Abstract: According to statistics, failure events of wind converters dominate in wind farm historical failure records with regard to both failure rate and downtime. Hence it is valuable to develop advanced CMDP (condition monitoring, diagnosis and prognosis) algorithm for wind converters. In this paper, regression based supervised machine learning technology is applied to analyze the on-line data measured from diversified sensors embedded in ABB’s wind converters. Four real wind turbines data collected from a real wind farm is tested to verify the algorithm performance. The experimental results reveal that the proposed approach based on regression machine learning is capable to successfully detect the defective wind converters from a fleet before developing into real failure.