Cleanliness prediction of rusty iron in laser cleaning using convolutional neural networks

As laser cleaning is a nonlinear physical process, modeling it is always a challenging problem for technicians and researchers. In this paper, we successfully introduce the use of computer vision technology to predict the cleanliness of rusty iron in laser cleaning. Statistical approaches, i.e., machine learning methods, can be used to construct a model of the process directly from images of the laser cleaning samples, which may avoid understanding the physical principle of laser cleaning. To apply a convolutional neural network (CNN), we not only build a laser cleaning dataset, but also propose an effective cleanliness evaluation index. We introduce the details of them on the rusty irons, for example. Then, we use the CNN to model laser cleaning successfully. A LeNet-5 network achieves a final accuracy of 82.45% while predicting the cleanliness, which shows the potential of this approach for modeling a nonlinear process from experimental data. Thus, for a given pre-cleaned image and laser parameters, laser cleaning cleanliness can be predicted.