Wear detection of WC-Cu based impregnated diamond bit matrix based on SEM image and deep learning

https://doi.org/10.1016/j.ijrmhm.2021.105530Get rights and content

Highlights

  • Mask R-CNN method was applied for wear detection of WC-Cu matrix composites based on deep learning.

  • A SEM wear image datasets were established and utilized for wear detection model training.

  • The trained model showed good performances in wear recognition and evaluation of the matrix samples.

  • Wear mode (the combination of the three wear types) of the SEM wear image was obtained by modifying the algorithms.

Abstract

The working efficiency and lifetime of impregnated diamond tools are closely related to their wear conditions, among which different wear modes of the metal matrix play an essential role. Since traditional qualitative description cannot meet the requirement of mathematical relationship establishment, a deep learning method, Mask R-CNN, was applied for the quantitative determination of the matrix wear based on scanning electron microscope (SEM) images. A series of WC-Cu based metal matrix composite (MMC) samples had been prepared by hot-pressed sintering, followed by a pin-on-disc wear test to obtain the wear surface images, and the datasets were established based on a normal wear classification principle where classification is of four basic types: abrasive wear, adhesive wear, fatigue wear and corrosion wear (corrosion wear is not involved in this study). After training, validation, and test based on the SEM wear image datasets, the wear segmentation results from the trained model indicated that Mask R-CNN could automatically identify the wear of metal matrices efficiently and accurately, which was in good agreement with the results obtained by manual labelling. By modifying the algorithm codes, the masks of abrasive, adhesive, and fatigue wear were extracted and counted for model effectiveness evaluation. Moreover, the wear condition values (i.e., wear region areas) obtained from extracted masks would be easily applied for correlation analysis between cutting tool qualities and drilling efficiencies in future research as well. In comparison with statistic results by artificial cognition, the three types of wear showed an average wear region mask IoU over 70%, and an average wear region area loss of less than 3%. In the process of wear detection on similar wear images in published work, the Mask R-CNN model also presented good performances. All related codes and SEM image datasets are available at https://github.com/sunwucheng/IDB_matrix_wear.

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)

Cited by (20)

  • Enhanced drilling performance of impregnated diamond bits by introducing a novel HEA binder phase

    2024, International Journal of Refractory Metals and Hard Materials
  • RobustFlow: An unsupervised paradigm toward real-world wear detection and segmentation with normalizing flow

    2023, Tribology International
    Citation 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 Materials
    Citation 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].

View all citing articles on Scopus
View full text