Research On Axial Stress Identification Method Of Ultrasonic Steel Components Based On Machine Learning

The axial stress of steel components is one of the important indicators for evaluating the safety of steel structures,Th e axial stress identification method based on ultrasonic characteristics is the mainstream method at present.However,the ultrasonic characteristic data is very susceptible to the influence of the internal microstructure of the steel component and environ mental factors,and the data is highly discrete,and the axial stress information inside the steel component is highly dependent on the ultrasonic characteristic data.Therefore,this paper introduces machine learning technology,combined with ultrasonic critical refraction longitudinal wave acoustic characteristics and ultrasonic transverse wave spectrum characteristics to establish an ultrasonic dual-feature model to reduce the influence of the internal microstructure and environmental factors of the steel component on the axial stress information data,thereby improving the ultrasonic steel component accuracy and robustness of axial stress recognition.The main research contents are as follows:Establishment of ultrasonic feature database for steel member axial stress recognition.Ultrasonic characteristic data is very susceptible to internal and external factors,and the establishment of the database needs to consider the characteristics of axial stress recognition.Therefore,this paper optimizes the extraction method of ultrasonic critical refraction longitudinal wave acoustic time and ultrasonic transverse wave spectral characteristics,and uses multi-point random sampling method to collect the axial stress information ultrasonic characteristic data of steel components,which reduces the influence of internal microstructure and environmental factors on steel components Influence of axial stress information data;The sample data was collected through the uniform experimental design method,and it was foun d that the characteristic gradient value of the axial stress uniform experimental design must be much larger than the sampling characteristic fluctuation value to ensure the learning efficiency of the database,and a test method for establishing an ultraso nic characteristic database was proposed.The establishment of a machine learning model based on Gaussian process regression algorithm.The selection of the covariance function of the Gaussian process regression algorithm needs to consider the characterist ics of the ultrasonic characteristic data and the algorithm for axial stress recognition.Therefore,this paper builds the ultrasonic critical refraction longitudinal wave acoustic time and ultrasonic transverse wave spectrum feature acquisition hardware system to obtain ultrasonic feature data,and conducts preliminary axial stress identification and comparison experiments to select the optimal Matern covariance function as the core of the Gaussian process regression algorithm,At the same time,Python was used to build a machine learning model and developed modules for feature extraction,database storage,database calling,feature input,and axial stress recognition.Axial stress identification of steel components based on ultrasonic feature model.On the basis of the above research,it is necessary to collect ultrasonic characteristic data and input the model to identify the axial stress.Therefore,this paper uses the proposed database establishment test method to carry out the loading test,and obtains the ultrasonic critical refraction longitudinal wave acoustic time characteristic,the ultrasonic transverse wave frequency spectrum characteristic and the ultrasonic acoustic time and frequency spectrum dual characteristic training database;The applicabi lity of the machine learning model based on the Gaussian process regression algorithm was verified by the single feature and dual feature model axial stress recognition of the steel plate;By comparing with traditional methods,it is found that the dual fe ature model of ultrasonic acoustic time and frequency spectrum can greatly improve the accuracy of axial stress identification;Through the identification of the axial stress of steel plates of common thickness,the robustness of the method of identifying the axial stress of steel members with dual characteristics of ultrasonic acoustic time and frequency spectrum based on machine learning is verified.