A body welded joint quality prediction method incorporating reconstruction error

With the continuous development of artificial intelligence technology, the application of machine learning in the prediction of the quality status of body welded joints is becoming more and more widespread. However, due to the strict setting of production parameters, the proportion of abnormal weld joints is relatively small, resulting in unbalanced data distribution. This poses a challenge to accurately assess welded joint quality and effectively identify abnormal welded joints. To address this problem, a quality prediction method incorporating reconstruction errors is proposed, aiming to accurately predict the quality of welded joints and effectively identify abnormal welded joints. In this paper, noise-reducing autoencoder (DAE) is used to reconstruct the feature data. It not only highlights the key features, but also extracts the reconstruction error as a new feature used to help the model identify abnormal weld joints. Because the reconstruction error of abnormal welded joints is bigger. Then, the reconstruction error and reconstruction features are used as inputs, and the welded joint quality score is the output. The support vector regression (SVR) model incorporating particle swarm algorithm (PSO) is used to construct the mapping relationship between input and output to achieve the prediction of welded joint quality status. The experimental results show that the method is highly reliable in the prediction of welded joint quality status, and the prediction accuracy is significantly better than other models. Compared with other methods, the accuracy of abnormal welded joint identification is not more than 82%, and the accuracy of abnormal welded joint identification of this method is more than 92%. This study provides technical support for the intelligent online inspection of welded joints in automobile bodies.