Shape characterization and depth recognition of metal cracks based on laser infrared thermography and machine learning
Haoze Chen,Zhijie Zhang,Wuliang Yin,Guangyu Zhou,Luxiang Wang,Yanfeng Li,Chenyang Zhao,Chao Wang
DOI: https://doi.org/10.1016/j.eswa.2023.122083
IF: 8.5
2023-10-21
Expert Systems with Applications
Abstract:Due to the operation of modern industrial equipment under high pressure, high speed and high load conditions, cracks inevitably appear on the surfaces of metal components. Crack detection on metal surfaces is increasingly becoming a focal point in the field of non-destructive testing. In order to locate and quantitatively analyze surface cracks on metallic materials as early as possible to reduce the cost in industrial production, this paper achieves rapid localization of surface cracks by establishing a line laser thermography scanning system. An unsupervised edge detection algorithm of principal component analysis combined with K-means (PK) is proposed to achieve accurate crack edge detection by combining with a point-pulse laser thermography system. The accuracy of the PK algorithm is confirmed by comparison with other edge detection algorithms. After that, accurate identification of crack depth is achieved by selecting temperature features through particle swarm optimization algorithm and optimizing hyper-parameters of support vector machine and k-nearest neighbor through Bayesian search method. The experiments show that the best recall and accuracy of PK for crack detection are 82.9% and 86.2%, respectively; Combining the particle swarm optimization algorithm with the BS method, the mean square error of SVR and KNN for crack depth recognition is 0.092 and 0.089, respectively.
computer science, artificial intelligence,engineering, electrical & electronic,operations research & management science