On-board Predictive Maintenance with Machine learning
To be published on April 2, 2019 by SAE International in United States
Field Issue( Malfunction) incidents are costly for manufacture’s service department. Especially for commercial fleet customers, the downtime can be the biggest concern. To reduce the warranty cost and improve customer’s confidence in our products, preventive maintenance provides the benefit of 1.fixing the problem when it is small; 2. reducing downtime of scheduled targeted service time to reduce downtime. However, normal telematics system has difficulty in capturing useful information even with pre-set triggers. Some malfunction issue takes weeks to find out route cause due to the difficulty of repeating the error in a different vehicle and engineers to analyze large amount of data. In order to solve above challenges, a machine learning based predictive software/hardware system has been implemented. Multiple machine learning techniques, including CNN, has been utilized in the proposed pipeline to: 1) decide when to record data. 2) decide what variables to record for what period of time 3) root cause diagnostics on the spot based on time series data analysis. The system utilizes both histogram data and time series data. For the time series data, a novel technique has been proposed to solve the lack of training data for the root cause analysis neural network. The root cause analysis will be further reviewed by Engineers through an expert knowledge feedback system to guide the on-board AI. A centralized fleet level system will be discussed for further development. In the paper, the overall on-board preventive maintenance system will be introduced and validation results with real vehicle data will be shown.