| The basic force and thermal test of materials is to conduct different types of mechanical and thermal property tests for energy-containing materials,including over 40 types of compression test,tearing test,thermal diffusion coefficient test,and linear expansion coefficient test.Various types(such as numerical values,text,images,audio,etc.)and large-scale test data will be generated in the test,and efficient management and utilization of such massive test data is the key to improving test efficiency and accelerating material research and development.Currently,in the management and use of this type of test data,there are mainly the following problems:(1)Due to human error,systematic error and other reasons in the test process,abnormal test data will be generated,which will affect the quality of the test report.(2)In the automatic generation of test reports,various standards are required to provide knowledge of test methods,test equipment,data analysis methods and other knowledge to support the intelligent processing of reports.(3)For the accumulated force-thermal test data,a data mining tool for material performance prediction is lacking to mine the characteristics of the test data and predict the performance of energy-containing materials.To address the above issues,this paper focuses on the following:(1)Abnormal test data screening.For abnormal test data caused by human error,system error,test equipment,and data variation.The k-means based screening method for abnormal test data was adopted to select the point with the greatest tightness between data as the initial clustering center to solve the problem of randomly selecting the initial clustering center,and finally,according to the clustering results,the abnormal data points that deviate from the overall test data were identified.Verification was performed on the vibration and compression test data to screen out abnormal test data and ensure the quality of the test data.(2)Construction of knowledge base for force-thermal tests.Aimed at providing the underlying knowledge support for test report generation and data mining,the scattered test knowledge is integrated.The test standard knowledge graph is constructed based on BERT+ Bi LSTM+ CRF.Firstly,the test standard documents such as national standards are formatted and manually annotated,and then the entity-attribute extraction model of the standard documents is trained to extract the entities and attributes in the test standard documents to form a triad,and finally the triad is stored in the database after knowledge verification and visualized and managed by using ECHARTS.And the quality evaluation was conducted on the test standard knowledge map,and the F1 value of the comprehensive query reached 0.93.(3)Material force and thermal test data mining tool.Aimed at providing analysis functions for force-heat test data and mining the characteristics of test data.The test data characteristics mining tool was developed based on python,providing pre-processing of raw test data and more than ten machine learning methods such as classification,clustering,regression and association rules,as well as visual display of data mining results.And the Brazilian experiment with more than three thousand sampling points was used for data mining to verify the preprocessing,algorithm selection and visualization functions of the tool.(4)Intelligent processing platform for force and thermal test data of material foundation.For the intelligent processing of force thermal test data,the TOG AF standard design platform architecture was adopted,and test data structured extraction tool,test data intelligent processing system,test data characteristic mining tool,etc.were developed.It has realized functions such as automatic data capture,intelligent report generation,material performance mining,and test standard knowledge map management for more than 40 tests,has been successfully applied in a certain research institute. |
PDF Full Text Request