1

CHAPTER 1 INTRODUCTION
1.1. AIM OF THE PROJECT Online monitoring of grinding wheel loading and grinding wheel dressing To determine the percentage of loading at certain intervals of time To determine the optimum interval of grinding wheel dressing time based on wheel loading and application.

1.2.

NEED FOR THE PROJECT Grinding is an abrasive machining process that uses a grinding wheel as the cutting

tool. It can produce very fine finishes and very accurate dimensions. Grinding is one of the final machining processes that determine the surface quality of machined products. After a long period cycles times of the grinding process, removed chips may stick in the space between abrasive grains or weld on the top of cutting edges. Factors such as wheel loading and wheel wear contribute to the deterioration of the working surface and also its cutting capacity. When the loading and wear is severe, dressing of grinding wheel has to be carried out in order to bring the wheel to its best state. Determining the timing to dress the grinding wheel is extremely important in order to prevent flaws in products. Therefore monitoring the grinding process for wheel loading and wheel dressing is critical to ensure the surface quality of the machined component as well as the efficiency of the grinding process. 1.3. SCOPE OF THE PROJECT To overcome the difficulties of experience-dependent dressing, a systematic dressing method which can measure the status of the working wheel surface and evaluate the dressed wheel surface is necessary. Much research concerning the monitoring of grinding process has

been conducted using AE, a touch-trigger probe inspection method , an optical triangulation method, Eddy currents, laser methods and ultrasound techniques. Advances in the computer vision technology have led to the investigation of its application in the monitoring of the grinding process. Visual information has the advantage, that can be interpreted very easily and due to its high information content is the first choice to investigate typical surface forms, which cannot be extracted from indirect measurement signals.

CHAPTER 2 LITERATURE SURVEY

Literature forms the basic backbone of all the work. Surveying the literature helps to gain knowledge about the work done previously and steps to be taken to move forward in particular fields of interest. Stephane LaChance, Andrew Warkentin, and Robert Bauer (2003) measure the wear flats by analyzing digital images. This system is mounted on the grinding machine and automates wear flat measurement by using computer control to automatically position the wheel and capture digital images of the wheel between grinding cycles. Image processing software is used to automatically analyze the digital images and measure wear flat area in the images. The proposed measurement system was validated using a scanning electron microscope. Experiments were performed on a Brown & Sharpe Micromaster 824 surface grinder to examine the relationship between wear flat area and normal force. The results agree with the literature. Wen-Tung Chang, Ting-Hsuan Chen and Yeong-Shin Tarng (2011) measures the characteristic parameters of form grinding wheels used for microdrill fluting. With the aid of the indirect duplication of wheel contours and by using computer vision, this paper presents a systematic process for the wheel contour measurement. The measuring process includes five sequential steps: the edge detection, the straight line detection, the contour separation, the circular arc fitting, and the circular arc angle evaluation. To test the proposed measuring process, a measuring apparatus was built, and experiments measuring the characteristic parameters of diamond grinding wheels used for microdrill fluting were conducted. It showed that the proposed measuring process was feasible to measure the characteristic parameters of certain form grinding wheels used for microdrill fluting. J. C. Su and Y. S. Tarng (2006) measures the grinding wheel wear using machine vision system. The vision-aided measuring system comprises a CCD coupled with a

telecentric lens, back lighting board and frame grabber. Measuring the image of the specimen with a grinded gap substitute directly captures the image of the actual grinding wheel. Using this method makes the 3 D of the topography of the grinding wheel into the 2 D of the contour of the grinding wheel. The results show that this developed system achieves a repeatable accuracy of 3 m for the measurement of the grinding wheel contours. Z. Feng and X. Chen (2006) detects and identifies the chip loading and cutting edge wear of a grinding wheel using the image processing toolbox of MATLAB. The different optical characters of the metal chips and the abrasive grains are analysed. The Sobel operator is adopted to make edge detection. A sensitivity threshold based on the global condition is used to decrease the noise. Image dilation and erosion processes are used to ensure the edge of each loaded chip is covered by a continuous section. The ratios of chips are calculated and displayed to monitor the wheel surface working status. K.C. Fan, M.-Z. Lee and J.I. Mou (2002) proposes an on-line non-contact method for measuring the wear of a form grinding wheel. A CCD (charge coupled device) camera with a selected optical lens and a frame grabber was used to capture the image of a grinding wheel. The analogue signals of the image were transformed into corresponding digital grey level values. Using the binarisation technique, the images of background and the grinding wheel were segmented. Thus the grinding wheel edge was identified. The mapping function method is used to transform an image pixel coordinate to a space coordinate. An auto -focus technology is also developed. The statistics of pixels are used as the focusing index. The signal was sent through an 8255 control card to drive a d.c. motor, and then to control the lens focusing movement to acquire the focal plane. The images before and after the grinding process were captured. The position deviation of the grinding wheel edge was analysed. Then, the grinding wheel wear was evaluated. Bernard C. Jiang, Chung-Li and Tsung-Chi Chen (2001) proposed a machine vision system to determine the protrusion rate of a diamond tool. The method developed is a noncontact method without manual judgment. Three sets of field samples were used to demonstrate the proposed method for determining protrusion rate.

C.K. Huang, Y.S.Tarng, C.Y. Chiu and A.P. Huang (2009) proposes using machine vision technique to assist the double-wheel grinding mechanism of moving-drill and fixed wheel to achieve resharpening function. First find and adjust the posture of micro-drill from vision system for grinding, then detect the line equation of cutting edge of primary facet grinded micro-drill and rotate cutting edge to be horizontal; and then grind the secondary facet of micro-drill. All grinding parameters for resharpening process are found automatically with machine vision technology. T. Warren Liao, Chi-Fen Ting, J. Qu and P.J. Blau (2006) presents a wavelet-based methodology for grinding wheel condition monitoring based on acoustic emission (AE) signals. Grinding experiments in creep feed mode were conducted to grind alumina specimens with a resinoid-bonded diamond wheel using two different conditions. During the experiments, AE signals were collected when the wheel was sharp and when the wheel was dull. Discriminant features were then extracted from each raw AE signal segment using the discrete wavelet decomposition procedure. An adaptive genetic clustering algorithm was finally applied to the extracted features in order to distinguish different states of grinding wheel condition. The test results indicate that the proposed methodology can achieve 97% clustering accuracy for the high material removal rate condition, 86.7% for the low material removal rate condition, and 76.7% for the combined grinding conditions if the base wavelet, the decomposition level, and the GA parameters are properly selected. Amin A. Mokbel and T.M.A. Maksoud (2000) uses an imprint of the profile of the grinding wheel was used to measure the surface condition of the wheel. An acoustic emission sensor with a high frequency sampling of 1.25 MHz was attached to the mild steel specimens to monitor the wheel condition. The raw AE signals generated from the grinding wheel/specimen contact were then analysed using a fast Fourier transform. The AE spectral amplitude of different grinding wheel bond types, grit sizes and their conditions represented by grinding wheel/truing speed ratios were then compared with the surface roughness (Ra) of the ground mild steel specimens.

SunHo Kim and Jung Hwan Ahn (1999) describes a systematic approach to deciding a proper dressing interval and an optimal dressing depth for the working grinding wheel. An eddy current sensor and a laser displacement sensor are used to measure the loading on the working wheel surface and the topography of the dressed wheel surface respectively. The dressing interval can be decided properly through the relational locus between the level of loading and the machined surface roughness. An optimal dressing depth to ensure less wheel loss and greater wheel surface quality is decided through the analysis of the variance of topography for the dressed wheel surface, which decreases at three different rates according to the accumulated dressing depth. Pawel Lezanski (2001) uses the neural network and fuzzy logic to classify the condition of the grinding wheel cutting abilities for the external cylindrical grinding process. For each measuring signal a few statistical and spectral features are calculated and used as an input for data selection andclassification procedures. First, a feed forward back propagation neural network was implemented to perform feature selection task from the multiple sensor system. Next, a neural network based fuzzy logic decision system for sensor integration in grinding wheel condition monitoring is discussed. 2.1 SUMMARY Much research concerning the monitoring of grinding process has been concentrated on wheel wear only. As wheel loading can equally contribute to the quality as well as productivity of the grinding process, monitoring of wheel loading and wheel dressing is also important. This work concentrates on monitoring the wheel loading and relating it to surface finish, thereby determining the optimum wheel dressing time.

CHAPTER 3 GRINDING PROCESS

(Courtesy: http://nptel.iitm.ac.in/courses/Webcoursecontents/IIT%20Kharagpur/ Manuf%20Proc%20II/pdf/LM-27.pdf, LM-28.pdf, LM-29.pdf)

3.1

GRINDING
Grinding is the most common form of abrasive machining. It is a material

cutting process which engages an abrasive tool whose cutting elements are grains of abrasive material known as grit. These grits are characterized by sharp cutting points, high hot hardness, and chemical stability and wear resistance. The grits are held together by a suitable bonding material to give shape of an abrasive tool. Figure 1 illustrates the cutting action of abrasive grits of disc type grinding wheel similar to cutting action of teeth of the cutter in slab milling.

Figure 3.1. Cutting action of abrasive grains

3.2

GRINDING WHEEL Grinding wheel consists of hard abrasive grains called grits, which perform the cutting

or material removal, held in the weak bonding matrix. A grinding wheel commonly identified by the type of the abrasive material used. The conventional wheels include aluminum oxide and silicon carbide wheels while diamond and CBN (cubic boron nitride) wheels fall in the category of super abrasive wheel. 3.2.1 SELECTION OF GRINDING WHEELS Selection of grinding wheel means selection of composition of the grinding wheel and this depends upon the following factors: 1) Physical and chemical characteristics of the work material 2) Grinding conditions 3) Type of grinding (stock removal grinding or form finish grinding) 3.3 TYPES OF ABRASIVES

Aluminum oxide Aluminum oxide may have variation in properties arising out of differences in chemical composition and structure associated with the manufacturing process. Pure Al 2O3 grit with defect structure like voids leads to unusually sharp free cutting action with low strength and is advantageous in fine tool grinding operation, and heat sensitive operations on hard, ferrous materials. Regular or brown aluminum oxide (doped with TiO 2) possesses lower hardness and higher toughness than the white Al2 O3 and is recommended heavy duty grinding to semi finishing.Al2O3 alloyed with chromium oxide (<3%) is pink in colour. Monocrystalline Al2O3 grits make a balance between hardness and toughness and are efficient in medium pressure heat sensitive operation on ferrous materials. Microcrystalline Al2O3 grits of enhanced toughness are practically suitable for stock removal grinding. Al2O3 alloyed with zirconia also makes extremely tough grit mostly suitably for high pressure, high material removal grinding on ferrous material and are not recommended for precision grinding.

Microcrystalline sintered Al2 O3 grit is the latest development particularly known for its toughness and self sharpening characteristics. Silicon carbide Silicon carbide is harder than alumina but less tough. Silicon carbide is also inferior to Al2O 3 because of its chemical reactivity with iron and steel. Black carbide containing at least 95% SiC is less hard but tougher than green SiC and is efficient for grinding soft nonferrous materials. Green silicon carbide contains at least 97% SiC. It is harder than black variety and is used for grinding cemented carbide. Diamond Diamond grit is best suited for grinding cemented carbides, glass, sapphire, stone, granite, marble, concrete, oxide, non-oxide ceramic, fibre reinforced plastics, ferrite, graphite. Natural diamond grit is characterized by its random shape, very sharp cutting edge and free cutting action and is exclusively used in metallic, electroplated and brazed

bond.Monocrystalline diamond grits are known for their strength and designed for particularly demanding application. These are also used in metallic, galvanic and brazed bond. Polycrystalline diamond grits are more friable than monocrystalline one and found to be most suitable for grinding of cemented carbide with low pressure. These grits are used in resin bond. CBN (cubic boron nitride) Diamond though hardest is not suitable for grinding ferrous materials because of its reactivity. In contrast, CBN the second hardest material, because of its chemical stability is the abrasive material of choice for efficient grinding of HSS, alloy steels, HSTR alloys. Presently CBN grits are available as monocrystalline type with medium strength and blocky monocrystals with much higher strength. Medium strength crystals are more friable and used in resin bond for those applications where grinding force is not so high. High strength crystals are used with vitrified, electroplated or brazed bond where large grinding force is expected.

10

Microcrystalline CBN is known for its highest toughness and auto sharpening character and found to be best candidate for HEDG and abrasive milling. It can be used in all types of bond. Grit size The grain size affects material removal rate and the surface quality of work piece in grinding. Large grit- big grinding capacity, rough work piece surface Fine grit- small grinding capacity, smooth work piece surface. Grade The worn out grit must pull out from the bond and make room for fresh sharp grit in order to avoid excessive rise of grinding force and temperature. Therefore, a soft grade should be chosen for grinding hard material. On the other hand, during grinding of low strength soft material grit does not wear out so quickly. Therefore, the grit can be held with strong bond so that premature grit dislodgement can be avoided. 3.4 GRINDING WHEEL LOADING There is another type of wheel wear phenomenon that has a disastrous effect on grinding performance. Loading occurs when the work piece material adheres to the tips of the abrasive grains and is brought into repeated contact with the material. Loading also occurs if long work piece chips fill the pores of the abrasive and are retained there. The consequences of loading and clogging are extremely poor surface texture of the work piece, increased grinding forces and increased grinding wheel wear. To avoid loading, it is important to use ample coolant with effective lubrication properties. 3.4.1 EFFECTS OF GRINDING WHEEL LOADING Work piece damage High grinding wheel wear High Grinding forces Increase in Grinding wheel work piece Interface Temperature

11

High vibrations 3.4.2 FACTORS MAINLY CAUSE GRINDING WHEEL LOADING

Soft work piece materials Coarse grade wheel material High depth of cut Grinding huge length materials with low width wheel Operator skill Very high transverse and longitudinal table speed Insufficient coolant supply Layout of machine 3.5 DRESSING OF GRINDING WHEEL Dressing is the conditioning of the wheel surface which ensures that grit cutting edges are exposed from the bond and thus able to penetrate into the workpiece material. Also, in dressing attempts are made to splinter the abrasive grains to make them sharp and free cutting and also to remove any residue left by material being ground. Dressing therefore produces micro-geometry. The structure of micro-geometry of grinding wheel determines its cutting ability with a wheel of given composition. Dressing can substantially influence the condition of the grinding tool. Truing and dressing are commonly combined into one operation for conventional abrasive grinding wheels, but are usually two distinctly separate operation for superabrasive wheel.

12

CHAPTER 4 MACHINE VISION AND IMAGE PROCESSING

(Courtesy: Understanding and applying Machine Vision , Hello Zeuch) 4.1 MACHINE VISION Machine vision involves the acquisition of image followed by processing and interpretation of data using computer for some useful application. Machine vision (MV) is the technology and methods used to provide imaging-based automatic inspection and analysis in the field of automatic inspection, process control, and robot guidance in industry. Sophisticated manufacturing systems require automated inspection and test methods to guarantee quality. Methods like machine vision can be applied in all the following manufacturing processes: incoming receiving, forming, assembly, and warehousing and shipping. However, hardware alone cannot be the main factor. The data which is being obtained from such machine vision systems is the foundation for computer integrated manufacturing. It ties all of the resources of a company together - people, equipment and facilities. Machine vision, or the application of computer-based image analysis and interpretation, is a technology that has demonstrated its caliber to contribute significantly in improving the productivity and quality of manufacturing operations virtually in every

industry. In many industries (semiconductors, electronics, automotives), some products can not be produced without machine vision as an integral technology on production lines. Machine Vision is the term associated with the merger of one or more sensing techniques and computer technologies. Fundamentally, a sensor (typically a television-type camera) acquires electromagnetic energy (typically in the visible spectrum; i.e., light) from a scene and converts the energy into an image which can be used by the computer . The computer extracts data from the image (often first enhancing or otherwise processing the data),

13

compares the data with previously developed standards, and gives the results usually in the form of a response. The three main operations in Machine vision are 1. Image acquisition and digitization 2. Image processing and analysis 3. Interpretation 4.2 FUNCTIONAL BLOCK DIAGRAM OF MACHINE VISION SYSTEM

Figure 4.1. Block diagram of Machine Vision System 4.3 IMAGE PROCESSING In imaging science, image processing is any form of signal processing for which the input is an image, such as a photograph or video frame; the output of image processing may be

14

either an image or a set of characteristics or parameters related to the image. Most imageprocessing techniques involve treating the image as a two-dimensional signal and applying standard signal-processing techniques to it. Image processing usually refers to digital image processing, but optical and analogue image processing are also possible. Image processing may occur in either the hardware or software. Image processing hardware makes sense when large numbers of images are to be processed repetitively by the same set of algorithms. Hardware implementation is faster than software execution but with less flexibility. Most systems perform some image-processing operations in hardware and some in software. Image processing is generally performed on most images for basically two reasons: to improve or enhance the image and, therefore, make the decision associated with the image more reliable, and to segment the image or to separate the features of importance from those that are unimportant. Enhancement might be performed, for example, to correct the non-uniformity in sensitivity from photo site to photo site in the imaging sensor, correct distortion, correct non-uniformity of illumination, to enhance the contrast in the scene, correct perspective, etc. Image processing is typically considered to consist of following steps Image Acquisition and digitization Enhancement/Preprocessing Segmentation Code/Feature Extraction Image Analysis/Classification/Interpretation 4.3.1 IMAGE ACQUISITION AND DIGITIZATION Image Acquisition is accomplished using a video camera. The camera is focused on to the object of interest and image is obtained by dividing the viewing area into a number of pixels in which each element has a value that is proportional to the light intensity of the portion of the scene. Each pixel is converted into its equivalent dgital value by ADC.

15

The two types of cameras are most commonly used are Vidicon Camera Solid state Camera Solid state cameras have several advantages such as physically smaller more rugged and image produced is more stable. 4.3.2 Enhancement/Preprocessing Enhancement techniques transform an image into a "better" image, or one more suitable for subsequent processing to assure repeatable and reliable decisions. There are three fundamental enhancement procedures Pixel or Point transformations Image or Global transformations Neighborhood transformations 4.3.3 Segmentation Process of separating objects of interest (each with uniform attributes) from the rest of the scene or background, partitioning an image into various clusters. Two of the most common segmentation techniques are 1. Thresholding 2. Edge detection 3. Morphology Thresholding Thresholding is the process of assigning "white" (maximum intensity) to each pixel in the image with gray scale above a particular value, while all pixels below this value become "black". That particular value is the threshold and is a gray scale value. Areas that are lighter than the threshold become white; areas darker than the threshold become black. The resulting image, consisting of only black and white, is called a binary image. Thresholding was the first segmentation technique used, and almost all systems use it to some extent. It has

16

a simplifying effect on the image. The number of pixels in the image does not change, but each pixel can now have one of only two values, usually written 1 or 0. There are mainly two types of thresholding Global thresholding Adaptive thresholding Global thresholding means using one threshold for entire image. Thresholding is called adaptive thresholding when a different threshold is used for different regions in the image. This may also be known as local or dynamic thresholding. Edge Detection Edge detection is a fundamental tool in image processing, machine

vision and computer vision, particularly in the areas of feature detection andfeature extraction, which aim at identifying points in a digital image at which the image brightness changes sharply or, more formally, has discontinuities. Many edge-segmenting systems are based on detecting patterns of increasing and decreasing intensities or gradients generally found at the edges of objects. Since they are based on gradients, they are less sensitive to illumination variations and can handle lower contrast scenes. Morphology Another approach in segmentation of regions or edges involves Boolean logical operations on images using set theory concepts adapted to images and is known as mathematical morphology. Systems based on mathematical morphology efficiently perform operations that involve treating each pixel in a set identically resulting in a new or transformed image. These image transformations fall into three categories: unary (one image in, one image out), dyadic (two images in, one image out), and information extraction (image in, numbers out). Within each of these categories, the operations on binary images are either geometric or logical. Two key operations in Morphology are dilation and erosion.

17

The dilation operation between two sets A and B involves transforming each individual pixel in the A image by each pixel in the B image. In one definition of dilation, the transformed image that results is characterized as the outermost image made up of the center point of all the B images (typically the structured element) added to the A image. Erosion is the opposite of dilation and is essentially a containment test. The erosion operation between two sets A and B (typically the structured element) results in a transformed image that is the universe of all center points of set B, where set B is fully contained in set A. 4.3.4 Code/Feature Extraction Feature extraction is the process of deriving some values from the enhanced and/or segmented image. These values, the features, are usually dimensional but may be other types such as intensity, shape, etc. Some feature extraction methods require a binary image, while others operate on gray scale intensity or gray scale edge-enhanced images. Code/Feature extraction are grouped into three sections Miscellaneous Scalar Features, including dimensional and gray level values; Shape Features Pattern Matching Extraction. 4.3.5 Image Analysis/Classification/Interpretation For some applications, the features, as extracted from the image, are all that is required. Most of the time, however, one more step must be taken; classified interpretation. The most important interpretation method is conversion of units. Rarely will dimensions in "pixels" or "gray levels" be appropriate for an industrial application. As part of the software, a calibration procedure will define the conversion factors between vision system units and real world units. Most of the time, conversion simply requires scaling by these factors. Occasionally, for high accuracy systems, different parts of the image may have slightly different calibrations (the parts may be at an angle, etc.). In any case, the system should have separate calibration factors in X and Y.

18

CHAPTER 5 MONITORING OF WHEEL LOADING AND DRESSING THROUGH MACHINE VISION SYSTEM
5.1 MONITORING OF WHEEL LOADING AND WHEEL DRESSING ON MILD STEEL SPECIMEN Initially the grinding wheel is dressed in order to bring the wheel to its best working condition. Speed of the grinding wheel is set at 2000 rpm, feed at 0.2 mm/rev and depth of cut as 0.1 mm. Surface grinding operation had been carried out on Mild steel specimen. As a result of surface grinding operation the wheel got loaded. Images of the grinding wheel were taken at regular intervals of time. The captured image was transferred to computer and was processed using Matlab software by using Global Thresholding technique. Specifications of the Experiment Machine Speed Feed Depth of cut Work piece Camera Mega Pixel Optical Zoom Shutter speed Image processing software Hydarulic Surface Grinding Machine 2000 rpm 0.1 mm/rev 0.2 mm Mild steel Kodak Easyshare CD14 8.2megapixel 3x upto 1/1000sec Matlab

19

5.2

PROGRAMME FOR GLOBAL THRESHOLDING WITH A THRESHOLD RANGE USING MATLAB

clc; clear all; clf; I=imread('F:\project I=rgb2gray(I);k=0;y=0; I=imresize(I,[280 280]);k=0; y=0; for i=1:280 for j=1:280 if(I(i,j)>200 & I(i,j)<250); I(i,j)=256;k=k+1; else I(i,j)=0;y=y+1; end end end k=k y=y z=k/(k+y);z=z*100 imshow(I); title('processed image'); 5.3 RESULTS AND DISCUSSIONS FOR EXPERIMENT CONDUCTED ON MILD STEEL Surface grinding operation was carried on Mild Steel specimen. Images of grinding wheel were taken after every 5 minutes of operation with constant settings. These images were vipin\phase 2\grinding wheel images\digital camera photos\real\good\DSCF5200.JPG');

20

processed using Global Thresholding technique. The original image and processed image of the fully dressed wheel is as shown in figure 5.1. Original image and processed image of the grinding wheel after every 5 minutes of operation is as shown in figure 5.2. On the processed image the white portion represents the loaded portion of the wheel.

Figure 5.1 Original image and processed image of the fully dressed wheel

Figure 5.2 Original images and processed images of loaded wheel while machining Mild steel specimen at various time intervals (continued)

21

Figure 5.2 Original images and processed images of loaded wheel while machining Mild steel specimen at various time intervals (continued)

22

Figure 5.2 Original images and processed images of loaded wheel while machining Mild steel specimen at various time intervals

23

The percentage of loading obtained after every 5 minutes of operation is given in table5.1. Plot with time along X-axis and percentage of loading along Y-axis is shown in figure 5.3. Table 5.1 Percentage of loading obtained after every 5 minutes of operation on Mild Steel Specimen Sl No 1 2 3 4 5 6 7 8 9 10 11 12 Time (minutes) 5 10 15 20 25 30 35 40 45 50 55 60 Number of white pixels 6 14 20 27 27 41 43 61 64 136 159 191 Number of black pixels 78394 78386 78380 78373 78373 78359 78357 78339 78336 78264 78241 78209 Percentage of loading 0.007653 0.017857 0.02551 0.034439 0.034439 0.052296 0.054847 0.077806 0.081633 0.173469 0.202806 0.243622

24

0.3 0.25
Percentage of loading
0.2

0.15 0.1 0.05 0 5 10 15 20 25 30 35 40 45 50 55 60

Time (Minutes)
Figure 5.3 Percentage of loading Vs Time From this graph it is clear that percentage of loading increases with time which indicates a positive relationship between time and wheel loading. 5.4 MONITORING OF WHEEL LOADING ON HCHCR STEEL SPECIMEN Initially the grinding wheel is dressed in order to bring the wheel to its best working condition. Speed of the grinding wheel is set at 2000 rpm, feed as 0.2 mm/rev and depth of cut as 0.1 mm. Surface grinding operation had been carried out on HCHCR (High Carbon High Chromium) Steel specimen. As a result of surface grinding operation the wheel got loaded. Images of the grinding wheel were taken at regular intervals of time. The captured image was transferred to computer and was processed using Matlab software, by using Global Thresholding technique.

25

Machine Speed Feed Depth of cut Work piece Camera Mega Pixel Optical Zoom Shutter speed Image processing software 5.5

Hydarulic Surface Grinding Machine 2000rpm 0.1mm 0.2mm HCHCR Steel Kodak Easyshare CD14 8.2 megapixel 3x upto 1/1000sec Matlab

RESULTS AND DISCUSSIONS FOR EXPERIMENT CONDUCTED ON HCHCR STEEL Surface grinding operation had been carried on HCHCR (High Carbon High

Chromium) Steel specimen. Images of grinding wheel were taken after certain intervals of time with constant settings. These images were processed using Global Thresholding technique. Original images and processed images of the grinding wheel after various trials is as shown in figure 5.4. On the processed image the white marks represent the loaded portion of the wheel.

Figure 5.4 Original images and processed images of loaded wheel while machining HCHCR (High Carbon High Chromium ) Steel specimen at various time intervals (continued)

26

Figure 5.4 Original images and processed images of loaded wheel while machining HCHCR (High Carbon High Chromium ) Steel specimen at various time intervals (continued)

27

Figure 5.4 Original images and processed images of loaded wheel while machining HCHCR (High Carbon High Chromium ) Steel specimen at various time intervals

28

The percentage of loading obtained after certain intervals of time is given in table 5.2. Plot with trial along X-axis and percentage of loading along Y-axis is shown in figure 5.5. Table 5.2 Percentage of loading obtained after certain intervals of time on HCHCR Steel specimen Trial no Number of white pixels 1 2 3 4 5 6 7 8 9 10 11 0 7664 20027 30102 34630 40760 41911 44375 45029 61940 66995 Number of black pixels 78400 70736 58373 48298 43770 37640 36489 34025 33371 16460 11405 Percentage of loading 0 9.77551 25.54464 38.39541 44.17092 51.9898 53.45791 56.60077 57.43495 79.0051 85.45281

29

100

Percentage of Loading

80

60
40

20
0 5 12 20 26 30 34 38 40 43 50 55 Time (Minutes)

Figure 5.5 Percentage of Loading Vs Time

From this graph it is clear that percentage of loading increases with time which indicates a positive relationship between time and wheel loading.

30

CHAPTER 6 MONITORING OF SURFACE ROUGHNESS (WITH RESPECT TO WHEEL LOADING WITH THE AID OF MACHINE VISION SYSTEM)
6.1 OPTIMISATION OF DEPTH OF CUT AND FEED FOR SURFACE GRINDING ON HCHCR STEEL In order to establish a correlation between surface finish and wheel loading, the experiments have to be carried out at optimized cutting conditions. So the first step was to optimise the cutting conditions for surface grinding on HCHCR Steel. The Design of Experiments (DOE) is an effective approach to optimise the parameters in manufacturing related process. The various parameters which commonly affected the surface finish are Speed, Feed, Depth of cut, Tool material, Work material, Wheel loading etc. In this experiment tool material and work material are made constant and the speed of grinding machine also maintained constant at 2000 rpm. To optimise Depth of cut and Feed full factorial experiment was carried out. Three levels of depth of cut and feed which are frequently followed in many industries were selected and surface roughness was taken as the response. Totally Nine experiments were conducted. The three levels of depth of cut and feed selected were given in table 6.1. Table 6.1 Three levels of depth of cut and feed Depth (mm) Level 1 Level 2 Level 3 0.05 0.1 0.15 Feed (mm/rev) 0.1 0.2 0.3

31

6.2

RESULTS FOR OPTIMISED CUTTING CONDITIONS Surface grinding operation was carried out on HCHCR steel specimen keeping the

speed as constant. Nine experiments were carried out for all the possible combination of 3 levels of depth of cut and feed. The surface roughness for all the combinations were measured and shown in table 6.2. Table 6.2 - Full Factorial Experiment Sl No Depth of Cut (mm) Feed (mm/rev) Surface Roughness (m)

Ra1

Ra2

1 2

0.05 0.05

0.1 0.2

0.58 0.52

0.59 0.54

3 4 5 6 7 8 9

0.05 0.1 0.1 0.1 0.15 0.15 0.15

0.3 0.1 0.2 0.3 0.1 0.2 0.3

0.38 0.52 0.6 0.59 0.65 0.61 0.7

0.36 0.53 0.58 0.59 0.62 0.62 0.69

32

Annova table for the full factorial experiment is shown in table 6.3. This analysis is carried out for a significance level of 5% i.e., for a confidence level of 95%. Table 6.3 - Annova Table for Full Factorial Experiment Source Degree of Freedom 2 2 4 9 17 Sum of squares 0.21778 0.08444 1.24889 0.08000 1.63111 F(0.05, 9, 2) = 4.25 Mean squares F

Depth Feed Depth*Feed Errors Total

0.10889 0.04222 0.31222 0.00889

12.25 4.75 35.13

From the Annova table, it is apparent that the calculated F is greater than the tabulated F value (4.25). This clearly indicates that depth of cut and feed has significant effects on surface roughness. When the depth of cut is increased surface roughness also increases and when the feed is increased surface roughness decreases. So a depth of cut of 0.05 mm and feed of 0.3 mm/rev seems to be the optimal one. The main effects plot and interaction plot for surface finish is shown in the figure 6.1 and figure 6.2 respectively.

33

Figure 6.1 - Main effects plot for Full Factorial Experiment

Figure 6.2 - Interaction effects plot for Full Factorial Experiment

34

From the analysis, optimized parameters for surface grinding on HCHCR steel are given in the table 6.4. Table 6.4 - Optimised parameters for Surface Grinding On HCHCR steel Depth of cut Feed Speed 0.05 mm 0.3 mm/rev 2000 rpm

6.3

EXPERIMENT CONDUCTED TO RELATE SURFACE ROUGHNESS AND WHEEL LOADING In order to relate surface roughness and wheel loading, experiments were conducted at

optimised cutting conditions. Optimised working conditions are Speed = 2000rpm, Feed = 0.3 mm/rev, Depth of cut = 0.05 mm. 18 sets of readings were taken at optimized working conditions. Veho Usb Microscope with a magnification of 25X was used for capturing images. The captured image was transferred to computer and was processed using Matlab software. The photographic view of surface roughness measuring equipment is as shown in figure 6.3. Machine Speed Feed Depth of cut Work piece Camera Mega Pixel Magnification Image processing software Hydarulic Surface Grinding Machine 2000 rpm 0.3 mm/rev 0.05 mm HCHCR (High Carbon High Chromium) Steel Veho Usb Microscope 2 mega pixel 25X Matlab Global Thresholding

Image processing technique -

35

Figure 6.3 : Photographic view of Surface Roughness Measuring Equipment 6.4 RESULTS LOADING Surface grinding operation was carried out on HCHCR steel specimen at optimised working conditions. The experimental data were shown in table 6.5 FOR RELATING SURFACE ROUGHNESS AND WHEEL

36

Table 6.5 - Experimental data for correlating Surface Roughness and Wheel Loading Sl No 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Surface Roughness Ra (m) 0.34 0.34 0.35 0.38 0.4 0.43 0.47 0.5 0.52 0.55 0.57 0.58 0.6 0.61 0.63 0.62 0.64 0.66 Wheel Loading (%) 0 0.122449 0.602041 1.191327 1.992347 2.517857 3.11352 4.21231 5.519133 7.195153 8.260204 9.960459 11.61735 12.66071 14.86352 17.26148 21.47321 24.95153

37

The graph plotted with surface roughness along X axis and Wheel loading along Y axis was shown in figure 6.4.

.Figure 6.4 - Wheel Loading Vs Surface Roughness

From the graph it is clear that surface roughness increases with percentage of loading but not in a linear way. So it is necessary to develop an ANN model to predict surface roughness corresponding to wheel loading.

38

CHAPTER 7 DEVELOPMENT OF ANN (ARTIFICIAL NEURAL NETWORK) MODEL TO PREDICT SURFACE ROUGHNESS
(Courtesy: Neural Network Toolbox, Howard Demuth and Mark Beale) 7.1 MONITORING OF SURFACE ROUGHNESS THROUGH ANN MODEL Artificial Neural Networks have been studied for many years in the hope of achieving the human-like performance in the field of its application. These neural networks are composed of many non-linear computational elements operating in parallel. Neural Networks, because of their massive nature, can perform computations at a higher rate. Because of their adaptive nature using the learning process, neural networks can adapt to changes in the data and learn the characteristics of the input signals. The functioning of ANNs depends on their physical structure. A neural network usually consists of an input layer, a number of hidden layers, and an output layer. BackPropagation algorithm is utilized for the prediction of surface roughness. In back-propagation neural network, the learning algorithm has two phases. First, a training input pattern is presented to the network input layer. The network then propagates the input pattern from layer to layer until the output pattern is generated by the output layer. If this pattern is different from the desired output, an error is calculated and then propagated backwards through the network from the output layer to the input layer. The weights are modified as the error is propagated. The neural network computational model coding is built using MATLAB 2012a software. Before developing an ANN model to predict surface roughness, it is very much important to identify the input and output parameters of the network. The forecasting capability or interpolation capability of an Artificial Neural Network (ANN) model strongly depends on the appropriate selection of input-output parameters. Since the experiment is carried out at optimised speed, feed and depth of cut there is no need to vary these three

39

parameters. Wheel Loading and Surface Roughness are taken as input and output parameters respectively in this ANN model. In this study, several machining tests were carried out and thus 18 pairs of input-output data set were obtained during the machining trials. 7.2

NEURAL NETWORK DESIGN AND TRAINING

The network architecture topology or features such as number of neurons and layers

are very important factors that determine the functionality and generalization capability of the network. The selection of the activation function and training algorithm also plays a significant role to obtain better forecast of response variable. In this work, standard feedforward back-propagation hierarchical neural network has been considered for the prediction of surface roughness. The neural network has been designed with MATLAB 2012a software. The back propagation algorithm is a gradient decent error-correcting algorithm which updates the weights in such a way that network output error is minimized. The feed forward back propagation network usually consists of an input layer (where the inputs of the problems are received, the inputs are the activity of collecting data from the relevant sources. These data are fed to the neural network) one hidden layer (where the relationship between the inputs and outputs are established represented by synaptic weights) and an output layer which emits the outputs of the network. The number of hidden layer may vary depending on the nature, complexity and non-linearity of the data at hand, but single hidden layer is sufficient to deal with this work. The input layer has one neuron corresponding to wheel loading and output layer also had one neuron corresponding to surface roughness. There is no fixed rule for determining the number of neurons in the hidden layer. The number of neurons in this layer must be large enough to provide non-linear evaluation space in the network. Training of an ANN plays a significant role in designing the direct ANN-based prediction. The accuracy of the prediction depends on how well it has been trained. The training of the neural network using a feed-forward back propagation algorithm has been carried out in the work.

40

The ANN configuration is represented as 1-7-1 that is input layer consists of one input neurons; the hidden layer consists of seven neurons and the output layer consisting of one output neurons. The architecture of neural network is as shown in figure 7.1.

Figure 7.1 - Architecture of Neural Network The network performs two phases of data flow. First the input information is propagated from the input layer to the output layer and, as a result it produces an output. Then the error signals resulting from the difference between the networks predicted values and the actual values are back propagated from the output layer to the previous layers for them to update their weights accordingly. The update of weights continues until the network error goal is reached. The performance of the network was evaluated by mean squared error (MSE) between the experimental and the predicted values for every output nodes in respect of training the network. The feedback from that processing is called the average error or performance. Once the average error is be low the required goal or reaches the required goal, the neural network stops training and is, therefore, ready to be verified. MATLAB 2012a has been used for training the network architecture which was developed to predict surface roughness. The input output dataset consists of 18 data pairs. These 18 data pairs were used for training the neural network. The algorithm used for the neural network learning is the backward propagation algorithm. This training algorithm offers higher accuracy in function

41

approximation. It also facilitates faster training. After the training is completed, the network is stored in a separate file. The various training parameters used for training the ANN are Training Perfomance Number of Hidden layers Epoch Time Maximum fall Gardient 7.3 = = = = = = = tarinlm (Levenberg-Marquardt) Mean Squared error 7 (trial and error) 1000 iterations Infinity 7 1.00e-0.005

RESULTS OF ANN MODEL The ANN configuration is represented as 1-7-1 that is input layer consists of one input

neuron; the hidden layer consists of seven neurons and the output layer consisting of one output neuron. In this study, several machining tests were carried out and thus 18 pairs of input-output dataset were obtained during the machining trials which are given in the table 7.1.

42

Table 7.1 - Experimental data set for training ANN Sl No 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Wheel Loading (%) 0 0.122449 0.602041 1.191327 1.992347 2.517857 3.11352 4.21231 5.519133 7.195153 8.260204 9.960459 11.61735 12.66071 14.86352 17.26148 21.47321 24.95153
Surface Roughness Ra (m)

0.34 0.34 0.35 0.38 0.4 0.43 0.47 0.5 0.52 0.55 0.57 0.58 0.6 0.61 0.63 0.62 0.64 0.66

43

The performance plot and regression plot for training ANN are shown in the figure 7.2 and 7.3 respectively.

Figure 7.2 : Perfomance plot for ANN

Figure 7.3 - Regression plot for ANN

44

7.4

VALIDATION OF ANN MODEL Once the training of neural network is done, it is possible to predict the surface

roughness for a given value of wheel loading. By using the online monitoring system the percentage of wheel loading can be determined. Once the percentage of wheel loading is fed as input parameter to the ANN model it will predict the surface roughness corresponding to wheel loading. Thus it is possible to monitor the surface roughness, but in offline condition. Five random values of wheel loading were given to the neural network and it predicts the surface roughness corresponding to that wheel loading value. Validation test was also carried out. Five random values of wheel loading, the predicted value of surface roughness, experimental values of surface roughness and percentage of error are given in the table 7.2. Table 7.2: Predicted values of Wheel Loading for a given Surface Roughness Measured Surface Roughness Ra (m) Predicted Surface Roughness Ra (m) 0.3558 Percentage of Error (%) 1.166667

Sl No

Wheel Loading (%)

0.89

0.36

2.8

0.45

0.45923

2.05111

7.2

0.53

0.55756

5.2

11.5

0.61

0.59453

2.388525

16.12

0.63

0.6258

0.666667

45

CHAPTER 8 ONLINE MONITORING OF WHEEL LOADING AND WHEEL DRESSING

8.1 DEVELOPED PROGRAMME FOR ONLINE MONITORING OF WHEEL LOADING AND WHEEL DRESSING A programme is developed in Matlab to automate the image capturing and analysing process. Once the program is given a start it will automatically capture the image, transfers it to the computer, process the image and gives the percentage of loading as well as the processed image within 3 seconds. 8.2 PROGRAMME FOR ONLINE MONITORING PURPOSE USING MATLAB

vid = videoinput('winvideo', 1, 'RGB24_640x480'); src = getselectedsource(vid); vid.FramesPerTrigger = 1; vid.ReturnedColorspace = 'grayscale'; vid.ROIPosition = [518 98 549 620]; start(vid); stoppreview(vid); imwrite(getdata(vid),'C:\Users\vipin\Desktop\orgnalimage.jpg'); I=imread('C:\Users\vipin\Desktop\orgnalimage.jpg'); figure, imshow(I); title('Original Image'); [b,c]=size(I);k=0; y=0; for i=1:b for j=1:c

46

if(I(i,j)>40 & I(i,j)<255) I(i,j)=256;k=k+1; else I(i,j)=0;y=y+1; end end end k=k y=y z=y/(k+y);z=z*100 imshow(I); title('processed image'); 8.3 VALIDATION OF THE PROGRAMME USING VEHO USB MICROSCOPE This experiment was conducted on surface grinding machine in which Al2O3 grinding wheel is used. HCHCR steel is used as work piece material. Veho Usb Microscope is used for capturing the images. With this microscope a magnification upto 200X can be obtained. This experiment was conducted with a magnification of 25X. With these magnified images errors can be reduced very much as compared to digital camera images. Since lighting is provided using the LEDs within the microscope, a constant lighting is available for all conditions. A provision is given in the Matlab program for setting the region of interest to avoid the resizing of captured images. Machine Speed Feed Depth of cut Work piece Camera Hydarulic Surface Grinding Machine 2000 rpm 0.3 mm/rev 0.05 mm HCHCR (High Carbon High Chromium) Steel Veho Usb Microscope

47

Mega Pixel Magnification Focus Image processing software

2 megapixel 25X Manual Matlab Global Thresholding

Image processing technique -

8.4

RESULTS AND DISCUSSIONS Veho Usb Microscope with a magnification of 25X was used for capturing images.

Images of grinding wheel were taken after every 5 minutes of operation and are processed to find out the percentage of loading. Image processing was carried out using Global Thresholding technique in which a binary image is created with loaded portion in white pixels and rest of background in black pixel. Using developed Matlab program, the whole image capturing and analysing process was automated. Original image and processed image of the fully dressed wheel is as shown in figure 8.1. Original image and processed image of the grinding wheel after every 5 minutes of operation is as shown in figure 8.2. On the processed image the white marks represents the loaded portion of the wheel.

Figure 8.1 - Orginal Image and processed image of fully dressed wheel

48

Figure 8.2 Original images and processed images of loaded wheel for the machining of HCHCR (High Carbon High Chromium ) steel specimen at various time intervals (continued)

49

Figure 8.2 Original images and processed images of loaded wheel for the machining of HCHCR (High Carbon High Chromium ) steel specimen at various time intervals (continued)

50

Figure 8.2 Original images and processed images of loaded wheel for the machining of HCHCR (High Carbon High Chromium ) steel specimen at various time intervals

51

The percentage of loading obtained after every 5 minutes of operation is given in table 8.1. Plot with time as X-axis and percentage of loading as Y-axis is shown in figure 8.3. Table 8.1 Percentage of loading obtained after every 5 minutes of operation on HCHCR Steel Specimen Time (minutes) 0 5 10 15 20 25 30 35 40 45 50 55 60 Number of white pixels 1

Sl No

Number of black pixels

Percentage of loading

1 2 3 4 5 6 7 8 9 10 11 12 13

78399

0.001276

26 124 153 1229 1935 2683 3389 5395 5805 9082 12269 13418

78374 78276 78247 77171 76465 75717 75011 73005 72595 69318 66131 64982

0.033163 0.158163 0.195153 1.567602 2.468112 3.422194 4.322704 6.881378 7.404337 11.58418 15.64923 17.1148

52

Figure 8.3 Percentage of Loading Vs Time From this graph it is clear that percentage of loading increases with time which indicates a positive relationship between time and wheel loading. 8.5 COST ESTIMATION The cost estimation of the hardware for the online monitoring system for grinding wheel loading and dressing is as shown in table 8.2 Table 8.2 Cost Estimation for the Online Monitoring System Sl No 1 2 3 4 Item Veho Usb Microscope Laptop USB Extension Chord Microscope Stand Quantity Required 1 1 1 1 Cost (Rs) 4500 25,000 50 250 Total Cost = Rs 29,800

53

CHAPTER 9 CONCLUSION
The experiments carried out show the feasibility of using Machine Vision System and Image Processing techniques in determining the wheel loading and wheel dressing. Global Thresholding with a threshold range is suitable for determining wheel loading. Experiment conducted on Mild Steel and HCHCR Steel for determining the wheel loading shows that the percentage of wheel loading increases with time. Since most of the industrial applications are concerned majorly with surface finish, time to redress the wheel depends upon the surface finish of the machined component. ANN was developed to predict the surface roughness corresponding to wheel loading. ANN model developed has shown good results with 5.2 % of error in predicting the surface roughness. A programme was developed on Matlab to automate the image capturing and analysing process. The whole machine vision system is made useful for online monitoring purpose with this programme. Programme developed for online monitoring of wheel loading and wheel dressing has shown good results with zero percentage of error. Implementation cost for the online monitoring of wheel loading and wheel dressing is Rs 29,800. (Exclusive of programme cost and Matlab software cost). Present scenario of wheel dressing is based on human prediction and experience. This may cause wheel dressing before or after the set wheel loading, which may result in loss in productive time and quality respectively. This system will result in reduction in non productive time because wheel dressing time is application oriented and not based on human judgment.

54

9.1

SCOPE FOR FUTURE WORK It is suggested that the output of the online monitoring system i.e., the percentage of

loading has to be fed as input to the ANN model, so that the ANN model will predict the surface roughness corresponding to the percentage of loading. This will make the monitoring of surface roughness also online. So once the surface roughness required for a particular application is reached, an alarm will be activated which indicates the dressing operation has to be carried out. Carry out Surface Grinding operation

Determine the Percentage of loading using online monitoring system

Use ANN Model to predict surface roughness corresponding to wheel loading obtained

Is Predicted Ra >= Maximum attainable value

Yes

NO
Initiate an alarm to carry out dressing operation

55

REFERENCES:
1. A.Hosokawa, H. Yasui, Characterization of the grinding wheel surface by means of image processing (1st report), J. JSPE 62 (9) (1996) 12971301. 2. Amin A. Mokbel, T.M.A. Maksoud, Monitoring of the condition of diamond grinding wheels using acoustic emission technique, Journal of Materials Processing Technology 101 (2000) 292297. 3. Bernard C. Jiang, Chung-Li ,Tsung-Chi Chen, Machine Vision Inspection for the Protrusion Rate of a Diamond Tool Journal of Manufacturing Systems Vol. 2a/No. 5 2001. 4. C.K. Huang, Y.S.Tarng, C.Y. Chiu, A.P. Huang, Investigation of machine vision assisted automatic resharpening process of micro-drills Journal of Materials Processing Technology 209 (2009) 59445954. 5. Hello Zeuch, Understanding and applying Machine Vision. 6. Howard Demuth and Mark Beale, Neural Network Toolbox. 7. http://nptel.iitm.ac.in/courses/Webcoursecontents/IIT%20Kharagpur/ Manuf%20Proc%20II/pdf/LM-27.pdf, LM-28.pdf, LM-29.pdf. 8. J. C. Su , Y. S. Tarng, Measuring wear of the grinding wheel using machine vision International Journal for Advanced Manufacturing Technology (2006) 31: 5060. 9. K.-C. Fan, M.-Z. Lee, J.-I. Mou, On-line non-contact system for grinding wheel wear measurement International Journal Advanced Manufacturing Technology (2002) 19:1422. 10. Katsushi Furutani, Noriyuki O hguro , Nguyen Trong Hieu , Takashi Nakamura, In-process measurement of topography change of grinding wheel by using hydrodynamic pressure , International Journal of Machine Tools & Manufacture 42 (2002) 1447 1453. 11. Pawel Lezanski, An intelligent system for grinding wheel condition monitoring, Journal of Materials Processing Technology 109 (2001) 258263.

56

12. Stephane Lachance, Robert Bauer, Andrew Warkentin, Application of region growing method to evaluate the surface condition of grinding wheels International Journal of Machine Tools & Manufacture 44 (2004) 823829. 13. T. Warren Liao, Chi-Fen Ting, J. Qu , P.J. Blau, A wavelet -based methodology for grinding wheel condition monitoring, International Journal of Machine Tools & Manufacture 47 (2007) 580592. 14. Wen-Tung Chang, Ting-Hsuan Chen, Yeong-Shin Tarng , Measuring characteristic parameters of form grinding wheels used for microdrill fluting by Computer Vision Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 3, 2011. 15. Z. Feng . X. Chen, Image processing of grinding wheel surface International Journal Advanced Manufacturing Technology (2007) 32: 2733.