Research On Remaining Useful Life Prediction Of Mechanical Equipment Based On Monitoring Data

After entering the21st century, with the increasingly serious energy shortages and environment problems, the machinery manufacturing technology develops to higher parameters and larger scales, and the trend of automation, large-scale, high parameter, high energy reserve in production process makes security problems of mechanical products with hitherto unknown importance. As the complexity and uncertainty of mechanical products increasing, the physical model is difficult to determine. Due to its high cost, small batch, the traditional statistical methods based on large sample are not suitable for this type of machinery and equipment with rare test sample characteristics. How to assess and predict remaining useful life under small sample condition is a great challenging task, so it is necessary to reform the relevant theories and methods. The data-driven approach in life prediction provides a feasible ways to solve these problems. The support vector machine model and state space model method in data-driven approach are studied in this paper. It aims at the important scientific problems in maintenance of machinery and equipment operation combing with the development plan of the national machinery and manufacturing science. The main research contents of the thesis are listed as follows:Firstly, remaining useful life prediction methods of mechanical equipment are reviewed. The life prediction methods are divided into four categories:the method based on physical model, the method based on experience statistical, the method based on knowledge and the method based on data-driven. Each of these methods is analyzed respectively, and the advantages and disadvantages of them are compared. The basic concept, development history and content of machinery fault prediction are introduced, the rules of fault evolvement are analyzed, and the definition of remaining useful life is given. The above contents lay a foundation for the establishment of remaining useful life prediction model of mechanical equipments.Secondly, in the case study of support vector machine model, RMS value is calculated as degradation feature from the vibration signals of double row bearing. Wavelet transform is introduced into the SVM model to reduce the influence of irregular characteristics and simultaneously simplify the complexity of the original signal. WT-SVM model is constructed and trained based on the degradation feature data and the1-step and multi-step prediction is analyzed. Besides Hazen plotting position relationships is applied to describe the degradation trend distribution and a95%confidence level based on t-distribution is given. The single SVM model and neural network (NN) approach is also investigated as a comparison.Thirdly, in the case study of state space model, wear data of the milling cutter in the machining process is measured, and the linear wiener process with random drift effect is adopted to describe the tool wear process and a state space model is established. This model is combined with particle filter algorithm, and the unknown parameters of the model are obtained based on the bootstrap re-sampling method. The degradation trend and remaining useful life of the milling cutter are predicted, and optimal change time is analyzed according to the corresponding decision modelIn summary, the research of the thesis may be used as reference for remaining useful life prediction research of high-end mechanical equipments.