Fault detection and diagnosis in chemical and biochemical processes

Abstract

Since, Chemical and bioprocesses are complex nonlinear systems and are difficult to model in real life, it becomes quite important to study process dynamics of these systems, as small changes in initial conditions can give dramatic output product quality, which is not desirable, e.g., in case of bioreactor, small change in pH can affect the growth kinetics or small changes in temperature might damage the cells. Chemical industries also suffer from various accidents mainly due to equipment failure and operator error. Robust fault detection and diagnosis systems are always required for such type of processes. The objectives of fault detection and diagnosis systems are early detection and diagnosis of fault, fast process recovery, avoid abnormal event progression and to satisfy environment and safety regulations. At the same time, it will reduce the product rejection rate, improve the quality of products and reduce the number of accidents in the industry. So a proper fault detection and diagnosis system will provide an economical and safe process. newlineThe idea of computerized on-line monitoring has been broadly researched over the past numerous years. Various approaches to deal with the fault detection and diagnosis have been developed in this period. These methodologies can be ordered into three categories: Methods based on mathematical models of physical systems; Methods make use of available information and knowledge of a physical system; Methods require neither first principles nor qualitative knowledge however rather a lot of historical data that contain the typical trends and fault information. Historical data-based methods are more suitable for such type of processes. All the machine learning techniques, for example, artificial neural networks, support vector machines, random forest, etc., belong to the class of historical data-based methods. In the past researchers have implemented and evaluated the performance of many techniques. Regardless, none of the already reported works gives adequate certainty.