The objective of this award is to develop general methodologies for modeling and making inferences from multichannel nonlinear sensing signals, to support quality improvement in complex manufacturing processes. Specifically, a new variable selection method using hierarchical regularization will be developed to extract informative sensing signals and signal features in modeling of the relationship between product quality and massive process sensing signals. Process variations will be analyzed by nonparametric mixed-effect models to consider the underlying cross-correlation among heterogeneous sensing signals and the inevitable between-profile variations within each sensing signal. When validated, the models will be incorporated in the development of new process monitoring and diagnosis methods. This will permit variation reduction and quality improvement in complex manufacturing processes.
If successful, the results of this research will provide an effective means to enhance an online monitoring and diagnosis system with the advanced capability of automatically analyzing and assessing massive online sensing signals to improve manufacturing process operations. Implementation of the methodologies in manufacturing is expected to provide significant cost reduction as an outcome of improving product quality and reducing production downtime. Consequently the research has the potential to generate broad economic impacts by improving the competitiveness of the U.S. manufacturing industry. Moreover, multidisciplinary academic training, outreach, and broad dissemination through publications and industrial collaboration will lead to wide application of the developed methodologies to many other sensor data fusion applications that are of vital importance to the nation's economic growth.
