Laser Ultrasonic Defect Detection Based on Local Outlier Factor and Isolated Forest
In response to the issue of artifacts in the maximum amplitude images in laser ultrasonic (LU) defect detection,principal component analysis (PCA) was integrated with two unsupervised machine learning algorithms including local outlier factor (LOF) and isolated forest (IF) to perform unsupervised anomaly detection on LU data. Firstly,the PCA algorithm was used to reduce the dimensionality of the LU data,simplifying its complexity. Secondly,the LOF and IF algorithms were employed to identify outliers in the data,and the thresholds for these outliers were determined using the cumulative distribution function and kernel density estimation. Finally,a comparison of the detection results from the LOF,IF algorithms,and the maximum amplitude images revealed that the LOF algorithm offered superior defect detection precision and a lower false positive rate.
万方数据
维普
