A parallel strategy for predicting the quality of welded joints in automotive bodies based on machine learning

Abstract

Due to the complexity of the resistance spot welding process, it is still a challenge to accurately know the operating status of the welding robot under the current parameter settings and to assess the welding quality of electrode caps under different types of plates in real time with large data sizes. To solve this problem, this paper classifies the overall data set and proposes a parallel strategy method for predicting the quality of weld joints using machine learning for subsets of the data with different distribution patterns. Firstly, the PCA dimensionality reduction model was used to set the number of principal components to reduce the dimensionality of the welding process feature value dataset and reduce the difficulty of classifying the data subgroups, and the elbow method was used to set the number of clustering centers to complete the classification of the sub-datasets by applying the k-means model on the basis of the dimensionality reduction data. Finally, the feature parameters of each sub-dataset are used as input for machine learning, and a parallel prediction strategy for weld joint quality is developed based on the data distribution characteristics of each sub-dataset. The test results show that the model in this paper outperforms the static BP neural network in predicting the quality of all types of welded joints, the machine learning parallel strategy tailored to the characteristics of the data population works well with more complexly distributed welded big data. This paper provides accurate and effective estimation of body resistance welding condition, which can provide some guidance for online inspection of body resistance spot welding quality in automotive production lines.