Intelligent fault diagnosis is a promising tool, which can process the acquired signals quickly and efficiently and provide accurate diagnosis results. In traditional intelligent diagnosis methods, features are extracted manually based on prior knowledge and diagnostic knowledge. Based on this, this paper proposes two learning methods for machine intelligence diagnosis. The first method consists of two stages. In the first stage, unsupervised sparse filtering is used to learn features directly from mechanical vibration signals. In phase 2, Softmax regression is used to classify the health of industrial control systems. The second method uses the fault diagnosis and prediction model of automatic control system based on deep learning technology, converts the time series data into image data, and processes the image data through convolutional neural network. The two methods are verified, and the experimental results show that both have high accuracy and stability. The results show that the proposed method achieves high diagnostic accuracy and is superior to the existing motor bearing data set methods. Due to the adaptive learning feature, the method reduces the need for manual labor and makes it easier for intelligent fault diagnosis to handle big data.