Fault self-healing: A biological immune heuristic reinforcement learning method with root cause reasoning in industrial manufacturing process
JiaYi Tian,Ming Yin,Jijiao Jiang
DOI: https://doi.org/10.1016/j.engappai.2024.108553
IF: 8
2024-05-18
Engineering Applications of Artificial Intelligence
Abstract:Stable fault self-healing ensures the smooth operation of intelligent industrial manufacturing processes. However, current research, while capable of inferring fault propagation paths or individually controlling manufacturing processes, still faces some challenges and limitations in addressing issues within manufacturing processes. In particular, it is easy to overlook the impact of fault variables on normal causal relationships in the manufacturing process and how to achieve adaptive fault repair from the source of faults. The biological immune system possesses sensitive nodes and a neuro-endocrine-immune network with interconnected functionalities, bearing remarkable resemblance to the sensor arrays, information networks, and intelligent controller systems in modern self-healing control systems. These problem domains have significant practical implications in industrial manufacturing processes and urgently require new methods and technologies to address them. To address the issue of root cause fault self-healing in intelligent industrial manufacturing processes, this paper proposes an innovative method that utilizes root cause inference and bio-inspired heuristic reinforcement learning models to achieve fault self-healing in industrial processes. First, innovatively introduce the Variational Autoencoder (VAE) into the Time Convolutional Network (TCN) to construct the TCN-VAE network. By performing feature reconstruction, the network's feature extraction capability is enhanced, further exploring the relationships between latent variables, and consequently building a causal graph of faults. A multi-head attention mechanism is introduced into the network, and the inference process is quantitatively evaluated, thereby improving the generalization and accuracy of root cause inference. Root cause determination rules are established to identify the fundamental reasons for fault occurrence, ensuring self-healing from the source of faults. Next, we developed for the first time a fault self-healing model based on reinforcement learning, utilizing the immune repair process of effector T cells to characterize the fault self-healing process. It effectively adapts to eliminate various levels of faults based on the state transition process of antigens in T helper cells. Additionally, stability control analysis is performed on the proposed model. The average reward function curve and Sobol sensitivity index demonstrates the model's strong robustness in practical applications. The validation results from case studies on the Tennessee Eastman (TE) and Continuous Stirred Tank Reactor (CSTR) show that the average improvement in fault repair F1 score was 0.0941 and 0.105 respectively, and the fastest improvement in fault repair response time was 0.266s and 0.258s respectively. It was confirmed that the root cause inference and fault repair results align with the actual mechanisms of the manufacturing processes. The proposed method has been verified to achieve stable and accurate root cause fault repair in real-world intelligent industrial manufacturing systems, making a significant contribution to achieving stable and accurate root cause fault repair in intelligent industrial manufacturing systems.
automation & control systems,computer science, artificial intelligence,engineering, electrical & electronic, multidisciplinary