Research On Qualitative Trend Analysis Based SDG Fault Diagnosis And Its Application In Industry

As the development of production technology, the chemical industry becomes more and more automatic and the scale becomes larger, but at the same time the complexity is increasing. Abnormal situations or faults occur due to sensor drifts,equipment failures,changes in process parameters or manual misoperation, leading to heavy economic lost and casualties. So it is an essential problem to find fault causes timely and avoid more heavy lost and serious accident.SDG based fault diagnosis is one of methods based on qualitative model. SDG based fault diagnosis can arrive at completeness,good robustness,explanation facility and so on.So in recent years, it becomes a focus in the research area.However, there are still some disadvantages in fault diagnosis based on SDG model, which restrict its application in real chemical industry. There are several methods of building SDG model. All of them have advantages and disadvantages. It is lack of a unified method for building SDG model; SDG model is a kind of qualitative models. In the model, the relationships between nodes are determined by "+" and "-". So the model can not represent the states of nodes and relationships between nodes in detail. In addition, most of the SDG models are based on initial response. They can only reveal the relationships between nodes during the range of initial response and can not reveal the relationships at different ranges after fault occurs.There are also disadvantages in fault diagnosis based on above SDG model. For example:The states can not reveal the trends of nodes, so it can not do early fault diagnosis. The SDG model based on initial response makes the fault diagnosis method get different results at different ranges and miss the real fault cause. When multiple faults occur, the nodes influence each other and the states of nodes may change at any time. It makes the traditional SDG based fault diagnosis can not find the real fault causes; diagnosis resolution is poor.In addition, the research about SDG based fault diagnosis focus on improving in some aspects such as the thresholds setting,improving the resolution and so on. There are few frameworks which can overcome the disadvantages and make it applicable in chemical industry.For these disadvantages and other problems in the application of SDG base fault diagnosis in the real chemical industry, the research is carried out in the thesis. The main contributions of the thesis are as follows:(1) The research is carried out on building SDG model. The method of SDG modeling based on qualitative trends relationships between nodes is proposed based on previous methods. The method includes the definition of basic patterns describing qualitative trends relationships between nodes and the whole procedures of modeling. Using algebraic equations,differential equations and experience, the relations between nodes are represented by patterns or the combination of patterns. The relations are stored in the SDG model in stead of incremental effect and reduction effect. The SDG model is not only based on initial response but can also represent the relationships between nodes at different ranges after fault occurs. The model makes preparation for fault diagnosis.(2) The SDG based fault diagnosis framework is proposed. This framework is based on SDG model which is based on qualitative trends relationships between nodes. There are three parts of the framework:data preprocessing,abnormal state monitoring and fault diagnosis. In the data preprocessing unit lifting scheme and median method are used to eliminating the random and gross error and supply more reliable data.Abnormal state monitoring unit is used to monitoring the state of system. Once abnormal situation or fault occurs, the unit will notice the fault diagnosis unit. In the abnormal state monitoring unit, the method based on qualitative trend segment and threshold is used. Qualitative trend segments used "Increasing","Decreasing "and "Steady" to represent the states of nodes instead of "+","-" and "0". The method of qualitative trend segments is used to find abnormal nodes in the early stage. When the variables are in steady state, the method of threshold is used to find abnormal nodes to make the abnormal nodes found in every stages.In the fault diagnosis unit, SDG and qualitative trend analysis based fault diagnosis is used. The method based on a sliding window and bidirectional fitting is used to extract and identify the qualitative trends of nodes. The qualitative trends of nodes are identified as "Increasing","Decreasing" and "Steady". And then the qualitative trends of nodes are transformed into qualitative trends relationships between nodes and after segments merging, the final qualitative trends relationships between nodes are identified. The qualitative trends relationships between nodes are compared with these in the SDG model to determine whether consistent and find the real fault cause. At last the method will sort the results to improve the resolution.The framework can do early fault diagnosis,avoid missing the real fault cause,can do multiple fault diagnosis and have good diagnosis resolution. The framework solves the problems of SDG based fault diagnosis and makes good preparation for application in the real industry.(3) Based on the SDG fault diagnosis framework, the abnormal situation information guidance system is studied and developed. And the system is applied in fault diagnosis of an atmospheric distillation unit of Harbin refining plant. The system provides SDG modeling and configuration environment for the whole process,on-line consulting for abnormal situation,kinds of data acquiring methods and kinds of denosing methods. The other problems in the application, such as data acquiring,denosing,window width of qualitative trend extracting,sample rate,switch between different production programmes and verificaiton are all studied and solved.(4) The results of case study and application prove that the abnormal situation information guidance system can do early fault diagnosis,avoid missing the real fault cause,can do multiple fault diagnosis and have good diagnosis resolution. It obtains the practicability,effectiveness and reliability and can be applied in the fault diagnosis of real industry.