The Mechanical Fault Diagnosis Of Numerical Control Machine Feed System Based On Multi-information Fusion

Numerical control machine is an automated machine tool which is equipped with numerical control system. It can change tools automatically and achieve complex curves or surfaces machining with high precision, stable machining quality, improved production efficiency. Numerical control machine has become the key equipment in modern manufacturing technology. The level of technological development and number of numerical control machine represents the manufacturing ability of a country. In terms of structural-composition, numerical control machine is a complex integrated system including mechanics, electronics, hydraulics etc. Any part fault inwill affect the normal operation of a machine, especially failures in mechanical part will take a long period to maintainance time and therefore cause a huge economic loss. The key mechanical parts are maintained or replaced with a regular routine. It causes great waste of manpower and material resources and on the other hand the abrupt failures can still not be avoided. Therefore research on condition monitoring and fault diagnosis of numerical control machine is very necessary for switching the current maintenance method to a preventive type.Based on the information fusion technology, the strategy of condition monitoring and fault diagnosis, the method of signal processing & feature extraction, the model of pattern recognition and the fusion method of decision level are researched and the diagnosis system based on multi-level information fusion is built.Propagation path of cutting force is analyzed firstly. According to the fact that numerical control machines work with variable load, high frequency impact condition and the character of diversity of processing, it is concluded that mechanical parts in the transmission path of cutting force are more likely to fail. The research object and fault types are determined on the base of fault sources. The fault diagnosis principle of vibration, temperature, servo motor current and other parameters are studied. The condition monitoring system is constructed by external sensors, internal information, program parameters and alarm information which can reflect the operating status from aspects of machine tool, cutting tool wear, workpieces and processing quality. Sufficient information for subsequent diagnosis process is prepared by the condition monitoring system.Failure, especially compound failure can not be effectively distinguished by traditional time domain and frequency spectrum analysis. The signal processing method based on wavelet packet and EMD is proposed. Firstly the signal is de-noised by wavelet pocket. The reconstruction signal is divided into high frequency part and low frequency part. These two signals are decomposed by EMD method and energy of each IMF is calculated. The energy characteristics, time domain parameters and frequency domain parameters constitute the multi-domain mixed feature set. The feature extraction method based on correlation analysis and fuzzy clustering is proposed to achieve sensitive feature subset.Due to the characteristics of numerical control machine fault diagnosis, the hierarchical diagnosis strategy is put forward. It is divided into two levels: fault location and fault pattern recognition. The main network which finds fault location is responsible for the subnet aggregation model results. The subnet task is to diagnose fault type and degree. All network structures are simplified through task division and collaboration. Back Propagation(BP) neural network model, Radial Basis Function Neural Network(RBF) model and support vector machine(SVM) is constructed of numerical control machine fault diagnosis. Those three kinds of models are trained and tested using the same samples, then the diagnosis ability of different models can be compared.In order to improve diagnosis capabilities, the global diagnosis models based on fuzzy comprehensive evaluation and based on weighted D-S Evidence Theory are established respectively. The single stage diagnosis model based on fuzzy comprehensive evaluation is established, combined with preliminary diagnosis of multiple classifiers which builds the mapping form the characteristic space to the diagnosis space. The classifier classification ability is evaluated by recognition rate and diagnosis accuracy. Then the evaluation function is constructed based on information entropy and the method of allocation for classifier weights is researched. Since the fault identification rate is different among various fault types, so the method is improved by constructing classification capability evaluation matrix which can solve this problem effectively.The rules of evidence theory synthesis can not deal with high conflict evidence which may make the conclusion be inconsistent with the facts. The diagnosis model based on weighted evidence theory is put forward. The original evidence is weighted based on correct diagnostic rate of classifier, which reduces the rate of evidence conflict effectively and has been tested in experiments.