Research And Implementation Of Weld Defects Detection Based On Faster-RCNN

With the development of science and technology,pressure vessels have been widely used in construction,chemical,wine,food,petroleum and other industries.The detection of weld defects in pressure vessels is an important part of ensuring the quality of pressure vessels.If this can be solved The problem of defect detection in the link can not only effectively improve the production quality of pressure vessels,but also effectively avoid the harm to the environment and humans caused by the leakage of harmful gas and liquid caused by quality problems of pressure vessels.In addition,X-ray imaging technology has the advantages of low cost,easy production of X-ray films,short imaging time,and high weld defect positioning accuracy.Therefore,X-ray inspection is currently the main method for weld defect detection.For the detection of X-ray weld image defects,the initial stage is mainly realized by manual inspection.Manual inspection has some disadvantages due to subjective reasons,such as time-consuming,low efficiency,and low accuracy.With the development of deep convolutional neural networks,the method of using deep convolutional neural networks to detect weld defects gradually replaces manual inspection.In order to improve the efficiency and accuracy of weld defect detection,this paper proposes a weld defect detection algorithm based on Faster-RCNN(Faster-Regions with Convolutional Neural Networks).The main work of this paper is as follows:(1)Due to the large sample size and unobvious sample target in the pressure vessel weld X-ray image data set,this will seriously affect the detection effect of the deep convolutional neural network.In response to these problems,the data set is preprocessed.This article organizes,preprocesses and annotates the original weld seam data set to create a unified data set.(2)In the weld defect detection part,this paper proposes a weld defect detection algorithm based on Faster-RCNN.First,analyze the detection principle of Faster-RCNN algorithm model in detail.In order to further improve the detection accuracy of Faster-RCNN,an improved method of Faster-RCNN algorithm model is proposed:combined with the characteristics of the data set,first introduce an attention mechanism into the network to improve network performance,and at the same time adopt a self-regular non-monotonic activation function(Mish)Improve the nonlinear mechanism of the network,and fine-tune the hyperparameters of the RNP(Region Proposal Network)network for the detection of small-size weld defects.(3)Faster-RCNN algorithm model before and after improvements and SSD300(Single Shot Multi Box Detector)algorithm model were tested and compared on the pressure vessel weld data set.This paper has obtained phased research results through experiments.The experimental results show that the Faster-RCNN algorithm is better than the SSD300 algorithm for the detection of weld defects.The improved Faster-RCNN algorithm has a better detection accuracy(mAP)for weld defects 0.86,which verifies the effectiveness of the improved scheme.