Research On Interpretable Prediction Of Resistance Welded Joint Performance And Welding Defect Identification Metho

Resistance welding is commonly used in the connection process of thin sheet structures of lithium battery packs.With the development of intelligent welding technology,higher requirements are placed on the welding quality of lithium battery packs.The traditional destructive testing method of manual sampling after welding has the problems of low efficiency and poor feedback timeliness.The use of resistance welding quality monitoring technology can achieve non-destructive testing of welding quality.In order to realize the online full inspection of welding quality efficiently and intelligently,the research on sensor monitoring,joint performance prediction and welding defect identification method is carried out.First,based on modern sensing and detection technology,an embedded multi-source information monitoring device for resistance welding was developed,the establishment of welding test experimental platform to carry out different working conditions under the welding test,multi-source information of electro-force process was collected and processed,and the influence of welding process parameters on the performance of welded joints under different working conditions was studied,which laid a foundation for subsequent research.Then,based on the welding test data under standard working conditions,the characteristic parameters of multi-source information in the electric-force process were extracted,and the welding feature set was constructed.A method based on the Nondominated Sorting Genetic Algorithm-III to optimize the Explainable Boosting Machine was proposed.The performance prediction model of welded joints was constructed,and the performance of the model was explained by the interpretable reinforcement machine.The experimental results showed that the proposed method can improve the prediction accuracy and the effectiveness of model interpretation compared with the traditional algorithm,and realize the online adjustment process.Finally,in the case of unbalanced welding defect samples,based on the welding test data under fluctuating working conditions,the characteristic parameters of the electrical signal process information were extracted,and the unbalanced welding defect feature set was constructed.A mixed sampling method of Majority Weighted Minority Oversampling Technique and Tomek links was proposed.Based on the Non-dominated Sorting Genetic Algorithm-III to optimize the Explainable Boosting Machine to build a welding defect recognition model.The experimental results showed that the proposed method can improve the accuracy of defect recognition of unbalanced data.The proposed method provides theoretical support for the non-destructive testing of the performance prediction and defect identification of the welded joint of the power lithium battery sheet connection,and has practical significance for guiding the welding process.