Study Of Bulk Material Loading Control System Based On Machine Vision

Bulk material loading exists in many engineering fields. Bulk level detection and carriage speed control are key points for automatic bulk loading. However, ultrasonic ranging bulk level detection has some drawbacks such as that it can’t detect bulk level beneath discharge opening directly and get integrate bulk contour rapidly, and the detection result is easily affected by environment. Besides, bulk loading control realized by presetting of the carriage speed. To solve the above problems affecting bulk loading control and system reliability, this dissertation first proposed a machine vision based bulk material fast-loading control system, contour of bulk-stack in the carriage was extracted from image of loading operation, then discharge opening and carriage move was controlled. At last, continuous loading can be executed automatically. The major contents of the dissertation include following points:1.Design of system structure and test platform. Main and secondary function of bulk loading machine vision system were discussed in this section, and result shows that bulk level control can be realized under a monocular vision system through real-time detection of discharge opening and bulk level. Machine vision feedback control principle, vision module arrangement and test platform was designed. Mode of vision feedback movement was studied and a polynomial planar calibration method was proposed according to structure of bulk loading system. Finally, a image template matching algorithm for discharge opening detection and a texture segmentation algorithm for bulk-stack contour were proposed.2.Research of rotation and illumination invariant template matching algorithm for discharge opening real-time detection. A matching algorithm taking GMM parameters as matching feature was proposed to improve RPT algorithm’s locating performance. Because NCC algorithm is noneffective to GMM parameters when illumination changing, a GMM parameters update method carried out by linear contrast stretch was proposed to realize illumination-adaptation ability. Compared with pixel-wise linear contrast stretch, the proposed method need much lesser computation. Look-up-table and down-sampling strategy were used to reduce calculation in feature extraction process. At last, kalman-filtering was used to revise locating result. Result shows that the proposed algorithm obtain more remarkable feature similarity peak value, lesser error locating and is insensitive to number of RPT rings. Average locating error is within 5 pixels. Under search area of 1010? pixels and rectangle template of 30~100pixels, the matching speed is less than 30 ms.3.Research of real-time and illumination-robust bulk-stack contour detection algorithm. To avoid complex calculation in image enhancement process, a non-overlapped blocking local binary method was applied to uneven-illumination bulk-stack image. Study shows that under block size which is bigger than bulk-granule, gray bulk-stack image can be converted to an even binary granular-texture image without blocking effect, and the binary texture is different from other background’s in some degree. At the same time, benefiting from only 2 gray level, time consuming of GLCM feature abstraction is much lesser. In order to solve problems of inevitable SVM error-identification and cover by other object, a MRF segmentation model combined with SVM posterior probability and constraint of bulk-stack position was adopted to block-sub-image recognition result. The proposed segmentation algorithm avoided texture modeling and improved illumination robustness by utilizing inherent characteristics of the bulk-stack image. Finally, in each column, several overlapped blocks of certain spacing interval were recognized by texture feature, and the greatest changing position was chosen to fitted out the bulk-stack contour. Result shows that the SVM recognition rate is 95.2%, MRF segmentation accuracy reaches 98.8%. Average contour error is about 2.2pixels, and the standard deviation of the error is 6.1pixels. Average detection speed of each frame which is 270600 ? pixels is about 9.7ms, and at same condition, speed of standard ICM segmentation is about 20.7ms.4.An improved method of continuous bulk loading level control by carriage moving speed self-regulation approach. This section focused on coordination control of discharge opening and carriage moving. Based on analysis of system model, fuzzy-cascade level control method and manual bulk loading process, a carriage moving speed adaptive control method was proposed in order to solve the problem that the moving speed must be preset by carriage size. Fuzzy rules of discharge opening increment target, acceleration of moving speed and forward discharge opening increment target were designed. PID control algorithm was used to adjust discharge opening in inner feedback loop. Result shows that the proposed method can adapt to different bulk level target, and the carriage can reaches maximum moving speed corresponding to discharge capacity without presetting. Error of bulk-stack level control is within-6.2pixels, and standard deviation of the error is 4.3pixels.