Aluminium Resistance Spot Weld Quality Monitoring using Tensors and their Principal Components

Abstract

Resistance Spot Welding (RSW) is a sheet metal joining technique used in many manufacturing industries to create high strength joints. RSW uses a high electric current and the resistive-heating caused by the contact resistance between the sheet metal parts to generate sufficient heat to fuse the material together. RSW has been used for decades in the automotive industry to weld steel car body panels together in a quick and efficient manner. More recently, aluminium has been introduced in automotive body manufacturing to reduce the weight of the car while still maintaining high strength. The crashworthiness of monocoque vehicles created with spot welded sheet metal relies on the quality of the spot welds used to join the panels. Monitoring the quality of welds that leave the production line is therefore critically important to ensure vehicles are safe in the event of a crash. Assessing the quality of spot welds non-destructively is challenging because the spot weld is formed in between the layers of sheet metal which make a part. One way to achieve real-time, non-destructive quality monitoring of RSW is to monitor the electrical, thermal and mechanical signals throughout the process. Signal-based weld quality monitoring is relatively well understood for steel RSW, however the entire field of aluminium spot weld quality monitoring is particularly sparse in the academic literature. This thesis addresses the gap in the academic literature of aluminium RSW quality monitoring. A particular focus is placed on understanding the signal shapes and the way multiple process signals interact during the process to produce accurate models for aluminium spot weld quality monitoring. Principal Component Analysis (PCA) was used to represent the major signal shapes that are apparent in a set of steel welding signals. Using this method, training sets were designed in the laboratory to reveal the types of signal shapes that relate to weld quality. From the training data, models for weld quality and fault identification were developed using the signal shapes found with PCA. The models for weld quality were validated in industry using the equipment on an automotive production line. This allowed the technique to be tested and compared to the existing industry weld quality monitoring practices for steel RSW. To address the challenging problem of aluminium RSW quality monitoring, analysis of the interactions of multiple signals during the process was proposed. To analyse the interactions of multiple signal shapes, a new methodology is developed which involves representing process signal information as mathematical tensors. The tensors encapsulate the interaction behaviour of multiple signals using the outer-product of tensors. Using Tensor Principal Component Analysis (TPCA) the dominant tensor shapes in the training data could be analysed. Models for a The use of mathematical tensors to represent multiple process signals and their interactions allows for more accurate and robust models for weld quality to be developed in the laboratory. To test the technique in an industrial setting, the tensor based weld quality monitoring method was validated using significant amounts of industrial data. The industrial data, including multiple process faults and thousands of welds, allowed the tensor based weld quality monitoring technique to be thoroughly tested under a number of challenging conditions faced in industry. The tensor based approach proved to be more robust and accurate in a number of situations than similar models calculated from the information from an individual signal. This showed that the analysis of multiple signals and their interactions simultaneously provided a more robust approach to aluminium RSW quality monitoring than analysis of the information from a single signal. The implications of these findings are an important contribution to the academic literature in the field of aluminium RSW quality monitoring as little work has be