Prediction And Evaluation Of Equipment Health Condition Based On Signal Feature Extraction

The complicated underground mechanical or electrical equipment in the coal mine bring about the special failure features including nonlinearity, time-variability, concurrence as well as uncertainty. Furthermore, the underground noise interference, large number of equipment and the overbalanced distribution make it difficult to conduct failure prediction. On the basis of the existing study on signal processing, sorting algorithm, failure prediction and machine learning, this thesis conducts research on the accuracy of condition assessment, application range and practicability of the underground equipment. The main contents are as follows:1) The prediction and assessment of the equipment operation. The complicated underground conditions bring difficulties to conduct direct failure prediction. In view of the failure prediction is not practical for an unexpected accident, so the thesis puts forward the idea of evaluating the operation of the equipment by normal state assessment. The experiment result shows that this method is of high application value for the equipment with gradual faults.2) Characteristic study of underground equipment based on the feature extraction. The operation sound can reflect the current equipment condition. This thesis studies the sensitiveness of the sound to the equipment condition and conducts the preprocessing of framing and demising of the extracted sound and then compares the short-time energy, cestrum, Mel Frequency cestrum coefficient features of the signals. The result shows that it is feasible to evaluate the condition prediction of the equipment based on the feature extraction. For the same signal, the accuracy changes with the different extracted features.3) Application of machine learning in the condition prediction. Support Vector Machine(SVM) has the unique globally optical solution and excellent machine learning capacity and can solve the problems including small sample, nonlinearity and high-dimension. This thesis transforms the problem of condition prediction into the problem of processing the model of equipment characteristic classification and puts forward the approach of equipment condition prediction and assessment based on characteristic extraction and SVM.4) Evaluation research and experiment analysis of equipment condition. The thesis puts forward the concept of normal degree and verifies the rationality and accuracy by processing the acoustical signal of the equipment by the sound pickup or vibrating sensor which set near by the water pump. The experiment studies the influence factors of the prediction accuracy by combining the normal condition of the equipment and the results show that the predictions based on Mel Frequency cestrum coefficient feature and SVM have a relative higher accuracy.