Development And Application Of Vibration Forecasting Technique For Large Machine-sets

This thesis focuses on condition forecasting technique of large machine-sets,which many enterprises is crying for. An optimized forecasting method based onartificial neural network (ANN) is developed, and then it is put into practice. Theresult shows that this method is suitable for condition forecasting of machinery. First, various traditional forecasting methods are introduced in brief. Butthese methods are unfitted to non-linear objects. ANN has many advantages overthese traditional methods. It is pointed out that machine-sets condition can bepredicted by such non-linear methods as ANN. Then, this thesis discusses how to establish the ANN's forecasting model.Some key questions are analyzed , such as determining ANN's topologicalstructure, selecting ANN's training arithmetic, and so on. After establishing the forecasting model, ANN is introduced to forecastmachinery operation condition. Firstly, the concept of predictive stability ispresented. The main factors affecting ANN's predictive performance are discussed.The result demonstrates that the ANN's predictive accuracy is depended on theamplitude of noise in trained samples. Secondly, based on the concept of fractaldimension, an index is presented to evaluate whether a trained samples isdivinable. Thirdly, based on the conclusion above, a new optimizing approach isput forward. Experiments and applications prove that this method is applicable toall sorts of typical situation, and can meet the demand of production.