Intelligent integrated diagnostics: Process monitoring and diagnosis for on-line quality control of powder injection molding

This thesis addresses the problem of on-line quality control of the Powder Injection Molding (PIM) process. Intelligent Integrated Diagnostics (IID), an integrated paradigm of product and process monitoring, diagnosis, and control is proposed. This thesis concentrates on process monitoring and diagnosis, with specific reference to the PIM process. Process monitoring involves three steps: selection of process parameters, sensor data representation, and signal abstraction by pattern classification. The injection and cavity pressure profiles during a molding cycle are selected as the process parameters. Considering the nonoscillatory nature of the pressure, sensor signals of the process parameters are represented using spline wavelet transformation. Nine partitioned 3-layer feedforward networks perform the abstraction task of the signals from different time periods during a molding cycle. Diagnosis involves two steps: representation of the domain knowledge and diagnostic reasoning to solve a problem. A computational model representing the domain knowledge, causality network, is developed. The causality network represents the structural knowledge of manufacturing processes efficiently. The proposed diagnostic reasoning procedure includes the simplification of the causality network using network conversion, hypotheses generation using abductive reasoning, and verification of the hypotheses. Algorithms for network conversion and verification procedures are developed. This work presents unique methods for process monitoring and diagnosis in the PIM domain. The effectiveness of the developed methods are validated through real case tests using experimental data. However, the general methodology is applicable to any manufacturing process.