Research On The Fault Diagnosis For Assembly Of Rotating Components In The Heavy-duty Horizontal Lathe

The heavy-duty horizontal lathe belongs to the special machine tool, and is mainlyapplicable to the manufacturing of large-scale and heavy-duty units which is used inaerospace, marine, nuclear power and so on. As an essential part of the heavy-dutymachine tool, the main spindle box is responsible for the transmission of power,supporting of the rotational components and guaranteeing the rotary accuracy ofworkpiece. What’s more, the quality of assembly plays a significant role in guaranteeingthe reliable power supply and steady operation of machine tool, which is working underlarge load and cutting force. The size of heavy machine tool is much larger than thesimple one, which leads to the raise of difficulty at the assembling stage. Therefore, theassembly of such gearbox can’t be accomplish in one step, during such process manytimes of modulation that aims to eliminate the assembly faults and match therequirement of finished product is necessary. Consequently, diagnosing the kind of faultin assembly precisely makes great contribution to adjusting the assembly moreefficiently and guaranteeing the outgoing quality of such machine tool.In this dissertation, based on the research of the methods used for diagnosis and thecharacteristics of different malfunctions in gearbox, combing the practical experience ofQIQIHAR Heavy CNC EQUIPMENT CO.LTD in assembling the main spindle boxused in heavy-duty horizontal lathe, the common modes of failure in gearbox assemblyand the characteristics along with them are determined. In order to diagnose the faulttypes, PCB356A16accelerometer sensor is selected to pick up the vibration signal andthe LMS Lab.Test is used as the acquisition software, and the raw vibration data isobtained.Wavelet transform is used to process the signal. The applicability of differenttransformation methods and wavelet base functions are compared under the citation ofthe Shannon-entropy, amount of energy and E-S ratio. Because of the orthogonality,symmetry and the comparison result, bior3.5wavelet base function, which is organizedby spline interpolation, is selected to analyze the signal. By wavelet packet method, theraw signal is decomposed into different frequency ranges. Then the nodescorresponding to the frequency ranges which contain specific frequency points arereconstructed, and the calculated power spectrum can demonstrate the feature moreclearly without other interference. After analysis, the number of feature vectordimension is set as23, and structure of the23-dimension feature vector used for faultdiagnosis is determined. The feature vectors are extracted from the obtained signals by the method demonstrated above.At last, the three intelligence pattern recognition methods—SVM，k-NN andBP-ANNs are utilized to classify the fault modes respectively. And the key parametersof each method are optimized. What’s more, the three methods are compared in terms ofthe accuracy of classification and the complexity of calculation. After the comparison,SVM is evaluated as the most appropriate method，which can accomplish the faultmodes classification quickly and accurately.