Knowledge Discovery Method For Industrial Process Hybrid Monitoring With Application In Fault Diagnosis

Industrial processes usually are of discrete system characteristics involving state changes of discrete parameters such as motors,valves and switches,which have an important impact on the safety,stability and economy of the production process.A large number of hybrid industrial process production time series data are collected and stored in the monitoring database,which contains a wealth of process operation information.The knowledge discovery method of the hybrid process monitoring data can help harvest richer industrial process operation knowledge able to effectively monitor the production process.Regarding industrial process hybrid monitoring data,this thesis proposes a knowledge discovery method based on extended data logic analysis,which can be used to effectively mine industrial process operating knowledge.Subsequently,the approach is applied to solve process fault diagnosis problems by means of the resultant interpretable failure model.The main research content and achievements of the thesis are presented as follows.1.An approach of extended logical analysis of data is proposed for hybrid data mining of industrial processes with high dimensional and multiple features.Interpretable rules with time delays and variation characteristics are generated before cross-correlation functions are used to estimate time delays between variables,trend and fluctuation characteristics of continuous variables are extracted,and deep mining is used to extract patterns of discrete variables.The improvement of the hybrid data processing and mining methods enriches the content of patterns as well as improves the efficiency of patterns for hybrid data.2.A fault model reconstruction and representation method using conditional logic tree is proposed based on extended logical analysis of data in order to solve the fault diagnosis problem of the industrial systems with hybrid characteristics.Expert knowledge and grey relational analysis are used to determine the priority conditions of variables in the hybrid dataset.Patterns are reorganized to generate models and build the conditional logic tree.The effectiveness of the proposed method is verified by comparisons with traditional methods.3.The proposed method is applied to an industrial coal gasification drum process.The actual operating data of the coal gasification drum were collected to detect and diagnose steam drum level failures caused by hybrid variables.Hidden fault information in the data is found,achieving satisfactory results.The experiment verifies that the proposed method is feasible and effective for the fault diagnosis of hybrid industrial processes.