Research On Welding Forming Quality Detection Technology Based On Visual Sensing And Deep Learning In Fusion Welding And Additive Manufacturing Process

In the field of fusion welding and additive manufacturing,the traditional off-line welding quality inspection can not fundamentally prevent the occurrence of welding defects,so the realization of on-line welding quality monitoring has become a research hotspot in this field.Weld penetration and reinforcement are important indicators reflecting welding quality,aiming at the on-line detection of welding penetration state and weld reinforcement of cladding layer in the process of fusion welding and additive manufacturing,this paper mainly carries out the following research work:(1)Aiming at the problem that it is difficult to detect the temperature field of molten pool on-line due to the influence of arc light and weld pool emissivity in the cold metal transfer(CMT)process,according to the characteristics of welding current and colorimetric temperature measurement method,two sets of temperature field detection systems for molten pool are constructed by using monochrome CCD and color CCD respectively.Aiming at the problem that the low measurement accuracy of its temperature measurement system due to the spectral response bandwidth of color CCD,a temperature field detection algorithm for molten pool based on back propagation(BP)neural network model is studied in the CMT process,which further improves the temperature measurement accuracy of temperature measurement system based on color CCD,the temperature measurement accuracy is better than 1.09%,and the non-contact on-line detection of temperature field of molten pool in the CMT process is realized.(2)Aiming at the problem that it is difficult to detect the change of welding penetration depth along the welding direction on-line at present,the welding penetration depth detection system based on dual CCD cooperation is constructed,and the welding penetration depth detection method based on the cooperation of molten pool shape and temperature is proposed,the detection accuracy of welding penetration depth is better than 0.12 mm,realizing the online detection of welding penetration depth.Meanwhile,the welding penetration state detection method for thin plate based on deep learning is studied,the recognition accuracy of welding penetration state is higher than 97%,realizing the on-line detection of welding penetration state of thin plate.(3)Aiming at the problem that when detecting the welding penetration state,the current weld pool visual sensing method is affected by arc light,the welding penetration state detection method based on infrared thermal imaging is studied,firstly,the welding penetration state detection method based on the cooperation of improved local binary pattern(LBP)and improved histogram oriented gradient(HOG)is studied,secondly,the welding penetration state detection method based on the geometric moment of welding temperature field image is studied,finally,the welding penetration state detection method based on welding temperature field and deep learning is proposed.The recognition accuracy of the above three detection methods for welding penetration state is higher than 95%,93% and 99% respectively,and the generalization ability of the proposed welding penetration state recognition network models is verified.(4)Aiming at the problem that it is difficult to detect the weld reinforcement of cladding layer on-line due to the influence of the remelting zone of cladding layer in the wire and arc additive manufacturing process,the weld reinforcement of cladding layer detection system based on dual CCD cooperation is constructed,the weld reinforcement of cladding layer detection method based on sequential molten pool image is proposed,the detection accuracy of weld reinforcement of cladding layer is better than 0.13 mm.Meanwhile,the weld reinforcement of cladding layer detection method based on multi-source information fusion and deep learning is proposed,the detection accuracy of weld reinforcement of cladding layer is better than 0.18 mm,realizing the on-line detection of weld reinforcement of cladding layer.Finally,by changing the cooling time between adjacent cladding layers,the generalization ability of the proposed weld reinforcement detection network models is verified.The welding forming quality detection technology based on visual sensing and deep learning in fusion welding and additive manufacturing process proposed in this paper realizes the on-line detection of welding penetration and weld reinforcement of cladding layer,which can be used as a reference for the research of welding quality detection.