Study And Implementation Of Fast Process Modeling And Monitoring For Multi-mode Process

There exist a lot of industrial processes which have multiple operating conditions during manufacturing. When the process enters into a new operating condition, in general, the well-developed process model for the previous operating condition cannot be directly used for the new condition. Building an accurate process model for new operating conditions will consume a lot of time and resources definitely. Moreover, it is difficult to assure process safety and product quality during process condition switchovers, since there is no accurate process monitoring or quality prediction model available. Fast modeling and online process monitoring is an urgent need for real industrial processes with multiple operating conditions. Model migration theory is an effective way to solve this problem.This thesis takes a tandem mill with flying gauge change control strategy as an application example, which can be characterized as a multi-mode, varying-mode and multi-phase operation process with high dimensional coupled process variables. Considering the above characteristics, multivariable statistical modeling, model migration and phase division techniques are adopted to achieve fast modeling and accurate online process monitoring and quality prediction for a new operating condition.Firstly, the thesis focuses on the multiple operating condition of a rolling process. After comprehensive analysis of process data under different operating conditions, we found that the process mechanism is unchanged or kept similar even though process data show significant differences for different operating conditions. Based on this cognition, base modes are defined according to the historical database and multivariable statistical process monitoring models are developed for base modes. Consequently, a principle component analysis(PCA) similarity indicator is introduced which divides new operating modes into gradually-changed modes and abruptly-changed modes. For the gradually-changed case, new model is developed by screening the similar data of base modes and augmenting them into the modeling data set for the new mode. For the abruptly-changed case, new model is obtained by reconstructing the base model via parameters' shifting and scaling. Online flatness prediction can be conducted based on the developed model of the new operating mode.Then, considering that a rolling process is inherently a multi-phase process too, a novel fast modeling and monitoring strategy is developed based on model migration and phase division. A density clustering method is used to recognize the stable operating phase; and a local process model is developed and used for phase division. Again, model migration theory is used to extract the common characteristics between the current new operating mode and historical operating modes for typical products to find the most similar samples for initial model migration. A model updating strategy is then proposed to iteratively improve the model for better accuracy and robustness. Based on the obtained multi-phase process model, an enhanced process monitoring method can be achieved.The experimental results conducted on a 2030 mm tandem mill in Baosteel can show that, the proposed strategy can develop a new process model using very little or no prior information, and the model is accurate for online process monitoring and flatness prediction to meet the need of a steel rolling plant.