Research On Gearbox Fault Diagnosis Method Based On Cascaded Black-Hat And White-Hat Transform Morphological Filtering And Bnn

The healthy operation of mechanical equipment is a necessary condition for ensuring economic benefits and avoiding casualties.Rolling bearings are important components of industrial machinery and are related to the health of the entire mechanical system.It is particularly important to develop equipment condition monitoring and fault diagnosis techniques for bearings.Due to the complex transmission path of the vibration signal,the signal acquisition signal is nonlinear,non-stationary,and environmental noise interference,etc.,the signal feature extraction becomes more and more difficult.Mathematical morphology is based on random set theory and integral geometry,and is suitable for nonlinear and non-stationary analysis.This paper considers the analytical ability of morphological filter and morphological spectrum in the fault signal of rolling bearing,and combines it with the width learning network to obtain good machine diagnosis.The paper mainly studies the following:(1)An improved differential filter is proposed for extracting the impact component of the rolling bearing and adaptively selecting the impact direction.Conventional differential filters only consider the extraction of the impact component without considering the direction of the impact,which results in frequency aliasing when there is coupling of the impact frequency.Since the overall energy of the signal remains basically the same,the frequency coupling causes the energy to be dispersed and is more likely to be submerged in the noise.In this paper,a new method is proposed to identify the impact direction by white cap transformation and black cap transformation,and to correct the impact direction of the differential filter.Simulation experiments show that the improved differential filter extracts the frequency of the impulse component consistent with the original signal,avoiding the frequency coupling phenomenon of the differential filter.Combining it with iterative morphology and passing the rolling bearing fault diagnosis test,the results show that the spectral components are clearer,which proves that the method avoids the energy dispersion caused by frequency confusion and effectively extracts the fault component information.(2)A bearing defect detection method combining the Pattern Spectrum(PS)and the Broad Learning System(BLS)is proposed.The existing deep learning establishes the recognition model in the multi-layer stack information processing module,and the parameterupdate needs to be corrected layer by layer.Deep learning requires a complete retraining process when the structure is not sufficient to model the system.The feature extraction of the signal can compress the data on the one hand,reduce the data dimension and improve the training speed of the model;secondly,it can reduce the noise interference and improve the training precision of the model.In this paper,a defect identification method combining morphological spectrum and width learning is proposed.The morphological spectrum or low-dimensional feature mapping of vibration signals is used,and the constructed data is used to establish a BLS fault diagnosis model.The rolling bearing fault diagnosis test results show that the method can effectively extract the signal fault component,identify the fault type and improve the recognition accuracy.