A NOVEL ANN AND PEEPHOLE LSTM TECHNIQUE FOR FAULT DETECTION AND CLASSIFICATION IN CYBER-PHYSICAL SYSTEM

Abstract

Cyber-Physical Systems (CPSs) size failure detection and isolation are usually low-scale operations such as critical infrastructure, smart buildings/towns, and the Internet of Things. Deficient sensors and actuators are degrading the CPS output and giving the controller or other CPS subsystems incorrect details. This problem will be solved by the new Artificial Neural Network (ANN) and peephole modified long-term memory (pmLSTM) technique-based fault detection and diagnostic system (FDDS) by identifying and alienating faults consisting of a large number of sensors in CPSs, taking into account the problems of scalability. By monitoring the temporal and spatial correlations among sensor data, the large-scale CPS needs to effectively learn the marginal fault-free conditions according to this approach. In the image collected from CPS sensors, ANN is used to evaluate key ways to reduce the data dimension to less computational complexity. Process engineers would know which variables cause the failure of the system. The goal is to develop a time series system based on the ANN-pmLSTM network to enhance real-time monitoring of the device. Creation of mechanisms for identifying and isolating different sensors defects and disambiguating the cyber layer of CPS inspections around sensor faults and physical phenomena of time variability for ANN and peephole LSTM. Using the current SVM-FTASL-LSTM, the efficiency and effectiveness of the proposed FDDS have been experimentally tested.