A cost-sensitive statistical fault detection method based on correlation analysis

A cost-sensitive statistical fault detection method based on correlation analysis is proposed for the detection of industrial robot joint metal powder faults in the case of small samples. First, position and torque data are collected from one normal robot and two faulty robots, and the position error signal is determined by data analysis for correlation detection. Next, a sliding window is utilized to slide progressively in chronological order and the maximum autocorrelation coefficient on normal data and cross-correlation coefficient between normal and faulty data are calculated within each window. Then, the distribution of the maximum correlation coefficients is statistically characterized to learn the detection threshold which can be further optimized according to the cost ratio. Finally, the fault state is determined by the detection threshold, and when the correlation coefficient exceeds the threshold, the variable is considered to be in a normal state; otherwise, the variable is considered to be in an abnormal state. Experiments show that the method can effectively distinguish between normal and fault states, achieving more accurate fault detection under small sample conditions.