Accurate screw detection method based on faster R-CNN and rotation edge similarity for automatic screw disassembly

Li, Xinyu; Li, Ming; Wu, Yongfei; Zhou, Daoxiang; Li, Tianyu; Hao, Fang; Yue, Junhong; Ma, Qin

doi:10.1080/0951192x.2021.1963476

Cited by 34 publications

(15 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Deep neural networks have gained wide popularity for 2D machine vision applications, thanks to their high accuracy and feature extraction ability [16]. In recent years, DNNbased vision technology found increasing application in the fields of manufacturing [17][18][19] and remanufacturing [20][21][22][23][24][25].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Primitive shape recognition from real-life scenes using the PointNet deep neural network

Zheng

Castellani

2022

Int J Adv Manuf Technol

View full text Add to dashboard Cite

In many industrial applications, it is possible to approximate the shape of mechanical parts with geometric primitives such as spheres, boxes, and cylinders. This information can be used to plan robotic grasping and manipulation procedures. The work presented in this paper investigated the use of the state-of-the-art PointNet deep neural network for primitive shape recognition in 3D scans of real-life objects. To obviate the need of collecting a large set of training models, it was decided to train PointNet using examples generated from artificial geometric models. The motivation of the study was the achievement of fully automated disassembly operations in remanufacturing applications. PointNet was chosen due to its suitability to process 3D models, and ability to recognise objects irrespective of their poses. The use of simpler shallow neural network procedures was also evaluated. Twenty-eight point cloud scenes of everyday objects selected from the popular Yale-CMU-Berkeley benchmark model set were used in the experiments. Experimental evidence showed that PointNet is able to generalise the knowledge gained on artificial shapes, to recognise shapes in ordinary objects with reasonable accuracy. However, the experiments showed some limitations in this ability of generalisation, in terms of average accuracy (78% circa) and consistency of the learning procedure. Using a feature extraction procedure, a multi-layer-perceptron architecture was able to achieve nearly 83% classification accuracy. A practical solution was proposed to improve PointNet generalisation capabilities: by training the neural network using an error-corrupted scene, its accuracy could be raised to nearly 86%, and the consistency of the learning results was visibly improved.

show abstract

Section: Related Workmentioning

confidence: 99%

“…A total of 1496 bounding boxes around the screw samples had to be manually labelled in the images. Li et al [22] used a fast region-convolution neural network to detect screws on motherboards of mobile phones for disassembly. The training procedure needed the manual acquisition of 488 images.…”

Section: Related Workmentioning

confidence: 99%

Primitive shape recognition from real-life scenes using the PointNet deep neural network

Zheng

Castellani

2022

Int J Adv Manuf Technol

View full text Add to dashboard Cite

show abstract

“…The recycling [4] and dismantling [5] of EOF EVBs have become a hot topic. EVBs, a class of electronic waste (E-Waste), contain heavy metals and other harmful substances, which can contaminate the soil and water, thus damaging the environment and affecting human health [6][7][8][9]. The safety of the battery itself also needs to be considered [10].…”

Section: Introductionmentioning

confidence: 99%

“…The screw detection task aims to detect the screws in the sensory field while obtaining information about the category of the screw and the accurate position for the subsequent disassembly task. In recent years, artificial intelligence technology has been developing rapidly in the field of object detection [19][20][21][22][23], and it only needs some annotated images for training the model to achieve high target localization accuracy [6,[24][25][26]. However, the result of the direct application of these methods of screw detection is not as good as expected.…”

Section: Introductionmentioning

confidence: 99%

“…Poschmann, H. et al [42] used a Fast R-CNN deep learning model structured by Inception ResNet V2 based on a stereo vision system to identify and locate different types of screws in images. Li, X. et al [6] used Faster R-CNN to initially locate and classify the screws in 2D images and generate a series of prediction boxes, which were then adjusted using rotated edge similarity (RES) to accurately locate the centre of the screws. Although these Fast R-CNN model-based methods achieve high accuracy in screw detection tasks, they still have some drawbacks.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

An Accurate Activate Screw Detection Method for Automatic Electric Vehicle Battery Disassembly

Zhang

et al. 2023

Batteries

View full text Add to dashboard Cite

With the increasing popularity of electric vehicles, the number of end-of-life (EOF) electric vehicle batteries (EVBs) is also increasing day by day. Efficient dismantling and recycling of EVBs are essential to ensure environmental protection. There are many types of EVBs with complex structures, and the current automatic dismantling line is immature and lacks corresponding dismantling equipment. This makes it difficult for some small parts to be disassembled precisely. Screws are used extensively in batteries to fix or connect modules in EVBs. However, due to the small size of screws and differences in installation angles, screw detection is a very challenging task and a significant obstacle to automatic EVBs disassembly. This research proposes a systematic method to complete screw detection called “Active Screw Detection”. The experimental results show that with the YOLOX-s model, the improved YOLOX model achieves 95.92% and 92.14% accuracy for both mAP50 and mAP75 positioning after autonomous adjustment of the robotic arm attitude. Compared to the method without autonomous adjustment of the robotic arm, mAP50 and mAP75 improved by 62.81% and 57.67%, respectively. In addition, the improved YOLOX model improves mAP50 and mAP75 by 0.19% and 3.59%, respectively, compared to the original YOLOX model.

show abstract

Design and Implementation of a Multifunctional Screw Disassembly Workstation

Zhang,

Wang

et al. 2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Accurate screw detection method based on faster R-CNN and rotation edge similarity for automatic screw disassembly

Cited by 34 publications

References 28 publications

Primitive shape recognition from real-life scenes using the PointNet deep neural network

Primitive shape recognition from real-life scenes using the PointNet deep neural network

An Accurate Activate Screw Detection Method for Automatic Electric Vehicle Battery Disassembly

Design and Implementation of a Multifunctional Screw Disassembly Workstation

Contact Info

Product

Resources

About