Envelope Analysis For Gear Faults Detection And Classification Using Artificial Neural Network

Gears has been widely used as integral parts for many equipments and machines in various industries, The rapid growth and development of modern technology has caused the equipments to become more sophisticated and therefore imposing a very challenging task in monitoring their operating conditions. Being part of these complex mechanical systems gearbox should be monitored closely and effectively so that any defect can be detected and timely corrected in its early stages.The envelope analysis has long been widely used for bearings and gear fault diagnosis basically involving the use of residue signal obtained after synchronous averaging before band pass filtering around the significant gear meshing components. However in some applications the synchronous averaging of gear vibration signal can hardly be achieved and therefore there need to be other technique to facilitate envelope analysis for gear fault diagnostic. This study is therefore aimed at solving the above mentioned problem by introducing a demodulation technique, which involve band pass filtering around the exited structural resonance away from any significant gear meshing harmonics, to detect and classifying each of the two types of local gear faults (gear tooth crack and spall) with different severity levels. It is known that a local fault on gear tooth will generates a low-level impulse every time the fault contacts another gear tooth surface during meshing causing a modulation effect on the gear vibration signal and possibly exciting the structural resonances. These low-energy impacts are always overwhelmed by the high energy vibration signal from other sources in the gear box, so the main idea is to develop a suitable technique which can extract these defect induced impulses from which the presence and severity of the fault can be revealed. The envelope analysis proposed here was accomplished by;â‘ Observing the presence of defect induced resonances using spectrum of the original vibration signal collected from the gearbox and then selecting a suitable demodulation band from which the impulse caused by a local gear faults impacts can be extracted and used for diagnostic purpose.â‘¡Choosing the center frequency for the band pass filtering around the excited structural resonance away from any significant gear meshing harmonic with the band width covering the whole resonance range, this was done with the help of original signal spectrum and was observed to be effective in removing the dominant effect of gear meshing components.â‘¢Using Hilbert transform for demodulation process, the analytic signal obtained from the Hilbert transform of the band pass filtered vibration signal is used to extract the modulating signal component, the resulted signal envelope only contain the components that are related to gear fault frequencies. The envelope spectrum is then obtained by taking FFT of the envelope signal from which the featrures related to conditions of the gears in the gearbox can be extracted.â‘£Exproling the the envelope spectrum by observing the charecteristcs of different frequency components in the spectrum (i.e the meshing frequency and its associated sidebands) to extract the basic discrimating features from each of the tested gears envelope spectrum and use them as the input to the neural network classifier for classification.The features considered were the amplitudes of the meshing frequency and the two pair of sidebands around it, the neural network classifier used in this study was back propagation neural network.Good diagnostic results for both gear tooth crack and gear tooth spall faults considered in this study were obtained using the proposed method which underline the effectiveness of the proposed method for gear fault diagnosis.