Due to the late response to process condition changes, forging processes are normally exposed to a large number of defective products. To achieve online process monitoring, multichannel tonnage signals are often collected from the forging press. The tonnage signals contain significant amount of real time information regarding the product and the process conditions. In this paper, a methodology is developed to detect profile changes of multichannel tonnage signals for forging process monitoring and to classify fault patterns. The changes include global or local profile deviations, which correspond to deviations of a whole process cycle or process segment(s) within a cycle, respectively. The principal curve method is used to conduct feature extraction and discrimination of tonnage signals. The developed methodology is demonstrated with industry data from a crankshaft forging processes.