CAREER: Dynamic Process-Attribute-Data-Performance Modeling to Enable Smart Ultrasonic Metal Welding

点击翻译按钮获取中文摘要

This Faculty Early Career Development (CAREER) grant will support fundamental research on ultrasonic metal welding (UMW). 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, making it a promising joining technology for the advanced manufacturing of electrified and lightweight vehicles. Nevertheless, UMW has a relatively narrow operating window and is very sensitive to unpredictable, uncontrollable environmental conditions. This longstanding knowledge gap in the underlying process mechanisms makes the prediction and control of joint quality difficult, which limits its use. This project will take advantage of the emergent information-centric transformation of manufacturing science by leveraging advances in process physics, microstructural analysis, and data science. By establishing dynamic, stochastic relationships between process conditions, microstructural weld attributes, online sensing data, and weld performance, the research will advance the fundamental understanding of process mechanisms in UMW. The knowledge gained will be used to establish a suite of machine learning-based decision-making tools that will ultimately enable smart UMW. This grant will also support diverse educational and outreach activities that contribute to the education of the U.S. smart manufacturing workforce. It is a widely accepted hypothesis that UMW process conditions influence the joining performance via the dynamic evolution of micro-scale weld attributes and the weld formation process generates a signature, as reflected in parameters that can be sensed online. Nonetheless, there exist no studies to date that adequately model or quantify the inherent dynamic, stochastic process-attribute-data-performance (PADP) relationship. The overarching goal of this research is to create a PADP modeling framework that consists of innovative machine learning and statistical models. The framework will be completed in two steps. First, spatiotemporal models incorporating uncertainty quantification will be built to characterize the process-attribute-performance relationship. Second, a tensor-based correlation and regression analysis will be performed to investigate the attribute-data relationship. This framework will be further employed to develop a series of physics-aware, machine learning tools for process control, including process optimization, online quality monitoring, and real-time control. Finally, the project will investigate the use of a transfer learning methodology to provide a cost-effective way to build PADP models and decision-making strategies for related products or product families. This learning capability will be an essential component in the cloud intelligence that enables the smart manufacturing paradigm.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.