Anomaly Detection in Fleet Vehicle Data and Statistical Approach to Develop Engine System OBD Thresholds

On-Board-Diagnostics (OBD) are crucial for ensuring the proper functioning of Engine’s emission control system by continuously monitoring various sensors and components. When the failure is detected, the Check Engine Light is triggered on Vehicle’s dashboard, alerting the driver to seek professional service to address the issue. However, the task of developing the monitoring strategies and performing robust calibration is challenging and time consuming. Model in loop (MIL) Simulation and testing is a technique used to understand and estimate the behavior of a system or sub system. The diagnostics model can be tested and refined within the model-based environment allowing a complex system to be efficiently regulated.

MIL framework could be explored at various stages of development from early in the design phase to later stages of series developments through vehicle fleet data. This framework allows early identification and correction of errors and bugs in a standalone dependent environment. In this paper, the data visualization dashboard is used to detect anomalies in Fleet Vehicle data and understand the relativity with different systems. A data driven MIL Simulation set up is established to validate and iterate for robust calibration. The drift monitor controller model is discussed as a case study to replicate the behavior and resolve the real time calibration issue. This approach on OBD systems to assess the reliability of models and make robust calibration through Monitor Performance Index (MPI) is carried out in Simulink environment. The key emphasis of this paper is to discuss a framework to identify anomaly from fleet data and resolve through MIL simulation to optimize monitor performance. Different combinations of calibration input data and Fleet Vehicle data are tested for validation to ensure repeatability of performance and decisions related to fault detection criteria.