Aiming at the problem that diesel engine misfire faults couldnot be accurately diagnosed when there were fewer data samples in practical applications, a misfire fault diagnosis method based on extreme learning machine (ELM) was proposed, which was verified by SC5S122D instantaneous speed signal of diesel engine. The acquired instantaneous speed signal was extracted in the time-frequency domain, and then the feature selection was carried out according to the Pearson correlation coefficient, and the feature parameter set composed of selected feature was used to diagnose the diesel engine misfire fault. Finally, the small sample dataset was divided into four cases and used to train ELM to evaluate the diagnostic effect of ELM when the data sample was small. At the same time, the small sample data was expanded, and ELM was used to diagnose diesel engine misfire faults on the expanded dataset. Analysis of the experimental results demonstrates that the ELM exhibits superior performance in misfire fault diagnosis compared to both PNN and BPNN, as evidenced by higher accuracy, precision, recall, and F₁ score metrics. Consequently, ELM proves capable of achieving accurate and effective diesel engine misfire diagnosis even with limited training data samples.