Research On Light-Rail’s Screws Looseness Fault Diagnosis

Chongqing’s straddle-type monorail transportation is first in China. Anchoring screw is key light rail force components. Once they happen to be loose and are not found in time, this situation will lead to disastrous consequences. So working out a fast and accurate method to find out which Anchor Screw is loosed is one important embranchment of the total monitoring system. According to the actual situation of Chongqing rail transit, we use signal acquisition device to get signals and study the characteristics of the signal. In view of low detection accuracy and extreme lack of fault samples problem, we put forward a new effective diagnosis method, accomplished the fault diagnosis of anchor screws.First of all, author briefly introduced the anchor screws’ signal acquisition system. We use the improved wavelet threshold denoising method to denoise the collected signal. Denoised signal can facilitate the signal feature extraction.Then, according to the characteristics of screw signal, authors extracted 71 dimensional features from time domain, frequency domain and time-frequency joint domain using the ensemble empirical mode decomposition, smoothing Wigner- Ville distribution, wave packet decomposition method.Then, In the classifier, features are not the more the better, too many features not only can produce large amount of redundant data, increase the calculation time and cost, but also may produce overfitting and reduce the detection accuracy. In order to solve this problem, Genetic simulated annealing algorithm is introduced to select features.Finally we select an optimal feature subset.Finally, normal samples can be got easily, but fault samples can be obtained hardly in the fault diagnosis of anchor screws, the traditional binary classification method can not build up only according to the existing sample. In order to solve this problem, SVDD is introduced to build up a classifier only with normal sample. We uses optimal features set to train and test SVDD classifier, and results show this method is feasible. We compared the difference between SVDD and other commonly used one-class classifier in anchor screws looseness fault diagnosis, and results indicate SVDD classification is the best result.The above mentioned method has been applied into the anchor screws looseness fault detection. Experimental results indicate that the fraction of outliers which is accepted is zero, the accuracy rate of monitoring system is above 93%, it meets the requirement of fault detection.