Convolutional neural network based bearing fault diagnosis of rotating machine using thermal images

The bearings are the crucial components of rotating machines in an industrial firm. Unplanned failure of these components not only increases the downtime, but also leads to production loss. This paper presents a non-invasive thermal image-based method for bearing fault diagnosis in rotating machines. Thermal images of rolling-element bearing in six conditions have been considered, including one healthy and five faulty conditions, and then a comparison based on classification performance has been done using shallow and deep learning approaches incorporating artificial neural network (ANN) and convolutional neural network (CNN). The CNN used in this work is based on the LeNet-5 structure and has proved to be a better than the ANN. It has been concluded that infrared thermography can be used in a non-contact way to automatically identify the faults that help to detect early warnings, irrespective of speeds and hence ensures reduced system shutdowns causing by bearing failure.