Study On Mine Ventilation Fault Diagnosis Based On Knowledge

Mine ventilators are the key equipments in coal mining safety production, which undertake vital responsibilities for infusing fresh air to the underground, exhausting dust and dirty flow, and ensuring the safe production of the coal mines. As ventilator is the power source of mine system, an unexpected failure of it may cause significant economic and casualty losses. Therefore, it has important theoretical significance and practical application value to the research on the on-line monitoring and fault diagnosis of mine ventilation system.Geographically distributed databases in different mining areas are taken as analysis objects in this dissertation, through which a managemnt model of distributed data is proposed based on grid and interoperability techniques and the structure of ventilator fault diagnosis system is put forward. In this structure, remote data are colleted and preprocessed to build local dynamic data warehouse, from which sample set is selected; Meanwhile fault features are extracted through time-freqency analysis on vibration signals of ventilators, so as to be adapted to adjustment parameters of multi-class classifier of support vector machine; Then the state similarity kernel function is presented and fuzzy compensation factor is introduced to analyze and model, so that multiple attributes are integrated effectively and valid diagnosis of multi-concurrent faults is realized; Finally, a knowledge-based fault diagnosis system of mine ventilator is developed.The research of this Dissertation mainly includes:(1) Grid architecture and hierarchical model of interoperability are studied, and the model, services and working process of distributed data management are put forward according to the objective conditions of coal mine production system and the requirements of fault diagnosis system. On the basis of the model, the framework of ventilator fault diagnosis system on distributed environment is proposed. Local data and remote data are combined together, operating state variables are taken as sample space and eigenfrequency of vibration signals as parameters, and multiaspect data are integrated to diagnosis by using pattern recognition technology of support vector machine.(2) Composition and mapping relation between structured data and unstructured data are studied, and the pattern transformation method among heterogeneous data is also analysed. The algorithm of building data warehouse directly by XML documents is proposed, which realizes the integration of remote data. Moreover, data pre-processing techniques dependent on density estimation is analysed, in order to put forward the data processing method based on error-adjusted density estimation and micro-dataset, which can execute smoothing estimation on uncertain data of data warehouse.(3) Some commonly used time-frequency analysis methods are compared such as short-time fourier transform, Wigner-Ville distribution and wavelet transform, and systematically research on the principle of empirical mode decomposition is also described. By analyzing and verifying some problems of Hilbert-Huang transform, such as decomposition stopping criterion, boundary treating methods and envelope fitting methods, suitable processing methods are chosen. Aiming at the characteristic of ventilator vibration signal, fault features extraction method is proposed based on approximate entropy and Hilbert-Huang transform, and applied to simulation signal and real signal. Compared with the second generation wavelet transform, experimental result demonstrates validity of this method.(4) The working principle and algorithms of support vector machines is researched, and aiming at the limitations of directly constructing multi-class classifier of support vector machine, a new fuzzy compensation multi-class support vector machine is presented based on state similarity kernel function. This method takes fault features extracted from time-frequency analysis of vibration signals as fuzzy compensation factors, integrates multiple characteristic parameters effectively, and uses state similarity kernel function to pattern recognition.In this Dissertation, there are fifty-two figures, nineteen tables, and one hundred and fifty-one reference documents.