State Predicting & Performance Monitoring Of Fermentation Processes

Fermentation processes are widely used in process industry and play important role in the production such as pharmaceuticals, biochemical products. Generally, it is difficult to get on-line measurements of some key state variables such as biomass and product concentration, which makes it difficult to implement my optimal scheduling strategies. Therefore, it is of great importance to predict these key state variables and fault detection for fermentation processes. Through monitoring the state of the production process, detecting the fault, process upsets and other abnormal events promptly, locating and removing the factors causing such event, the safety of production process will be assured and the quality of the product will be improved. With the development of the DCS (Distribute Control System), a great amount of process data can be sampled and collected. How to fully utilize this deep-level information to improve the performance of the process monitoring has been gradually becoming one of the focuses in the field of process control.This thesis involves around artificial neural network (ANN),multi-way principal component analysis (MPCA) and multi-way partial least square regression(MPLS). In view of the characteristics of different industrial processes, some improvements of traditional MPCA and MPLS have been made at different degree, and some new monitoring algorithms are also proposed. All the strategies don't require the mathematical model, but are based on data-driven technique. The main contributions are as follows:1. A method called Rolling-Learning-Prediction (RLP) that is based on artificial neural network (ANN) is employed to obtain the estimations of these variables five steps ahead. Simulations proved the accuracy and robustness of the RLP method. The key state variables such as total product, product concentration and sugar consumption are predicted. A rolling learning-prediction procedure is applied to deal with the time variant property of the process, which is also demonstrated to be beneficial to improve the prediction accuracy. The accurate prediction of the product formation enables the on-line evaluation of the economic performance of a batch and makes optimal scheduling possible.2. The profit function is defined and on-line predicted. Compared to product concentration and other state variables, the profit function is a more direct and sensitive variable to evaluate fermentation process from the economic point of view. Taken penicillin cultivation as an example, detailed calculation procedure of the profit function is given. The amount of product formation required for the prediction of the profit function is derived from the predicted product concentration. It can classify a batch into three categories, namely good, bad and normal. Based on the classification result, the novel method of optimal scheduling based on the basic idea of extending the cultivation period of a good batch as long as possible and terminating the cultivation of a bad or a faulty batch as early as possible can set up.3. Two improved MPCA approaches are proposed to monitor the fermentation process. Data from commercial-scale penicillin fermentation process are used to develop the rolling MPCA. Using the moving data windows technique, a series of MPCA is set up to avoid pre-estimating the unknown part of the process variable trajectory deviation from the current time to the end, which extends the static MPCA for dynamic process performance monitoring and detect the abnormal quicker than the traditional MPCA.4. Using abundant process measurements, two data-driven prediction and fault detection methods, the rolling multi-way partial least square regression (RMPLS) and support vector machines multi-way partial least square (SVM-MPLS) were proposed in the paper. They are especially suitable to deal with complicated nonlinear fed-batch bioprocesses. The fed-batch penicillin was studied, which indicates the proposed methods are able to rapidly predict and detect abnormalities or fault of the process. With these methods, the faults are detected in real time and the responsible measurements are directly identified through inspection of a few simple plots (t-chart, SPE-chart and T2 chart). Thus, the presented methodology allows the process operator to actively monitor data from several cultivations simultaneously.5. An on-line prediction and process monitoring software system using advanced process control technology (BioAPC) has been developed for fermentation processes. The software system is designed based on theoretic research in this dissertation and VC++ programming language. It is able to realize on-line prediction of the key state variables such as total product, product concentration, computation and prediction of the profit function, real-time classification of the present interesting batches, and optimal scheduling for multi-reactor system. In addition, the software system can be also used for process monitoring and fault detection. Finally, some conclusions and further study areas are given.