Research On Mechanical Fault Feature Extraction Based On Dynamic Behavior And Signal Morphology

Mechanical fault diagnosis technology is crucial to monitoring conditions of continuously operating mechanical equipment and conducting fault diagnosis and prognostics for the purpose of ensuring the safe operation of the mechanical equipment.Fault feature extraction is the key to realizing fault monitoring and diagnosis.Fault feature extraction techniques based on vibration signal analysis have always been a leading edge and hotspot in the field of fault diagnosis.In recent years,as equipment maintenance management has developed toward the on-condition maintenance and predictive maintenance in condition monitoring,incipient fault detection,performance degradation prediction,and condition assessment of mechanical equipment have become problems demanding prompt solutions in fault diagnosis.Therefore,research on feature extraction has considerable practical significance for the overall development of fault diagnosis and prognostics.In the case of malfunction of critical kinetic components of mechanical equipment,particularly the rotational components such as gears and bearings,vibration signals often contain transient impact responses,which can substantially influence dynamic behaviors of the entire system.Moreover,fault vibration signals of different facilities,different types or different degrees generally have different morphological characteristics.Considering this,based on the existing studies,research on fault feature extraction was implemented in this work from the perspective of analyzing the dynamic behaviors of failure system and the morphology of fault signal.The primary tasks accomplished in this work are introduced below:1)Investigation on the dynamic behavior feature extraction of the backlash fault and hunting tooth problem of gear transmission system was performed.The calculation method of dynamic backlash in backlash faults was determined.The characteristics of hunting tooth problem were simulated using the variations of transmission error and meshing stiffness.Accordingly,a dynamic model of the failure system was established.By conducting numerical simulations and experimental studies,features that can characterize the behaviors of backlash fault system,such as oscillation and modulation frequencies,dynamic transmission error frequencies,and state of impact,were obtained.Causes of hunting tooth problem were discussed,and the corresponding dynamic behaviors were analyzed.The investigation also indicates that the approach of calculating transmission errors by using the angular displacements of driving gear and driven gear collected from high-resolution encoders can efficiently identify some fault feature frequencies.2)Detecting and extracting the transient components in fault signals that contain a large amount of fault information is the base for behavior feature extraction and condition identification of fault,especially for incipient and slight fault detection.In this regard,associated method was introduced and improved adaptively by considering the characteristics of fault vibration signal analysis,based on the idea of detecting salient voices in noise environment by using the mechanisms of human auditory attention in the field of voice signal analysis.A new approach of transient component detection based on auditory saliency calculation was proposed,which was subsequently validated using the fault signals that contain transient components.3)A method to extract the impact feature of mechanical fault based on signal sparse representation theory was developed.In order to construct a redundant dictionary more compatible with the impact component in fault signal,based on the idea of system mode parameter identification to obtain the actual impulse response of the fault system,two new approaches of dictionary construction were proposed:1)an approach using Laplace wavelet obtained from the combination of ensemble empirical mode decomposition(EEMD)and correlation filtering as the basis function and 2)an approach using the measured impulse response excited at the fault point based on experimental mode analysis method as dictionary atom.Over-matching of sparse decomposition algorithm was overcome by conducting threshold de-noising on the coefficients of sparse representation.On this basis,a method of impact feature extraction based on mode identification dictionary for sparse representation modeling was created.Application cases indicate that more accurate and sparse reconstitution of fault impact component can be obtained using this approach to further realize the effective extraction of fault features.4)A method to distinguish different faults according to the morphological features of fault signals was developed.Based on the multi-scale morphology theory,by using a morphological corrosion operation to replace the opening operation,the pattern spectrum algorithm was improved.A mathematical morphology algorithm named corrosion pattern spectrum and a fault feature extraction method based on this algorithm were proposed.Analysis results demonstrate that morphological component content of fault vibration signals in various analysis scales can all be described using this method.Compared to the feature extraction method based on pattern spectrum,this method has superior capability of quantitatively distinguishing the fault states as well as strong stability and high calculation efficiency.5)A fault prognostics-oriented feature extraction approach for mechanical equipment was developed.Performance degradation caused by fault generation and deepening is likely to affect the distribution and variation of morphological components in fault signals.Based on this consideration,a new method for signal morphological complexity measurement,corrosion pattern spectrum entropy,was proposed,and its application as a feature parameter in prognostics feature extraction was discussed as well.Analysis demonstrates properties and variation patterns of the pattern spectrum,pattern spectrum entropy,corrosion pattern spectrum and corrosion pattern spectrum entropy in the process of performance degradation.The efficiency of the prognostics feature extraction based on corrosion pattern spectrum entropy was verified.Moreover,a deeper degradation degree of equipment performance could lead to more complex signal morphology and a higher value of corrosion pattern spectral entropy.At the end,as a summary to the entire dissertation,the developing tendency of fault feature extraction as well as the emphases and hotspots of future exploration of associated topics was also discussed.