A Study On Operation Safety Analysis And Online Assessment Of Industrial Processes

In modern industry, in order to lower the production cost, reduce the emission and energy consumption, it is necessary to keep the process operations in safe, reliable and steady states. In all of them, the safety issue is the most important. In practical industries, process faults or malfunctions are the main causes of unsafe threats and hazardous situations. The research of process monitoring is focused on the solution of fault detection, isolation, diagnosis, process recovery, etc. And it is now more of the important branch of industry automation. According to different monitoring models, the methodology of process monitoring can be usually classified into three categories:the approaches based on analytical models, data-driven, and based on knowledge or qualitative models. Compared with the fault detection and diagnosis, the approaches of safety analysis and assessment provide more particular descriptions on process operation or equipment safety degree based on the process characterizations. The automation in fault diagnosis, fault-recovery control, online safety assessment, and safety control is one of the future directions in process automation. And the step to achieve these is to obtain the online assessment results of operation safety.The studies on process monitoring & fault diagnosis, safety assessment have developed greatly, both on theories and practical applications. However, there are many unsolved issues. In this thesis, the impact of control strategy on the fault detection performance is discussed firstly. Then, a method is proposed to address the issue of complex fault isolation based on structured residuals. As an explorative work, two types of safety index are then constructed, characterizing the operation safety based on the process variable constraints and dynamic stability properties. Since the area of process monitoring & fault diagnosis and safety assessment is very vast, it is impossible to list and discuss all issues detailedly in the thesis. The main innovative work of this paper is as follows.(1) Some introductory discussions on the complex phenomena in the industrial processes. The practical industries are always involved with complicated mass and energy exchanges and meanwhile accompanied with physical or chemical reactions. And the process systems usually have the features like large-scale, high dimension of variables, nonlinearity, strong coupling, and correlation. The introduction of process control loops changes the input-output relations and dynamic characterizations of the system. The process operations may have multiple modes in practical situations due to parametric drifts, disturbance, feedstock, and environmental impacts. For nonlinear systems, there might be some complex phenomena, like bifurcation, which may have impacts on the system stability properties. It is stated that the complexity operation states in industrial process bring a great of difficulties and challenges to study and application of the process monitoring and safety assessment. Taking the Tennessee Eastman (TE) process as an example, some preliminary discussions on the impact of control strategies on the fault detection performance.(2) The principle of fault isolation approach based on structured residuals lies in the decoupling of the fault features by designing appropriate structured residuals. However, many faults in industry processes are too complicated to be decomposed. Under the principal component analysis (PCA) model, a multi-level strategy is proposed to extract the complex fault features, and assign the faults on different levels, where the structured residuals are designed to achieve best isolation performance. Each residual is expected to be insensitive to one fault and most sensitive to others. Lastly, a general incidence matrix is obtained to represent all residual response to faults, based on which all faults are isolated. The approach is applied in the TE process simulation results, which demonstrates the effectiveness.(3) A method based on multiple-mode identification is proposed in order to give an online assessment on the process operation. The Gaussian Mixture Model (GMM) is adopted to characterize the historical operation modes and identify the current state. A probabilistic safety index (SI) is constructed to denote the distance from the current operation to the safety limits. The calculation of the SI can be seen as two successive mappings. The original process variable space can be segmented into several subspaces, which are corresponding to different SI levels. In the D space, which have lower dimensions, the boundaries between different subspaces are obtained, and the current operation margin in a certain SI level can be calculated. The method is applied to the TE process and an industrial polypropylene (PP) process, and the results show that the SI can effectively show the variations in the operation safety, and provide preliminary advice on the manipulating directions to improve the process operation safety.(4) Considering the dynamic stability of process system, an approach is proposed to assess the process operation safety based on Hopf bifurcation analysis. Hopf bifurcation equations are adopted to characterize the curved surface of the Hopf points in the parameter space. A dynamic safety index (DSI) is constructed to represent the distance of the current operation state to the safety limits, i.e., the Hopf surface. The DSI is applied in a simulated gas-phase fluidized polyethylene reactor process. The results demonstrate that the DSI can effectively characterize the distance of the operation state to the Hopf bifurcation, and it is also able to give alerts on the unstable oscillations in advance, which is very beneficial to the practical maintenance of process operation.In the last chapter, conclusions of the thesis are summarized. Some unsolved problems and the prospects on the future study areas are pointed out and preliminarily discussed.