Abstract: 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.