| In aluminum alloy pulsed gas tungsten arc welding (GTAW) process, both real-time penetration information and well controller are important for stable weld shape closed-controlling. Because aluminum alloy is special, the welding process information sensing and control algorithm are still bottlenecks in automatic welding. Moreover, gap is inevitable in welding process while it effects on weld shape seriously. So systematic research is done in aluminum alloy pulsed GTAW process information sensing and intelligent control.According to the requirements of aluminum alloy the GTAW control system, multi-light-route visual sensing subsystem is designed to catch weld pool image from top-back, top-front and back directions simultaneous. The sensor provides powerful tools for the investigation of the relation between the top and the back weld pool parameter.Narrowband multiple filter technology is developed after analysis on the are sensing mechanism and experiments in aluminum alloy GTAW process. Dirrent ultra-red filter glass and neural density filter glass are selected for the three light-routes, separately. The effects from obtaining images time, base level current, peak level current on visual sensing are studied carefully under several technical conditions. Lastly, the clearly and stable aluminum alloy weld pool images are obtained while the welding technics and visual sensing parameters are confirmed. Based on the experiments, general weld pool image and four weld pool defects images are collected. Extensively, the defects characters are summarized, separately.Combined with welding pool variation in different penetration, several parameters about weld pool geometry and welding directions are defined. For top-back image, wavelet transform and Canny operator advantages are used to get weld pool edge points. After noise removing and calibration, subsection polynomial curve fitting method are used to recove the whole weld pool edge. For the top-front image, degrade recover, automatic threshold segmentation, noise removing, thinning and Hough transform are used to computing gap size and welding direction. For the back image, algorithm including noise removing, automatic threshold segmentation, edge detection is used to recover the whole back contour of the weld pool. From these recoved whole weld pool edge and the weld pool parameters'definitions, the values of the parameters can be computed.Not interface with the above image processing flow, weld pool defect can be inspected on-line by the image processing software and these defects character evaluation functions are defined for them. For weld wire misdirection image, it can be judged by comparing with the gray value difference of two sides of the weld pool because wire only appears on one side each time. For misalignment image,there are obvious misalignment of its abscissa value in the front up and down part of the pool edge after calibration. So the different of the average gray values in the up and down sides separately may be used to judge the defect. For burning through image, its back image is quite bright. It means the average gray value of the special region may show if there is burning through. For incomplete penetration, its back wide of the weld pool is the best direct paramenters.To study the dynamic relation between weld pool characters and weld technics parameters, several single input and single output transfer function models are established through step response identification experiments. The inputs include peak current, wire feeding rate, gap size, weld speed. And the outputs are weld pool parameters. By analyzing effects of welding parameters on weld shaping, it s discovered that arc welding is characterized as multi-variables, strong coupling, nonlinear, time varying. The simple conventional models are not enough to describe the complex system.To establish the welding process dynamic models, the appropriate arrangements of welding current, wire feeding rate, gap are determined. Stochastic experiments are done to extract the shape parameters of weld pool when the three welding parameters are stochastic variable. After data extension and noise removing, two prediction models are constructed by using RBF neural net. The models are and predictions model , and predictions model . Through these two models, offline simulation experiments are applied to analyze the dynamic process of weld pool shape. The simulation result agrees to the actual manual experiments. It proves that the two models are right and feasible.For the workpiece shape, groove shape and technics conditions provided for the experiments, the characters of the technics and weld shaping are analyzed carefully. Also SISO control experiments are carried out. Here gap size in [0, 0.5mm] is regarded as disturbance and set as 0.3mm. To compare with the capability of different controller, PID controller and MS-PSD controller are designed, separately. Simulations and welding experiments proved that, PID controller works not well when varied heat-sink and varied gap appears. For the shortcomings of PID, MS-PSD is designed to obtain better weld quality. It can adjust its parameters on-line by itself. Simulation and welding experiments proved that, MS-PSD controller can ensure both varied heat-sink and varied gap workpieces welding process stable. At the same time, the experiments also show that the topside of the weld pool is not in control. The topside surface sinks severely, which effects on weld shape.Avoiding the limitation of traditional intelligent controller, rough set (RS) theory is introduced into welding process control considering both intelligence and comprehension. After data extension, noise removing, data discrete preprocessing, attribute reduction, attribute value reduction and rule reduction, SISO RS controller is designed. Simulation and welding experiments proved that, the control result is acceptable for back width of varied heat-sink or varied gap workpiece.To control both top-front width and back with of the weld pool, double inputs and double outputs RS&MS-PSD multi-controller is designed. In this controller, wire feeding rate is used to control back width and weld current is used to control top-back width. Simulation and welding experiments proved that: the controller can insure weld shape from both sides of the weld pool. Moreover, topside height is also better than ever since wire feed rate is adjusted.Welding experience shows that: weld defection increases quickly when gap size is greater than 1.5mm in aluminum alloy GTAW with groove but no backing. The welding technics can not ensure stable welding shaping. So the gap size arrangement is confirmed as [0,1.5mm].After analyzing the relation between gap size and wire feeding rate under idea state, a wire compensation formula based on experience and workpiece groove style is raised up when there is big gap in real welding process. Extensively, new RS&MS-PSD controller with parameter preset is designed. Two types of workpieces with different big gap are provided to check the validation of the controller. When gap changes gradually, the controller is effected on both sides of the weld zone. The welding shaping is uniform. When gap changes suddenly, the shaping in the position where gap just appears is worse than elsewhere. For those big gap workpiece, the multi-controller with parameter preset can ensure both sides shaping of the weld zone. When gap is 0, the controller equals to the RS&MS-PSD multi-controller.
|
PDF Full Text Request