微间隙焊缝磁光成像识别模型研究

发布时间：2018-07-31 14:41

【摘要】：激光焊接技术具有激光功率大、光斑直径小、光束质量优良、热影响区域小、大深宽比、可实现异种材料之间的连接并且焊接质量优良等优点。激光焊接过程中,控制激光束实时准确地对准焊缝中心位置是保证获取良好焊件的关键。由于激光束光斑直径小(一般小于200μm),对焊缝间隙大小敏感,要求焊缝间隙尽可能小。传统的结构光视觉法利用结构光横跨于焊缝位置所产生的突变特征实现焊缝识别,但无法识别小于0.10mm间隙的焊缝。在实际工业焊接现场存在大量的烟雾、飞溅及等离子体等干扰影响,普通摄像机无法清晰捕捉焊接区域熔池和微间隙焊缝位置的准确信息,且激光焊接过程中存在剧烈的热能转换效应,对焊接工艺参数及工件的装配、固定精度要求极高,微小的变化即可导致严重的焊接缺陷甚至报废,因此,精确控制激光束使其始终对正并跟踪焊缝是保证激光焊接质量的前提。论文综合比较了现有焊缝识别与跟踪方法的优缺点,结合实际工业需求,重点研究激光焊接微间隙(不大于0.20mm)焊缝磁光成像焊缝识别技术。针对激光焊接等厚、无坡口、紧密对接、肉眼难以分辨的微间隙焊缝,根据法拉第磁光效应原理构成磁光成像传感器获取焊缝磁光图像,参与设计并搭建了激光焊接不锈钢紧密对接焊缝磁光成像试验平台(第二章),研究激光焊接前微间隙焊缝的磁光成像特征和机理,实现微间隙焊缝位置检测,为后续激光焊接过程中焊缝识别与跟踪奠定基础,保证激光焊接质量。首先,将焊件放置于伺服工作台上,在焊缝下方放置磁场发生器,通过调节磁场发生器的励磁电压大小,改变焊缝周围的感应电流及感应磁场强度。根据法拉第电磁感应效应及法拉第磁光效应,当涡流在流动路径上存在焊缝间隙时,其流动受到影响,涡流会在焊缝位置处产生畸变,畸变的涡流会产生畸变的涡流磁场,从而引起该位置处垂直磁场分布的变化。通过磁光传感器将涡流磁场变化转换成相应的光强变化,实现焊缝的实时成像,研究微间隙焊缝磁光成像特征与焊缝位置的关联。结果表明,改变励磁电压、磁光传感器与焊件距离、焊接速度、焊缝间隙大小等参数,对微间隙焊缝磁光成像的变化较为敏感。其次,分别利用微间隙焊缝磁光图像特征(灰度特征、灰度梯度特征、彩色空间特征和纹理特征),探索各特征与微间隙焊缝位置之间的规律。分析微间隙焊缝磁光图像灰度和灰度梯度分布特点,通过全局阈值和边缘算子,可以提取出焊缝过渡带轮廓并将其中心位置认作焊缝中心位置坐标,但阈值的选取不具有通用性,当微间隙焊缝磁光成像试验参数改变,需要多次反复试验选取合适的阈值。利用焊缝磁光图像的彩色空间特征,在RGB和HSV颜色空间分别计算各彩色分量图的灰度分布特征后提取焊缝位置坐标,彩色颜色空间焊缝位置测量精度较灰度图的焊缝位置测量精度高。最后,分析微间隙焊缝磁光图像序列特性,利用图像序列中各像素数据的时域变化和相关性确定各像素位置的运动和焊缝位置坐标。研究了光流法和梯度矢量流模型在微间隙焊缝磁光图像识别中的应用。另外,利用人工神经网络、粒子滤波和卡尔曼滤波算法构建描述焊缝位置的识别模型,最终实现微间隙焊缝识别、跟踪和预测,保证焊接质量。通过这一系列的研究工作,取得了如下主要研究成果：(1)研究焊缝磁光成像与焊缝实际状况及其它因素间的内在关系。对于一个特定的磁光传感器,影响微间隙焊缝磁光成像的主要因素为法拉第磁光效应,相关因素包括：励磁电压、激励线圈与焊件的距离(提离度)、焊缝形态以及焊接速度等。试验表明：针对同一微间隙焊缝,改变励磁电压大小,焊缝周围的磁感应强度随之改变,焊缝位置处磁感应强度介于两边磁场(N极和S极)强度的对称中心。焊缝磁光图像中的焊缝过渡带区域随励磁电压改变而上下移动,但对于两个不同的磁场强度下所获得的磁光焊缝位置的偏移量是恒定的,即对于特定的磁场强度环境下此偏移可以忽略不计,不影响实际的焊缝识别和跟踪。针对同一微间隙焊缝,改变焊接速度,磁光成像焊缝位置测量值基本恒定不变,即焊接速度仅仅对工件的焊透情况和图像采集帧数有影响、对微间隙焊缝位置的检测无影响。同一励磁电压下,改变焊缝间隙大小,间隙越大,磁光成像中焊缝位置过渡区域越小,焊缝区域成像越清晰。同一焊缝间隙和励磁电压下,磁光传感器与焊件距离(提离度)越近,磁光成像中焊缝位置过渡区域越小,焊缝区域成像越清楚。(2)微间隙焊缝磁光图像特征的提取与分析。微间隙焊缝磁光图像特征包括：灰度特征、灰度梯度特征、纹理特征和图像序列特征。焊缝介于两块母材的中间,垂直扫描焊缝位置,焊缝左右两边母材的灰度分布存在明显的差别,可利用灰度分布在焊缝位置处的差异性检测焊缝中心位置。扫描微间隙焊缝磁光图像所有列的灰度梯度分布曲线,寻找灰度梯度极大值所对应的行作为焊缝过渡带的上、下边缘坐标,计算焊缝过渡带上下边缘的中心坐标作为焊缝位置坐标。分别在图像的焊缝区域和母材区域(焊缝上、下区域的母材)提取三个相同尺寸的子图像,计算各个子图像的纹理特征(包括：平均亮度、标准差、平滑度、三阶矩、一致性、熵1、能量、相关、熵2、逆差矩等),利用纹理特征差异,将焊缝和母材区域进行分割。(3)微间隙焊缝磁光成像焊缝位置检测。采集微间隙焊缝磁光连续图像序列,利用采样时间重叠的图像序列光流场分布,根据图像序列的时域特性和相关性,提取光流场中的u分量峰值处对应的像素点作为焊缝位置。H-S(Horn,Schunck)光流法焊缝识别方法所计算出的焊缝位置大部分与焊缝位置实际值吻合,由于焊缝磁光图像存在噪声干扰影响,光流法焊缝提取方法在若干部位出现较大的波动,在工程实践中,可根据工程经验,设定一个阈值将波动较大的值视为粗大误差,予以剔除。同时,分析了梯度矢量流场在微间隙焊缝磁光图像分割中的应用,将感兴趣区域的焊缝边缘看作具有能量的不间断曲线,在控制点所具有的能量控制下使得活动轮廓产生变形,活动轮廓在控制点的内力、外力以及图像力的共同作用下,向焊缝目标区域进行伸缩,最终实现焊缝位置检测。(4)微间隙焊缝位置识别模型的建立。分别建立了基于BP神经网络、基于Elman神经网络、卡尔曼滤波预测焊缝位置模型。设计了一种前馈型神经网络焊缝位置预测模型,通过对前一时刻的焊缝位置和焊缝位置差值来估算当前时刻焊缝位置。比较了BP神经网络和Elman网络焊缝位置预测的精度。结果表明：BP神经网络的预测能力比Elman神经网络更强,能有效地进行焊缝位置的预测,且测量精度优于Elman网络。利用Kalman滤波对微间隙焊缝进行跟踪和预测,在已知焊缝位置测量信息的前提下,获取系统状态的最优估计,最后实现焊缝位置的最佳预测和估计。卡尔曼滤波后,噪声干扰得到较大地抑制,能有效提高焊缝跟踪精度。
[Abstract]:Laser welding technology has the advantages of large laser power, small spot diameter, good beam quality, small heat affected area and large depth width ratio. It can realize the connection between different kinds of materials and excellent welding quality. In the process of laser welding, it is the key to ensure the accurate alignment of the weld center by controlling the laser beam in real time. The light spot diameter of the beam is small (generally less than 200 m), which is sensitive to the size of the weld gap, and requires that the weld gap be as small as possible. The traditional structure optical vision method uses the abrupt characteristics of the structure light across the weld position to realize the weld recognition, but can not identify the weld less than the 0.10mm gap. There are a lot of smoke in the actual industrial welding site. With the influence of the interference of splash and plasma, the ordinary camera can not clearly capture the accurate information of the weld pool and the micro gap weld position, and there is a severe heat transfer effect in the process of laser welding. The welding process parameters and the assembly of the workpiece and the fixed precision are very high, and the small change can lead to serious welding defects. As a result, it is a prerequisite for accurate control of the laser beam to make it always positive and tracking weld is the prerequisite for ensuring the quality of laser welding. This paper compares the advantages and disadvantages of the existing welding seam recognition and tracking methods. Combined with the actual industrial demand, the paper focuses on the study of the laser welding micro gap (less than 0.20mm) weld recognition technology of magneto optic imaging weld. Light welding with equal thickness, no slope, close butt, undistinguishable micro gap weld, magneto-optical imaging sensor based on Faraday magneto-optical effect principle to obtain magnetic and optical images of weld, and to design and build a laser welded stainless steel close butt weld magneto-optical imaging test flat (second chapter), to study the micro gap weld before laser welding. The characteristics and mechanism of magneto-optical imaging can be used to detect the position of the micro gap weld. It lays the foundation for the seam recognition and tracking in the follow-up laser welding process and ensures the quality of the laser welding. First, the welding parts are placed on the servo worktable, and the magnetic generator is placed under the weld, and the excitation voltage is changed around the weld line by adjusting the excitation voltage of the magnetic field generator. According to the Faraday electromagnetic induction effect and the magnetic field intensity of the induction magnetic field, the flow is affected when the eddy current exists in the weld gap on the flow path, and the eddy will distort at the weld position, and the distorted eddy current will produce the distorted eddy current magnetic field, resulting in the distribution of the vertical magnetic field at this position. Change. The change of eddy magnetic field change into corresponding light intensity change through magneto-optical sensor, real-time imaging of weld seam, the correlation between magneto optic imaging characteristics of micro gap weld and weld position. The results show that the excitation voltage, the distance between magneto optic sensor and welding part, welding speed, weld gap size and so on, and the micro gap weld magnetism The change of optical imaging is more sensitive. Secondly, the characteristics of the magneto-optical image of the micro gap weld (gray feature, gray gradient feature, color space feature and texture feature) are used to explore the regularity between the features and the position of the micro gap weld. The characteristics of the gray degree and gray gradient distribution of the magneto optic images of the micro gap weld are analyzed, and the global threshold and edge are used to get the edge of the weld. The edge operator can extract the contour of the weld transition zone and recognize its central position as the position coordinates of the weld center, but the selection of the threshold is not universal. When the magnetic imaging test parameters of the micro gap weld are changed, many repeated tests are needed to select the appropriate threshold. Using the color space features of the weld magneto-optical image, the color space of RGB and HSV is empty. The weld position coordinates are extracted after calculating the gray distribution characteristics of each color component, and the precision of the weld position measurement accuracy is higher than that of the gray map. Finally, the characteristics of the magneto-optical image sequence of the micro gap weld are analyzed, and each pixel is determined by the time domain change and correlation in the image sequence. The application of optical flow method and gradient vector flow model to the recognition of magneto optical image in micro gap weld is studied. In addition, the recognition model of weld position is constructed by artificial neural network, particle filter and Calman filtering algorithm. Finally, the recognition, tracking and prediction of micro gap weld seam can be realized, and the welding is ensured. Quality. Through this series of research work, the main achievements are obtained as follows: (1) the study of the intrinsic relationship between the weld magneto-optical imaging and the actual conditions of the weld and other factors. For a specific magnetic and optical sensor, the main factor affecting the magneto-optical imaging of the micro gap welds is the Faraday magneto-optical effect, and the related factors include the excitation electricity. Pressure, the distance (lift off degree), weld shape and welding speed of the excitation coil and the weldment. The test shows that the magnetic induction intensity around the weld changes with the size of the excitation voltage changed for the same micro gap weld, and the magnetic induction intensity at the weld position is at the symmetry center of the magnetic field (N pole and S pole) on both sides. The transition zone of the weld transition zone moves up and down with the change of the excitation voltage, but the offset of the magneto optic weld position obtained under two different magnetic field strength is constant, that is, the offset can be ignored in a specific magnetic field intensity environment and does not affect the actual weld recognition and tracking. Speed, the measurement value of the weld position of the magneto optic imaging is basically constant, that is, the welding speed is only influenced by the penetration of the workpiece and the number of the image collection frames, and there is no influence on the detection of the position of the micro gap weld. The more clearer the image is. The closer the distance between the magneto-optical sensor and the weldment is, the closer the distance between the magneto-optical sensor and the weldment is, the smaller the weld position transition region is, the more clear the image of the weld area is. (2) the extraction and analysis of the characteristics of the magneto-optical image of the micro gap weld. The characteristics of the micro gap welding seam magneto optical image include the gray feature and the grayscale gradient special. Signs, texture features and image sequence characteristics. The weld is in the middle of two pieces of parent material, and the weld position is scanned vertically. There is a significant difference in the gray distribution of the left and right sides of the weld. The gray distribution at the weld position can be used to detect the position of the weld center. The gray gradient distribution of all the magneto optical images of the weld is scanned. Curve, the line corresponding to the maximum value of grayscale gradient is used as the upper and lower edge coordinates of the weld transition zone, and the center coordinates of the upper and lower edges of the weld transition zone are calculated as the position coordinates of the weld. Three subimages of the same size are extracted in the weld area of the image and the base material on the base of the weld, and the sub images are calculated. The texture features (including: average brightness, standard deviation, smoothness, three moment, consistency, entropy 1, energy, correlation, entropy 2, inverse moment, etc.), use the difference of the texture features to divide the weld and the base area. (3) the detection of the weld position of the micro gap weld magneto optical imaging. The continuous image sequence of the micro gap weld is collected, and the overlap time is overlapped. According to the time domain characteristics and correlation of image sequence, the corresponding pixel points at the peak of the U component in the optical flow field are extracted as the weld position.H-S (Horn, Schunck) method for welding seam recognition. Most of the weld position is in accordance with the actual value of the weld position, because the magnetic and optical image of the weld is noisy. The welding seam extraction method of the optical flow method has a large fluctuation in several parts. In engineering practice, according to the engineering experience, a threshold value is set to remove the fluctuating value as a rough error. Meanwhile, the application of the gradient vector flow field in the segmentation of the magneto-optical image in the micro gap weld is analyzed, and the weld seam in the region of interest is used. The edge is regarded as an uninterrupted curve with energy. Under the control point energy control, the active contour is deformed. The active contour is expanded to the weld target area under the joint action of the control point, external force and image force, and the weld position detection is finally realized. (4) the establishment of the recognition model of the micro gap weld position. Based on the BP neural network, based on the Elman neural network and the Calman filter to predict the weld position model, a prediction model for the weld position of the feedforward neural network is designed. The weld position at the present time is estimated by the difference between the weld position and the weld position at the first time. The weld position of the BP neural network and the Elman network is compared. The results show that the prediction ability of the BP neural network is stronger than that of the Elman neural network, and it can predict the weld position effectively, and the measurement accuracy is better than that of the Elman network. The Kalman filter is used to track and predict the micro gap weld, and the optimal estimation of the system state is obtained on the premise of the known weld position measurement information. Finally, the optimal prediction and estimation of the weld position are realized. After Kalman filter, the noise interference is greatly suppressed and the tracking accuracy is effectively improved.
【学位授予单位】：广东工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TG441.7;TP391.41

【相似文献】