Digital twin for risk and uncertainty analysis in complex industrial control and automation systems using artificial intelligence and machine learning
Year: 2023
Type: dissertation
Abstract: Industrial control systems play a crucial role in enabling advanced manufacturing operations. However, these systems are inherently susceptible to failure. Detecting faults at an early stage is of paramount importance, as it can prevent the occurrence of fatal and catastrophic consequences resulting from equipment failures. Moreover, timely detection and resolution of faults can save significant costs and time for organizations. The failure of these systems not only poses risks to operators but can also lead to substantial delays in the advanced manufacturing process, imposing substantial financial burdens on organizations. Therefore, a methodology is needed that can be used to avoid the adverse effects of equipment failure of industrial control systems to achieve smooth advanced manufacturing operations. To achieve this, the methodology should be able to detect the abnormal behaviour of the system at very early stages for predictive maintenance. This methodology can be designed using an extremely popular concept known as the Digital Twin, which has gained significant importance in the era of Industry 4.0. In this research, artificial intelligence techniques will be employed to develop a highly accurate and detailed digital twin model. This model will serve as a valuable tool for predictive maintenance in complex industrial control systems, facilitating the achievement of smooth and uninterrupted advanced manufacturing processes. Also, the performance of the proposed Digital Twin model will be compared with state-of-the-art anomaly detection approaches. The digital twin, utilizing the proposed algorithms, will not only be able to detect anomalies but also quantify their severity, classifying them into different levels such as minor, severe, and faulty operations. Furthermore, the research addresses the generalization challenges faced by state-of-the-art approaches, showcasing the digital twin's ability to effectively classify unseen data as healthy or anomalous. The results obtained from the analysis and comparison of state-of-the-art approaches with the proposed algorithms clearly demonstrate the methodology's capability to detect anomalies, quantify their level, and classify them accurately and effectively in real-world data. This validation underscores the robustness and reliability of the developed methodology, further solidifying its potential as a valuable tool for predictive maintenance in complex industrial control systems.
Author: Siddiqui, Muhammad
Institution: Federation University
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7056046205
