Research On The Identification Of Welding Defects In Aluminum Alloy Parts Based On Arc Sound And Electrical Signals

The welding process of melting gas shielded welding is accompanied by fluctuations in the welding electric signal and arc acoustic signal,and the identification of welding defects can be achieved by extracting the feature parameters related to the welding defects from the signal.The feature set consisting of welding electrical signal and arc sound signal features is more comprehensive in terms of information description than a single feature,but there are redundant features in the high-dimensional feature set,which need to be dimensionally reduced.In addition,the actual available samples of weld defects are less than the well formed samples,and the identification of weld defects is an unbalanced data classification problem.Therefore,Thesis takes a typical aluminum alloy part,a new energy vehicle battery pack chassis,as the research object,extracts and selects its acoustic and electrical signal features,and constructs a welding defect recognition model considering sample imbalance,and the main work is as follows.First,in response to the lack of comprehensiveness of a single feature to describe the information of the welding process,this paper extracts the acoustic and electrical signal features of the welding process to achieve a multi-angle analysis of the welding process.For the non-smooth characteristics of the electrical signal,the use of adaptive noisecomplete empirical modal decomposition and Pearson correlation coefficient method to select the intrinsic modal function closely related to the welding defects,and extract its time-frequency features,information entropy features totaling 44 dimensions;wavelet threshold denoising of the acoustic signal,and extract its Meier frequency cepstrum coefficients and its difference coefficients totaling 36 dimensions.Then,based on the extracted feature parameters to construct the high-dimensional feature sets,and for the problem of redundant and noisy features in the feature space,a multi-objective optimization-based feature selection method is proposed for dimensional simplification of the high-dimensional space.The method analyzes the relevance,redundancy and complementarity of feature sets based on information theory to construct a multi-objective feature selection model.Then the third generation of non-dominated ranking genetic algorithm is improved and the dimensionality reduction of the highdimensional feature sets is achieved based on this solution model.The algorithm utilizes an adaptive probabilistic variation operator based on information theory to guide the variation process of the population in order to reduce the influence of invalid features and improve the convergence efficiency.The experimental results show that the proposed algorithm can obtain better classification accuracy while reducing the feature dimensionality than the classical algorithm.Finally,an unbalanced data classification method based on the optimized oversampling technique of the cuckoo search algorithm is proposed in consideration of sample imbalance,and a welding defect recognition model is constructed based on this to achieve welding defect recognition.The method uses the cuckoo search algorithm to optimize both the sample oversampling rate and the support vector machine hyperparameters to avoid the influence of noisy samples in the oversampling process;and based on the optimal sampling rate combination,a few classes of samples are oversampled and the new sample data are added to the original training set to form a new training set;finally,a support vector machine with optimized parameters is used for sample training to construct a welding defect recognition model to realize the recognition of welding defects.Finally,the effectiveness of the algorithm was verified through experiments.Thesis provides some basic ideas for effective feature mining of the acoustic and electrical signals of the welding process of aluminum alloy parts,and provides methodological support for effective improvement of the identification of welding defects in aluminum alloy parts.