Geometry And Stability Monitoring In Welding Formation Processing Based On Multi-bands Vision And Transient Spectrum

Intelligent sensing and control is a hot research topic in the field of welding.Traditional manual and offline quality checking cannot fulfill the requirement of rapid increasing amount of manufacturing and complex welding processing.Thus,online quality monitoring has become more and more urgent in welding processing.Online multi-information acquiring and analyzing can provide real-time weld seam evaluation and control guidance for welding manufacturing,so as to realize the quality supervision and intelligent control of welding.Weld pool vision imaging characterizes the stability of the melting,flowing and the solidification process of metal.Arc spectrum characterizes the stability of the arc directly and further assists the quality monitoring of weld seam.This paper focuses on weld pool vision imaging and arc spectrum analysis.The main contributions of this paper are listed as follows.(1)Optimal imaging band selection method for weld pool imaging based on spectrum analysis.For the disunity of existing weld pool imaging band selection and lacking theoretical guidance,optimal imaging band selection method for weld pool imaging based on spectrum analysis is proposed.Through measuring the spectrum of self-emitted radiation of weld pool and arc spectrum,the proposed method can seek the optimal imaging bands.At the same time,an FPGA trigger module is designed to reduce arc interference based on the welding current and obtain high quality weld images with lower arc interference.On the basis of obtained experimental results,a coaxial multi-bands weld pool imaging system is designed and a multi-bands weld pool image fusion method is proposed for the first time,which include grayscale image fusion method based on non-subsampled wavelet transform and an improved Reinhard color image fusion method.The grayscale fused image has abundant details and clear texture,the color fused image has sharp contrast and harmonious color.(2)High-speed arc spectrum acquisition system with sub-Hadamard matrix.For the existing Hadamard transform spectrometer(HTS)cannot measure transient phenomena spectrum,a new snapshot and high sensitivity spectrometer is proposed.A specially designed extra imaging path is added in traditional HTS to collect light intensity distribution of the scenes.Both light intensity distribution of the scenes and the overlapped dispersed spectra are measured at the same time.Through reversible reconstruction,the spectrometer can realize snapshot.Both simulation and experimental results have demonstrated that the proposed snapshot HTS can obtain almost the same denoising capability and high spectral resolution arc spectrum with one shot,compared to traditional HTS.(3)Weld pool stability monitoring based on random subspace semi-supervised manifold classification.For the inaccuracy of existing single band of weld pool contour extraction,an online weld pool contour extraction and weld seam width prediction based on dual bands weld pool imaging is proposed,which can realize high precision weld pool contour extraction and weld seam width prediction.For the existing weld pool stability monitoring mainly based on artificial experience,random subspace semi-supervised manifold classification is proposed to monitor weld pool geometry.Local property of each feature is used to select the features,the selected probability of each feature in the original space is corresponding to its local property.In this way,the important feature has higher selected probability and improves the performance of the method.In many random subspace methods,all random subspaces are independent.Random subspace fusion method is proposed to fuse all random subspaces and improve the performance.Extensive experimental results with controllable welding speed and welding current have demonstrated the effectiveness of the proposed two methods,which can realize geometry stability monitoring of weld pool.(4)Arc stability monitoring based on discriminative sparse manifold regularization.For existing threshold based method cannot process drastic changing arc spectrum in pulsed welding processing,discriminative sparse manifold regularization is proposed to monitor arc stability in pulsed welding processing.In many semi-supervised classification with sparse representation,all samples share the same dictionary and do not update the dictionary in terms of the sample property.In the proposed method,the whole unlabeled sample set is used to reconstruct the mean value of each labeled class and the dictionary of the corresponding sample is updated according to the obtained sparse coefficients.In this way,the speed of sparse representation is accelerated which obtains better classification performance.Extensive experimental results with controllable shielding gas flow rate have demonstrated the effectiveness of the proposed methods,which can realize arc stability monitoring of the weld pool.