inferring missing values or utilizing other methods to maintain prediction accuracy. Additionally, time-series data can be subject to outliers or spikes that either reflect real operational problems or are caused by sensor errors [87]. It is crucial to identify and manage these outliers to prevent incorrect predictions [88].

Despite these challenges, AI applications can greatly enhance the utility of sensor data in manufacturing. Predictive maintenance, for example, leverages AI models to analyze time-series data and predict potential machine failures before they occur, enabling proactive maintenance scheduling and minimizing unplanned downtime. Anomaly detection is another key AI application, where machine learning algorithms, particularly unsupervised models, detect unusual patterns in sensor data [89]. This helps identify early signs of equipment malfunction or production inefficiencies, allowing for quick corrective actions. AI can also be used for process optimization by analyzing sensor data in real time to adjust machine settings and production parameters, improving efficiency and reducing waste. Overall, AI-powered applications like predictive maintenance, anomaly detection, and process optimization are essential for transforming raw sensor data into actionable insights that drive manufacturing improvements [90].

4.2. Image and Video Data (Vision Data)

Manufacturers are increasingly relying on computer vision systems to monitor product quality, inspect assembly lines, and detect defects in real time. Vision data, typically in the form of images and video, provides valuable insights into the physical characteristics of products and components during production. Cameras and machine vision systems capture visual data that can be analyzed to assess product quality, machine alignment, and operational efficiency. However, working with vision data comes with several challenges. Variability in lighting, camera angles, and changes in the appearance of objects on the production line can significantly affect image quality. These variations introduce inconsistencies, making it difficult for AI models to accurately interpret the data. Additionally, vision data is high dimensional and high resolution, meaning it requires significant computational resources to process. Recognizing small defects or complex features demands advanced AI techniques, such as deep learning, further complicating the process. Moreover, in supervised learning models, images often need to be manually labeled to indicate defects or specific features of interest, which can be both time-consuming and costly.

Despite these challenges, AI applications in vision data have proven invaluable in manufacturing. One of the primary uses of AI-powered vision systems is defect detection. These systems can automatically inspect products for defects, such as scratches, dents, or misalignments, ensuring that only high-quality products make it to consumers. Computer vision is also crucial for assembly line monitoring, as it tracks the movement of parts and materials along the production line, detecting any deviations from the expected assembly process that could indicate issues [91]. Additionally, AI-driven vision systems, when combined with robotic arms or drones, can perform autonomous inspections of hard-to-reach areas or complex components, enhancing efficiency and reducing the need for manual labor. Recent studies have shown that vision-based AI systems can outperform human inspectors in both speed and consistency. For instance, AI models have achieved defect detection accuracies exceeding 95%, compared to 80–90% for manual inspection, while processing thousands of items per hour with minimal fatigue or variation [92]. These results highlight the potential for AI to supplement or enhance traditional quality control methods. In summary, computer vision systems, despite their challenges, are transforming the manufacturing process by automating quality control, improving efficiency, and reducing the risk of defects.

4.3. Textual and Log Data

Manufacturing systems generate large volumes of textual data, often in the form of logs from machines, enterprise systems, and software applications. These logs can provide valuable insights into machine status, operator actions, production line conditions, and system alerts. Although textual data may initially appear unstructured, it holds crucial information that AI models can leverage to enhance manufacturing operations [93]. However, there are several challenges when working with this type of data. One major hurdle is its unstructured nature. Log files, for instance, can vary widely in format, terminology, and data types across different systems, making it difficult to extract actionable insights without the use of preprocessing and natural language processing (NLP) techniques [94]. Additionally, the sheer volume of logs generated by multiple systems, coupled with the complexity of interpreting them, can overwhelm traditional data processing methods. As a result, AI systems must be capable of efficiently handling large datasets and extracting meaningful information from them [95].

Despite these challenges, AI applications for textual data in manufacturing are highly valuable. One such application is sentiment analysis, where NLP techniques can be used to analyze maintenance logs, operator notes, or system feedback to detect patterns of dissatisfaction, recurring issues, or areas in need of improvement [96,97]. Additionally, AI models can aid in root cause analysis by processing logs to identify recurring problems and pinpoint the underlying causes of system failures, which helps in recommending corrective actions [98,99]. Another key application is process monitoring, where AI systems can continuously analyze system alerts and maintenance reports in real time, automatically flagging potential issues before they negatively impact production [100,101]. In summary, while handling textual data from manufacturing systems presents challenges, AI models have the potential to unlock valuable insights, improve decision-making, and prevent costly downtime [102].

4.4. Audio Data

Although less commonly used than other data modalities, audio data from sound sensors or microphones is gaining traction in manufacturing for monitoring machine health and detecting anomalies. Acoustic sensors capture the sounds produced by operating machines, and AI models can be trained to recognize irregular sounds that indicate potential malfunctions, such as bearing failure, motor imbalances, or fluid flow issues. However, working with audio data presents several challenges. One significant issue is background noise, which is prevalent in manufacturing environments and can obscure subtle sounds associated with machine malfunctions. To address this, advanced noise filtering and sound classification techniques are required to isolate relevant acoustic signals. Another challenge is signal variability; the acoustic signature of a machine can change based on factors such as its condition, environment, and load. This variability makes it difficult to develop a universal model that can be applied across different machines and production lines [103].

Despite these challenges, audio data has several valuable AI applications [104]. For example, condition monitoring is one area where AI models can analyze acoustic signals to detect early signs of equipment failure, enabling predictive maintenance and minimizing downtime. Another application is anomaly detection, where AI can identify deviations from normal sound patterns, helping operators spot issues such as loose components or abnormal vibrations before they lead to more significant problems [105].

5. AI-Powered Use Cases for Smart Manufacturing

The integration of AI into manufacturing processes represents a shift in how data is utilized to optimize production, improve quality, and drive efficiency [106,107]. As manufacturing environments become increasingly complex, AI provides the tools necessary to process and analyze vast amounts of data in ways that were previously unattainable [108,109,110]. At its core, AI enables manufacturers to transition from traditional, manual methods of decision-making to data-driven, automated systems that are capable of learning from historical and real-time data to make informed predictions, detect patterns, and optimize operations [110].

In data-driven manufacturing, AI functions as the key enabler of intelligent systems that leverage large datasets generated by sensors, machines, and other digital tools [111,112]. These AI systems can interpret and act on data in real time, making them indispensable for the modern smart factory [113,114]. The ability of AI to analyze and process data with speed, accuracy, and scalability enables manufacturers to address several critical challenges, including operational inefficiencies, quality control, predictive maintenance, and supply chain optimization [65,115]. We explore some of the most popular use cases prevalent in manufacturing as shown in Figure 4, showcasing how various AI techniques and data modalities are being applied across different domains.

5.1. Predictive Maintenance and Equipment Health Monitoring

One of the most widely recognized applications of AI in data-driven manufacturing is predictive maintenance [88,116]. Traditionally, maintenance in manufacturing environments has been reactive (i.e., responding to equipment failures) or scheduled based on fixed intervals (i.e., time-based maintenance). These methods can be inefficient, costly, and disruptive to production. AI-driven predictive maintenance, on the other hand, uses real-time data collected from sensors embedded in machines to monitor equipment health continuously [117,118]. AI models can analyze this data to identify early warning signs of potential failures, such as unusual vibration patterns, temperature fluctuations, or excessive wear and tear [90,119].

By predicting when a machine is likely to fail, AI allows manufacturers to take corrective actions before a breakdown occurs, reducing unplanned downtime, extending equipment life, and optimizing maintenance schedules [120]. Predictive maintenance also enables manufacturers to replace or repair parts only when necessary, reducing waste and lowering the costs associated with unnecessary maintenance activities.

5.2. Quality Control and Defect Detection

Maintaining high product quality is a essential for manufacturing success [121]. AI plays a crucial role in enhancing quality control through automated visual inspection systems and real-time quality monitoring [122]. Traditional manual inspection processes are often slow, inconsistent, and prone to human error. In contrast, AI-powered systems, particularly those using computer vision and deep learning, can rapidly analyze images and detect defects in real time with high accuracy [123]. These systems can identify issues such as surface defects, incorrect dimensions, or misalignments that may not be visible to the human eye [57].

Machine learning models trained on large datasets of labeled images can continuously improve their ability to detect subtle defects that may vary depending on production conditions, material types, and product variations [124]. Moreover, AI systems can provide actionable insights for root cause analysis, helping manufacturers understand the underlying factors contributing to quality issues. By integrating AI-driven quality control into the production line, manufacturers can achieve higher consistency, reduce scrap rates, and improve customer satisfaction [125].

5.3. Production Optimization and Scheduling

AI can significantly improve production efficiency through the real-time optimization of production schedules [126]. In manufacturing, production lines are often subject to delays, bottlenecks, and unpredictable variations in demand [127]. AI-driven systems can process large volumes of data from machines, inventory levels, supply chains, and workforce schedules to generate optimized production plans [128]. For example, AI models can automatically adjust the production schedule based on real-time data, such as machine performance, material availability, and operator availability, ensuring that production is continuous and optimized for maximum throughput [129].

Furthermore, AI can anticipate and mitigate disruptions by recognizing patterns in production processes that lead to inefficiencies, such as machine downtime or supply chain delays [130]. By predicting these issues ahead of time, AI enables manufacturers to proactively adjust their operations to avoid costly delays, ultimately improving productivity and reducing lead times. Machine learning algorithms can also analyze historical data to identify the most efficient process configurations, further enhancing productivity in the long term [115].

5.4. Supply Chain and Inventory Optimization

The role of AI extends beyond the factory floor and into the broader supply chain. Supply chain optimization is another area where AI can significantly improve manufacturing efficiency. AI-powered systems can predict demand more accurately by analyzing historical sales data, customer behavior, market trends, and even external factors such as weather or economic conditions. This ability to forecast demand allows manufacturers to optimize production schedules, ensuring that resources are allocated efficiently and products are delivered to customers on time [131].

In addition to demand forecasting, AI can optimize inventory management by ensuring that stock levels are maintained at optimal levels. AI algorithms can identify potential bottlenecks or shortages in the supply chain and recommend proactive measures to prevent disruptions [132]. Furthermore, AI systems can identify opportunities for cost savings by analyzing transportation routes, warehouse operations, and supplier performance, allowing manufacturers to reduce excess inventory and minimize logistics costs [133].

5.5. Energy Efficiency and Sustainability

Sustainability is an increasingly important concern for manufacturers, and AI can play a key role in improving energy efficiency and promoting sustainable production practices [134]. AI-driven systems can analyze energy usage patterns across manufacturing facilities, identifying areas where energy consumption can be reduced without affecting production quality or speed [135]. For example, AI can suggest adjustments to machinery operations or lighting schedules to minimize energy waste during periods of low activity [136].

In addition to energy optimization, AI can help manufacturers reduce waste by optimizing raw material usage and recycling. By analyzing production data, AI can identify opportunities to reduce scrap rates, improve material yield, and recycle excess materials back into the production process [137]. These efforts not only help companies reduce operational costs but also contribute to their overall sustainability goals by lowering their carbon footprint and minimizing resource consumption [138,139].

5.6. Real-Time Analytics and Decision-Making

The real-time capabilities of AI-driven systems provide manufacturers with the ability to make informed decisions rapidly [140]. In traditional manufacturing environments, decision-making often relies on periodic reports, historical data, or manual oversight, which can lead to delays in responding to problems or changes in production conditions [141]. AI, by contrast, can analyze data in real time, enabling manufacturers to quickly adjust operations in response to emerging issues [142].

For example, AI-powered systems can immediately detect a drop in production quality, a machine malfunction, or a supply chain disruption and suggest corrective actions [143]. By integrating AI into operational workflows, manufacturers can respond dynamically to changing conditions, ensuring that production processes remain efficient, cost effective, and adaptable [144]. Furthermore, AI-driven decision-making can be integrated into broader enterprise systems, such as ERP, to ensure alignment across departments and to provide actionable insights for managers at every level of the organization.

5.7. Human–Machine Collaboration

AI is not intended to replace human workers in manufacturing but rather to augment human capabilities, enabling more intelligent collaboration between machines and humans [145,146]. AI can assist operators in tasks such as machine setup, process adjustments, and maintenance by providing real-time insights, recommendations, quality inspection, and predictive alerts [147]. For example, AI systems can guide operators through maintenance tasks based on real-time machine health data or suggest adjustments to machine settings to optimize production conditions [148].

By empowering human workers with AI-driven tools, manufacturers can enhance workforce productivity, improve decision-making, and reduce the cognitive load on operators [149]. This collaboration between human expertise and AI technology creates a more adaptive and resilient manufacturing environment [150].

6. Data-Centric Framework for Intelligent AI Systems in Manufacturing

To fully harness the potential of AI and big data technologies in manufacturing, companies must address a range of challenges, including data heterogeneity, scalability, quality, integration, and security [151,152] as shown in Figure 5. Overcoming these obstacles requires a comprehensive framework that combines both technological and organizational strategies [153,154]. This framework should focus on integrating data from diverse sources, such as IoT sensors, production logs, and machine vision systems, into a unified platform to ensure seamless data flow and improve accessibility [155]. Additionally, it should prioritize scalability by utilizing cloud, edge, or hybrid computing solutions to efficiently manage large datasets, along with robust data cleaning, preprocessing, and validation methods to maintain data quality [156]. Furthermore, the framework must incorporate AI models into manufacturing workflows to enhance real-time decision-making and optimize processes like predictive maintenance and quality control [157,158]. Security is also crucial, and the framework should implement strong encryption, access controls, and other protective measures to safeguard sensitive manufacturing data [159]. Data visualization plays a pivotal role in this framework, as it helps to transform complex data into actionable insights, allowing decision-makers to quickly interpret and respond to operational trends [160]. Additionally, a continuous feedback loop is essential, ensuring that AI models remain adaptive to evolving conditions and improving over time [161]. By adopting this multi-layered framework, manufacturers can improve operational efficiency, enhance product quality, and maintain competitiveness in an increasingly technology-driven industry, unlocking the full potential of AI to drive innovation and long-term success. The details of each layer of this framework, including data acquisition, storage, processing, model development, security, visualization, and feedback mechanisms, are discussed subsequently in Section 6.1, Section 6.2, Section 6.3, Section 6.4, Section 6.5, Section 6.6 and Section 6.7.

6.1. Data Acquisition Layer

The data acquisition layer is a foundational element in building a data-driven AI system in manufacturing, as it collects and aggregates data from various sources to provide a comprehensive view of the entire production environment [162]. Its primary function is to ensure that the AI system receives high-quality, relevant, and up-to-date data, which is essential for accurate decision-making and real-time optimization. This layer gathers data from a wide range of systems, devices, and sensors distributed across the manufacturing facility, ensuring a holistic understanding of operations. Beyond merely collecting data, the data acquisition layer plays a critical role in integrating diverse data types, enabling the AI system to make well-informed decisions based on a unified data ecosystem. It systematically integrates data from IoT sensors that monitor machine performance, equipment status, and environmental conditions, as well as from vision systems that capture visual insights for quality control. The layer also collects operational data, such as production schedules, raw material usage, labor availability, and process flows, along with maintenance records, logs, and textual information from operators. This data is stored in a centralized platform, ensuring seamless access and maintaining data accuracy and consistency. By consolidating data from various sources including IoT sensors, machines, logs, and enterprise systems including ERP, the data acquisition layer enables smooth communication across systems and enhances AI models’ ability to make real-time decisions based on reliable, high-quality data from multiple sources. This holistic integration supports predictive maintenance, process optimization, and continuous improvement in manufacturing operations.

IoT sensors are a primary data source, offering real-time insights into machine health by monitoring critical metrics such as temperature, vibration, pressure, and humidity. These sensors help detect early signs of malfunctions, which is vital for predictive maintenance and performance optimization [64]. However, the data acquisition layer must also incorporate other data types, such as vision data from cameras and machine vision systems, which inspect products for defects, monitor assembly lines, and capture visual data to identify inefficiencies and improve product quality [155]. Logs from both machines and software systems also provide valuable operational data, including system events, performance logs, and error notifications, which are crucial for troubleshooting and enhancing system performance. Additionally, textual data from maintenance records, production schedules, and operator notes offer contextual information, supporting decision-making and helping manufacturers adjust processes or schedules to enhance operational efficiency [163]. Audio data, such as sound recordings from machines, can further aid in detecting mechanical anomalies or signs of malfunction, providing another layer of valuable insight that might otherwise go unnoticed.

Moreover, operational data plays an integral role in optimizing manufacturing processes. This data, which includes production schedules, raw material usage, labor availability, and process flows, helps manufacturers adjust operations in real time, ensuring optimal resource allocation and preventing delays during fluctuations in demand or supply chain disruptions [163]. Maintenance data is equally important, as historical records of machine breakdowns, repairs, and routine servicing provide insights into the lifecycle of machinery. By identifying patterns in this data, manufacturers can predict future maintenance needs, reducing unplanned downtime and optimizing machine lifecycle management [152].

Finally, integrating data from ERP systems enhances the value of the data acquisition process. ERP systems consolidate business-critical information such as inventory levels, procurement activities, and supply chain management [155]. When combined with machine, operational, and maintenance data, ERP data offers a more complete picture of the production process, allowing AI systems to make informed, data-driven decisions that improve overall efficiency, reduce waste, and ensure the timely availability of resources.

6.2. Data Storage Layer

After acquiring data, the next step is storing it in a way that guarantees accessibility, consistency, and scalability. Manufacturing environments generate massive volumes of data, so modern data storage solutions must be capable of handling this large influx of information. One commonly used solution is the data lake, which serves as a centralized repository for storing raw, unstructured, and structured data in its native format. Data lakes provide flexibility by allowing manufacturers to store different types of data (such as sensor data, operational logs, and machine performance data) together, making it easier to analyze large and diverse datasets [164,165].

Cloud storage is another vital component in the data storage layer, offering scalable and accessible solutions. Cloud storage enables manufacturers to store vast amounts of data and retrieve it whenever necessary, regardless of their physical location. This is particularly important, as manufacturing systems increasingly rely on real-time decision-making and remote monitoring [166,167]. Cloud-based solutions also offer the advantage of reducing infrastructure costs while providing flexibility in scaling storage capacity as data volume grows [168].

For scenarios requiring quick decision-making and low-latency responses, edge computing is becoming increasingly important in manufacturing. Edge devices process data locally, on the factory floor, without having to send the data to a centralized cloud server [169,170]. This is crucial for applications where real-time analytics and immediate action are necessary, such as in automated quality inspections or machine condition monitoring [171]. By processing data at the edge, manufacturers can reduce delays, improve system responsiveness, and maintain high levels of operational efficiency [172].

6.3. Data Processing Layer

The data processing layer is essential for transforming raw data into actionable insights that optimize manufacturing processes and inform decision-making. The initial phase of data processing involves cleaning, as raw data frequently contains noise, errors, or missing values. Data cleaning and preprocessing techniques are crucial for ensuring data accuracy, consistency, and readiness for further analysis [173,174]. This step typically includes the removal of outliers, imputation of missing values, and correction of inconsistencies arising from sensor malfunctions, human errors, or data collection issues. Clean data is a prerequisite for any subsequent analysis, as it guarantees the reliability of insights derived from it.

After the data cleaning process, the data is often transformed into a standardized format to facilitate seamless integration across various systems and data sources. Data transformation techniques, such as normalization, scaling, or aggregation, are applied to harmonize data from diverse sources, including IoT sensors, maintenance logs, ERP systems, and operational databases [175,176]. These techniques ensure consistency across datasets, thereby promoting smooth data flow and simplifying subsequent analysis. For example, operational data may be recorded in different units or formats across multiple systems, and standardizing this data enables more effective cross-system analysis and decision-making.

Feature engineering is another critical component of the data processing layer, wherein new features or metrics are created to enhance the predictive capabilities of AI models [177]. For instance, raw sensor data may be transformed into meaningful features, such as moving averages, trends, or rate-of-change metrics, to better capture patterns in machine behavior. Feature engineering enables AI models to discern relationships within the data more effectively, improving their ability to detect anomalies, predict failures, and optimize processes [178].

Furthermore, data enrichment plays a significant role in the data processing layer, which involves integrating external data sources with internal manufacturing data [179]. For example, external information such as weather data, market demand forecasts, or supply chain data may be incorporated to provide a more comprehensive context for decision-making. This added context enhances the AI models’ capacity to make informed decisions that account for external factors influencing production, such as fluctuating raw material prices or shifts in demand [180,181].

By ensuring that data is clean, standardized, and enriched, the data processing layer establishes the foundation for effective AI-driven decision-making in the manufacturing domain.

6.4. AI Model Development, Training, and Deployment

The development of AI models for manufacturing requires customization to handle complex data and provide actionable insights specific to the needs of the factory floor. The initial phase in this process involves tailoring AI models to address the particular challenges encountered in manufacturing environments. For instance, predictive maintenance models must be designed to detect early signs of machine failure, such as unusual sensor readings associated with vibration, temperature, or pressure [182,183]. Conversely, quality control models need to be trained on diverse image datasets to detect defects under varying lighting and environmental conditions [171,184]. Customizing AI models to these specific use cases ensures that the insights provided are accurate, relevant, and aligned with the operational requirements of manufacturers [50,185].

Real-world data in manufacturing settings is often limited. It is either imbalanced or sparse, particularly in the case of rare events like equipment failures. To address these challenges, data augmentation and simulation techniques play a critical role in enhancing model performance [186,187,188]. Synthetic data generation and simulation models can be utilized to supplement datasets, particularly for rare but critical events, such as anomalies or equipment malfunctions. Simulated data provides a valuable source of information, allowing AI models to be trained on scenarios that are infrequent but essential to model. This approach enhances the robustness of the AI system and improves its ability to handle edge cases [189,190,191]. The use of such techniques is particularly beneficial when historical data is limited or underrepresented.

To maintain the effectiveness of AI models and ensure their adaptability to evolving conditions, continuous model training is essential. Active learning is a key approach, where the AI system periodically updates and retrains itself using new, real-time data as it becomes available [192,193,194]. By incorporating data from the factory floor into the model training process, manufacturers can ensure that AI systems remain responsive to changes in operations, new production patterns, or the introduction of new machinery. Continuous training not only enhances the accuracy of predictions but also allows the AI system to adapt to unforeseen conditions that may arise [195,196].

Lastly, explainable AI techniques are critical for ensuring that the decisions made by AI systems are transparent, understandable, and interpretable by human operators and decision-makers [25,197]. For example, when a predictive maintenance system flags a potential failure, it should provide a clear rationale, such as the identification of unusual patterns in temperature or vibration. This transparency fosters trust in the AI system’s decisions, allowing operators to take informed action when necessary. Explainable AI is also vital for meeting regulatory requirements, ensuring that decisions made by AI systems are auditable and comprehensible, thus promoting wider adoption in environments where transparency in decision-making is required [198,199].

In conclusion, the integration of custom AI models, data augmentation through simulation, continuous model training, and the incorporation of explainability ensures that manufacturers can develop robust AI systems. These systems enhance operational efficiency, improve quality control, optimize maintenance strategies, and support more informed decision-making, ultimately leading to better outcomes in manufacturing environments.

6.5. Data Security and Privacy Layer

In manufacturing, the protection of data is important to safeguard intellectual property, production secrets, and other sensitive operational information. To preserve the integrity and confidentiality of this data, the implementation of end-to-end encryption is essential for all forms of data, whether it is in transit or stored at rest [159]. By applying industry-standard encryption algorithms, manufacturers can ensure that data remains unreadable and secure from unauthorized access, even in the event of interception during transmission or unauthorized attempts to access stored data [200]. This level of protection is critical in an era where cyberattacks and data breaches are increasingly common [201].

Another fundamental aspect of data security is the adoption of role-based access control (RBAC) [202,203,204]. By enforcing robust access control policies, manufacturers can restrict access to sensitive data—such as proprietary designs, production methods, or financial information—to authorized personnel only [205]. In addition to RBAC, the integration of multi-factor authentication (MFA) further strengthens security by requiring additional layers of verification, such as biometric data or one-time passwords (OTPs), before access is granted [206,207]. This dual-layer authentication mechanism mitigates the risk of unauthorized access, even in cases where employee credentials are compromised [208,209].

For situations in which sensitive data must be shared with third-party analytics services or cloud providers, data masking and anonymization techniques offer an effective means of protecting proprietary information [210]. These techniques allow key data elements, such as production processes or design specifications, to be obscured while still enabling external parties to conduct necessary analyses. This approach ensures that sensitive data remains secure, even during collaborations with third-party partners, while facilitating the extraction of valuable insights from the data.

Lastly, regular security audits and vulnerability assessments are critical for identifying and mitigating potential weaknesses in the system [211]. Manufacturers should perform these assessments on a periodic basis to ensure the ongoing effectiveness of their data security measures and to address emerging threats. In addition, aligning security practices with established industrial standards including the IEC 62443 series [212] for industrial automation and control systems, and the NIST SP 800 series for information security and risk management provides a robust foundation for implementing cybersecurity measures that meet industry expectations. Compliance with industry standards for information security management, as well as adherence to regulations such as the General Data Protection Regulation (GDPR), should also be prioritized to meet regulatory requirements and foster trust with customers and partners [213]. By embedding these security practices into the manufacturing data infrastructure, companies can protect sensitive information, maintain operational integrity, and minimize the risks associated with cybersecurity threats [214].

6.6. Data Visualization and Decision-Making Layer

The final layer of an AI-driven manufacturing system focuses on the visualization and interpretation of insights generated by AI models. This layer is critical because it enables decision-makers to interpret complex data and make informed decisions [160]. Tools such as dashboards, reports, and visualizations present key performance indicators (KPIs), trends, and other crucial insights in an accessible format [215]. By transforming raw data into clear, actionable information, these visual tools allow manufacturers to monitor processes, track performance, and identify areas for improvement [216].

In addition to visual tools, this layer may include automated decision-making systems that leverage AI to directly control manufacturing processes. For example, supervised learning models can predict machine failures or product defects based on historical data, triggering automated maintenance schedules or quality control checks. This integration of AI into real-time decision-making helps manufacturers optimize production processes, reduce downtime, and improve product quality.

The use of AI-driven visualizations and automated decision-making extends to areas like inventory management, supply chain optimization, and production scheduling [198,217]. By analyzing past production data, manufacturers can create more efficient scheduling systems or optimize inventory levels based on predictive insights. This layer, combining visualization tools with automated actions, enables the continuous improvement of operations, enhancing productivity while reducing costs and errors.

6.7. Continuous Improvement and Feedback Loop

To ensure ongoing success in AI-driven manufacturing processes, fostering a culture of continuous improvement is critical. This is achieved through the creation of a feedback mechanism that allows operational outcomes such as product quality, yield, and downtime to be analyzed, and insights to be fed back into the AI models for further refinement [161]. By incorporating real-time data and performance results into the AI system, manufacturers can iterate on their models to improve predictive accuracy and adapt to changing conditions. This cyclical process of data collection, analysis, feedback, and adjustment creates a dynamic system that constantly evolves, enabling the manufacturing process to become increasingly efficient and responsive over time [218].

In addition to the feedback loop, collaboration across teams is essential for maximizing the value of AI insights. The successful integration of AI requires close cooperation between data scientists, operational engineers, and decision-makers. This collaboration ensures that the insights generated by AI are not only technically sound but also practical, actionable, and aligned with broader business goals [219]. By involving key stakeholders throughout the process, organizations can ensure that AI solutions are tailored to meet operational needs and improve key areas of the production process.

To measure the success of these AI-driven systems, it is important to establish performance metrics that track the effectiveness of AI models. These KPIs might include predictive accuracy, uptime, throughput, and energy consumption. Regularly assessing these metrics enables manufacturers to gauge the impact of AI on their operations and identify areas that require further optimization. Continuous tracking of these performance measures ensures that AI models are always aligned with business objectives and contributes to ongoing operational efficiency. Through this iterative process of refinement, collaboration, and measurement, manufacturers can maintain a steady path toward improved productivity and competitiveness.

By adopting this comprehensive framework, manufacturers can effectively tackle the challenges of data quality, heterogeneity, scalability, integration, and security. This will empower them to unlock the full potential of AI and big data technologies, driving operational optimization, enhanced product quality, and increased overall efficiency. With the right data governance, infrastructure, and AI models in place, manufacturers can create resilient, intelligent systems capable of adapting to the demands of modern production environments, paving the way for sustained growth, innovation, and competitive advantage in the industry [220,221].

7. Discussion

The proposed unified framework for integrating AI in manufacturing offers a structured approach to address the key challenges faced by the industry, including data heterogeneity, quality, system integration, and scalability. By providing a comprehensive solution that incorporates data collection, storage, preprocessing, and analysis, the framework aims to enhance the integration of AI systems in manufacturing environments. One of the key strengths of the framework is its emphasis on real-time data processing, enabled through scalable cloud-based solutions and edge computing. This ensures that AI systems can efficiently handle the large volumes of data generated by IoT devices and other sensors, facilitating timely insights for applications such as predictive maintenance and quality control.

However, as discussed in Section 3, the rapid growth of data in manufacturing environments poses a significant challenge for scalability. As new IoT devices and interconnected systems are introduced, the volume of data generated is expected to increase exponentially. To address this, the framework must remain adaptable to accommodate future data expansion. This requires investments in infrastructure that can scale efficiently, such as distributed storage solutions, advanced data compression techniques, and AI models capable of processing large, dynamic datasets. Emerging solutions such as edge/cloud orchestration models can further optimize the distribution of computational workloads, improving latency and resource utilization across heterogeneous environments. Additionally, federated learning offers a promising approach for training AI models across decentralized data sources while preserving data privacy. The integration of technologies like 5G networks and next-generation AI algorithms will also be crucial to ensuring that the framework can meet the growing demands of manufacturing environments.

An important consideration in the implementation of this framework is the role of human workers alongside AI systems. The adoption of AI should be seen as a means of enhancing, rather than replacing the workforce. The framework must prioritize the development of AI models that provide transparent and interpretable insights, allowing operators to understand the rationale behind AI-driven decisions. This transparency is essential for building trust among workers and decision-makers, ensuring that AI systems support human expertise and facilitate informed decision-making. As AI systems evolve, it is crucial to ensure that they complement human skills, enabling collaboration and ensuring that workers remain central to the manufacturing process.

Another critical aspect of the framework is the need to address data security and privacy concerns. As manufacturers collect and process large volumes of sensitive data, including proprietary designs and production methods, safeguarding this information becomes important. The framework must incorporate robust security measures, such as end-to-end encryption, multi-factor authentication, and compliance with data protection regulations. Furthermore, secure data sharing protocols are essential when collaborating with third-party service providers or utilizing cloud-based solutions. Ensuring the confidentiality and integrity of manufacturing data throughout its lifecycle is essential to maintaining trust and operational security. By addressing these challenges, manufacturers can ensure that AI-driven insights lead to improved operational efficiency, enhanced decision-making, and the overall optimization of manufacturing processes.

8. Conclusions

This paper emphasizes the critical role of data-driven architectures in enabling the successful integration of AI into manufacturing environments. It explores the challenges of managing diverse data types, such as time-series, visual, log, text, and sensor data, and stresses the need for advanced technical frameworks and AI learning methods to support the efficient collection, storage, and analysis of large-scale industrial data. The paper also highlights key use cases, including real-time analytics, predictive maintenance, and operational optimization, illustrating how AI can potentially enhance manufacturing performance and efficiency. Additionally, it addresses key challenges related to data quality, heterogeneity, system integration, and data governance, providing strategies to overcome these obstacles and ensure successful AI implementation. By emphasizing the importance of scalability, data security, and the protection of sensitive information, the paper proposes a unified, data-centric framework that could enable manufacturers to fully capitalize on AI and data-driven technologies. Ultimately, this work offers a roadmap for fostering long-term growth, improving operational efficiency, and driving innovation in the manufacturing sector. However, the framework has not yet been validated in real-world settings, and its adaptability across different manufacturing environments remains to be explored. Future research should focus on practical implementation, performance evaluation, and refining the framework based on industrial feedback.