Background: Given that the device possibly suffers the abnormal condition before it finally
reaches the critical failure threshold in the degradation process.
Methods: In this paper, an abnormal failure judgment method of the device is proposed based on the
performance degradation data itself. In this method, the device degradation process is considered to be
an independent incremental process by probabilistic approximation, and then according to Grubbs
criterion, all of degradation quantity is divided as two population, and an abnormal failure occurs only if
a new observed sample belongs to abnormal class, and conversely, and no abnormal failure occurs.
Results: The standard deviation values of all the degradation quantity can be then obtained by dealing
with the performance degradation dates between adjacent moments, statistically. In addition, we still
estimate the cumulative failure probability of the device, and consider the accumulated cost losses
arising from the continuous degradation, and then the expected maintenance cost model is established.
Conclusion: Finally, a practical case on a certain type of Electric Motor Unit (EMU) brake pad
thickness analysis shows that the method presented is quite effective and feasible.