The wide range of applications bearings have in various mechanical and electrical equipment, with failure of its core component having serious consequences. Nevertheless, manual adjustment is the most common parameter adjustment method; however, this has its drawbacks, such as relying on existing data, being prone to local extremes, not being able to reach global extremes, and requiring a large amount of energy. For this reason, this project intends to combine wavelet transform and convolutional neural network, and use ADAM stochastic optimisation algorithm to optimise it, to establish a new bearing fault diagnosis model. To begin, a continuous wavelet transform (CWT) was employed to evaluate the vibration signal in time-frequency and draw out the feature details for training and assessment. Subsequently, a convolutional neuralnetwork was utilized to adjust its parameters, and the ADAM optimisation algorithm was utilized to modify its parameters. Examining the proposed method by taking the bearing database of Western Reserve University as an example, it is compared to the traditional intelligent fault diagnosis algorithm for bearings. Demonstrating its effectiveness through example analysis, this thesis presents a novel innovation: the integration of convolutional neural network and wavelet transform with ADAM technology to create a full rolling bearing fault diagnosis system.