Vision sensors adopted for automatic seam tracking in fusion welding are generally based on optical triangulation. They basically have an auxiliary light source and preview distance - the distance between welding and sensing position. It can generate some tracking error in automatic seam tracking, especially in micro welding of small parts. On the other hand, vision systems adopted for the weld pool monitoring are generally equipped with a high-speed camera and intense illumination.
In this study, a vision sensor was applied to find the weld seam by monitoring the weld pool in pulsed Nd:YAG laser welding. The vision sensor with a 810nm bandpass filter and 1/30000 shutter speed observed the weld pool and the weld seam was extracted from the image of monitored weld pool. No auxiliary illuminating light source was used for monitoring the weld pool. Pulsed Nd:YAG laser welding is characterized by the interaction of multiple laser pulses, while the shape and brightness of the weld pool change temporally even in one pulse duration. Therefore, the shutter of the CCD camera was synchronized with the start of each pulse, and opened with a specific time delay from the laser pulse start. Experiments were carried out to find the appropriate moment of shutter triggering for relatively clear images.
The principle of seam tracking using the developed vision sensor is based on the separation angle between the vision sensor and Nd:YAG laser focusing optics to find the 3D information of weld pool centerline. The CCD camera and Nd:YAG laser beam, instead of the laser diode plane as in the generally used vision sensor, have a separation angle and the weld pool is imaged in CCD. As a result, the cross-section of Nd:YAG laser beam in a developed vision sensor can be regarded as the plane of laser diode in other vision sensors using a structured light. Consequently the weld pool centerline, where the cross-section of laser beam intersects the weld pool, is acting as the laser stripe on workpiece.
Weld pool image contains the information about the weld penetration. The information about a penetration can be extracted from the geometry and brightness of weld pool. Weld pool parameters that represent the characteristics of the weld pool image were selected based on the geometrical appearance and brightness profile. In order to achieve accurate prediction of the weld penetration, which is non-linear model, neural network with the selected weld pool parameters was applied. The tests for the proposed monitoring and prediction method were performed to the various weld conditions and workpiece conditions. As a result of the test, the proposed parameters including the information from the brightness profile resulted in the improvement of prediction because of the more accurate description of the weld pool image. Finally, penetration control was successfully performed to the workpiece of continuous width change. As a specific target, the previous penetration prediction and control test were applied to 200 μm edge joint.
