Fault Detection And Diagnosis Of High Sulfur Gas Desulphurization Process Based On Data-driven

With the development of complexity systems and automation systems in modernindustry, procedure under control occurs physical and chemical reactions as well asphase response simultaneously, involving material transformation and energy transfer,integrated by people, machines, ring, material, law and other uncertain factors. Theentire production process presents uncertainty, nonlinear, strong-coupling, dynamiccharacteristics. Traditional fault detection and diagnosis methods based on mechanismand process characteristics were extremely restricted. Process monitoring based ondata-driven mines system inherent laws, detects and identifies fault during the operationand working conditions online, and traces the root causes of failure according toanalysis a variety of monitor data by data processing method and tools. It providesintelligent decision for troubleshooting and system recovery, and ultimately ensures thereliability and safety of complex systems running.There exist three mian questions in high sulfur gas desulphurization as followscurrently. First, the fluctuations of natural gas processing load can causedesulphurization system model parameters to migrate, resulting in static model failureto identify the normal condition monitoring adjustment and false alarms. Second, datastructure is nonlinear, non-Gaussian nature and timing of autocorrelation characteristicsin high-sulfur natural gas desulphurization process monitoring, leading to it extremelydifficult to extract key parameters driven by desulphurization mechanism. Third, thecontribution of original parameters can’t be traced by extracting the key parameters ofthe desulphurization process, which make it difficult to fault diagnosis.The thesis discussed process monitoring methods by principal component analysisand independent component analysis. Tennessee Eastman model was used for thestandard library to test performance of various methods of fault detection and diagnosis.These methods were applied to solve fault detection and diagnosis problems in practicalhigh sour natural gas purification process.The main results were as follows:First, hypothesis testing and dynamic order determining algorithm was proposed todetermine order of Auto-Regression (AR) model aiming at the question that conditionadjustments was not recognized for the static model, which resulted in high false alarm.Dynamic principal component analysis and dynamic independent component analysis were studied to monitor process performance. Second, fault detection and diagnosisbased on dynamic kernel independent component analysis method was brought up torealize complex industrial process monitoring aiming at non-linear, non-Gaussian anddynamic industrial processes. Third, first-order partial derivatives of monitoringstatistics to the original parameters was used to measure the contribution, andcontribution graph based on statistics of the first order partial derivatives was putforward for fault diagnosis.Finally, taking high-sulfur natural gas desulphurization process for example,dynamic kernel independent analysis was used for fault detection and diagnosis, andmonitor performance achieved good results. Security controls were proposedappropriately to solve the problem of abnormal parameters for fault diagnosis.