No abstract

To implement the quality prediction scheme for batch processes, long short-term memory (LSTM) neural network is a feasible tool to handle with the process dynamics and nonlinearity. However, a global LSTM soft sensor suffers a decline in performance facing batch-to-batch variations. To overcome the batch diversity problem and take advantage of LSTM model, a multivariate trajectory based ensemble just-in-time learning strategy is proposed in this paper. Different trajectory based similarity measurements are designed to extract historical batch trajectories which share similar spatial positions and trends. For each selected trajectory, an online local LSTM soft sensing model is constructed and the real-time quality prediction result for each local model can be obtained. Then, a weighting parameter is determined for each model by cross validation. Bringing together quality prediction results from different local models, the ensemble prediction result can be finally figured out. Two case studies are carried out to prove the effectiveness of the proposed methodology including a fed-batch reactor and the fed-batch penicillin fermentation process.INDEX TERMS Batch production systems, Ensemble just-in-time learning, long short-term memory, multivariate trajectory analysis, soft sensor, quality prediction.

A soft sensor is a key component when a real-time measurement is unavailable for industrial processes. Recently, soft sensor models based on deep-learning techniques have been successfully applied to complex industrial processes with nonlinear and dynamic characteristics. However, the conventional deep-learning-based methods cannot guarantee that the quality-relevant features are included in the hidden states when the modeling samples are limited. To address this issue, a supervised hybrid network based on a dynamic convolutional neural network (CNN) and a long short-term memory (LSTM) network is designed by constructing multilayer dynamic CNN-LSTM with improved structures. In each time instant, data augmentation is implemented by dynamic expansion of the original samples. Moreover, multiple supervised hidden units are trained by adding quality variables as part of the layer input to acquire a better quality-related feature learning performance. The effectiveness of the proposed soft senor development is validated through two industrial applications, including a penicillin fermentation process and a debutanizer column.

To predict key variables of complicated batch processes, the long short-term memory (LSTM) soft sensor is developed to deal with both data nonlinearity and dynamics. To extract proper historical samples and implement the real-time modeling scheme with model updating strategy, the just-in-time learning (JITL) algorithm is widely used at the data selection stage of LSTM soft sensor. However, the multiphase issue of batch processes are not considered for the conventional JITL-LSTM soft sensor. In this paper, a multiphase Mahalanobis distance based JITL framework is developed to integrate the phase recognition strategy into the similarity measurement and data selection scheme, by which an extra step of phase identification can be avoided and the accuracy of JITL can be significantly improved. Thus, batch samples from different operating phases can be recognized without an additional phase identification step. By the use of the Mahalanobis Distance based JITL-LSTM Soft Sensor, the probability of data mismatch can be significantly reduced so that the accuracy of quality prediction can be promoted. Two simulation cases are provided to verify the effectiveness of the proposed method consisting of a fed-batch reactor process and the penicillin fermentation process.

This paper proposes a new method combining the multivariate trajectory analysis and the principal component analysis (PCA) for multiphase batch process monitoring. To handle the uneven length problem, the trajectories of process variables are calculated instead of the original samples. For online monitoring, similar trajectories are extracted by just-in-time learning (JITL) with historical trajectories and the PCA model is constructed, which can deal with the missing data problem as well. Furthermore, to acquire a more reliable monitoring performance, a new distance-based measurement is proposed to show the location of samples. For performance evaluation, case studies of a numerical example and a simulated penicillin fermentation process are provided, with detailed comparisons to traditional methods.