Defect Recognition For The Welding Rotor Of Steam Turbine Based On Ultrasonic Signal

As the core component of large power rating nuclear power generating units or steam turbine with high steam parameters,the welding rotor needs to withstand high temperature,high stress,high-speed and other harsh working conditions.The welding quality of its weld and newborn defects during operation directly affect the safe operation of the unit.It is of great significance to identify the defects of girth weld by means of ultrasonic nondestructive testing technology.Recognition of defects in traditional ultrasonic testing generally observes the echo signal by inspection personnel.But this method has a great subjectivity,low detection efficiency and high false positive rate.With the development of artificial intelligence and instrumentation technology,it is more and more important to use computer aided recognition.This paper focuses on the key point of defect recognition and mainly studies the signal de-noising,feature extraction,feature selection and pattern recognition in ultrasonic detection technology.The main contents are as follows:(1)Research on de-noising method of ultrasonic echo signal.Aiming at the problem of difficult to remove the structural noise in ultrasonic testing signals,this paper presented a de-noising method which combined variational mode decomposition(VMD)and wavelet energy entropy threshold.Firstly,the characteristic of entropy increase in noisy system and the distribution feature of structural noise in different period were analyzed.Then,the state of signal with noise was characterized by wavelet energy entropy and the threshold of wavelet decomposed in different scales ware determined according to the wavelet energy entropy.Simulation and experimental results show that the de-noising method(VMD-WEET)in this paper can restrain the noise and restore the accurate waveform to verify its effectiveness.(2)Feature extraction of defect signals in girth weld of welding rotorThe ability to identify representative and valid feature information from defect echo signals directly affected the accuracy of defect type identification.In view of this,this paper identified the multi-feature extraction frame of defect signal.Firstly,different types of defect signals in the time domain were analyzed.Then,form factor and crest factor of the time domain waveform were extracted and used as the eigenvalues of the time-domain waveform.Furthermore,the feature extraction of time-frequency domain was accomplished by wavelet packet analysis with good time-frequency analysis ability.Finally,the complexity and uncertainty of the signals of different defect types were analyzed from the perspective of information entropy and the variational modal entropy was extracted as the eigenvalue that identified the signal of different defect type.(3)Automatic identification of girth weld defects in welding rotorAiming at the difficulty of identifying defect types in small sample,a pattern recognition method based on support vector machine was proposed.Firstly,the classification principle of SVM was introduced.Then,particle swarm optimization(PSO)algorithm was proposed to optimize SVM parameters.Furthermore,the principal component analysis(PCA)was used to reduce the dimension of the eigenvector set.Finally,the defective identification of girth weld in welding rotor was accomplished by using the reduced feature vector set and the established PSO-SVM classifier.