Online process monitoring in ultrasonic welding of automotive lithium-ion batteries is essential for robust and reliable battery pack assembly. Effective quality monitoring algorithms have been developed to identify out of control parts by applying purely statistical classification methods. However, such methods do not provide the deep physical understanding of the manufacturing process that is necessary to provide diagnostic capability when the process is out of control. The purpose of this study is to determine the physical correlation between ultrasonic welding signal features and the ultrasonic welding process conditions and ultimately joint performance. A deep understanding in these relationships will enable a significant reduction in production launch time and cost, improve process design for ultrasonic welding, and reduce operational downtime through advanced diagnostic methods. In this study, the fundamental physics behind the ultrasonic welding process is investigated using two process signals, weld power and horn displacement. Several online features are identified by examining those signals and their variations under abnormal process conditions. The joint quality is predicted by correlating such online features to weld attributes such as bond density and postweld thickness that directly impact the weld performance. This study provides a guideline for feature selection and advanced diagnostics to achieve a reliable online quality monitoring system in ultrasonic metal welding.