Abstract
Surface roughness of specimens is an important area of research since it influences the performance of machined parts. Meanwhile, employing a vision system to judge the roughness of the machined surface of specimens via captured images acquired from the specimen is an innovative and extensively used method. In this investigation, a vision system is used to capture the SEM images of the machined surface. The two-dimensional images of the machined surface of the Nimonic263 alloy are used to approximate the profile of the surface of specimens in finish turning. Surface roughness was detected in simulated images of specimens in a variety of machining conditions using the imaging technology. In this research work, the surface texture is extracted using a technique that combines 2D surface images and wavelet transform approach. The 2D wavelet transform has the capability to disintegrate a machined surface image into multiresolution depiction for several surface characteristics and can be utilized for surface evaluation. The difference in the histogram frequency of an illuminated region of interest (ROI) from turned surface images was analyzed to aid in the evaluation of surface roughness with an average prediction error of less than 3.2%.
1. Introduction
In the production and development of industrial parts, both the microstructure of the surface on a μm-scale and integrity strongly influence the characteristics and functions of components, such as wear, precision, anticorrosion, fitting, friction, and antifatigue attributes. Direct methods and indirect methods are the two types of techniques available for surface roughness evaluation. Using a toolmaker’s microscope, 3D surface profiler, optical microscope, or SEM (offline method) or a CCD camera, significant procedures evaluate the surface profile of machined surfaces. These observing systems are often based on a comparison of an optimum cutting process reference signal with the actual process signal [1, 2]. Electrical discharge machining was recently used to machine Nimonic super alloy (EDM) [3]. The measurement scale () and root mean squared roughness () are determined to comprehend the surface property based on scaling analysis [4]. Researchers have developed a method to determine the degree of tool wear by evaluating the machined surfaces’ texture using an image processing methodological approach to the image of the machined surface to overcome these limitations. Tool wear measurement, surface quality control, workpiece surface texture measurements, and other machining processes can all benefit from digital image processing (DIP) with machine vision. Image processing techniques were used to determine the surface roughness of large-scale SEM images [5, 6].
Response surface methodology (RSM) was employed to do mathematical modelling of surface roughness and flank wear. ANOVA is assessing the impact of each individual parameter on responses. Between the simulated and experimental data, there was a maximum inaccuracy of 13% and 7% in flank wear and roughness of the surface, respectively [7]. Machining of Nimonic alloy and Inconel alloy is regarded as critical, and it is necessary to investigate the impact of machining factors on machining responses in order to identify the best machining/control factors and their levels in order to improve the machined surface’s integrity/machinability [8–10]. The majority of image processing-based machined surface roughness evaluation approaches depend on feature extraction processes [11].
A review of tool condition monitoring based on image processing approaches was presented by Dutta et al. The findings of Dutta et al. suggest that in a single experimental setup, direct approaches will be employed to validate the outcomes of indirect procedures. The tool wear was evaluated based on the workpiece’s rough surface. The gray-level cooccurrence matrix (GCLM) is an efficient technique for selecting the best texture based on entropy metric [12]. Shahabi and Ratnam [13] proposed surface roughness evaluation by capturing the images of the machined surface in cycle using computer vision in this investigation. Grayscale imaging was used to detect simulated pictures of specimens in a variety of machining conditions. The results demonstrated that this method may be utilized to mimic and analyze a specimen’s surface profile in finish lathe machining as a tool tip fingerprint.
Dutta et al. [14] designed an improvised GLCM algorithm to monitor the tool condition with an offset parameter. According to the findings, the GLCM analysis focussed on the pixel pair spacing (PPS) value selection that is entirely reliant on the variances of feed marks having been turned [15]. Using response surface methodology, the impact of drilling parameters like tool rotational speed, geometrical parameters, flank wear, and feed rate on surface roughness was investigated. GLCM-based texture analysis of workpiece surface images was employed to obtain waviness information through spatial correlation of pixels [16]. DWT-based texture analysis has also been used to machine surface images to gather information on modification in surface texture due to the increase in tool flank wear via Ga, GRMS, and energy in space-frequency localization.
Liu et al. [17] investigated the explored flank face surface texturing on the WC/Co carbide tools with various geometrical features. The tools with microscale grooves parallel to the primary cutting edge on the flank face had the best flank wear resistance. The tool wear was assessed using the DWT method on machined surface images [18]. The DWT was used to perform a microscale study of a turned surface with average coefficient of determination values of 0.957 and 0.953, respectively; an exponential association of GRMS and energy values with progressing flank wear of the tool is discovered. Illumination-compensated pictures were suggested by John and Arunachalam to evaluate the surface roughness of turned surfaces in the grinding process [19].
Gandla et al. investigated surface roughness using incremental forming [20]. Taylor Hobson Talysurf is used to measure the surface roughness of manufactured parts, which varies from 0.61 to 3.61 mm. Few images were acquired from each formed part and saved in an image dataset for machine vision-based surface roughness evaluation. The wavelet transform method, Euclidean distance method, and Hamming distance approach were used to classify these images into three classes predicated on the range of surface roughness. The wavelet-based technique has a maximum classification performance of 95.4 percent, according to the results. The classifying effectiveness of the Hamming and Euclidian distance approaches is 78.39 percent and 81.48 percent, respectively. Ali and Dhar [21] proposed an ANN technique for forecasting surface roughness in turning. This model may be used to optimize the cutting process for efficient and economical manufacturing by predicting tool wear and surface roughness in the turning process. CNN algorithms have gained popularity in the assessment of surface roughness in recent years [22–25]. Because feature extraction is included into the network during the convolution phase, this technique eliminates it. Five loss functions are selected and evaluated for the prediction models based on their potential application and accuracy. The actual and expected surface roughness values are compared using a stylus-based profilometer.
The vibration and communication particle swarm optimization (VCPSO) algorithm, developed by Xu et al., included self-random vibration and interparticle communication processes [26]. When estimating tool wear, the ANFIS learned by the VCPSO algorithm (ANFIS-VCPSO) outperformed other intelligent models. The VCPSO algorithm was tested using benchmark functions, and the results showed that it is capable of global optimization. All the while, the VCPSO algorithm was used to determine the best milling parameter combinations under a variety of tool wear conditions. The technique proposed in this article extracts the surface texture by combining two-dimensional surface photography and a wavelet approach. Then, the future surface roughness is predicted by extracting the time delay parameters, the embedding dimension, and the false nearest neighbour of the produced surface texture. All offline and online approaches can estimate surface roughness parameters based on evidence from the surface image [27]. The stylus tracing (ST) technique became the most widely known way of evaluating the surface properties of components in recent decades, due to the implementation of tactile profilometers [24, 25]. The ST involves measuring the texture of the surface and calculating the roughness parameter but in contact with the surface. The noncontact approach provided in this work, on the other hand, can predict surface texture and roughness characteristics more precisely and with less prediction error by using DWT histogram equalization and Laplacian filtering. This technology represents a significant advancement in the development of smart surface roughness measuring devices that may be used in the smart manufacturing sector in the near future.
2. Experimental Setup
In this experiment, a super alloy Nimonic C263 workpiece was machined in dry cutting mode with CBN inserts. After cutting the entire length of the workpiece, the machined surfaces were imaged as -pixel gray-level image data using a CCD camera connected to a personal computer with machine vision capacity at 5 different locations on the surface of the workpiece. The photos were taken with a diffused lighting setup online. With relation to the workpiece, the camera and lighting source were put in place. For the most part, the illumination source was managed to hold at a 30-degree angle. The average flank wear of the insert ( average) was also determined using a Leica S6D microscope at 10 magnification and Leica QWin-V3 image capturing software.
Simultaneously, the surface roughness of the turned surface () was measured at the appropriate locations, where stands for average surface roughness. Table 1 summarises the cutting conditions of turning operations performed on Nimonic 263. For each of the machining conditions, three sets of experiments were carried out (as listed in Table 1). Table 2 summarises the camera setup and surface roughness tester specifications. Each image was cropped to pixels before processing. The MATLAB® environment was used for all image processing (version 7.8.0.347 R2021).
2.1. Materials
Nimonic C263 alloy with 150 mm length and 70 mm diameter with hardness of 32 HRc was considered the work material. It is a material with good resistance to high temperature and oxidation owing to its Ni, Cr, and cobalt content. The following are the chemical components of the workpiece material (in % weight): 0.48 Al, 2.94 Ti, 0.15 W, 0.04 Nb, 0.02 C, 0.02 V, 0.007 Ta, 0.001 S, 0.19 Si, 52.49 Ni, 20 Cr, 0.46 Mn, 0.07 Cu, 6.29 Mo, 16.7 Co, 1.0 Fe, 0.48 Al, and 1.94 Ti.
2.2. Methods
The experimental study was carried out on a NAGMATI175 lathe. Figures 1(a)–1(d) show the experimental setup, insert, tool holder, and flank wear. The Sandvik makes Cubic Boron Nitride (CBN) inserts to the specifications of CNGA 120 408S01030A, 7025 grade was chosen as the insert for the tests, and a PCLNR 2020 K12 tool holder was being used to fix the insert. Back rake angle (BRA) (6°), side rake angle (SRA) (6°), end relief angle (ERA) (6°), side relief angle (SRA) (6°), end cutting edge angle (ECEA) (5°), side cutting angle (SCEA) (5°), and nose radius (NR) (0.8 mm) are the tool signatures of the insert. The cutting parameter’s effect on surface roughness in turning the Nimonic C-263 alloy was investigated. All turning trials were carried out by a new cutting edge. The experiment was carried out using an orthogonal array. The machining factors such as feed rate, cutting speed, and depth of cut were considered, and their ranges are listed in Table 3.
A tool maker’s microscope was used to measure the surface roughness offline at regular intervals. The measuring range is X-200 mm and Y-100 mm, the microscope stand tilting range is 12, the maximum distance between centres is 700 mm, the maximum diameter accommodating between centres is 100 mm, the plane stage area is, and the resolution is 0.2 lm (linear) (1 minute). Table 4 shows the orthogonal array and surface roughness measured using conventional and unconventional methods.
3. Methodology
Machine vision systems by default initiate with the image acquisition stage. Machine vision-based methods are extremely recommended for safe evaluation in in situ investigation of machined surfaces. Turned surface images are captured utilizing a high-end vision camera to evaluate the surface texture. The machine vision camera acquires the machined surface images after turning operation, to examine the surface texture. The captured machined surface images undergo normalization to cope up with lighting changes that might affect the image quality. The overall process is encapsulated in the block diagram as shown in Figure 2.
Normalization is performed as given below:
The normalized image matrix is represented by . The intensity of each pixel in the image matrix is represented by . The image matrix’s minimum and maximum pixel intensity values are denoted as and , respectively. The machined surface images obtained after turning operation at varied cutting conditions had an enormous flank wear resulting in variation in workpiece textures. This increased flank wear () reduces the cutting insert noise radius, which affects the machined surface. Machine vision system-refined machined surface images contain flaws such as nanoparticle settlings and feed marks.
The digital camera output is where is the original image and is the added noise.
Wiener filtering is utilized to retrieve the images disrupted by noise.
where is the local mean intensity surrounding every pixel. Also, denotes -by- neighborhood of every pixel, and is the location of every pixel in the local neighborhood mask . Also, variance of intensity surrounding every pixel is obtained by
The Wiener filter reduces the noise depending upon the statistical variables by lowering the variance of the neighboring pixels. In stage 3, the image segmentation is performed to segment the object from its surrounding pixels. This is utilized to set the cutting tool pixel intensity to 0 and the background to 1; it is explained in the below equation: where denotes the actual image of the tool tip after noise removal in the algorithm step 2 and denotes the tool tip image after segmenting the cutting tool image. Also, is the threshold value set by default utilizing the MATLAB command. The machine details and machining conditions considered for the study are given in Table 5.
Preprocessing is required to improve image quality through contrast stretching, histogram equalization, noise reduction through filtering, and inhomogeneous illumination compensation, among other things. Low-pass filtering (median filter) is very effective at reducing noise in machined surface images, particularly cutting tool images, because noise can occur due to dirt, oils, and precision machining dust and dirt on the surface of an object. The Laplacian filter is a second-order temporal high-pass filter that is used to sharpen machined surface images so that the feed marks can be seen clearly. The segmentation method and feature extraction are carried out after preprocessing. By calculating the image gradient and direction, the feature extraction ensures the information of the edges of the workpiece surface images. Because of its noise immunity and ability to detect actual edges with minimal flaws, the Canny edge detector is used in the field of machine vision. Convolution is performed between both the machined surface image and a Gaussian smoothing filter with a standard deviation of in the Canny edge identification technique. The gradient computation for the smoothened image is the next step. This technique can be used to fetch the edges of a workpiece surface texture. To capture turned images with good contrast, a dispersed lighting prototype such as fibre optic-guided light with a DC-regulated light source and infrared interference filter is used. Segmentation is the process of dividing an image into multiple sections according to a set of rules. The feature-state method collects pixel characteristics into feature vectors, which are then used to assign pixels to classes by selecting a threshold value. Because the 2D wavelet transform can disintegrate a workpiece surface impression into multiresolution depictions for several surface morphologies, it can be used to evaluate surfaces. Signal vectors derived from feed mark images are used to represent the image’s grayscale intensity. Use the 2D wavelet transform and the key local intensity variation technique to decompose the image signal vector. Wavelet transform is a time-frequency-conversion mathematical procedure that is frequently used. The frequency analysis of the signal into scalar analysis is dealt with by wavelet analysis.
where seems to be the scale parametric quantity for varying frequency and “” is the interpretation parametric quantity and is the wave function. Surface roughness is an important parameter for evaluating and controlling manufacturing quality because it can affect the friction coefficient, creep life, and fatigue strength of a machined part. Surface roughness results due to enhanced tool wear; it serves as an indication to replace the tool. Surface roughness evaluation procedures based on image processing can be performed online through E-evaluation. However, this facility is not available in traditional procedures. A rough surface tends to scatter more quantity of light. This causes larger dark regions to form that result in decrease in the image signal RMS value. A rough surface ends up with huge speckle patches that in turn reduce the variance, within the fixed evaluation length. The histogram mapping of an illuminated area of interest (ROI) from machined surface photographs was studied to see if there was any fluctuation in histogram frequency, which helps with surface roughness evaluation [28].
3.1. Evaluation Parameters
The roughness average, also known as the arithmetic average height (), is a roughness parametric parameter that is often used in quality control.
“” is the number of samples for a certain evaluation duration and denotes how far the profile deviates from the average line. The root mean square slope () and arithmetic mean slope () hybrid parametric quantities are used.
is the slope at point “” and “” is the total number of points.
4. Results and Discussion
4.1. Characterization of the Machined Surface Utilizing Image Histogram
The surface profile of the workpiece surface is used to measure surface roughness in this paper. The MATLAB software is used to create an intensity histogram for the workpiece surface image. The histogram is a visual representation of how the intensity of gray levels increases as the surface quality of the machined surface changes. The left portion of histogram denotes small intensity values, and the right portion denotes larger intensity values. The machined metal surface has high reflectivity; hence, it is convenient to monitor the right side of the histogram. Therefore, gray level of 125 is taken as the reference to evaluate the number of histogram frequency. The reflectivity is enhanced if the machined surface is smooth, thereby resulting in greater frequency values. The histogram for the machined surface is obtained by taking the gray-level intensity values on the -axis and surface roughness values taken on the -axis. The machined surface is smooth possessing a surface roughness value of and gray-level intensity of 125, at a frequency of 600 Hz. In the second case, the machined surface is coarse possessing a surface roughness value of and gray-level intensity value of 100 at a frequency of 400 Hz. By analyzing the histogram of the machined surface image, it enables one to determine if a given machine surface is coarse or smooth. The work involves observing and inspecting the machined surface while turning the Nimonic 263 material with coated carbide inserts at various spindle speeds and feed combinations. Variations in the histogram statistical characteristics that aid in evaluating the surface finish were explored in the histogram form of an illuminated area of interest (ROI) from turning machine surface images. The results clearly show that the turning procedure’s cutting specifications have no effect on surface roughness. Figure 3 shows the relationship between machining time and histogram frequency for various feed rates and a steady depth of cut of 0.2 mm at 450 and 510 rpm spindle speeds. The graphs displayed give a general sense of the machined surface’s histogram frequency. Only the lowest and maximum feeds at two different speeds are shown in the graphs. Again for entire cutting considerations, the machined surface’s surface roughness remains constant. Many investigations have determined that the value changes in a stable range, just as the histogram frequencies of the machined surface obtained with a machine vision system do. The results in Table 6 show that, while the cutting speed and feed rate vary, they have no effect on the machined surface’s histogram frequency. For a consistent change in the histogram profile, changing the feed rate causes a change in the form of the feed marks over the machined surface. Various cutting speeds and feed rates were used in the technique.
Figure 4 depicts the original machined surface image and its histogram equivalent. The histogram graph is fetched by taking the surface roughness values on the -axis and feed rate on the -axis. An image histogram is a graphical representation of a digital image’s tonal distribution. It assigns a pixel value to each tonal value. A viewer can quickly assess the entire tonal distribution by looking at the histogram for a specific image.
Figure 5 depicts the reconstructed machined surface image and its corresponding histogram equivalent. The histogram graph is fetched by taking the surface roughness values on the -axis and the feed rate on the -axis. Image reconstruction is the process of putting together 2D and 3D images from scattered or insufficient data.
Figure 6 depicts the reconstructed approximation at level 1 and its corresponding histogram equivalent. The histogram graph is fetched by taking the surface roughness values on the -axis and the feed rate on the -axis. The cumulative histogram is a histogram in which the vertical axis displays not only the number for a single bin but also the number for that bin and all bins with lower response variable values.
Figure 7 depicts the synthesized image at level 1 and its corresponding histogram equivalent. Image synthesis is the technique to create novel images from the image specifications. The histogram graph is fetched by taking the surface roughness values on the -axis and the feed rate on the -axis.
Figure 8 shows the decomposition of the original machined surface image and the equivalent generated image using discrete wavelet transform (DWT). Image decomposition segregates an input image as structural and textual components.
The average flank wear enhances with cutting time, as depicted in Figure 9(a), and the surface roughness is depicted in Figure 9(b). It demonstrates that, under the similar machining conditions, increasing cutting speed increases flank wear progressively. Higher cutting speeds cause the material to pass away in a short period of time, allowing the machining interface to become an adiabatic system. The cutting tool softens even as temperature increases, resulting in increased tool wear. The influence of feed rate on flank wear was measured using a 498 m/min cutting speed and a 0.1 mm cut depth.
(a)
(b)
The predicted value is obtained by utilizing discrete wavelet transform (DWT) and the measured value from experimental investigation.
As seen in Figure 10 and Table 7 from the above analysis, it is clear that the average prediction error obtained using discrete wavelet transform is 3.16% which is very less when compared to conventional algorithms and techniques.
5. Conclusion
In this research, noncontact surface roughness estimation method has been proposed and investigated. This unconventional method involves steps of noise removal (median filtering), sharpening using the Laplacian filter, image segmentation, Canny edge detection, 2D wavelet transform, and histogram equalization. An improved DWT-combined histogram equalization is used to evaluate the surface roughness with minimum complexity and high accuracy. The following are the significant findings of this experimental study and analysis: (i)Histogram equalization has been shown to be a reliable method for determining surface roughness(ii)The proposed method is noncontact, with no additional surface damage to the work item and less complexity than standard surface roughness assessment methods(iii)Surface texture estimation on machined surfaces is used to verify the effectiveness of the suggested new method(iv)The results show that this approach predicts surface roughness with a 3.16 percent average prediction error
Abbreviations
| SEM: | Scanning electron microscopy |
| CCD: | Charge coupled device |
| EDM: | Electrical discharge machining |
| DIP: | Digital image processing |
| RSM: | Response surface methodology |
| ANOVA: | Analysis of variance |
| GCLM: | Gray-level cooccurrence matrix |
| PPS: | Pixel pair spacing |
| DWT: | Discrete wavelet transform |
| : | Root mean square acceleration |
| ANN: | Artificial neural network |
| CNN: | Convolutional neural network |
| VCPSO: | Vibration and communication particle swarm optimization |
| ANFIS: | Adaptive neuro fuzzy inference system |
| ROI: | Region of interest. |
Data Availability
The data used to support the findings of this study are included within the article.
Conflicts of Interest
The authors declare that they have no conflicts of interest regarding the publication.