Affected by interference factors such as Gaussian noise, the traditional methods have the problems of inaccurate diagnosis results of unsteady vibration signals, high uncertainty of fault diagnosis, and low overall fault diagnosis accuracy. In this paper, a fault diagnosis algorithm of vehicle internal combustion engine based on instantaneous speed and machine learning is proposed. The instantaneous speed is measured by the hardware method. According to the processing results of instantaneous speed, the unsteady vibration signal of the vehicle internal combustion engine is analyzed, and the principal components of unsteady vibration are separated to suppress the interference of Gaussian strong noise. The running state of the vehicle internal combustion engine is identified by the wavelet transform method. According to the identification results, the fault diagnosis of the vehicle internal combustion engine is realized by the twin support vector machine classification algorithm in machine learning. The experimental results show that the minimum uncertainty coefficient of fault diagnosis in this algorithm is 0.08, the accuracy of the unsteady vibration signal diagnosis is higher, and the overall accuracy of fault diagnosis is lower.