Aiming at the actual needs of on-line monitoring and intelligent fault diagnosis and early warning of reciprocating compressors, a normalization process of indicator diagrams of reciprocating compressors and an intelligent fault diagnosis process of reciprocating compressors based on convolutional neural network machine learning are proposed. Using isoparametric normalization method to process the indicator diagram can eliminate the factors that cause the change of the indicator diagram, such as process changes and environmental changes. The processed samples are conducive to the subsequent neural network intelligent recognition, with higher accuracy and universality. Through convolution neural network classification and recognition, data self-learning and fault intelligent diagnosis of reciprocating compressor can be realized. Through simulation and test fault data verification, 480 test samples are tested, and the diagnostic accuracy is 97.99%. Aiming at the confusion of some faults, the frequency domain signals of bonnet vibration are further fused to achieve 100% classification diagnosis, which can be effectively applied to the fault identification and health management of reciprocating compressors.