Predictive Maintenance in the Internet of Vehicles Using Transformer-Based Deep Learning Models

The recent advancements in the development of the IoV have contributed to the need for developing accurate and efficient PM models to improve the dependability and efficiency of cars, management of operating costs, as well as traffic safety. What has been used in the traditional maintenance plan is the inspection, which may be scheduled or based on a failure, or the more common method, the reactionary method that leads to expensive repairs when a failure occurs. This paper focuses on the use of deep learning based on the transformer for the development of predictive maintenance for IoV systems. Thus, by using real-time data about connected vehicles, our approach introduces self-attention mechanisms capable of identifying component degradation symptoms timely. Through the historical record of the maintenance data and other multiple source telemetry data, the proposed model can accurately anticipate the failure and thus perform preventive actions on them. The use of the transformer-based model is more advantageous than other machine learning methods since it is more efficient in feature representation, especially for a large number of IoV datasets. A number of empirical based assessments show the increases in prediction performance, decrease of false signals, and an optimal schedule for the maintenance of equipments. The paper examines the need and benefits of deep learning for predicting the vehicle's lifecycle, reducing downtime, and ensuring safety on the roads. Thus, the work presented in this thesis provides the base for intelligent maintenance solutions that can be beneficial for the development of self-driving and connected vehicles.