A novel predictive maintenance system for automotive industry using enhanced optimization and weighted fused feature-based cascaded deep learning network
The previous predictive maintenance (PdM) systems can summarize the required information for estimating the period of the failures in the machine, the powerful impacts of the issue, or the need to prevent production from executing the repair functionalities. But the logs for the maintenance are apt to have unbalanced distributions. In this work, a new PdM system is developed to detect equipment failures in the automotive industry using deep learning techniques. It is used to improve productivity. The data is collected from the internet. The collected data is given to the data cleaning phase to remove unwanted disruptions from the data. Then, deep feature data are extracted using restricted Boltzmann machine (RBM), autoencoder, and one-dimensional convolutional neural network (1DCNN). Then, the extracted features are given to the deep-weighted feature fusion phase. Here, the weight is optimized using the developed enhanced American zebra optimization (EAZO). Then, the deep-weighted fusion features are given to the prediction phase. The cascaded residual long short-term memory network (LSTM) with AM (CResLSTM-AM) is used to predict the equipment faults effectively. The developed PdM system in the automotive industry is contrasted with conventional models with performance measures to show accurate outcomes.