Automobile Maintenance Prediction Using Integrated Deep Learning and Geographical Information System

Abstract

In the industry, predictive maintenance, or PdM, has gained popularity as a means of lowering maintenance costs and achieving efficient operational oversight. The essence of PdM is to anticipate the occurrence of the next breakdown, allowing for the timely scheduling of maintenance activities prior to its actual manifestation. The objective of this work is to develop a Time-Between-Failure (TBF) forecasting framework using a data-driven methodology. Forecasting the Remaining Usable Life (RUL) is a vital issue in PdM. The objective of the current research is to include Geographic Information Systems (GIS) data into TBF (Time Between Failures) modeling and investigate their influence on automotive TBF via the use of deep learning techniques. Initially, an information fusion technique has been investigated to address the disparity in information category and sample rate between the maintenance information and GIS information. The Cox Proportional Hazard Model (Cox PHM) has been employed using the combined information to create the Health Index (HI). This research introduces an Integrated Deep Learning (IDL) architecture that aims to provide a unique perspective on PdM. This design consists of an input layer, a Long-Short Term Memory (LSTM) layer, a Dropout layer (DO) followed by another LSTM layer, a hidden layer, and an output layer. The Genetic Algorithm (GA) has been employed to discover the optimal number of periods and batch size for the design. The activation function is utilized after the output level and the DO ratio, and the optimization method enhances the loss function established using Grid Searching (GS). Research utilizing an extensive record of vehicle upkeep from a fleet firm demonstrated the efficiency of the suggested method and the influence of GIS parameters on the investigated automobiles.