Abstract: The coupling of bending vibration and torsional vibration often exists in the actual operation of rotating machinery. This paper considers the bending and torsional coupling of unbalanced rotors under different complex conditions, and utilizes the advantages of deep learning technology to construct a diagnosis model based on one-dimensional convolutional neural networks. An intelligent fault diagnosis method for handling the bending, torsion and bending torsional coupling vibration of the unbalanced rotors is proposed. The influence of data input type and L2 regularization on the diagnosis is analyzed, and the diagnosis model is optimized to improve the diagnosis accuracy. The research results indicate that this method can realize intelligent diagnosis of single or multiple composite faults when bending torsional coupling vibration occurs at different speeds, and achieve better diagnostic results than other methods.