Because the traditional machine learning algorithm, which is limited by input variables and easy to have or not to learn, a random forest algorithm with good accuracy is proposed to detect the fault of car bearing without limits of input variables. Firstly, the collected sample data was filtered to suppress the noise in the signal. Secondly, random forest algorithm was used to classify and label the collected time domain signals and to determine the signal sequence containing fault information. And then the signal was converted to the frequency domain, and the signal in the frequency domain was detected by using random forest to determine the fault frequency. Finally, the experimental data were collected to verify the proposed algorithm. The results show that the random forest algorithm has fast response speed and high accuracy as compared with the traditional machine learning algorithm.