Identification Of Characteristic Areas And Parameters Of Mechanical Test Fractures Intelligent Evaluation Study

The Charpy Impact and Quasistatic Fracture Toughness tests are two important mechanical tests for assessing the quality of pipeline steel,and this paper focuses on the identification of the characteristic areas of these two mechanical specimen fracture images and the intelligence of the associated characteristic parameters.For the evaluation of Charpy impact specimens,our current standard gives a variety of assessment methods with the comparison method and vernier caliper measurement as the main measurement method,so as to calculate the fibre section rate of the Charpy impact specimen fracture,as an indicator to assess the quality of steel.For the quasi-static fracture toughness test,China's current standard gives a nine-point method as the main measurement method of assessment,so as to achieve the quasi-static fracture toughness specimen of the average crack extension length calculation,as an indicator of the assessment of the quality of steel.Due to the complexity of the fracture section characteristics and the nature of the industrial assessment,the current industrial assessment has a series of drawbacks such as the high professional quality of the assessor,the difficulty of reproducing the assessment results and the tedious assessment process.With the continuous development of deep learning technology,it is possible to use deep learning technology to realise the above assessment process,thus freeing the technicians from the repetitive and tedious assessment work.The research idea of this paper is: based on the convolutional neural network technology in deep learning,multiple typical convolutional neural networks are trained using pre-produced data sets of specimen fracture images from Charpy impact test and quasi-static fracture toughness test,and the trained models are used to identify the feature regions in the specimen fracture images of the two mechanical tests respectively,and the optimal model is selected based on the identification results,and the optimal model is Based on the recognition results,the optimal model is selected,and the optimal model is optimised to further improve the recognition effect.On this basis,the subsequent calculation and evaluation of the characteristic parameters of the specimen steel is completed,and a complete system for the recognition of the characteristic regions and intelligent evaluation of the parameters of the mechanical test fractures is finally established,providing a more efficient and accurate evaluation method for the recognition of the characteristic regions and calculation of the characteristic parameters of these two types of mechanical tests.The details of the research are as follows.1)The images of the fractures of the two mechanical test specimens were collected in the field using professional equipment,and a series of pre-processing work such as acquisition,annotation and data enhancement of the data set in the early stage of this study were completed to produce the image data sets of the fractures of the two mechanical test specimens.2)The U-Net network with VGG-16 as the backbone feature extraction network,the PSPNet network with MobileNetv2 as the backbone feature extraction network and the Mask R-CNN network were used to train the datasets respectively,and the recognition effects of the above typical network models on the two types of mechanical test image datasets were verified by adjusting some hyperparameters.The results show that the U-Net network has a good recognition accuracy for the fracture images of the two types of mechanical tests.3)In order to improve the recognition rate of the U-Net network for the two mechanical test fracture images and optimise the recognition effect of feature region edges,this paper introduces a hybrid attention mechanism on the basis of the U-Net network,and further improves the accuracy of the network for the two mechanical test fracture images without affecting the recognition speed by adding an attention module to the enhanced feature extraction network to achieve the recognition of The network achieves accurate recognition of intersection regions,edge regions,discrete regions and small regions.The improved algorithm model has been evaluated by experts in the field and the prediction error of the feature regions in both types of mechanical test fracture images is within 5%.4)In order to evaluate the toughness of the two mechanical test specimens and the static crack extension ability,this paper proposes a depth learning-based method for calculating the fibre section rate of Charpy impact specimens and a depth learning-based method for measuring the crack extension length of quasi-static fracture toughness specimens.The method proposed in this paper has been evaluated by experts in the relevant fields,and the errors in calculating the characteristic parameters of both types of mechanical specimens are within 3%.5)In order to ensure the ease of use of the mechanical test fracture region identification and parameter intelligent evaluation system,simplify the evaluation process and reduce the technical threshold of this research,intelligent identification and analysis software for Charpy impact test and quasi-static fracture toughness test has been developed,and clear images of specimens collected by specific equipment have been used to improve the efficiency of specimen evaluation while ensuring the accuracy of specimen identification and evaluation results.This provides an end-to-end solution for the practical deployment of this research in industrial projects.