Research On Automatic Generation Method Of Geometric Tolerance Types Based On Machine Learning

With the release of a new generation of product geometric technical specifications,the position of geometric tolerances relative to dimensional tolerances has become more important.In most cases,designers must manually specify geometric tolerance types when designing mechanical products.For the same nominal geometric shape,different designers may specify different geometric tolerance types.With the continuous progress of artificial intelligence,machine learning has given new development impetus to traditional industries.In order to reduce the uncertainty of tolerance design and automatically realize tolerance specifications,this paper introduces machine learning into computer-aided tolerance design,and proposes a tolerance specification method based on machine learning.Specifically,the work completed in this article is as follows:(1)In response to the requirements of the new generation of product geometric technical specifications,this article splits the assembly and uses the geometric elements as the research object for modeling.The factors that affect geometric tolerances are comprehensively studied,including the geometric shape of the elements,matching relationships and benchmarks,and these factors are put into the database for machine learning training,and the realization of tolerance specifications is changed from rule-driven to data-driven.(2)A method of tolerance specification based on machine learning is proposed.The method first regards the past tolerance specification schemes as cases,and builds the cases into the tolerance specification database.In this paper,the database is used as a training set for training,and machine learning is used to turn the tolerance specification problem into an optimization problem,which greatly simplifies the complex process of tolerance specification.(3)Aiming at the specific scenario of tolerance specification,this paper performs feature engineering on the training set,including feature fusion,data up-sampling and other operations,so that the data can better reflect the essence of the tolerance specification problem.At the same time,this article also improved the training method.In the training process,multi-algorithm fusion and training cross-validation are used to prevent the algorithm from overfitting and greatly improve the performance of the model.(4)The geometric tolerance generation system is constructed using Python language combined with Scikit-learn library,and the working process of the system is analyzed with specific examples.