Adaptive Detrended Fluctuation Analysis As A Fault Diagnosis Method To Study Gear’s Vibration Signal

Gear, as transmission parts, because of its high transmission efficiency, transmission ratio,transmission and other characteristics of the acclaimed accurate, is widely used in variousmechanical devices. With the constant changes in technology, machinery and equipmentadvances, emerging, gear position is becoming more and more important, the significance of thegear fault diagnosis is particularly prominent.Since the gear system is nonlinear and complex, determines its presentation ofnon-stationary vibration signals and multi-scale features, differentiate similar failure mode isespecially vital to fault diagnosis. From the nature of the gear vibration, requires the use ofreasonable parameters detailed description of the vibration behavior, studies have shown thatmulti-scale analysis is more consistent to the nature of the gear vibration analysis, and scalingexponent is an important parameter to describes the multi-scale behavior.Scaling exponentresponse to gear fault condition is very sensitive and can be used to characterize the gear faultcondition. Detrended fluctuation analysis using non-scaling behavior of gear vibration signals toextract reliable results, over time scales, the scaling behavior of the gear double-scaling features,the scaling exponent corresponding intercept consisting of two-dimensional feature vector has aclear physical meaning, being characterized as a state characterized by the amount of gearfailure.With researching detrended fluctuation analysis，we could find that the algorithm itselfdoes not overlap interval.Being obtained from the empirical formula, there is no reasonableproof. In fact, it would influence the scaling exponent result in deep. Proposed adaptivedetrended fluctuation analysis method, using the knowledge of information theory, combinedwith sliding windowing algorithms, while achieving reasonable selection of adaptive scalingexponent extracted, and the using of neural network algorithm successfully distinguish thenormal wear and tear, as well as pitch error four kinds of failure modes broken tooth gear.Studies show that this method,as a new theory,can improve extraction scaling exponent for faultdiagnosis of gear provides,becoming more efficiency and more accuracy.