Artificial intelligence (AI) techniques, specifically machine learning (ML) methods, are increasingly critical in the industrial automation landscape as AI-powered machines can perform complex industrial automation functions, such as predictive maintenance. AI methods also facilitate greater automation in the manufacturing processes, leading to significantly improved process efficiency and increased production of high-quality final products at lower costs. This article discusses the role and applications of AI in industrial automation.
Image credit: Gorodenkoff/Shutterstock
Importance of AI in Industrial Automation
The rising uncertainty and evolving dynamics in the manufacturing sector have increased the need for constant innovation, quick response to market volatilities, and changes in customer behavior without affecting product quality and increasing costs.
Thus, the use of AI industrial automation is becoming crucial in manufacturing as manufacturers implement different steps to improve their operations' safety, efficiency, and flexibility to adapt to a rapidly changing environment effectively. AI can enable faster, data-driven decisions, better production outcomes, greater scalability, reduce operational costs, and facilitate product innovation.
Several challenges in the manufacturing sector, such as information overload, integration issues, decision complexities, and lack of expertise, can be addressed using AI. In industrial automation, AI and ML can create intelligent digital twins for visual inspection, predictive maintenance, and production planning.
ML methods, including supervised ML, unsupervised ML, and semi-supervised ML, can enable manufacturers to quickly implement manufacturing processes corresponding to customer behavior changes regarding product customization and quality.
AI Methods in Industrial Automation
Industrial AI models can obtain useful/actionable insights from the substantial amounts of data generated daily. These insights can enable designers/engineers to identify new ways to update and improve manufacturing processes based on the latest technologies.
Computer vision can improve service and product quality by increasing efficiency, production, and manufacturing plant safety and security. Similarly, cognitive computing and data-driven deep learning (DL) can effectively handle conceptual data and improve manufacturing techniques.
DL utilizes artificial neural networks (ANN)-based techniques to extract valuable insights from raw data, while cognitive computing can handle conceptual data as it possesses an advanced comprehension and reasoning capability.
Intelligent digital twins and AI-driven collaborative robots (cobots): Cobots and intelligent digital twins can be employed to significantly increase the safety and productivity of manufacturing processes. Cobots can work with humans to pack, analyze, inject, place, and pick items/products.
These human-sized autonomous systems can track their motions to avoid collisions/accidents. AI-driven cobots deployed on assembly lines and humans have demonstrated higher productivity than fully manual or automated assembly lines.
These systems can also be used for sorting, pressing, grinding, welding, fixing screws, and moving heavy objects. Intelligent digital twins can reduce the downtime and the costs of deploying robotic systems.
Self-learning Industrial Robots: Self-learning ability can be imparted to industrial robots that use cameras to monitor their surroundings, such as localizing parts required for manufacturing, using AI techniques such as unsupervised learning and reinforcement learning.
The self-learning ability enables them to perform their tasks and autonomously correct the errors detected in their surroundings/production conditions. Thus, this ability substantially reduces the need for lengthy programming processes as programmers are not required to consider several possible incidents or build these incidents into industrial robots.
Additionally, self-learning robots can be utilized rapidly and flexibly for different individual tasks and applications, eliminating the need to create a dedicated and new robot for every task/application. These robots can play a significant role in the small-batch production and manufacturing of personalized products, such as shoes and clothes.
Self-learning robots can eliminate or minimize the need for humans in materials handling tasks that are repetitive and/or dangerous and can learn and efficiently complete the execution of nonlinear processes that are difficult to automate, such as deburring, sanding, cutting, and polishing.
Moreover, ML-powered specialized robots with several arms can move into unusual areas and positions and independently decide on the action that will be executed by each arm without requiring extensive programming.
Condition Monitoring: AI methods can be used for several condition monitoring applications, including machine performance monitoring depending on sensor data-based condition analysis, intelligent asset monitoring involving real-time automated monitoring of all physical assets/goods, predictive quality assurance using predictive analytics algorithms to resolve and detect potential quality issues, and workplace safety using self-learning algorithms that analyze the sensor data and implementing protection measures based on the analysis to prevent accidents and ensure worker safety.
Predictive maintenance employs AI, primarily supervised learning-based algorithms, in combination with sensor technologies for continuous evaluation and monitoring of production equipment.
Self-learning algorithms can analyze data from different systems and machines and detect patterns indicating impending machine failures. Manufacturers can use this information to schedule maintenance cycles cost-effectively to prevent outages/decline in performance and avoid downtime and expensive repairs.
Unsupervised learning algorithms can analyze and identify rare observations, events, and incidents deduced from substantial data discrepancies. Data parameters and types corroborate these findings by identifying the factor that induced the problem.
Process Management: Augmented reality (AR) coupled with AI can be used to increase the flexibility of management software solutions in the operations process chain to account for the process-related data, including real-time incident information, sensor data, and details about material, machines, and people, that was previously unavailable or extremely difficult to program.
AI-enabled AR glasses/intelligent AR glasses can be extremely beneficial for the support staff during their training and task execution. The DL algorithms in these glasses can precisely recognize and classify objects using computer vision and user vision data. Additionally, the technology utilizes natural language processing in multiple languages in real-time.
Computer-aided situation assessments with AI-validated data capture and image data display ensure fewer errors/wrong decisions in critical situations. Moreover, experienced workers can assist a less experienced/new worker as a real-time, virtual shadow without being physically present on the site.
Recent Studies
In a paper recently published in the journal Procedia CIRP, researchers proposed a framework for intelligent/AI-enhanced industrial automation systems that can meet four attributes of an intelligent system, including perception and observation of information, analysis, and storage of this information, reasoning based on the analysis results, and execution of the reasoning results.
Researchers also presented an architecture for an intelligent digital twin to realize the AI component characters within this framework. The proposed framework was based on the concept of an AI component improving a conventional industrial automation system through new interfaces.
In the framework, the AI component is software located in the cyber part of a cyber-physical system, either locally or in a cloud service accessible through a global area network. The AI component collects the data sent by its information application programming interface (API) and the industrial automation system through a data acquisition API.
Moreover, the networking API of the AI component can offer additional information by providing access to other entities, such as users, machines, or environment representations, through common network interfaces. Subsequently, the intelligence process is performed within the AI component using all available information, and the results are relayed to the industrial automation system through the feedback API for execution.
Additionally, the AI2AI API provides a direct interface for cross-domain/in-domain communication between various AI components, allowing knowledge sharing and improving overall system performance. The AI-enhanced industrial automation system framework was realized by implementing a modular production system/real asset and its intelligent digital twin using different technologies.
The implemented intelligent/AI-enhanced modular production system using the intelligent digital twin enabled the system to automatically react to new customer requirements on new products through the automatic new control code generation for the system based on environmental parameter analysis, indicating effective automatic configuration and control of the real system by the intelligent digital twin.
References and Further Reading
Jazdi, N., Ashtari Talkhestani, B., Maschler, B., Weyrich, M. (2021). Realization of AI-enhanced industrial automation systems using intelligent Digital Twins. Procedia CIRP, 97, 396-400. https://doi.org/10.1016/j.procir.2020.05.257
Uyttersprot, M. (2021). Best practices and use cases for machine learning in industrial automation [Online] Available at https://www.avnet.com/wps/portal/silica/resources/article/best-practices-for-machine-learning-in-industrial-information/ (Accessed on 18 August 2023)
Li, Q. (2021) Application of Artificial Intelligence in Industrial Automation Control System. IOP Conference Series: Earth and Environmental Science. https://doi.org/10.1088/1755-1315/647/1/012043
Artificial Intelligence in Industrial Automation: A Primer [Online] (Accessed on 18 August 2023)