Research On Multimode Process Monitoring Method Based On Local Tangent Space Alignment

In modern industrial process, the complexity, large-scale and automation of the whole production system has become an annually increasing trend. Once faults occur in the production process, there will be huge damage to the economic efficiency of enterprises and human security. Therefore, it is necessary for monitoring the process to detect the abnormal situations as early as possible.However, the whole process usually consists of multiple operating modes and the process data of different modes will present different statistical properties, in addition, the actual industrial process includes not only process data but also some quality data. And there is certain correlation between each other. In order to solve these problems above, researches have been done in this paper. The main contributions are listed as follows:(1) Based on existing process monitoring methods and techniques, this paper proposes a new multiple modeling and monitoring method based on common subspace separation. The main idea is to use Manifold Learning method to extract common information changes between multiple operating modes. The common information can represent the variation characteristics which are shared by all operating modes. Conversely, the specific information reflects the unique variation characteristics respectively. Then the common subspace and specific subspaces which are separated already are decomposed by KPCA method. The operating mode can be judged by monitoring the common and specific models and the fault can be detected effectively once it turns up in the process.(2) Because the available modeling data includes not only process data but also some quality data, in order to make full use of the data information of system, the data space is further divided by introducing quality variables. The new subspaces consist of quality-related common subspace, quality-related specific subspaces and the quality-unrelated residual space. And modes are built and monitored in these subspaces respectively. The simulation results of the electro-fused magnesium furnace show that the proposed method can improve the faults detection sensitivity and can effectively reduce the false alarms.