Improved roughness measurement method using fiber Bragg gratings and machine learning

•Roughness is measured using fiber Bragg gratings and machine learning.•Eight strain features of FBGs are used as inputs of machine learning for sensing.•Polynomial regression with principal component analysis is used for prediction.•Improved decision tree algorithm has a high roughness classification performance. [Display omitted] An improved roughness measurement system is proposed that uses fiber Bragg gratings (FBGs) and machine learning. One sensor is a stylus profiler fabricated from FBG and silicone, and the other is a strain sensor fabricated from FBG. Eight statistical strain features of FBGs are extracted for intelligent sensing. Experimental results clearly showed that most strain features changed monotonically with the roughness and could be used individually to measure roughness. Roughness prediction and classification were realized with a polynomial regression algorithm using principal component analysis and a decision tree, respectively. The polynomial regression algorithm had a better performance compared with support vector regression. Corresponding optimized mean square error and coefficient of determination for the roughness prediction were 0.0035 µm and 0.9950, respectively. The macro-averaged F1 score for roughness classification after optimization was 0.98932.