Knowledge Discovery Research On Spindle Component Fault Of CNC Machine Tools

Fault diagnosis and maintenance of CNC machine tools are important parts of machine debugging and processing, which is one of the main factors restricting CNC machine tools playing a normal role.Simple fault of components which are connected to system as well as common electrical system fault can be self-diagnosed presently. As to high-frequency mechanical failure which leads to declining machining quality, self-diagnosis is still of blind zone. Spindle component fault accounts for a considerable proportion of such fault, which has been puzzling the field of fault diagnosis for CNC machine tools at home and abroad.From the perspective of Soft Computing Theory, the dissertation discusses and investigates key issues regarding fault knowledge acquisition data preparation stage and fault knowledge discovery stage.Two tasks are done in data preparation stage of knowledge acquisition.In the first place, the two major components of the CNC machine tools spindle system, rolling bearings and gears, are selected as research objects. By comparative study of general mechanical vibration mechanism with vibrations of rolling bearings and gears, it can conclude that the main fault of rolling bearings are surface wear and spalling while the main fault of spindle gears derives from tooth surface uniform wear and localized spalling. As to the fault of rolling bearings, the corresponding amplitude about base frequency as well as its integer or fraction multiple frequency are selected as characteristic parameters in the process of knowledge acquisition. As to the fault of gears, vibration signal meshing frequency as well as combined spectrum produced in sidebands can be used in fault diagnosis. In regard to measurement point arrangement optimization, finite element modeling and harmonic response analysis can be introduced in determining the theoretical optimum position of spindle box vibration measurement points. The fault simulation experiment system of rolling bearings and gears is established for original data acquisition.In the second place, from the perspective of data collecting and processing, a third-order low-pass Butterworth filter and a third-order high-pass Butterworth filter are used to establish a band-pass filter for filtering. Besides, the filtering process is mathematically analyzed. In order to achieve sensor acquisition data fusion, data fusion method of arithmetic mean and recursive estimation is proposed regarding single sensor data fusion timing problem. Thus, measurement results which are more reliable than arithmetic mean are acquired. As to the spatial problem of multi-sensor data fusion, a multi-sensor data weighted fusion algorithm is proposed. Based on measurement result of each sensor, different sensors look for corresponding weights in an adaptive way under optimal conditions that the total mean square error is minimum. In this way, the fused data are optimal. Besides, information entropy is used to evaluate the effect of data fusion.In the stage of fault data knowledge discovery, different soft computing methods are used to acquire fault knowledge rules based on rolling bearing and gear fault respectively, through which self-diagnosis is achieved.Based on acquired data of rolling bearing fault experiment, algorithm combining equally spaced clustering with attribute importance reduction as well as algorithm combining k-means clustering with discernibility matrix reduction are applied to acquire knowledge and rules of surface wear, spalling failure and normal state mode respectively.A kind of BP neural network of three-tier network architecture model is established according to typical CNC machine tools spindle gear fault diagnosis. After training and simulation of the experimental data samples, the result shows that the method can be used to recognize gear tooth surface even wear failure, tooth surface localized spalling failure and normal state.