Main bearing that plays the role of supporting and making the cutter to rotate and tunnel is the core part of TBM. Because of the harsh working conditions and complex changeful construction, the axial and radial load and environmental factors such as temperature of TBM main bearing are changing to make the fault of main bearing presenting randomness, and then the not easy identified fault may be produced. The traditional neural network model can not dynamic consider the cause of the reasons, parts and types, BP neural network fault diagnosis model based on fault reasons—signs matrix is presented in this paper. Firstly, the faults are screened through the fault reasons—symptom matrix, and the neural network structure is designed according to the screening result, then, the fault type is identified by way of model training. According to TBM main bearing fault symptoms data provided by a heavy enterprises practical engineering and MATLAB simulation validation, the feasibility and superiority of this model method are proved.