Acquisition And Recognition Analysis On The Ultrasonic Testing Of Rail Crack Data

With electronic and computer technology develop rapidly in recent years,NDT(nondestructive texting) field also enhanced continually with digital andintelligent, in order to improve accuracy of detection and stability of diagnose.Ultrasonic texting technology was used and developed immensely for the fileddue to the detection equipment and method in search of low cost. It isincreasingly significant to maintain and text for the rails using, with the railtransportation development desired. Under the background, rail crack detectionand recognition researched in the paper, rail cracks data were sampled withdigital ultrasonic texting technology, and were recognized by mature and stablearithmetic.Firstly, feature of ultrasonic and operating principle of crack detection wereintroduced. The feasibility and major parameters of pulse transmitted circuit fortransducer verified under Simulink tool and analyzed the influence of the circuitto ultrasonic testing, the ultrasonic testing system for steel rail based onultrasonic pulse reflection method was proposed. Pulse transmitted circuit andecho signal reception circuit were researched. With the digital programmable ofFPGA, data timing of system was controlled and processed, rail cracks data wereassembled correctly, and the detection system proposed was great accuracy andlow cost.Finally through further research based on rail cracks data collected andanalysis of data recognition. The echo signal was divided into IMF with EMD,which the arithmetic can divide non-stationary signals into different frequencycomponents of the IMF. Rail crack signal was represented accurately by sixcharacteristic values, which they are zero crossing amount, signal area, centerfrequency, signal energy and maximum amplitude in time and frequency domain.There is better capacity of signal decomposition than traditional arithmetic. Combined with the neural network algorithm for samples were trained repeatedly,and the defect signal characteristic values normalized as input of the neuralnetwork. Through it was simulated and analyzed by neural network kit underMatlab tool, simulation results show that non-stationary signals split by EMDand neural network algorithm were used for effective recognition of the defectsignal. The system achieved the desired effect, and can complete the acquisitionand recognition of rail crack data.