Research On Crack Detection Method Of Infrared Thermal Imaging Steel Plate Based On CNN

Regular inspection of steel plate damage in industrial equipment is essential to ensure safe operation.The detection results of the existing steel plate crack detection methods are mostly dependent on manual evaluation.And the detection efficiency of cracks under complex surfaces is relatively low.This paper studies a method of crack detection of infrared thermal imaging steel plate based on convolutional neural network.The specific research contents are as follows:(1)The main techniques of steel plate crack detection are studied.And compare the advantages and disadvantages of each technology.Among them,the infrared thermal imaging non-destructive testing method is characterized by its large detection area,high detection efficiency and strong visualization effect.(2)Aiming at the abnormal emissivity of the steel plate surface,this paper proposes the horizontal heat transfer method to thermally excite the surface of the steel plate.The heat transfer mechanism is studied.A rolling electric heating device is developed as an infrared heat shock source.This method suppressed the interference of abnormal surface emissivity to the detection results to some extent.(3)The temperature variation law is studied.The differences of temperature and time between different cracks and non-crack regions are compared.The influence of heating direction on the detection results is studied,and the differences in temperature characteristics of different crack defects are explained.(4)The infrared thermal image preprocessing algorithm is studied.A hybrid filtering algorithm combining morphological filtering and Gaussian filtering is proposed.The histogram equalization is used to enhance the contrast of the image,and a Canny edge detection algorithm suitable for infrared thermal image is optimized.(5)Many infrared thermal images containing cracks are collected and made into data sets.Aiming at the problems of the Faster R-CNN network model in the dataset of this paper,the VGG-16 network is used for migration learning,and the feature maps of multiple levels in the network are merged,and the anchor frame selection scheme of the RPN network is adjusted.This paper is closely related to the experimental results from the theory,combining infrared thermal imaging non-destructive testing methods with computer vision algorithms.Finally,the effectiveness of the proposed method is verified by a large number of experiments.The accuracy of crack detection on the test set reaches 96.82%.This paper explores a more efficient and intelligent method for steel plate crack detection,and lays a good foundation for practical use on site.