ECU Software Abnormal Behavior Detection Based On Mahalanobis-Taguchi Technique

To confirm the correct operation and detect the potential errors in the ever more complicated automotive ECU application software are very challenging. This paper presents a new approach to detect potential ECU application software abnormal behavior based on the Mahalanobis Distance, the Mahalanobis-Taguchi System, and vehicle driving data playback capability with a simulator. Vehicle driving data is recorded by instrumentation calibration tools and played back on the test bench to stimulate the ECU. In our study, the normal behavior is characterized by the Mahalanobis Distance (MD), which is calculated from the data set logged while playing back the recorded vehicle maneuvers while the ECU is flashed with “baseline software” that was believed to be error free. Then the MDs were calculated from a new data set logged while playing back the same vehicle maneuvers while the ECU was flashed with the new software. The large MDs resulting from the new software indicate times when the new software behaved abnormally. An MD from the data set that is above a threshold indicates a potential software error, which must be investigated in detail by an expert. Some encouraging experimental results using the method on Electronic Throttle Control (ETC) sub-system tests are presented and indicate the capability for this method to detect potential software errors.