Predictive Maintenance for Autonomous Vehicle Sensors

Abstract

Autonomous vehicle sensors are becoming more advanced with each vehicle model. Not only that, but sensor numbers are growing at an exponential rate: there are approximately 22 sensors per SAE L2 vehicle and approximately 29 per SAE L4 full automation vehicle, with further expected increases projected for the future. Therefore, manufacturers must employ an advanced maintenance schedule to ensure vehicle safety and driving performance. Unfortunately, current maintenance strategies such as 'replacing parts when they fail' fall behind other machine learning or artificial intelligence strategy-based approaches because modern sensor-cockpit interfaces prevent meaningful data flow to either a Condition Monitoring System or the cloud. This paper will examine and improve upon the latest predictive maintenance algorithm, residual signal, for use with autonomous vehicles.

Autonomous vehicle sensor systems are compounding in complexity, making individual sensor reliability much more important. Advanced Driver Assistance Systems must also have a 99-percent reliability rate to achieve the safety improvements that are predicted. Autonomous Vehicle-Car2X sensors must be able to transmit information quickly and effectively to prevent vehicle damage in low-visibility circumstances. This makes predictively maintaining these sensors vital to vehicles and the roads. Dense sensor systems are becoming necessary to transport passengers and drive goods. With improvements in technology, problems that arise throughout a system must be corrected as soon as possible to maintain safety. Having a resilient preventative maintenance strategy in place can drastically reduce the amount of vehicle downtime. Some logistics companies have autonomous vehicles in operation, where routes are dictated daily to their perception systems. If a vehicle should fail, route assignments are less reliable because they cannot begin to repair the vehicle until it returns.