Rotating Machinery Fault Diagnosis Method Based On Acoustic Features And Its DSP Implementation

The acoustic signal contains a large number of fault information when mechanical equipment failure occurs.The use of acoustic signal in monitoring mechanical equipment technology does not need to set up contact sensors,and thus reduces the complexity of the system.Aiming at the phenomenon of rotating mechanism failure which often occurs in industrial production practice,this paper proposes a dimensionality reduction method which combines the acoustic signal characteristics with the machine learning data to realize the fault diagnosis of gears.Firstly,the common failure phenomena of rotating mechanism in industrial production are introduced,and the main forms of gear failure and its mechanism are analyzed.Because the fault acoustic signal is often mixed with other acoustic signals,which will cause interference to the acquisition of the target signal,the fault acoustic signal is firstly separated by using the multi-layer non-negative matrix decomposition method,and the fault acoustic signal is pre-weighted,windowed,and framed to extract the MFCC characteristics of the fault acoustic signal.Due to the high dimensional and nonlinear characteristics of fault acoustic signals,it is necessary to conduct dimensionality reduction processing on the acquired fault acoustic signal features,so as to reduce the complexity of the algorithm and identify fault acoustic signals more quickly and accurately.According to the acquired feature data set of acoustic signal,an isometric mapping algorithm is used to reduce the dimensionality of the acquired feature data set and compared with the LLE local linear embedding dimensionality reduction method.The simulation results show that the two methods have obvious dimensionality reduction effects on the high-dimensional feature data set.After the dimensionality reduction processing of the extracted high-dimensional MFCC sound signal characteristic parameters,the low-dimensional sound signal characteristic parameters were obtained.Then the k-means clustering analysis method was used to conduct clustering analysis on the sample points after dimensionality reduction to realize fault diagnosis and monitoring of the gear.The results show that the reduction of high-dimensional MFCC feature parameters by Isomap,combined with K-means clustering analysis,has a certain improvement in the accuracy of gear failure recognition after LLE dimensionality reduction.Because the fault sound signal information of the gear is mainly present on the high-frequency or detailed component of the signal,the wavelet decomposition of the detail signal is then subjected to Hilbert transform,and the high frequency component of the fault signal is analyzed to determine the gear's working conditions.The third layer detailed signal envelope spectrum under the four working conditions of the gear was analyzed according to the experimental results Under the gear pitting state,the power spectrum is abrupt when the frequency is 500 Hz;under the gear broken state,the power spectrum changes suddenly when the frequency is 400 Hz.Under the condition of gear wear,the power spectrum attenuates with the change of frequency to the minimum,and thus the failure state of pitting,wear or broken tooth of gear can be obviously judged.Finally,the DSP implementation method of fault diagnosis algorithm is presented.With CCS5.5 and TMS320 as the core,the development process of the whole system is designed,including the design of the hardware development environment and the design of the software system.