Research On Fault Diagnosis Of Cement Rotary Kiln Reducer Based On Support Vector Machines

Cement rotary kiln is the most important equipment of the cement productionline, the rotary kiln’s continuous and stable operation is an important way toimprove the production efficiency. Whereas the online monitoring and faultdiagnosis system of the rotary kiln is only available for parameters of the chemicalreaction process, as to supporting wheels and reducer, which also affect theoperation state of the rotary kiln, their maintenance is still being carried outmanually. Once these devices fail, the companies have to take the maintenance,resulting in large economic losses.The paper aims at the fault diagnosis technology, a module of the onlinemonitoring and fault diagnosis technology for the key components of cementproduction line, By the identification of mechanical equipment’s operation state,we hope to realize the detection of early failure, to quickly determine the cause offailure, and to avoid the failure to expand, thus to provide support for decisionmaking for the smart maintenance of the equipment.The vibration signal analysis techniques are currently the main method forgetting the failure information of rotating machinery equipment, the non-stationaryand non-linear characteristics of the vibration signal leave the outcomes oftraditional analysis methods, such as Fourier transform, wavelet transformmeaningless. Hilbert-Huang transform decomposes the signal to the intrinsic modefunctions, whose instantaneous frequencies are physically meaningful, thusHilbert-Huang transform is the effective way to analyze non-stationary signals.Hilbert-Huang transform is composed of empirical mode decomposition and Hilbertspectral analysis, with empirical mode decomposition the core of the Hilbert-Huangtransform. The paper does an analysis on the adaptive filtering characteristics of theempirical mode decomposition in detail and compares the Hilbert spectrum and thewavelet spectrum.Support vector machines roots its mathematical basis on statistical theory andis suitable for small sample classification and of greater generalization ability.Model selection is an important step in support vector machines’s training. the paperanalyzed the grid optimization and cross validation.The feature vector greatly affects the support vector machines’s classificationability. The singular values descript the essential characteristics of a matrix. Thepaper carries out the singular value decomposition of the Hilbert spectrum and usesits singular values as the feature vector, and make a detailed analysis whether thesingular values can reveal the intrinsic characteristics of the Hilbert spectrum or not. With the help of pattern recognition, it makes us avoid the influence ofsubjective factors of the engineering staff, and also provides a guarantee for theautomation of fault diagnosis.