An on-line model development and quality control methodology is presented for manufacturers in the process industries with the goal of enabling automated quality assurance. Given appropriate process instrumentation, the methodology starts with the characterization of statistical variation for the process while operating in steady state. Significant process conditions are then perturbed by six standard deviations to bound the expected long-term process variation including lot-to-lot variability of feedstock materials. If the process is found to be robust, acquired process data is used to model the process behavior using principle components analysis (PCA). The PCA model is then used to accept and reject manufactured parts given real-time process data. This methodology is applied to an instrumented injection molding process that is subjected to 12 common process faults. The results indicate that the methodology was able to detect every one of 33 defective molding cycles caused by eight of the imposed faults as well as two additional faults that did not result in observable defective products. The quality controller did not detect the two remaining imposed process faults that did not produce defects and also rejected three molding cycles (2% of the total) that appeared to produce acceptable products.
展开▼