Research On Multi-feature Index Of Welding Quality Of On-line Monitoring And Evaluation Technology

The welding process is a complex physical,chemical and metallurgical process.With the change of arc sound,pool,spectrum,current and voltage,extracting the characteristic parameters from the above signals and evaluating the welding quality has become a research hotspot in the field of welding technology.The current and voltage signals in the welding process are the direct manifestation of the state of the arc and the transition type of the droplet.In this paper,aiming at the characteristic relationship between welding current,voltage signal and droplet transfer and the stability of welding process,a sensor acquisition system for welding arc signal(current and voltage)has been developed for GMAW.In order to study the relationship between the signal and the welding quality,the statistical characteristics,time-frequency domain characteristics and nonlinear characteristics of electrical signals with different protective gas flow,current,voltage and welding speed are analyzed.On the basis of further to establish a prediction model of welding the quality of the information,and revealed about the quality of welding which contains in the signal waveform.A series of on-line monitoring system for welding status of welding workshop is developed.The system includes bottom device,terminal,electronic board,hardware and collection module,server database and client.The system has the support of a hundred level welding equipment network monitoring.It has the functions of real-time monitoring,welding time,protection gas,consumables management,historical data query,welding quality evaluation,report printing and other functional modules.Through the real-time sensing of the electrical signals in the short-circuit transition of GMAW welding,extracting statistical characteristics parameters,such as short-circuit time,arcing time,short-circuit current rising rate,and study the relationship between the characteristic parameters and the welding process stability.When the welding process is unstable,the number increases which short-circuit and arcing time is too small or too large,and may appear instantaneous short circuit.Meanwhile,the number also increases which the rising rate of short-circuit current is too large or too small.The translation invariant wavelet transform method is applied to filter and denoise the short-circuit current and voltage signal.Compared with the traditional Symlets function soft threshold wavelet filtering,the signal to noise ratio exponent increases and the local details of the signal are smoother.Empirical mode decomposition(EMD)and complete ensemble empirical mode decomposition(CEEMDAN)were used to study the distribution and variation of Hilbert-Huang spectrum and marginal spectrum under different technological conditions.When the protective gas shortage,the Hilbert-Huang spectrum distribution becomes uneven,and the marginal spectrum in frequency components of low frequency components increases.When the voltage is too low,the weld quality variation,compared with stable welding voltage,the Hilbert-Huang spectrum distribution becomes uneven,and the frequency components in the low-frequency components of the marginal spectrum becomes larger.When the current exceeds normal short circuiting transfer,the droplet transition state changes from short circuited transition to large droplet mixed with short circuit.The uneven distribution of Hilbert-Huang spectrum increases,and the frequency component of low frequency component increases in the marginal spectrum.On this basis,the time-frequency entropy and the marginal spectral index are introduced to quantify the distribution and variation of Hilbert-Huang spectrum and marginal spectrum.The overall trend of change is that the more stable the welding process is,the larger the time frequency entropy and marginal spectral index.The entropy value distribution of welding current signal is processed by multi-scale entropy,and the entropy distribution tends to a horizontal state with the increase of the scale.The overall trend of change is that the more stable the welding process is,the more small the multi-scale entropy.The multifractal analysis is used to analyze the characteristics of welding current.Because the multifractal spectrum of current signal presents a single convex distribution,it is concluded that the current signal has multifractal characteristics.On this basis,the multi fractal spectrum characteristic index is extracted.When the welding process is stable,the mean of the multifractal index becomes smaller and the variance becomes smaller.The detrend multifractal method is used to analyze the welding current signal,and the detrending multifractal spectrum index is extracted.When the welding process is unstable,the change of the mean value of the detrending multifractal spectrum index is not obvious,but the variance becomes larger.On this basis,the cause of multifractal generated by the welding signal is further studied.The feature vectors composed of Hilbert-Huang spectrum,marginal spectrum index,multiscale entropy,multifractal index and detrend multifractal index.The prediction results show that according to the welding quality(good forming,overlap,ripple,big spatter and gas hole)classification,genetic algorithm optimization neural network forecasting accuracy can reach 95.0%and the prediction of genetic algorithm optimization support vector machine accuracy can reach 96.1%;according to the droplet type classification(normal short-circuit transition,big droplet and short-circuit mixing transion,short-circuit and spray droplet mixing trasition),and hybrid genetic algorithm to optimize the neural network prediction accuracy of 100%,predictive genetic algorithm optimization support vector machine accuracy can reach 99.0%.