Statistical Modeling And Online Monitoring For Uneven-length Batch Processes

Recently, as the products of wide varieties, excessive specifications and high qualities are demanded by the society urgently, the batch process that produces product with small quantity and high added value has become one of an important industrial production mode. It is applied extensively in the field which is closely related to our daily life, such as food, fine chemicals, biotechnology, metal processing and so on. As a result of this, the safe and reliable operation of batch process is coming to the forefront and it is significant to monitor the batch process and conduct fault diagnosis. Due to its variety of advantages, the process monitoring method based on multivariate statistics has gain wide application, especially in the process monitoring of the industrial production.One of the outstanding features for batch process is that the process is repeated according to the pre-designed procedure. However, due to the influence of the weather, quality difference of the raw materials, non-time based data acquisition system and so on, the process can’t be repeated completely and therefore, the length of the collected data is no longer the same and this is known as the problem of data uneven-length in batch process. Due to this problem, the measured data can’t form a regular three-axis matrix, making it rather difficult for the monitoring and modeling of the batch process based on PCA.Dynamic time warping (DTW) is one kind of signal match method used in the field of signal match. Optimum matching signal is got by computing the matching distance between the measured signal and standard signal. The concept of bending route for computing the data matching distance in the DTW method has laid a theoretical foundation for its implementation in the process of uneven-length data. In this dissertation, the method of DTW based processing of uneven-length data is proposed. When the shortest matching distance between the standard data and measured data is obtained by optimization technique, a corresponding bending route is determined too. The length of the route is equal to the standard data and its projection in the measured data will stretch or compress the measured data so that their length is even and finally the uneven-length data is processed.Moreover, since the batch process is featured as multi-stage, this research proposed that the batch process should be divided into segments firstly and then the uneven-length data in different segments are processed based on DTW method. This not only ensures the even length of the overall modeling data in different batches but also keeps the length of the molding data in different period even, bringing great convenience for the monitoring and modeling of the batch process.The application to an injection process shows the effectiveness of the proposed method for uneven-length data process. Batches of uneven-length data are processed. Statistical modeling, on-line monitoring, and fault diagnosis is conducted. The simulation result shows that the model built on the basis of DTW processed data can monitor the injection process and conduct fault diagnosis effectively. So, the feasibility of the proposed method in this research is verified.