Research On Fault Diagnosis And Forecasting Methods Of Complex Process Industry System

As key techniques of Computer Integrated Process Operation System(CIPOS), faultdiagnosis and forecasting of complex process industry system are very important forsafely operation of process enterprises. In recent years, they are becoming a focus inindustry and research field, many interesting results have been got, however,disadvantages of these methods are obvious. They could not complete satisfy thepractical needs. Further researches are necessary. In order to apply fault diagnosis techniques to practice, on the basis of the results ofexisting theory, fault diagnosis and forecasting of complex process industry systemare researched from different directions by different methods. The main contents ofthis dissertation are as follows:1. The general situation of the fault diagnosis development is introduced , thecharacteristic and limitation of existing fault diagnosis methods are analyzed andevaluated. In the end, the rationale for diagnosis is introduced and existing problemsof diagnosis system are proposed.2. Research on identification of fault stateThe information entropy is introduced into fault diagnosis of complex processindustry system. A new computation method of Kolmogorov entropy based on generalentropy is used to extract the complexity of time series in order to characterize thesystem state, and is proved reliable by a case. Time-frequency analysis is applied tomonitoring the state of the complex process industry system in order to display thedetails of signals in different state. The results correspond to the characterizationresults of Kolmogorov entropy which was presented above. Study shows thatKolmogorov entropy is sensitive to fault state of complex process industry system.3. The notion of the fault diagnosis knowledge model according rough sets is givenfirst. Then the main steps of fault diagnosis knowledge modeling method areproposed , some theoretical problems involved in each steps are studied which includemodeling data acquisition, data preprocessing, continuous attribute discretization,condition attributes attribute value reduction, rule reduction . Finally, an example isalso illustrated to prove that the method is very effective.4. In this dissertation, multi-sensor information fusion technique is introduced tofault diagnosis of complex process industry system according to the characteristic ofhigh precision for fault diagnosis using multi-sensors. Based on the analysis of thefault diagnosis system, system faults are classified, diagnosis process and policy areproposed. The structure of the diagnosis system and reasoning methods are alsostudied.In view of the characteristic and limitation of the existing methods of faultdiagnosis , an integrated intelligent framework of fault diagnosis was proposed.5. On the basis of CIPOS and the proposed integrated framework, an intelligentdiagnosis system for MTBZ system is established, which could combine most ofexisting fault diagnosis methods. The stability and yield of MTBZ system areincreased by using the intelligent diagnosis system.6. Research on state forecastingState forecasting as an essential link of fault diagnosis plays a very important rolein production management and control. In this paper, according to the Chaoticproperty of complex process industry system, a local-forecasting model is built basedon the theory of phrase space reconstruction, which is applied to the prediction of timeseries of the pressure value in a chemical process system. Case study proves thevalidity of the model.Based on the chaotic local-forecasting model, the wavelet packed transformationtheory is introduced to the state forecasting. A wavelet packet-chaoticlocal-forecasting model is also established. The concrete example of time series ofpressures value is researched. From the research and apply of the waveletpacket-chaotic local-forecasting model, following conclusions can be drawn:(1) The forecasting precision of this model is higher than that of chaotic done.(2) The wavelet packet-chaotic local-forecasting model is fit for the weak faultstate forecasting.