Bitte wählen Sie ihr Lieferland und ihre Kundengruppe
This paper describes the research undertaken at a Tata Steel UK coke oven plant to develop a data-driven diagnostic system to detect and locate through-wall leakage using the process and stack emissions data. Preliminary investigations through visual inspection of the data provided valuable knowledge of the data variations and interactions, and enabled a procedure for the detection and location of through-wall leakage to be developed. The integrity of the diagnostic system depends on the quality of the PCA model training data. It is important to remove any data that are non-representative of normal process operation and ensure that the data representative of normal operation is not deleted. To provide coke oven plant engineers the means for developing and updating the PCA models for implementation in the G2 system, a MSPC performance monitoring toolbox based on MATLAB algorithms has been developed. A MSPC performance monitoring toolbox GUI has been developed to help coke oven plant engineers to develop and update the PCA models for implementation in the G2 system. The GUI can provide visualization of the data, and enables data pre-screening and outlier removal. The developed PCA model was implemented into a G2 knowledge based system its iMSPC toolbox. The models were run in real-time and simulated time, and provided appropriate displays and warning messages for operators, which enabled coke oven plant engineers to diagnose the source of the fault. Since it is expected that a process state changes with time, it is recommended that the PCA model is regularly updated and improved by using the most recent process and emission data. This will ensure that the reliability of detection and location of through-wall leakage of the G2 based diagnostic system is maintained.