针对V型坡口中厚板对接焊,研究一种噪声环境下应用径向基函数(Radial Basis Function,RBF)神经网络补偿卡尔曼滤波(Kalman Filter,KF)误差的焊缝跟踪方法。根据三角测量原理设计激光结构光视觉传感器并采集焊缝图像,对受噪声干扰的焊缝图像进行预处理。采用基于光强度分布特性的灰度平方加权重心法提取结构光条纹中心线,通过道格拉斯-普克算法(Douglas-Peucker algorithm)和最小二乘法相结合的方法提取焊缝特征点。建立描述焊缝中心位置的系统状态方程与测量方程,运用RBF神经网络补偿卡尔曼滤波模型误差及噪声统计不确定性造成的滤波误差,修正卡尔曼滤波估计值。试验结果表明,RBF神经网络补偿卡尔曼滤波能够减小噪声干扰的影响,提高焊缝跟踪精度,有效抑制卡尔曼滤波发散。

A weld seam tracking method based on radial basis function (RBF) neural network compensation for Kalman Filter(KF) error under noisy environment for V-type groove medium thick plate butt welding is studied. A laser structured-light vision sensor was designed with the triangulation principle, the weld images were acquired and the images of the weld that is disturbed by noise were preprocessed. The center line of structured-light stripes was extracted by the gray-level square weighted gravity center method based on the characteristics of light intensity distribution. By using Douglas-Peucker algorithm and least squares method, the weld feature points were extracted. The system state equation and measurement equation were established to describe the center position of the weld. The RBF neural network was used to compensate for the Kalman filter model error and the filtering error caused by the uncertainty of noise statistic characteristics and amend the estimation values of Kalman filtering. The experimental results show that the Kalman filter algorithm compensated by RBF neural network can improve the seam tracking accuracy, and reduce the interferences of noises and suppression of Kalman filtering divergence effectively.