Three-phase inverters are extensively utilized in industry, where the key to stable operation lies in accurate and rapid fault diagnosis. Given that inverters may operate under diverse operating conditions and different system configurations, fault diagnostic methods must possess robust generalization capabilities. This paper introduces a data-driven fault diagnostic method for inverters, applicable across different control strategies. The method employs Fast Fourier Transform (FFT) with motor rotational speed to precisely extract phase current features. Subsequently, the dimensionally reduced features are fed into an Extreme Learning Machine (ELM) network for fault detection and diagnosis. The key to this method is selecting vital features that are consistent across different control strategies. This allows a model trained under one control strategy to be directly applied to systems under other control strategies, thus significantly enhancing the model’s generalization capability. The substantial reduction in the number of features reduces the computational demands and complexity of the algorithm, thereby greatly improving the speed of training. Experiments and analyses conducted on three-phase induction motor drive systems under different control strategies confirm the method’s high accuracy, excellent generalization capability, and rapid training speed.