Kalman filtering optimized by multi-innovation theory for on-line weld detection

Amagneto-optical sensor was used to capture the images of butt joint weld of low carbon steel plate whose gap is less than 0.05 mm. In order to detect and track the weld,Kalman filtering which is optimized by a multi-innovation theory was applied to the magneto-optical image that contains the weld location information. Because of the noise interferences in acquiring magneto-optical images and data extracting, the standard Kalman filtering that is easily affected usually can not work efficiently. Therefore it is difficult to make an optimal estimation for the deviation in seam tracking by using the standard Kalman filtering. Based on the multi-innovation theory, an optimized Kalman filtering algorithm was proposed to detect the weld. When the present weld position is predicted, this optimized Kalman filtering algorithm was fully applied for the useful information of previous moment and historical data to estimate the weld location information. Experiments using different innovation values were carried out and the amount of calculation and filtering precision was considered. It was found that the optimized Kalman filtering algorithm could obtain better effect by using two innovation values. Experimental results showed that the seam tracking accaracy could be improved via using Kalman filtering optimized by multi-innovation theory.