In manufacturing systems, multiple faults often occur simultaneously and generate complex fault patterns. However, most of the previous research on multiple fault diagnosis assumes that faults either occur one at a time or have linearly additive fault patterns. In this paper, we present a new approach for multiple simultaneous fault diagnosis. A comprehensive pattern modeling approach is developed for prediction based on functional regression. With the complete set of fault patterns, a minimum distance pattern classifier is designed for diagnosis. The proposed approach is demonstrated to be successful with a simulation study. An example from the resistance spot welding process is also provided. In this case, an 88% success rate for fault classification is achieved based on the predicted fault patterns.