Active monitoring of CAN Bus data quality

In recent years, there has been an increase in the use of time series records for monitoring any device with a sensor. CAN BUS records, in particularly, are essential for tracking the life cycle of off-road vehicles and other equipment. Signals of this nature are extremely sensitive to transmission errors, so they may be contaminated or missing. The solution would be to design a system that can detect and correct an anomalous message. To build this system, supervised machine learning techniques are the best option since they do not require human intervention and provide more accurate results than unsupervised methods. Although few annotated data available, and the range of reported exceptions are wide, the analyses were restricted to certain scenarios. The focus of this research is to increase the scope of anomaly identification and analysis in an unsupervised setting. On the CAN bus time series, unsupervised anomaly detection techniques were applied. Methods of efficient retrieval of specific time series segments considered possibly important by experts in the field have been implemented. Finally, a decision support system was developed that allows specialists to analyze annotated series with anomaly predictions and extract the segments corresponding to the anomaly of greatest interest.