Online Detection Method Of Weld Depth For Aluminum Alloy Overlap Joint Based On Spectral Analysis

Laser welding is a new material joining technology,which has many advantages such as high energy density,small deformation,high automation level,and it has been widely used in the automotive industry.In the welding process,some weld defects are prone to occur due to uncontrollable factors such as assembly errors.One of the most serious defects is insufficient penetration depth,especially for the overlap joint,in which the penetration depth directly determines the quality of weldment.Therefore,there is a great significance to monitor the penetration depth of overlap joint to ensure welds quality in actual production.Among many kinds of penetration depth detection methods,the spectral diagnosis method has attracted extensive attention from scholars around the world because of its advantages such as a large amount of information,low cost,simple operation,etc.In order to detect the penetration depth of lap joints of aluminum alloy in remote laser welding,the metal vapor above the keyhole and surface morphology were set as monitored objects,and the following study was conducted based on spectral diagnosis method:In this paper,four variables of laser power,welding speed,defocus and gap between plates were designed to obtain welds with different penetration depth.The metal vapor radiation above keyhole was collected by Princeton spectrometer during welding process,and 3D information of weld surface morphology was scanned by Keyence after welding.In this paper,the effects of variables on metal vapor radiation and weld penetration are studied.It is found that the weld depth can be detected through the radiation intensity of metal vapor.For example,the radiation intensity of metal vapor was improving with the increase of the weld depth,when the laser power or welding speed varied;the radiation intensity was decreasing with the increase of weld depth when the defocus changed;the increase of gap between plates significantly weakened the radiation intensity,but the penetration depth didn't change with the gap.In addition,these characteristic lines with different wavelengths had different sensitivity to variables,as well as weld depth,thus the change of weld depth could also be detected by the intensity ratio or intensity difference between different characteristic lines.The penetration depth could be detected through observing the change of weld surface information.The effect of variables on weld profile was studied,and the result showed that the penetration depth and weld surface width were significantly affected by laser power,welding speed and defocus.If laser power increased or welding speed decreased,the penetration depth and weld surface width as well as weld depth-to-width ratio increased rapidly.When the defocus increased,the weld surface width increases first then decreases,while the penetration depth and depth-to-width ratio was always decreasing.The penetration depth decreased slightly when gap between plates appears,but if gap value continues to increase,the weld surface width and penetration depth changed little.Combined radiation signal and weld surface information to improve effective information source and extracted 16 indicators.There are four kinds of(four total)indicators extracted from weld surface information,which were surface width,surface contour area,average reinforcement,and maximum weld notch;five(12 total)kinds of indicators were extracted from radiation signal,which are the average intensity of characteristic lines,the intensity difference between different characteristic lines,the intensity ratio between different characteristic lines,the correlation coefficient between different characteristic lines,intensity ratio between characteristic lines to continuum.A weld depth detection system and an interface width detection system were built based on the BP neural network.These two systems can detect the value of penetration depth and interface width according to 16 indicator extracted from radiation signal and weld surface information.In addition,these two detection models were optimized from the activation function and the input indicators.Finally,the average error of optimized depth detection model is 0.14 mm,and the average error of optimized width detection model is 0.18 mm.