Quality monitoring in multistage manufacturing systems by using machine learning techniques

Manufacturing and production processes have become more complicated and usually consist of multiple stages to meet customers' requirements. This poses big challenges for quality monitoring due to the vast amount of data and the interactive effects of many factors on the final product quality. This research introduces a smart real-time quality monitoring and inspection framework capable of predicting and determining the quality deviations for complex and multistage manufacturing systems as early as possible; introduces a hybrid quality inspection approach based on both predictive models and physical inspection in order to enhance the quality monitoring process, save resources, reduce inspection time and costs. Several supervised and unsupervised machine learning techniques such as support vector machine, random forest, artificial neural network, principal component analysis were used to build the quality monitoring model with considering the cumulative effects of different manufacturing stages and the unbalance and dynamic nature of the manufacturing processes. A complex semiconductor manufacturing dataset was used to verify and assess the performance of the proposed framework. The results prove the ability of the suggested framework to enhance the quality monitoring process in multistage manufacturing systems and the ability of the hybrid quality inspection approach to reduce the inspection volume and cost.