Abstract Scope |
Ultrasonic metal welding (UMW) is a promising solid-state joining technology that enables innovative and sustainable manufacturing. Among the advantages of UMW over conventional fusion welding techniques are the ability to join dissimilar metals, energy efficiency, short welding cycles, and environmental friendliness. Nevertheless, UMW has a relatively narrow operating window and is susceptible to both internal and external disturbances. As such, industrial-scale UMW production calls for efficient, effective, and non-destructive joint quality assessment. This talk will present physics-informed machine learning methods that enable smart decision-making in UMW. Specifically, a hybrid physics-informed transfer learning method is developed for highly data-efficient response surface modeling. The learning method is applied to jointly model the UMW response surfaces for three different material combinations and achieves a substantial improvement in prediction accuracy. In addition, a physics-informed ensemble learning framework is created for online prediction of UMW joint strength. The framework models a global trend and a residual of joint strength using physical knowledge and online sensing data, respectively. Experimental case studies demonstrate that the framework can effectively account for the influence of physical conditions and uncontrollable process factors, thereby enabling significantly more accurate online joint strength prediction. |