For systems using rotating machinery, diagnosing the faults of the rotating machinery is critical for system maintenance. Recently, a machine learning algorithm has been employed as one of the methods for diagnosing the faults of rotating machinery. This algorithm has an advantage of automatically classifying faults without an expert knowledge. However, despite a good training performance of the deep learning model, there remains a challenge of performance degradation arising from noise when the model is applied in a real environment. In this study, to solve this problem, we identified the faults of a rotating machinery after applying the continuous wavelet transform (CWT) and then we extracted the images for detecting the faults of rotating machinery and apply them to the convolution neural network (CNN). Subsequently, we compared it with a commonly used artificial neural network technique according to load and noise. When we added the white noise from 1dB to 20dB to vibration signal, the proposed method converged to 100% accuracy from 8dB at no load, at 10dB at presence of load. we verified that the proposed method improved the performance in diagnosing the faults of rotating machinery.