A complete strategy for monitoring industrial batches processes using gray models is presented including fault detection and fault diagnosis tools. The use of gray models is a novel concept in batchprocess modeling a...
详细信息
A complete strategy for monitoring industrial batches processes using gray models is presented including fault detection and fault diagnosis tools. The use of gray models is a novel concept in batchprocess modeling and monitoring A gray model is a hybrid model, intermediate between hard (white) process models and soft (black) models, combining the advantages of both approaches. The principles of gray models are explained and it is shown how these models can be constructed. For this purpose an industrial batchprocess is available that is spectroscopically monitored, and an explanation is provided as to how the spectroscopic measurements are combined with prior process knowledge. To show the versatility of the strategy, two types of gray models are constructed and used for statistical batch process monitoring The two models are compared and validated for both on-line monitoring and post-batch analysis. For the latter, the batch consistency number (BCN) is introduced to have a fast and simple post-batch analysis. The, results show how these models help to detect and diagnose process upsets. The use of gray models for batchprocessmonitoring results in a fast detection of process upsets and a good fault diagnosis. (C) 2005 American Institute of Chemical Engineers.
This paper describes the concepts of statistical batch process monitoring and the associated problems. It starts with an introduction to processmonitoring in general which is then extended to batchprocessmonitoring...
详细信息
This paper describes the concepts of statistical batch process monitoring and the associated problems. It starts with an introduction to processmonitoring in general which is then extended to batchprocessmonitoring. The performance of control charts for batchprocessmonitoring is discussed by means of two performance indices: the overall type I error and the action signal time. Problems associated with the existing approach are discussed and highlighted. Improvements are suggested and checked with the performance indices. To evaluate the effect of the proposed improvements as well as to assess the performance of the existing approach, an industrial batch production process is used.
This paper discusses alternative methods for batchprocessmonitoring. Two alternative methods are investigated and compared to an existing one (the benchmark). A description of the models is given and the performance...
详细信息
This paper discusses alternative methods for batchprocessmonitoring. Two alternative methods are investigated and compared to an existing one (the benchmark). A description of the models is given and the performance is discussed by means of fault detection performance indices. The performance indices used are the overall type I error and the action signal time. In order to evaluate the performance of the models in terms of the overall type I error and action signal time, six different batchprocess data sets have been used. The data sets comprise four industrial data sets, one simulated data set and a laboratory spectral data set. (c) 2005 Elsevier Ltd. All rights reserved.
This paper discusses a method for warping spectral batch data. This method is a modification of a procedure proposed by Kassidas et al. [AlChE Journal 44 (1998) 864;Journal of process Control 8 (1998) 381]. This itera...
详细信息
This paper discusses a method for warping spectral batch data. This method is a modification of a procedure proposed by Kassidas et al. [AlChE Journal 44 (1998) 864;Journal of process Control 8 (1998) 381]. This iterative procedure is based on the dynamic time warping (DTW) algorithm. The symmetric DTW algorithm is discussed in this paper. Kassidas defined a certain weight that is received by every process variable in the DTW algorithm. However, high weights are received by process variables that contain no warping information. Therefore, a new definition of these weights is presented. These new weights take into account the amount of warping information of every process variable. The DTW algorithm using the new weights is compared to the procedure suggested by Kassidas. Furthermore, some aspects of this algorithm are optimized for speech recognition, but seem to be not necessary for warping batches. This concerns the normalization of the distance function. This step can therefore be omitted for warping batch data. (C) 2003 Elsevier B.V. All rights reserved.
Commercial biopharmaceutical manufacturing comprises of multiple distinct processing steps that require effective and efficient monitoring of many variables simultaneously in real-time. This article addresses the prob...
详细信息
暂无评论