Dual slide ferrogram maker is used to prepare ferrogram slides, which are observed under a bi-chromatic optical microscope to capture the wear particle images. These wear particle images are classified into various categories to provide predictive analysis of wearing condition of gearbox setup. The purpose of this research paper is to train a convolutional neural network model based on MobileNet to identify wear particles present in the gearbox setup. Five categories of wear particles are trained with the help of Python for image classification. Wear particle analysis using ferrography provided information related to root cause analysis of wear failure. The trained CNN model has achieved accuracy of 96.34% to detect wear particles correctly out of 967 tested images. To get the effectiveness of the proposed model, it was successfully tested on the unknown dataset of wear particle images collected from different gearbox setup. The CNN model used in this work is specially trained for the five categories of wear particles generated from the gearbox. Classifying these kinds of wear particles leads to early detection of failure of gearbox due to wearing out. This model can be implemented in real-life industrial case studies for faster prediction of wear particles as a predictive maintenance strategy.