A Data-Driven Framework for On-Board Vehicular Diagnostics: An Academia-Industry Case Study

Modern telematics systems churn a wealth of streaming data that can be leveraged for enhanced on-board monitoring and optimization of vehicle emissions, health, and performance. Yet, the complexities of raw telematics data, including their sheer volume and heterogeneity, hinder the utility of data-driven methods for vehicular diagnostics and prognostics applications. This paper presents an academia-industry joint effort to develop a data-driven framework to leverage on-board telematics data for improved vehicle diagnostics and prognostics. By directly acting on the multivariate time series generated by telematics systems, the proposed framework, referred to as the Structured Time Refinement and Intelligent Content Transformation (or in short, STRICT), performs a set of data analysis steps that are intended to handle the complexities of raw telematics data and extract a minimally sized, but information-rich subset of data that can be used for effective performance monitoring. Trained and tested using data from a large number of trips, the proposed framework is shown to accurately predict key diagnostic trouble codes several days in advance, with up to 95.7% accuracy, while maintaining a sensible balance of false and missed alarms. Potential benefits of this framework to enabling smarter vehicular monitoring and emission mitigation are discussed.