Similarity-based failure threshold determination for system residual life prediction

An accurate determination of the system failure threshold is an essential requirement in achieving an appropriate system residual life prediction and a reasonable planned maintenance strategy optimization afterward for degradation systems. This paper pro-poses a failure threshold determination method based on quantitative measurement of the similarity between the operating system and the historical systems. The similarity is formulated by a weighted average function and then calculated by a convex quadratic formulation to minimizing the variance between the operating system and the historical systems. With an accurate determination of the system failure threshold in real-time, a better prediction of the residual life for the operating system is achieved. Finally, a real case study for several power-shift steering transmission systems monitored using oil spectral analysis is adopted to illustrate and numerically compare the improved performance of the proposed method.