A Semi-Supervised Hybrid Machine Learning Framework for the Qualification of Resistance Spot Welds

Abstract Industries requiring high structural integrity, including automotive, aerospace, and construction, place considerable significance on weld quality classification. The challenge to classification model development is the scarcity of labeled data and imbalanced distributions in the data that are labeled. This work develops a new hybrid methodology that achieves clustering using KMeans++ together with supervised classification to overcome these challenges. In considering that clustering would be used to augment the training dataset by providing pseudo-labels for unlabeled samples, several feature subsets—spanning both sensor-derived and radiographic computed tomography-based features—were compared to determine the most influential features. The results reveal that sensor-based features ensure well-separated and coherent clustering, which consequently allows for marked enhancement in classification performance. The ensemble based classifiers were identified as optimal, with accuracy enhancements of up to 8% using the pseudo-labeled dataset. The proposed framework overcomes the major limitations of existing methods by combining unsupervised and supervised learning and, therefore, is a scalable and flexible solution in any data-constrained classification problem. The work provides practical insight into feature engineering and machine learning integration in industrial quality assurance applications.