A permanent magnet actuator (PMA) is a critical device for transforming, transmitting, and protecting electrical energy in renewable energy systems. The reliability of a PMA exerts a direct effect on the operational safety, stability, and reliability of renewable energy systems. An effective fault diagnosis and adjustments for manufacturing processes (MPs) are vital for improving the reliability of a PMA. However, the state-of-the-art fault diagnosis methods are mainly used for single process parameters, extensive sample data, and automated manufacturing systems under real-time monitoring and are not applicable to a PMA with low levels of automation and high human factor-induced uncertainties. This study proposes a novel fault diagnosis approach based on a surrogate model and machine learning for multiple manufacturing processes of a PMA with insufficient training data due to human factor uncertainties. First, a surrogate model that correlated the MP parameters with the output characteristics (OCs) was constructed by a finite element simulation. Second, the quality performance of the OCs under different fault combinations with the mean or variance of the shift of the MP parameters as typical patterns was calculated by the Monte Carlo method. Finally, using the above computations as the training data, a fault diagnosis model capable of identifying the fault pattern of the manufacturing process parameters according to the OCs was constructed based on machine learning. This approach compensated for the inadequacies of traditional fault diagnosis methods with complex analytical models or numerous processing data. The effectiveness and potential applications of the proposed approach were verified through a case study of a rotary PMA in smart grids.