Search

Automated rubber tapping not only frees the workers from heavy physical labor and harsh working conditions, but also reduces the dependence on the workers' skills and greatly increases tapping efficiency. The key technologies for tapping robots are the independent acquisition of operational information and servo control of the tapping position in unstructured environments. In this study, taking rubber tree in rubber plantations as object, incorporating robot kinematics, machine vision technology and multi-sensor feedback control technology, a modular prototype of a rubber tapping robot was developed. The rubber tapping robot was mainly composed of an orbital mobile platform, a multi-joint robotic arm, a binocular stereo vision system and an end-effector. The binocular stereo vision and structured light system were used to obtain the structural parameters of the rubber trunk and secant. A six-joint tandem robotic arm was used for the planning and realization of complex rubber tapping trajectories. An multi-sensor fusion end-effector was developed to complete the identification of the starting point, the measurement of cut compensation and the tapping operation. To address the technical difficulties in rubber tapping operations, such as complex and variable environment, superimposed interaction of operational information, similar target background features and sub-millimeter operational accuracy requirements, the spatial mathematical model of the rubber tapping trajectory was established to plan the robot's movement path for fast approaching and moving away from the operation space. The results of the field tests conducted at a natural rubber plantation in Hainan province showed that the accuracy in bark consumption was about 0.28 mm and the accuracy in cutting depth was about 0.49 mm when the rubber tapping robot cut 1 mm thick bark. Compared to manual operations, the continuity of the chips and the flatness of the rubber output surface were improved significantly. This research could provide a positive reference and development direction for exploring automated rubber tapping operations.

To detect the coal-carrying rate in gangue, a new method based on three-dimensional (3D) image features and gray wolf optimization-support vector machine (GWO-SVM) was proposed. First, region segmentation was performed for each sample image according to the particles' shapes automatically, and each region was identified as coal or gangue. For each region, 4 geometry features and 11 density features were extracted, respectively. In each sample image, the geometry features of the coal region were added by category, while its density features were averaged by category, and the gangue region performed the same operation. The initial feature set, i.e. a two-dimensional (2D) image feature set was obtained by dividing the gangue feature value by the coal feature value. Binocular images were then gathered to extract the third-dimensional (3D) feature, that is, the height feature of the particles. The final feature set consisted of the 2D image features and the third-dimensional feature. Extreme Gradient Boosting was applied to calculate the importance ranking of all the features, and the top nine features were selected as the best features. Then, the coal-carrying rate in gangue was modeled using the SVM optimized by gray wolf optimization. Finally, based on 16 complete features, 8 optimal two-dimensional features and 9 optimal three-dimensional features, three models were established respectively. The results showed that the proposed 3D feature model performed the best with an average relative error of 5.78%. [ABSTRACT FROM AUTHOR]