Wear detection of WC-Cu based impregnated diamond bit matrix based on SEM image and deep learning
Graphical abstract
Introduction
With ongoing worldwide exploring target converting to deep mineral deposits [1,2] and geothermal resources [3], the demand for deep hard rock drilling has risen sharply in recent years. Impregnated diamond bit (IDB), containing a working layer of diamond reinforced metal matrix composite, has been proven to be an efficient cutting tool against dense and hard rocks [4]. Apart from application in energy exploration, there are also impregnated diamond tools utilized in many other fields, such as construction [5], stone, and civil engineering [6] industry, where the wears of MMCs are all much concerned. For deep hard rock drilling, the rotary drilling bits would suffer from various wear damages, which may result in premature failure and even destruction of the whole set of drilling tools [7]. In exertion of the diamond bit, what plays the key role is the thrust along with the spin of synthetic diamond grits embedded in the IDB matrix. The composite matrices are expected to be worn synchronously with diamond edges to ensure the best drilling performance, involving penetration efficiency and service life. Hence there have launched multiple studies on the diamond impregnated bits [[8], [9], [10], [11], [12]], including research on diamonds [13], metal matrices [14], and the wear mechanisms [15]. Considering that the service life and working performance of diamond tools are largely determined by metal matrices for the diamond retention capacity [16], it is of great importance to pay close attention to the analysis of the wear mechanism of metal matrices. This study focuses on the quantitative evaluation of metal matrix wear in an accurate and efficient way.
Wear identification and detection have been in research for a long time [17]. Many technologies and methods have been applied for wear measurements of impregnated diamond bits, like optical profilometry [18], fractal theory [19], and statistical analysis [20]. However, correlation analyses may not meet the need for straightforward evaluation of the wear condition. To our knowledge, one of the most common means is to carry out a pin-on-disc wear test and to obtain the wear surface of test samples for SEM analysis. To be specific, the wear mode of MMCs could be described as a combination of several wear types characterized by some specific morphologies on SEM images. For instance, plow grooves due to scratching indicate abrasive wear, and smear implies adhesive wear [21]. However, qualitative description of the worn surface morphologies, illustrating the existence of some wear types and comparing the degrees of wear among different samples, is only on a primary stage. Moreover, with widely differing perspectives, the final results and conclusions are likely to vary from person to person, especially after the increase of the sample size.
A potential method is the convolutional neural network (CNN), which has exhibited excellent performance in automatic detection. For wear mechanism assessment, there was research proposing a two-level inference system made up of one wear mechanism CNN and three wear severity CNN for automated assessment of gear wear mechanism and severity [22]; and in a more common application, the deep learning model of CNN was used for wear debris classification, indicating the wear condition in an indirect manner [23]; Besides, in the form of amplitude-time evolutional curves, the acoustic emission signals in rock drilling [24,25] have the potential to be processed and analyzed with neural network method as well. As one of the most effective CNNs, Mask R-CNN is a very suitable method for instance segmentation to handle the issues in wear detection and evaluation. Since proposed in 2017 [26], Mask R-CNN has been widely used in many fields, such as concrete assessment [27,28], material detection [29,30], precision agriculture [31,32] and medical image [33,34]. In this work, Mask R-CNN was used for wear detection on SEM images, which had been proved feasible in other applications [35,36]. By changing proportions of the main components, WC and Cu, a series of WC-Cu based matrix samples in the shape of cuboid were manufactured by powder metallurgy and hot-pressed sintering for the requirement of wear image datasets first. Then the end faces of the prepared samples were worn against grinding wheels perpendicularly in a pin-on-disc test, and finally got captured by a scanning electron microscope. In this study, the wear mode on an image was described as a combination of different degrees of abrasive, adhesive, and fatigue wear (corrosion wear was exclusive in this research). By manual labeling of the wear regions, the obtained SEM images were divided into the training set, validation set, and test set for the establishment of the Mask R-CNN model. After the Mask R-CNN model was obtained, 100 images in the datasets would be selected for calculation of mask IoU, area loss, and some other wear indexes to make the effectiveness evaluation. Besides, wear images in previous research were added into the test set for verification of its detection effect. The whole computational process was conducted online with Google Colab (online development environment) and Detection2 (software system to implement Mask R-CNN) [37], which means all steps can be repeated easily.
Section snippets
Wear image preparation
To obtain the wear images of metal matrices, a series of WC-Cu based powder combinations were determined based on previous experience [38], whereafter powder metallurgy and hot-pressing methods were applied for manufacturing of the MMC samples. All involved powder mixtures and their physical properties in this work are listed below in Table 1.
The metal powders could be divided into three types from their existing states during the process of hot pressing, i.e., skeleton components with the
Wear detection performance
For a demonstration of wear detection making use of the trained Mask R-CNN model, six SEM images published in previous work [38] were added into the test set for prediction, or in other words inference, to output the information to be gained from them. The detection results and their quantitative statistics are presented in Fig. 10 and Table 5 respectively.
According to the wear classification principle prescribed in section 2.2, the Mask R-CNN model would make effective identification among
Conclusions
To evaluate the wear of metal matrix quantitatively and normatively, a series of WC-Cu based MMC samples were manufactured with hot-pressed sintering. After the pin-on-disc test, the SEM wear images of the worn surface were obtained, based on which the datasets for training, validation and test were established. In the utilization of Mask R-CNN algorithms, a wear detection model had been trained and showed excellent performance in recognition and evaluation of wear regions, and the main
Data and code availability
The codes and datasets are available at https://github.com/sunwucheng/IDB_matrix_wear.
The training, validation, and test processes are repeatable online with the codes open-sourced there.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This study was supported by the National Natural Science Foundation of China (contract no. 41972327) and the External Cooperation Program of Science and Technology Department of Hubei Province (no. 2019AHB051).
References (43)
- et al.
Geothermal energy in deep aquifers: a global assessment of the resource base for direct heat utilization
Renew. Sust. Energ. Rev.
(2018) - et al.
A new composite impregnated diamond bit for extra-hard, compact, and nonabrasive rock formation
Int. J. Refract. Met. Hard Mater.
(2014) Trends and market perspectives for diamond tools in the construction industry
Int. J. Refract. Met. Hard Mater.
(2000)- et al.
Diamond tools in stone and civil engineering industry: cutting principles, wear and applications
Diam. Relat. Mater.
(2002) - et al.
Significant damages of core diamond bits in the process of rocks drilling
Eng. Fail. Anal.
(2016) - et al.
The wear mechanisms of impregnated diamond bits
Wear.
(1994) - et al.
Wear response of impregnated diamond bits
Wear.
(2018) - et al.
The wear of diamonds in impregnated diamond bit drilling
Wear.
(1991) - et al.
Wear mechanisms of diamond-containing hardmetals in comparison with diamond-based materials
Int. J. Refract. Met. Hard Mater.
(2018) - et al.
Bit cutter-on-rock tribometry: analyzing friction and rate-of-penetration for deep well drilling substrates
Tribol. Int.
(2014)
Investigation on abrasion-corrosion properties of WC-based composite with fractal theory
Int. J. Refract. Met. Hard Mater.
Effects of MoS2 and WS2 on the matrix performance of WC based impregnated diamond bit
Tribol. Int.
Automated assessment of gear wear mechanism and severity using mould images and convolutional neural networks
Tribol. Int.
Integrated model of BP neural network and CNN algorithm for automatic wear debris classification
Wear.
Acoustic emission analysis for rock–bit interactions in impregnated diamond core drilling
Int. J. Rock Mech. Min. Sci.
FibeR-CNN: expanding Mask R-CNN to improve image-based fiber analysis
Powder Technol.
Probability-based Mask R-CNN for pulmonary embolism detection
Neurocomputing.
Deep learning-based method for SEM image segmentation in mineral characterization, an example from Duvernay Shale samples in Western Canada Sedimentary Basin
Comput. Geosci.
Image-based size analysis of agglomerated and partially sintered particles via convolutional neural networks
Powder Technol.
Special issue “deep mining,”
Rock Mech. Rock. Eng.
Deep mining: a rock engineering challenge
Rock Mech. Rock. Eng.
Cited by (20)
Effect of Cu-coated diamond on the formation of Cu–Sn-based diamond composites fabricated by laser-powder bed fusion
2024, International Journal of Refractory Metals and Hard MaterialsEnhanced drilling performance of impregnated diamond bits by introducing a novel HEA binder phase
2024, International Journal of Refractory Metals and Hard MaterialsIntelligent analysis method of dam material gradation for asphalt-core rock-fill dam based on enhanced Cascade Mask R-CNN and GCNet
2023, Advanced Engineering InformaticsThermal failure of diamond tools indicated by diamond degradation: Damage evaluation and property prediction on small image datasets
2023, Engineering Applications of Artificial IntelligenceRobustFlow: An unsupervised paradigm toward real-world wear detection and segmentation with normalizing flow
2023, Tribology InternationalCitation Excerpt :Some researchers have proposed a specific DL-based method for wear detection and segmentation. For example, Liu et al. and Peng et al.[23,24] introduced a way of detecting wear based on convolutional neural network (CNN) and support vector machine (SVM) combined with the ferrography images; Wang et al.[20] improved the Faster R-CNN model for detecting and locating the wear; Sun et al.[25] used Mask R-CNN and the scanelectron microscope (SEM) images to detect wear. In addition, Bergs et al.[26] extended the CNN coupled with fully convolutional network (FCN) to detect and segment the wear in tools.
The effect of Ti/Si ratio for introducing diamond into Cu melts based on Ti-Si-diamond reaction
2022, International Journal of Refractory Metals and Hard MaterialsCitation Excerpt :But the low hardness, poor wear resistance, deformation at high temperature and other shortcomings limit the application of pure Cu [1–3]. By adding one or more kinds of reinforcements, such as ceramic particles, to the Cu matrix, the mechanical properties and wear resistance of Cu can be effectively improved [4–6]. Diamond is the material with the best thermal conductivity which is 2200-2600 W·m−1· K−1 at room temperature, the lower thermal expansion coefficient, the highest hardness in nature, high melting point, high wear resistance and good chemical stability [7–9].