The president paper contributes to the development ways of expectation the failure gear. In this research, we are looking for systems that deal with plastic gear in commercial product. The challenge lies in how detect plastic gear failure which is extremely hard to find the amplitude of vibrations, on the contrary of metal gears. For that, it draw upon deep learning algorithm called CNN (Convolutional Neural Network) to detect the failure more efficiently. CNN has a characteristics, automatically feature extraction without expert advice. Essentially, CNN has successfully been applied to analyzing visual imagery. Depending on CNN characteristics, the time domain signal is transferred to frequency domain by use FFT then convert it as an images due to use it and classify it through CNN. One of the CNN properties is to visualize processes​ that happened in hidden layers which is made classification possible. At beginning of learning layers, it can not distinguish between them, however, at the end of learning process can differentiate between them effectively.