Machine Learning Prediction Model For B-pillar Drawing Defects Under Material Properties Fluctuation

Along with the rapid economic growth and continuous improvement of the modern industrial system,China has recently advanced in automotive industry by leaps and bounds and will continue to maintain a growing trend for a long time in the future.In the mean time,problems in environment,resource and road safety etc will be brought about with the development of the automotive industry.And the study of the lightweight of automobile matters most in order to solve the problems mentioned above.The extensive application of Advanced High Strength Steel to car body is one of the ways to realize lightweight of automobile.Compared to ordinary ones,the Advanced High Strength Steel is harder to be formed.Moreover,CAE simulation analysis software can be applied to predict the possible defects of in the forming process of Advanced High Strength Steel.However,with the automobile production enterprises’ requirements for the quality of body parts being higher and the accelerating speed of product update,some shortcomings of tra ditional CAE software are gradually highlighted.For example,stamping simulation of more complex parts requires much time and computing power,and too many parameters need to be set before simulation.In order to make up for the traditional CAE software,a stamping quality prediction system for car parts which is more efficient,occupies less resources,has better accuracy and practicability is needed.Furthermore,the continuous development of computer software and hardware provides the possibility for th e successful invention of this prediction system.Based on the basic theory of machine learning,two types of problems that need to be solved to realize quality prediction of drawing are established,among which the prediction of defect value belongs to re gression problem and the prediction of defect position belongs to classification problem.The yield strength and tensile strength of the Advanced High-strength Steel TRIP780 are selected as features.The maximum thinning rate and maximum failure value of t he automobile B-pillar after drawing,the maximum thinning parts and maximum failure parts are used as labels.The original data are obtained through Auto Form software simulation,and the data preprocessing is completed by means of feature engineering,etc.,to establish the original data set.The original data set is divided into training data set and test data set.The model was trained by using the training data set,and the model was optimized by means of grid search,cross-validation and model regularization method,and the B-pillar drawing-defect prediction model was established.Through the test data set,the correctness,reliability and prediction accuracy of the model are verified.When the TRIP780 high strength steel is used as the material for draw ing of automobile B-pillars,the model can still accurately predict the main defect value and defect location in the drawing process of B-pillar,and judge whether the forming quality of drawing parts is qualified even if the material property parameters o f TRIP780 fluctuate,providing a reference for designers,helping them early in the design of the corresponding parts to set aside enough safety margin,avoiding the defects of the practical production.Finally,the model is used to predict the maximum thi nning rate and maximum failure value of the 15000 sets of yield strength and tensile strength combinations of TRIP780 material.The safety material fluctuation range of TRIP780 material used for a certain type of B-pillar drawing was established.As long a s the fluctuation of material properties parameters of TRIP780 is controlled within this range,steel manufacturers can obtain qualified B-pillar drawing parts in actual production.