Detection Of Micro Gap Weld Using Neural Network And Kalman Filter Fusion Algorithm Using Magneto Optical Images

This paper mainly focuses on improved algorithm for detecting and locating micro weld seam by using magneto optical imaging technique, in which the weld joint width is less than 0.1 mm. Laser welding has been widely used for its advantage in narrowly focusing laser radiation to a small area and high intensity heat source, which is instrumental in realizing deep penetration and high-speed welding and improved mechanical properties. However, this narrow area of the laser weld fusion zone brings about problems of joint alignment and fit up. Small focus wandering off seam will result in lack of penetration or unacceptable weld. The weld position is determined by many parameters in the thermodynamic and physicochemical process, such as the weldment thermal deformation and the laser welding speeds and so on. If an electromechanical or purely mechanical system is used to drive the-weldment, many nonlinear factors will also be inevitably introduced to the welding process and influence the system performance. The ability of detecting the weld joint is of primary concern in a laser welding system. Therefore, an auxiliary sensor should be used to detect and track the weld joint in real-time during laser welding.The most popular technique used for weld detection and seam tracking is based on the principle of optical triangulation. A Structured light is the process of projecting a known pattern of pixels (often grids or horizontal bars) on the weldment surface in front of the laser focus, and the reflected scattered light is imaged back to a camera. The way, that these deform when striking weld and seam, allows vision systems to calculate the weld center information in the scene. The controller extracts information from the image and uses it for seam tracking. However, to detect the weld is a formidable task in that spatter and light disturbance are very strong. With regard to a laser butt joint welding where the seam gap is less than 0.1 mm, the reflected light deforms so small that the weld is difficult to identified. More, in the welding process, because there are much light radiation and spatter, it is difficult to accurately detect the position of the weld. At present, there is no other effective automatic detecting and tracking methods for a micro gap butt joint welding.For the laser micro gap butt joint welding, using MOI technology to detect the weld is a new method. According to the magnetic induction principle and Faraday rotation effect, it is necessary to configure a MOI sensor and a magnetic excitation device to detect the weld during laser welding. The magnetic excitation device is used to magnetize the weldment, and it was found that magnetic field distribution at the weld joint is different from other regions. The magnetized weldment is detected by using a magneto optical (MO) sensor and the magneto optical images of the weld joint are captured. By analyzing the weld MO image features and characteristics, the weld position can be obtained, the offset between the laser beam focus and the weld center can be measured in real-time. There are much interference such as spatter, plasma and other radiation to influence on weld detection during laser welding. Unlike other visual sensing methods, the method to detect the weld using magneto optical imaging technology is based on the induced magnetic field distribution on the weldment surface. Using the MOI method can avoid the above interference and is expected to become an effective detection method of micro gap weld.In the process for detecting micro gap weld, the magnetic field is usually weak, and the signal related to the weld joint is difficult to obtain in an environment of the magnetic domains and other noises. It is important to eliminate the noises in order to enhance the weld MOI image, therefore, under low contrast and strong noise conditions, the image processing techniques are applied for the weld MOI and calculate the weld joint center. The laser welding experiments are carried out at different welding speeds. Experimental results show that the MOI technology is effective to detect the micro gap weld joint.A magneto-optical imaging corrected technique is proposed based on BP neural network to improve the detection accuracy of micro-gap weld seam (seam width less than 0.1mm). The butt joint laser welding of low carbon steel was carried out. The magnetized weldments were detected by using a magneto-optical sensor and the magneto-optical images of weld seam were captured. By using BP neural network and processing weld seam MO images with low contrast and strong magnetic field noises, the weld seam center position could be extracted accurately. Experimental results at different welding speeds indicated that MOI compensated technique with BP neural network could be applied to detect the micro-gap weld seam accuratelyDuring weld tracking process, there are various kinds of noise interferences which greatly affect the accurate extraction of weld information. Therefore, an optimum estimation method for system status based on Kalman filter is proposed to make accurate prediction on weld offset. The system status equation and weld center measurement equation are set up based on weld center parameters. The optimal prediction of seam offset could be obtained through the Kalman filtering algorithm under the least squares condition. Experimental results show that kalman filter algorithm can effectively eliminate noise and improve weld tracking stability.In the actual weld process, the system process noise and measurement noise is colored noise, it is difficult to accurately obtain the statistical characteristics of the noise before welding. To solve this problem, The RBF neural network and the Kalman filter algorithm were combined, and a Kalman filtering algorithm compensated weld offset by RBF neural network is researched. A RBF neural network is established, using Kalman filtering parameters as its input and filtering error as its output. The Kalman filtering result were modified by the RBF network and the filtering error caused by the noise statistical characteristics was compensated. The experimental results verify the effectiveness of the method in improving the stability of filtering.