Abstract: In order to reduce the influence of manual subjective intervention on the diagnosis results in traditional fault diagnosis methods, a fault diagnosis method was proposed based on sparse auto-encoder (SAE), using an unsupervised learning method to extract the fault characteristics of hoist monitoring data. First, the failure mechanism of the brake system was analyzed, the monitoring data under the normal operation and failure simulation state of the hoist were collected, and a failure diagnosis datum set was generated. Then a SAE fault diagnosis model was established and optimized based on the Dropout and Adam algorithm. Finally, the performance of the model was tested using a test data set. The experimental results show that the presented method can better avoid the training error of sparse data, reduce the over-fitting phenomenon, and reduce the influence of the local optimum of sparse data. The average classification accuracy of fault types can reach up to 94%, the presented method can realize mine hoist fault diagnosis effectively.