Detecting Engine Anomalies Using Batteries

Yiqin Wang, Linghe Kong, Siyu Lin, Liang He

Abstract: The automotive industry is increasingly deploying software solutions to provide value-added features for vehicles, especially in the era of vehicle electrification and automation. However, the ever-increasing cyber components of vehicles (i.e., computation, communication, and control) incur new risks of anomalies, as evident by the millions of vehicle recalls by different automakers. To mitigate these risks, we design the <monospace>B-Diag</monospace> , a battery-based diagnostic system detects engine anomalies with a cyber-physical approach. The core idea of <monospace>B-Diag</monospace> is to diagnose engines using the physically-induced correlations between battery voltage and engine variables, which is captured as a customized 3-layer correlation graph and a set of data-driven norm models describing the edges thereof. The design of <monospace>B-Diag</monospace> is steered by a dataset collected with a prototype system when driving a 2018 Subaru Crosstrek in real-life for over three months. Our evaluation shows <monospace>B-Diag</monospace> to detect anomalies of 17 engine variables with a <inline-formula><tex-math notation="LaTeX">$&gt;$</tex-math></inline-formula> <inline-formula><tex-math notation="LaTeX">$86\%$</tex-math></inline-formula> (up to <inline-formula><tex-math notation="LaTeX">$100\%$</tex-math></inline-formula> ) accuracy.