Clamping Force Prediction of Different Fastener Connections Using Machine Learning

Fasteners are widely used elements for joining sheet metal. Real-time monitoring is desired to predict the quality respectively a sufficient clamping force between the joining partners. Usually, measured process signals are evaluated through expert systems or control windows. These methods are restricted to one specific use case and need to be reevaluated for every new joint combination. Often those systems can only determine significant deviations from the reference curve. The quality criteria of the joint, the clamping force, is usually not predicted and a process monitoring using ultrasonic testing is time consuming and highly cost intensive. Another possibility is quality prediction using machine learning algorithms to detect patterns in the process signals which correlate with the quality criteria. Many studies introduced machine learning algorithms to detect outlier curves or different failure types. Therefore, this paper presents an approach of unifying the shape of different torque-angle-curves from different fastener applications while simultaneously achieving high accuracy on the prediction of the clamping force with this novel method.