The Quality Inspection & Prediction Of Rail Flash Butt Welding Joints

The rail welding plays a significant role in the construction of continuous railways, for it directly determines the quality of rail welding joints and the security of railways transportation. Now, long rails can be welded in field or in factory, flashing butt welding is always used in factory, and gas pressure rail welding and mobile flash butt welding are mainly used in field. The voice of flash in rail flash butt welding process produced by in contact with the liquid lintel blasting, in the methods of quality of flash welding joints on-line monitoring, using flash acoustic signals for the flash welding process monitoring and quality of diagnosis is still in the blank stage.In this paper, synchronous data acquisition system was established, including welding current, welding voltage and flash acoustic signals in flash butt welding machine, and the data obtained by experiment. first, extract the flash rate of the characteristic parameters of flash stability based on welding current, comparative analysis with the stage of rail flash welding; second, use of modern signal analysis method, extract the value of characteristics in time-domain and frequency-domain based on the flash acoustic signals of flash stability, and comparative analysis with the characteristic value of the flash rate signal based on the welding current; finally, the use of wavelet analysis to extract flash acoustic signal band energy information, calculated the energy value of frequency range and the percentage of it's accounted for the total energy of the signal, establish the prediction model of neural network, the percentage of the energy structure feature vector as input vector, the flash rate of the characteristic parameters of flash stability as output vector.The results show that, the size and trends of flash rate reflected two very important stages'characteristics in rail flash welding process, that is low-voltage end of the period and speed up the flashing stage; the mean-square values and average power spectrum values of flash acoustic signal can be used for Eigen values of the stability of flash; the characteristic parameters are extracted by wavelet analysis as input, trained RBF neural network can be mapped the size of the flash rate to a certain extent.