The invention relates to a PMSM drive system inverter open-circuit fault online diagnosis method, which can carry out online fault diagnosis and positioning on various inverter open-circuit faults. The method comprises the steps: introducing a Luenberger observer to track a PMSM three-phase current so as to obtain a three-phase current observation residual error as a fault diagnosis basis; designing a sampling strategy, and obtaining an original sample set for training; in offline training, introducing a principal component analysis method to preprocess an original sample set to obtain a datadimension reduction model, and then acquiring a classification model by training a support vector machine; finally, operating the dimension reduction model and the classification model in a driving system, and achieving the online diagnosis and positioning of faults. According to the method, hardware equipment is not added, and driving control is not affected; meanwhile, the method is not affectedby load and model parameter changes; through machine learning of actual working conditions, the method does not depend on a diagnosis basis set manually, and the possibility of misdiagnosis is reduced.