In the railway system, the turnout machine is a key device to ensure the safe and smooth operation of trains. The fault diagnosis of the S700K turnout machine is crucial for preventing accidents and maintaining railway operations. To solve the shortcomings of traditional diagnostic methods in speed and accuracy, a diagnostic model that integrates deep residual shrinkage networks (DRSN) and bi-directional long short-term memory (BiLSTM) is proposed. First, the power curve of the turnout machine is preprocessed; then, DRSN is used to automatically learn features from the preprocessed data and compress the data length, improving the speed of diagnosis. Its attention mechanism and soft thresholding reduce the impact of noise features, and the DRSN network structure helps overcome network degradation and overfitting problems; subsequently, the bidirectional structure of BiLSTM is used to capture complex relationships in time series data; finally, the Softmax classifier is used for fault classification. Simulation results show that the accuracy, precision, and recall of the DRSN-BiLSTM model all exceed 98.3%. While ensuring the efficiency of the training process, the accuracy of fault diagnosis of the DRSN-BiLSTM model is at least 1.47% higher than that of DRSN, DNN and other models, especially showing excellent robustness in a noisy environment.