End-to-End Fault Tolerant Control of Discrete Event System Using Recurrent Neural Networks

Abstract

Industry 4.0 is concerned with how the global production and supply network operates through the ongoing automation of traditional manufacturing systems which depends on closed-loop control system components. These components are prone to faults, especially in sensors. so, a fault-tolerant control framework is needed to deal with these kinds of faults. Many researchers work on developing more intelligence data driven fault-tolerant control techniques instead of traditional ones which need a complete sophisticated model. This paper proposes a learning fault tolerant control of discrete event systems subjected to sensors faults, based on recurrent neural networks. The proposed network has both previous and current sensors signals as inputs and current actuators signals as outputs. The new data driven fault tolerant control replaces the diagnosis and control reconfiguration block function in the traditional one. A case study that integrates automated pneumatic material handling is used to demonstrate the proposed approach. Factory I/O simulator is integrated with MATLAB based on the digital twin concept to simulate and verify the proposed approach. Results demonstrate that the proposed method is capable of handling sensors faults in faultless and various faulty scenarios.