Intelligent fiber width detection using visible light and machine learning.

Area of Science:

• Engineering
• Computer Science
• Materials Science

Background:

• Accurate width measurement is crucial for quality control in manufacturing.
• Traditional methods can be time-consuming and lack precision.
• Developing automated, intelligent measurement systems is a key area of research.

Purpose of the Study:

• To develop an intelligent system for precise width measurements.
• To utilize optical fiber sensing, visible light detection, and machine learning for this purpose.
• To evaluate the performance of an ensemble machine learning model for width variation detection.

Main Methods:

• An ensemble machine learning model combining random forest regression, k-nearest neighbors regression, and decision tree regression was employed.
• Input features were derived from statistical analysis (first-order statistics, GLCM, DLS) of light spot images transmitted through optical fibers.
• Feature dimensionality reduction and hyperparameter optimization using quasi-Monte Carlo and tree-structured Parzen estimator algorithms were performed.

Main Results:

• The optimal ensemble model achieved a mean square error of 0.2084 mm² and a coefficient of determination of 0.9997 after five-fold cross-validation.
• The feature importance analysis guided the reduction of model input features.
• The developed system demonstrated high accuracy and reliability in width variation detection.

Conclusions:

• The proposed intelligent width measurement system is highly accurate and reliable.
• The ensemble machine learning approach effectively detects width variations using optical fiber sensing.
• This method holds significant potential for advancing intelligent manufacturing processes.