Rotary Machinery Non-stationary Feature Representation And Extraction In Variable Rotating Speed For Mechanical Fault Diagnosis

Rotating machine takes up the major part in all machine equipments and plays animportant role in industry enterprises. It is very important to ensure the running safetyof rotating machine for both enterprises and our social economy. Local faults in rotatingmachine components including bearings and gears tend to result in shocks and thusarouse transient impulse responses in the vibration signal. So the transient featurerepresentation and extraction is the most crucial problem for the reliability and accuracyin mechanical fault diagnosis. With the research target of rolling bearings and gears,which are the typical rotating machine components, the paper proposes thetime-frequency representation based on the fusion of time-frequency feature, themethod of time-frequency resampling, the synchronous enhancement algorithm ofperiodic transient feature and the automatic detection of transient vibration in variablerotating speed. The theoretical research and application research are studied in depth,respectively.The failure mode of bearing and gear and the kinematics of rolling bearing wereintroduced in this paper, to ensure that the theoretical research works were all validatedby the experiment analysis, bearings were tested under the localized fault condition andthe vibration signals were collected at constant rotating speed and in variable rotatingspeed respectively.The time-frequency analysis (TFA) methods contain two classes, the linear TFAand the bilinear TFA. The linear TFA methods have no cross-terms and low in resolution,the bilinear TFA methods are on contrary. In order to overcome the effect of thetime-frequency features detection from cross-terms, a novel scheme known as thetime-frequency fusion algorithm (TFFA), this is similar with the logic “AND”. It selectsa TFA method from linear TFA methods and an another TFA method from bilinear TFA methods, then apply the TFFA, and fine time-frequency features can be obtained, whichkeeps fine resolution and suppress the cross-terms. The effectiveness and theapplicability of the method in fault feature detection is tested and verified through theapplication in the simulation signal analysis and the vibration signal of both the bearingand the gear.At varying speed, the transient is not periodical anymore, but relating to therotating speed, or proportional to the order. In practice, at varying rotating speed, theorder analysis is employed for vibration signal feature analysis by using order asfrequency axis of the spectrum. The essence of order analysis is resample for the signalin the time domain, which transforms the non-periodical signal features in time domaininto periodical signal features in order domain to meet the need of power spectrum.However, the resample in time domain shall cause obvious distortion，especially forthe information in frequency domain, because the transient tends to be oscillatedamping vibration of high frequency. In time-frequency domain, the transient isrepresented as energy distribution, and there is no obvious oscillation, but slowvariable for the transient feature. So, for transient feature represented intime-frequency domain, and resample in time-frequency domain shall avoid distortion.Considering the above-mentioned facts, a new technology called time-frequencyfeatures resample (TFFR) is proposed, the basic principle of which is an extension ofthe resample for a signal in time domain to the resample for the time-frequency data intime-frequency plane that transforms the non-periodic transient features to periodictransient features.Through the proposed TFFR algorithm, the transient feature shall be periodicallyrepresented in the angular-frequency domain. However, it is still extremely tedious toobtain the relation between the transient period in angle domain and the fault type. Thispaper proposes the synchronous enhancement of periodic feature (SEPF) method,which is applied for detecting the bearing inner race fault, outer race fault and rollingelement fault caused by localized defect at varying rotating speed; the transient featuredetection result shows clearly that t is rather easy to discriminate the fault according tofeature distribution in the polar diagram. Considering the mechanical fault diagnosis in variable rotating speed, a novelscheme known as the detection of machinery transient features in variable rotatingspeed is proposed based on the proposed TFFA, TFFR and SEPF. First, two kinds ofTFA methods, the WVD and the WS, are applied for the target signal, which is invariable rotating speed, the fine time-frequency features are obtained by using theTFFA for the outcomes of the two TFA methods. And then, TFFR transforms thenon-periodic transient features in time-frequency plane to periodic transient features inangle-frequency plane. Finally, the method SEPF based on polar diagram makes theperiodic transient features map to a certain area on polar diagram, which detectstransient features of vibration signal in variable rotating speed by representation andenhancement.This research is supported financially by the Natural Science Foundation of China(No.50905021).