Connectivity oriented fast Hough transform for tool wear monitoring
Introduction
The automatic monitoring of tool wear helps to reduce costs, improve product quality and is a long-standing goal of the modern manufacturing industry. Recent research efforts are focused on using various sensory information, such as acoustic emission, tool tip temperature, vibration signatures (acceleration signals) and cutting force. A good review of these methods is provided by Byrne et al. [1]. However, such methods are not easily or economically realized for industrial use. On the contrary, machined surfaces provide valuable information about tool wear and the images are more easily captured and analyzed. Fig. 1 shows the textures of machined surfaces produced by sharp and dull tools during different machining processes. It can be seen that the pattern of texture changes significantly as the machining processes progress. These changes could be utilized to monitor the tool wear.
Texture analysis can be used to differentiate the surfaces produced by sharp tools and dull tools. There are two main approaches to texture discrimination: statistical and structural. Second-order statistical methods of texture analysis, such as run length matrix and co-occurrence matrix, are appropriate for this task because these methods make use of the spatial information of texture patterns. A successful application of run length statistics and column projection to monitor tool wear in turning operations is reported by Kassim et al. [2]. However, the statistical approach is sensitive to the image quality and the cutting conditions. In this paper, we use a structure-based approach to analyze textures of machined surfaces. Gray level images of machined surfaces are converted to binary edge images using the Canny edge detector[3], which is insensitive to poor illumination. In this way, important geometric features are highlighted.
The edge images of the surface textures of Fig. 1, shown in Fig. 2, can be approximated by line segments. When a tool is sharp, the texture of the machined surface is rather regular and the edge image consists of long line segments that are aligned in the same direction. As tool becomes worn, the texture is more irregular, resulting in shorter line segments in all directions. We propose a new connectivity oriented fast Hough transform to detect line segments in edge images. Several features such as the orientations and lengths of the detected line segments are used to estimate the flank wear of tools. The organization of this paper is as follows. In the next section, the basic Hough transform and its variants are discussed. The connectivity oriented fast Hough transform algorithm is introduced in Section 3 and its performance is analyzed in Section 4. The features obtained from the detected line segments are discussed in Section 5. A description of the experimental setup and results are presented in Section 6. Our conclusions are provided in Section 7.
Section snippets
Hough transform
Hough transform can detect straight lines in a binary image by mapping the feature points in – space to the parameter space. For example, all lines passing through the point can be expressed in the parameter space as follows:where is the perpendicular distance from the origin to the line passing through and is the angle between the norm of the line and the axis as shown in Fig. 3. The Hough transform maps all points in space to space,
An overview of the connectivity oriented fast Hough transform
In this paper, we present the connectivity oriented fast Hough Transform algorithm is proposed. There are a few remarkable differences between the new algorithm and its predecessors, such as WRHT and CRHT. First, after curve fitting in the window, there is no accumulation in the parameter space. Instead, the results of curve fitting, i.e., the parameters of the detected line segment, are stored in a table. Second, the line segment detected in the small window is extended to the whole image so
Performance analysis
There are several parameters that can be used to fine-tune the algorithm. First, the size of the window, m, determines the shortest lines that can be detected. In practice, instead of keeping m constant, an adaptive scheme is used to find long line segments first and m is gradually decreased to the minimum length m=3. Second, the fitting error tolerance (E1,E2) of the first and second curve fitting can be chosen as needed. Third, the connectivity condition (D1,D2) that controls the searching of
Feature extraction
In our work, we used 256×256 images of machined surfaces. The Canny edge detector is first applied to obtain the binary edge image and then the connectivity oriented fast Hough transform is used to detect all line segments in the edge image. Fig. 11 shows the distribution of the lengths and orientations of detected line segments, which are produced by a sharp tool and a dull tool, respectively. Clearly, when the tool is sharp, the line orientations are within a narrow range and line segments
Experimental procedure
Table 1 summarizes the different sets of machining processes that were carried out. Images of machined surfaces are captured during the breaks of the machining processes, and the actual flank wear of cutting tools are measured using a microscope at the same time. The images are numbered sequentially in the order that they are captured. The machining processes continue until the tools break or if the tool is severely worn out. Our connectivity oriented fast Hough transform is applied to the
Conclusion
In this paper, a structural approach is used to analyze textures of machined surfaces. Our connectivity oriented fast Hough transform is shown to be able to detect line segments in edge images of machined surfaces. This algorithm is more accurate and flexible than other Hough transform based methods. The features that are extracted from the detected line segments show high correlation with the tool wear. A MLP neural network is used to approximate the complicated relationship between the
About the Author—A.A. KASSIM received his B.Eng. (First Class Honors) and M.Eng. degrees in Electrical Engineering from the National University of Singapore (NUS) in 1985 and 1987, respectively. From 1986 to 1988, he worked on the design and development of machine vision systems at Texas Instruments. He went on to obtain his Ph.D. degree in Electrical and Computer Engineering from Carnegie Mellon University, Pittsburgh, in 1993. Since 1993, he has been with the Electrical & Computer Engineering
References (13)
- et al.
Tool condition monitoring (TCM)—the status of research and industrial application
Ann. CIRP
(1995) - et al.
Probabilistic and non-probabilistic Hough transforms: overview and comparisons
Image Vis. Comput.
(1995) - et al.
An extension to the randomized Hough transform exploiting connectivity
Pattern Recog. Lett.
(1997) - et al.
Machine tool condition monitoring using workpiece surface texture analysis
Mach. Vis. Appl.
(2000) A computational approach to edge detection
IEEE Trans. Pattern Anal. Mach. Intell.
(1986)- Pekka Kultanen, Lei Xu, Erkki Oja, Randomized Hough Transform (RHT), Pattern Recognition, Proceedings of 10th...
Cited by (53)
Pattern characterization using topological data analysis: Application to piezo vibration striking treatment
2023, Precision EngineeringTool wear monitoring based on multi-kernel Gaussian process regression and Stacked Multilayer Denoising AutoEncoders
2023, Mechanical Systems and Signal ProcessingCorrelation statistics of a Fourier transform feature with flank wear on different sections of turned surface images for real time monitoring applications
2023, Measurement: Journal of the International Measurement ConfederationMachine vision based condition monitoring and fault diagnosis of machine tools using information from machined surface texture: A review
2022, Mechanical Systems and Signal ProcessingCitation Excerpt :In 2000, Mannan et al. [112] proposed a meta-cutting TCM system based on texture analysis of machined surfaces and sound generated by the metal removal processes. In 2004, Kassim et al. [115] presented a connectivity oriented fast Hough transform for tool wear monitoring, which detects all line segments in binary edge images of machined surfaces. In 2005, Kang et al. [119] researched fractal analysis for in-process tool wear monitoring from machined surfaces.
About the Author—A.A. KASSIM received his B.Eng. (First Class Honors) and M.Eng. degrees in Electrical Engineering from the National University of Singapore (NUS) in 1985 and 1987, respectively. From 1986 to 1988, he worked on the design and development of machine vision systems at Texas Instruments. He went on to obtain his Ph.D. degree in Electrical and Computer Engineering from Carnegie Mellon University, Pittsburgh, in 1993. Since 1993, he has been with the Electrical & Computer Engineering Department at NUS, where he is currently an Associate Professor and Deputy Head of Department. Dr Kassim's research interests include image analysis, machine vision, video/image processing and compression.
About the Author—ZHU MIAN received his B.E. degree from Beijing University of Aeronautics and Astronautics, People's Republic of China, in 1996. He is currently a student in the Electrical Engineering graduate program at the National University of Singapore.
About the Author—M.A. MANNAN obtained his M.Sc (Hons) in machine tools and manufacturing engineering in 1975 from PFU, Moscow on a government of India scholarship. He joined the Royal Institute of Technology (KTH) in 1975 and has been affiliated with the same until 1994, where he held different positions including research engineer, research associate, research scientist and associate professor. In 1987, the academic title of Docent was conferred on him by KTH. Dr Mannan joined NUS in 1994 as a senior visiting fellow and is now an Associate Professor with the Mechanical and Production Engineering Department, National University of Singapore. Dr Mannan is a corresponding member of the CIRP and a member of ASME.