12. Method for monitoring state of driving devices using vibration analysis and machine learning

Early detection of driving device abnormalities is required to guarantee the safety and stability of transport. A method for monitoring the state of vehicle driving devices was therefore developed.

An effective way to check the state of traction motors and engines is to monitor vibrations. However, given the complex, changing operating environment of driving equipment, and vibrations generated by actual running of the train, it is difficult to detect abnormalities based simply on the amplitude of the vibration.
As such, machine learning was applied to the outcomes of vibrational octave-band analyses, in order to develop a method for detecting abnormalities based on differences with normal vibrations (Fig. 1).

The principal components of vibrational data obtained from octave band analyses were plotted as coordinates. In doing so, any abnormal vibration would produce an isolated plot, making it possible to detect an abnormality based on the distance from the points plotted for normal vibrations.

Then, by separating octave-band analyses results by frequency band and detecting abnormality separately, it is possible to predict the type of abnormality that has been detected (Table 1).
Furthermore, by estimating the abnormal vibration rate within a given timeframe, it is possible to evaluate the speed of progression of the abnormality. Trials were conducted with the developed monitoring system mounted on an actual train, which demonstrated that indications of an abnormality were detected 50 days before the auxiliary driving gear malfunctioned (Fig. 2).