Neural network based design of fault-tolerant controllers for automated sequential manufacturing systems

This paper presents a novel application of recurrent neural network (RRN) to fault-tolerant control (FTC) of automated sequential manufacturing systems (ASMS) subject to sensor faults. Two RRNs are employed: the first one acts as an I/O relations recognizer and is able to detect faulty sensors and the latter is used as an inverse model of the AMSM to compute the desired control action in a faulty case according to nominal specifications. The learning process of these networks is carried out based on training data generated from the healthy manufacturing system controlled by a programmable logic controller (PLC). Design of the proposed fault-tolerant control system (FTCS) scheme is based on utilizing the two RNNs, a reconfigurable controller and a fault decision subsystem. The design procedure of the proposed FTCS is introduced. The proposed FTCS has been implemented and tested experimentally for a benchmark industrial ASMS subject to single or multiple faulty sensors. Experimental results show the effectiveness of the procedure for a real simple plant. In addition, the results prove these features of the proposed FTCS: (a) effectively improving the faulty control system behaviors, (b) accomplishing its proper functionality in handling single and multiple sensor faults, (c) identifying the sensor faults, and (d) being advantageous in reducing the complexity of the hardware redundancy.