Robotic compliant grinding of curved parts based on a designed active force-controlled end-effector with optimized series elastic component

Haiqing Chen,Jixiang Yang,Han Ding

DOI: https://doi.org/10.1016/j.rcim.2023.102646

IF: 10.103

2024-04-01

Robotics and Computer-Integrated Manufacturing

Abstract:High-accuracy, fast-response, and low-overshoot force control are important to guarantee the material removal accuracy and surface quality of robotic compliant grinding. In order to achieve the above target, this study develops an active compliant force-controlled end-effector based on a series elastic actuator for the robotic grinding of curved parts. Firstly, a decoupled robotic grinding system composed of an industrial robot and an end-effector is developed, and a novel force-controlled end-effector is designed by adding an elastic component between the servo motor and the load to improve compliance. Secondly, the influences of the elastic component and grinding tool stiffness on the stability of the force-controlled end-effector system are analyzed by establishing a contact model of the compliant end-effector and the transfer function of the entire force control system. Then, the stiffness of the grinding tool and the elastic component are optimized with full consideration of the system cutoff frequency and the end-effector compliance. To improve the force tracking accuracies of the developed compliant end-effector, a proportional-integral (PI) controller with first-order differential force feedforward control is designed. Finally, grinding experiments are conducted for verification. The effects of spring stiffness and grinding tool stiffness on the force control system are tested, which match well with the theoretical analysis. Grinding results show that the maximum force control error of the compliant force-controlled end-effector is decreased by 70% compared to that of the rigid force-controlled end-effector, and the overshoot of the force control is reduced from 30% to almost 0 under the same response speed. The maximum and average absolute grinding depth errors using the designed end-effector are reduced by 57.2% and 58.6%, respectively, compared with those of the traditional rigid end-effector, and the average surface roughness of the finished part is reduced by 19.2%. Which demonstrates the effectiveness and advantages of the proposed compliant force-controlled end-effector.

robotics,computer science, interdisciplinary applications,engineering, manufacturing

What problem does this paper attempt to address?

The main goal of this paper is to propose an integrated framework that utilizes Convolutional Neural Networks (ConvNets) to achieve object classification, localization, and detection tasks in images. Specifically, the study addresses the following key issues:

1. **Effective Implementation of Multi-Scale and Sliding Window Methods**: The researchers demonstrate how to efficiently implement multi-scale and sliding window methods in convolutional neural networks, which is crucial for handling objects of different sizes.

2. **New Method for Localization**: A new deep learning method is introduced to predict object boundaries, thereby improving localization accuracy. By accumulating predicted bounding boxes rather than suppressing them, the confidence in detection can be increased.

3. **Shared Network for Multi-Task Learning**: It is shown that a single shared network can simultaneously learn different tasks (classification, localization, and detection), which enhances the performance of all tasks.

4. **Successful Application of the Integrated Framework**: The proposed integrated framework won the localization task in the ImageNet Large Scale Visual Recognition Challenge 2013 and also achieved very competitive results in detection and classification tasks.

5. **State-of-the-Art Detection Results**: In post-competition work, this method established a new benchmark for the detection task.

6. **Release of the OverFeat Feature Extractor**: The researchers also released a feature extractor named OverFeat, which comes from their best model.

In summary, this paper aims to demonstrate how to train convolutional neural networks to simultaneously perform classification, localization, and detection tasks, and validates the effectiveness of this approach through a series of experiments. Additionally, the study discusses how to combine multiple localization predictions for detection without the need for specially trained background samples, thereby improving efficiency and accuracy.