SmartShip: Data Decoding Optimization for Onboard AI Anomaly Detection

Maritime rescue operations rely on efficient and resilient sea vessels for responding to offshore emergencies. To ensure the availability and prompt deployment of rescue cruisers, system operators need to recognize (and resolve) the failure patterns before they occur onboard. However, many SAR organizations do not have sufficient computing resources for advanced anomaly detection and future fault prediction. This paper details our performant and cost-effective solution for sensor data acquisition, CAN bus decoding, and AI analysis, deployable on typical maritime edge devices. The main focus of our work is enhancing the decoding speed, which is a prerequisite for efficient AI analyses and anomaly detection. Besides elaboration on different optimization strategies, we provide the actual use case for AI anomaly detection based on anomaly scores of electrical, fuel, and cooling system sensors. As a result, we improved the decoding speed by more than two orders of magnitude, allowing edge outlier detection in minutes rather than days.