Knowledge-based Mill Fan System Technical Condition Prognosis

The task of diagnosis is to find an explanation for a set of observations and – in the case of prognosis – to forecast the course of events. Diagnosis can be broken down into anomaly detection and failure identification, depending on the desired granularity of information required. Prognosis is concerned with incipient failure detection, margin prediction, or overall performance prediction. The latter can be prediction of efficiency, current system status, etc. The outcome of diagnosis and prognosis processes drives planning and execution. Fault isolation task can only be realized if the fault to be isolated has been previously taken into account in the model. There are different approaches for the design of diagnostic observers: the geometric methods, algebraic methods, spectral theory-based methods and frequency domain solutions. In our paper a two-step procedure is commonly employed for data-driven fault detection. A model that represents the normal operation conditions is first developed; then fault detection is carried out according to the residual information or according the differences in the quality parameters of the transient process. The data-based models, usually black-box models, lie in the core of a modular diagnosis system concept which has been chosen as separate fault detection systems. Each of these systems is handling only partial information on the process. This is similar to different persons analyzing the same situation with different methods and/or different sources of information. In the paper are presented the studied industrial mill fan, models of the studied systems and corresponding controller design by implementing conventional and fuzzy logic-based approaches. Simulation results – transient processes in the closed loop are implemented for the knowledge-based fault detection and prognosis. Key-Words: Knowledge-based Fault Detection, Fault Diagnostic, Fault prognosis, Mill fan, Fuzzy Logic Controller design