Intelligent Fault Prediction and Diagnosis of Intelligent Networked Vehicle Based on Filter Pruning with Snapshot Stacked Ensemble Deep Neural Network

In recent days, the automotive industries has specified a large multidimensional data for the huge development in production sites as well as in after sale-services. Fault diagnostic system is one such type of services that are provided by automotive industries. In existing fault diagnosis prediction, Deep Symptoms Based Model (Deep SBM) is utilized but due to the large computational requirement, it has led to the poor performance. In this research, Filter Pruning (FP) with Snapshot Stacked Ensemble Deep Neural Network (SSEDNN) is proposed and named as FP-SSEDNN. Initially, the data is collected from historical diagnostic data, which is provided by congnitran Ltd. Then, the data is further processed into preprocessing with Word2Vec to convert the words into vectors to represent the vectors in high dimensional space. The vectors are further processed for prediction and FP is performed to reduce the computational resources requirements in a network. This pruned prediction model is further fine-tuned to increase the accuracy and reduce the training time. From the results, the proposed FP-SSEDNN achieved better results than existing Deep SBM in terms of accuracy (99.9%), precision (99.5%), Fl-score (97.9%) and Area Under Curve (AUC) (99.7%) respectively.