A hybrid machine learning approach to determine the optimal processing window in femtosecond laser-induced periodic nanostructures

Abstract

Surface nanostructuring could enhance surface properties such as strength, self-cleaning, anti-fog and anti-bacterial properties. Femtosecond laser-induced periodic surface structures (LIPSS) is a nanoscale structure created with laser technique. However, its quality is significantly influenced by the complicated interrelationship between the various laser processing and material parameters. Hitherto the selection of the appropriate laser parameters mainly depends on personal experience in conjunction with many time-consuming experimental trials. To have a simple, fast, and intelligent process, a hybrid machine learning method is proposed to determine the optimized processing window for femtosecond laser-induced nanostructures. Firstly, k-means clustering method was applied to automatically classify the laser-induced nanostructures into good and bad quality classes. Before clustering, dimensionality reduction methods were applied to reduce the high dimension of image data and to extract features. Different dimensionality reduction methods including principal component analysis (PCA), local linear embedding (LLE), t-random adjacent embedding (t-SNE) and transfer learning were explored. Transfer learning showed a much better result compared with other dimensionality reduction methods. Transfer learning VGG19 model achieved the highest accuracy of 90.6 %. After clustering, the image was labelled as good and bad clusters accordingly. The labeled image was trained using artificial neural network (ANN), random forest (RF), decision tree (DT), support vector machine (SVM), k-nearest neighbors (KNN) and Naive Bayesian Classifier (NBC) algorithms for the prediction of laser processing results. The results show that DT gives the best accuracy of 96.7 %. Finally, an optimal laser processing window for femtosecond laser-induced nanostructures was determined.