Imperial Journal of Interdisciplinary Research (IJIR) Vol-2, Issue-6, 2016 ISSN: 2454-1362, http://www.onlinejournal.in Surface Roughness Evaluation of Turned Surfaces Using Wavelet Packet Transform Srivatsa T 1, Ravi Keerthi C 2, Dr. Srinivas H K3 , Ravi Kumbar4 & T. Madhusudhan5 1 M.Tech Student, Department of Mechanical Engineering, SJBIT, Bengaluru, Karnataka, India. 2 Asst. Prof, Department of Mechanical Engineering, GAT, Bengaluru, Karnataka, India. 3,5 Prof, Department of Mechanical Engineering, SJBIT, Bengaluru, Karnataka, India. 4 Asst. Prof, Department of Mechanical Engineering, SJBIT, Bengaluru, Karnataka, India. Abstract: The surface roughness is a measure of the deformation of a surface resulting from various machining process. The measurement of surface roughness is categorized in two ways, the contact method and the non-contact method. The first method involves the direct contact of the measuring instrument with the specimen being measured. The latter method avoids contact with the specimen being measured such as vision system. In this paper using the non-contact method the prediction of surface roughness values of turned surfaces are explained with the application of wavelet packet transform and image processing. The paper also shows the comparison of the measured value and the predicted value. Keywords: Wavelet Packet Transform (WPT), denoising, feature extraction, mother wavelet. 1. Introduction Image processing is a method which gives detail information of a digital image and is used to enhance the quality. The operations used in the image processing include de-noising and filter of images etc. The objective is to remove the disruptions from the image which improves the quality of the picture. An image is a representation of pixels in a matrix format. The images are of two types namely the RGB image format and the gray scale image format. The RGB image format consist of three channels of red, green and blue whereas the gray scale image format contains one channel Image de-noising is used to obtain an image with better clarity and the objective of removing the noise in an image is to obtain important details and retaining the main characteristics with complete removal from an image. Most of the information are obtained from an image by applying mathematical transformations. Out of such transformations the wavelet packet transform are selected. The wavelets being mathematical functions are used to study data in terms of resolution or scale. The selection of levels Imperial Journal of Interdisciplinary Research (IJIR) and the selection of mother wavelet for de-noising and decomposing an image depend on the values of peak signal to noise ratio in wavelet packet transform In image processing the peak signal to noise ratio value of an image is in the range of 30-50db for an 8 bit data. Higher the peak signal to noise ratio, better is the quality of the image. The acquisition of an image for the feature extraction process is an important factor to be considered. The conditions in using the vision system such as camera settings and lighting arrangements are important to obtain better results. [6] The use of artificial neural network has been used in image processing that can be applied to different fields. The artificial neural network has provided solutions in pattern related problems. The artificial neural networks are models consisting of neural networks used for evaluating the objectives which are dependent on more number of inputs. 2. Literature review The method of confining a coloured image using the wavelet packet with best tree selection was studied by Dr. G. K. Kharate. In this study a method was proposed using the wavelet packet technique with the threshold entropy. The comparison of the image obtained from the wavelet technique with the JPEG 2000 file format was carried out. [1] The regeneration of an image obtained between wavelet transform and wavelet packet transform was studied by Sanjeev Chopra. In this study the different wavelets were chosen to confine the image. The process of filtering for each wavelet was carried out. Among the criteria’s the mean square error and peak signal to noise ratio were important. The image regeneration using the two criteria’s for both the wavelet technique was studied. [2] Anna Zawada-Tomkiewicz proposed the method of determining the surface roughness using the vision system with the wavelet packet transform and neural network with a surface roughness estimator. The Page 978 Imperial Journal of Interdisciplinary Research (IJIR) Vol-2, Issue-6, 2016 ISSN: 2454-1362, http://www.onlinejournal.in outcome of the estimator proved to be efficient to indicate the roughness quality for turned operations. [3] Paul Dan Cristea described the application of artificial neural network in image processing and concluded that the artificial neural network provide faster and good results. [4] Anna Zawada-Tomkiewicz proposed the correlation of energy vector entropy of decomposition tree coefficients and machined surface criteria and tool wear using wavelet packet transform. [5] 3. Wavelet packet transform In wavelet packet transform, the image is decomposed in to low frequency and high frequency parts. The image is decomposed in to trees known as packets. In each packet there is important information that can be extracted. The wavelet packet transform has an advantage over the wavelet transform. The decomposition of image in to levels give energy values based on the selection of level. In wavelet packet transform, each packet gives information on energy on the image. Also better information of edges or textures of the image can be achieved, hence a better detailed image can be obtained. Figure 1 below shows the decomposition structure of the trees at level 3. It is used for approximations images and the detail images. The energy of the packets will produce a better reconstructed image. corresponding level and the mother wavelet in the de-noised with the corresponding level are used for feature extraction. 4. Methodology In this paper the surface texture of a mild steel specimen is studied. Turning operations were carried out on the specimen on a HMT lathe machine with carbide tip tool. The parameters such as speed, feed and depth of cut were varied for each operation. The image of the surface texture of the specimen was captured and used for image processing. The vision system was used to capture the images. The vision system consists of a Digital camera, tripod stand, working table, lighting arrangements and a V-block as shown in figure 3and figure 4. The specimen dimensions of 40mm diameter and 95mm length. Nikon Digital SLR Camera D5500 and Simpex 888 tripod stand were used. The image acquisition of the images was carried out in a dark room simultaneously after each operation. Once the images were acquired, the images are cropped to a size of 400x200 as shown in figure 5 using Picasa software. These images are then fed to the decomposition procedure in which the selection of mother wavelet and levels of decomposition are carried out for both wavelet transform and wavelet packet transform. Fig 2. HMT Lathe Machine Fig 3. Vision system Fig 4. Specimen on V-Block Fig 5. Surface texture Fig1.Decomposition of wavelet packet transform with level 3 3.1 Selection of mother wavelet: The procedure is to perform the decomposition process to determine the levels of decomposition and selecting the mother wavelet for the images. Eight mother wavelets have been used. Among the selected mother wavelets, the mother wavelet with more peak signal to noise ratio with respect to the level are chosen. Four levels of decomposition were used. The peak signal to noise ratio indicates the condition of the original information present in the image during the reconstruction process. The MATLAB software 2013Ra is used to carry out the procedure. Based on the decomposition criteria, the mother wavelet selection de-noising with level 1 in terms of the peak signal to noise ratio is calculated. The mother wavelet in the decomposition with the Imperial Journal of Interdisciplinary Research (IJIR) Haar wavelet was chosen as the mother wavelet and the level of decomposition was fixed to level 1.The de-noised procedure was carried out using the wavelet bior2.4 with level 1. The first step involved was converting the RGB image in to greyscale image. The grey scale image was then de-noised by using the Gaussian noise. Page 979 Imperial Journal of Interdisciplinary Research (IJIR) Vol-2, Issue-6, 2016 ISSN: 2454-1362, http://www.onlinejournal.in Fig 6. Original image Fig 7. Greyscale image Mother Wavelet Level 1 sym1 23.7235 haar 23.7235 bior2.4 23.7375 bio6.8 23.7307 bior2.2 23.7374 db4 23.728 db15 23.7194 db20 23.7194 Table 2. De-noised level peak signal to noise ratio Fig 8. Noisy image Fig 9. De-noised image The above figures show the grey scale image, noisy and de-noised image from the original image of 250rpm, 0.06mm/rev and 0.5mm depth of cut. In figure 10 the decomposition level of the original image is shown in terms of peak signal to noise ratio for every level. The haar wavelet and level 1 were selected based on the peak signal to noise ratio value. Mother Level 1 Level 2 Level 3 Level 4 haar 37.8336 37.2596 37.1379 37.3253 sym1 37.8336 37.2596 37.1379 37.3253 bior2.4 35.8937 35.0785 34.7666 34.2663 bior 2.2 35.6542 34.7821 34.3845 33.9181 bior 6.8 37.0715 36.5153 36.4095 36.2595 db4 37.5506 36.8316 36.6299 36.5257 db15 37.4314 36.7219 36.5889 36.4849 db20 37.4076 36.7291 36.501 36.4359 wavelet Table 1. Decomposition level peak signal to noise ratio In table 2 the den-noised table for level 1 with values of peak signal to noise ratio are shown, the wavelet used is bior2.4 was used. Fig 10. Decomposition level 1 In figure 10 the decomposition for level 1 are shown, which shows the tree decomposition. From the tree shown below, the original image is split in to approximation and detail images. The reconstruction of images is based on the energy retention values of the image. In table 3, the values in each tree is shown with b1, b2, b3 and b4. b1 represents the approximation image and b2, b3 and b4 are the reconstructed images of the original image as shown in figure 11. b3 is used to obtain the important parameters as the energy value is more compared to b2 and b3. The parameters such as kurtosis, mean, variance, standard deviation and skewness are obtained. b1 98.0566 b2 0.4883 b3 1.1537 b4 0.3015 Table 3. Energy retention values Imperial Journal of Interdisciplinary Research (IJIR) Page 980 Imperial Journal of Interdisciplinary Research (IJIR) Vol-2, Issue-6, 2016 ISSN: 2454-1362, http://www.onlinejournal.in Fig 11. Reconstructed images 5. Artificial neural network The artificial neural network is a processing unit that is used to train and retain knowledge from the inputs and provide an efficient outcome. In this paper a multilayer perceptron model is used to predict the surface roughness values. The surface roughness values are experimentally found out using a stylus instrument called Mitutoyo surface roughness measuring tester SJ-210. The multilayer perceptron network consists of hidden layers of neurons in which the inputs are passed from one layer to another layer. The addition of one or more hidden layer results in better results. In this model the conditions such as the iterations and the number of neurons and the selection of algorithm to execute are important factors. In this work 85% of images are used for training and 15% are used for testing. The model will predict the surface roughness values in MATLAB 2013Ra software. In figure 12 shows the model used for prediction of surface roughness value. Fig 12. Multilayer perceptron model 6. EXPERIMENTAL RESULTS The results have been plotted with predicted versus experimental values of surface roughness Ra and necessary correlation has been obtained using the least square method Imperial Journal of Interdisciplinary Research (IJIR) Fig 13. Correlation between experimental values of Ra and predicted values of Ra for training data. Fig 14. Correlation between experimental values of Ra and predicted values of Ra for training data. Fig 13 and Fig 14 presents the overall correlation results for surface roughness prediction using WPT based feature extraction for surface images using ANN for Turned surfaces for both training and test data respectively. Sl. No. 1 2 3 4 5 6 7 8 9 10 11 Predicted Ra 3.461434 3.348804 2.707005 2.575648 2.581301 1.956348 1.988463 1.949999 1.614096 1.571566 1.583535 Experimental Ra 3.464 3.351 2.697 2.554 2.568 1.917 1.951 1.91 1.553 1.507 1.52 Table 4. Values of experimental and predicted surface roughness for turned surfaces Page 981 Imperial Journal of Interdisciplinary Research (IJIR) Vol-2, Issue-6, 2016 ISSN: 2454-1362, http://www.onlinejournal.in 7. Conclusion In this work the surface roughness parameter R a has been predicted for turned surfaces using features extracted from surface images. WPT has been used to extract the features. ANN has been used to correlate these statistical features with the surface roughness parameter Ra measured using conventional stylus based measurement. The developed ANN model has been able to predict surface roughness effectively. The use of vision system for surface roughness evaluation is effective for turned surfaces. Vol. 38, Nos 5-6. Page 685-690. 1998 i 1998 Elsevier Science Ltd. [8] D.E.P. Hoy and F. Yu, “Surface Quality Assessment Using Computer Vision Methods”, Journal of Materials Processing Technology, 28 (1991) page 265-274 Elsevier. . 8. Future work Comparison of WPT results can be made with WT techniques. Effective de-noising techniques with thresholding can be used in order to obtain improved results. References 1] Prof. Dr. G. K. Kharate, “Color Image Compression Based on Wavelet Packet Best Tree”, IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 2, No 3, March 2010, page 3135. [2] Sanjeev Chopra, “Comparative Analysis of Wavelet Transform and Wavelet Packet Transform for Image Compression at Decomposition Level 2”, 2011 International Conference on Information and Network Technology IACSIT Press, Singapore IPCSIT vol.4 (2011). [3] Anna Zawada – Tomkiewicz “Estimation of surface roughness parameter based on machined surface image”, metrology and measurement system, volume 17, 2010, page 493-504. [4] Paul Dan Cristea, “Application of Neural Networks in Image Processing and Visualization”, NATO Science for Peace and Security Series C: Environmental Security page 59-71. [5] Anna Zawada – Tomkiewicz, “Machined surface quality estimation based on wavelet packets parameters of the surface image”,Pomiary Automatyka Kontrola, 2010, page 606-609. [6] Omar Monir Koura, “Applicability of image processing for evaluation of surface roughness”, IOSR Journal of Engineering, Vol. 05, Issue 05 (May. 2015), ||V2|| Page 01-08. [7] M.B.Kiran, “Evaluation of surface roughness by vision system”, Int. J Mach, Tools Manufacturing. Imperial Journal of Interdisciplinary Research (IJIR) Page 982