Intelligent fiber width detection using visible light and machine learning

Intelligent width measurements were obtained using optical fiber sensing, visible light detection, and machine learning. An ensemble machine learning model was used to intelligently detect width variations by identifying visible light spots transmitted through optical fibers. In the ensemble model, random forest regression and k-nearest neighbors regression were used as base learners, and decision tree regression was used as the meta-learner. The input features were generated by applying arithmetic operations to the first order statistics, gray level co-occurrence matrix, and gray level difference statistics features extracted from the light spot images. The dimensionality of the model input features was reduced according to the feature importance. The quasi-Monte Carlo and tree-structured Parzen estimator algorithms were combined to optimize the model parameters. For the optimal ensemble model, the mean square error and coefficient of determination after five-fold cross-validation were 0.2084 mm² and 0.9997, respectively. The detection method has considerable potential for applications in the field of intelligent manufacturing.