APPLICATION OF MACHINE VISION TECHNIQUES IN SURFACE FINISH MEASUREMENTS OF MACHINED COMPONENTS AND TO EVALUATE THE FUNCTIONING OF THE COMPONENT

Transcription

1 Research Paper: Dhinesh Kumar and Rajaram, 2013: Pp APPLICATION OF MACHINE VISION TECHNIQUES IN SURFACE FINISH MEASUREMENTS OF MACHINED COMPONENTS AND TO EVALUATE THE FUNCTIONING OF THE COMPONENT Dhinesh Kumar, S and Rajaram Narayanan CMJ University, Shillong, India *MGR University, Chennai, Tamil Nadu, India ABSTRACT Intense international competition has focused the attention of manufacturers on automation as means to increase productivity and improve quality. The surface quality of any machined parts plays a very important role in the performance of good-quality machined components and significantly improves fatigue strength, corrosion resistance, creep life etc. Surface roughness also affects several functional attributes of parts, such as friction, wear, light reflection, heat transmission, ability of distributing and holding a lubricant, coating etc. Therefore, the desired surface finish is usually specified and appropriate processes are required to maintain the quality. Surface finish of machined parts plays an important role on mechanical properties. The surface finish of work piece is an issue of main concern to the manufacturing industry and the determination of surface roughness of the work piece using inspection is a very important technology. The traditional stylus method is the most widely used technique in industry. In this, a precision diamond stylus is drawn through the surface being examined and the perpendicular motion is amplified for further processing. The accuracy of stylus method depends on the radii of diamond tips used. When the surface roughness falls below 2.5 μm, the stylus instruments are affected by large system error. The disadvantage of such method is that they require direct physical contact and line sampling which may not represent the real characteristics of the surface. In recent times, computer vision technology has attained tremendous vitality in a lot of fields. New applications continue to grow and existing applications expand. Several investigations have been performed to inspect surface roughness of a work piece based on computer vision technology. It has been observed that surface roughness of a work piece is strongly characterized by the surface image. Machine vision has the advantage of grabbing the images on-line without accounting for factors like noise and vibrations of machine tool. Advances in the semiconductor industry has made machine vision field to find its way from crude binary systems that were extremely sensitive to changing light conditions to rugged industrial systems processing gray level images at the rate of several hundred millions of instruction per second. Keywords: Computer vision technology, light reflection, heat transmission, fatigue strength and machine tool. INTRODUCTION Machine vision has evolved to become a mainstream automation tool, enabling computers to replace human vision in high speed and precision manufacturing techniques. Machine vision systems are used to capture images, extract information using vision sensors and make intelligent decisions. The sequence of operations - image capture, early processing, region extraction, region labeling, highlevel identification, and qualitative/quantitative conclusion are the characteristic of image understanding and machine vision problems. Images processed using a computer is digitized first, after which it may be represented by a rectangular matrix with elements corresponding to the brightness at appropriate image locations. Images usually acquired through modern cameras may be contaminated by a variety of noise sources and decreasing intensity and in most cases the type of noise and state of lighting are also not known apriori. There is a need for design of recognition systems with capability to adapt to changing environments automatically. This requires computing architectures that are less complicated, highly flexible and more cost-effective as compared to the traditional ones, which calculates the coefficients of a general-purpose model. All machined surfaces, irrespective of their vetra flatness and smoothness posses surface irregularities. Surface texture indicates repetitive or random deviations from the nominal surface which form the three-dimensional topography of the surface. Surface texture includes roughness, waviness, lay and flaws as shown in Figure 1. The surface is obviously important in many practical situations. The proper functioning and service life of a machine part is in many instances largely dependent on the quality of its surface. A thorough understanding of surface quality is important in order to specify correctly those surfaces which will provide optimum properties for service, life, appearance, performance and other desired mechanical properties. For many years typical parameters used to describe the surface include the average value Ra, the RMS value Rq and various peak height estimates. Researchers hitherto used peak measurements rather than average values because they observed that peak measurement correlated more with tribological situations than did average values. Importance Of Surface Finish: It is proposed that (Rainer Brodmann and Gerd, 1985) the measurement of engineering surface roughness is 31 ISSN (Online): (Print) - Rising Research Journal Publication

2 becoming increasingly important. Surface roughness is a measure of the character of the surface. Along with the specified tolerances, surface finish is used to characterize the accuracy of a surface. The end function of a component generally determines the surface roughness of that component. Rainer Brodmann and Gerd (1985) performed a case study on video head and reported that the surface roughness of a video head has direct influence on the wear behavior of the video tape and on the quality of the picture production. The desired surface of the video head is mirror like and the specification of the surface finish needs to be less than 2 µin (rms). It is also a mass-produced component. Serope Kalpakjian (1989) reported that the friction and wear characteristics of mating surface are influenced to a large extent by their surface finish. Also, fatigue life of rotating components or parts can be improved by providing a finer finish, thus reducing the effects of surface imperfections. The manufacturing process for a component is many times selected based on the specified tolerance and surface roughness of that component. Serope Kalpakjian (1989) cites the range of surface roughness obtainable by various manufacturing processes. Sometimes the surface roughness is also indirectly determined or influenced by the dimensional tolerance specified for that component. It this case, the method of manufacturing is determined by the specified dimensional tolerance and the process so selected determines the surface finish. Of course, these two details can be superimposed and a common table enveloping all the three variables, namely surface roughness, tolerance, and manufacturing process, can be obtained. These reported works on surface roughness evaluation and its importance in realtime applications provides a base for evolving techniques that need to estimate the surface roughness of machined components with greater degree of accuracy. The roughness of a surface may be defined by more than one parameter. Thomas (1982) points out that roughness exists in two principal planes, at right angles to the surface, when it may be denoted by a factor of height and in the plane of the surface (identified as texture ) and represented by a factor of wave length (λ). Many methods of surface finish measurement have been developed, ranging from the simple touch comparator to sophisticated optical like techniques. The following are major categories of surface finish measurement methods: (i) Comparison based methods (ii) Direct measurement methods (iii) Optical methods In this paper, an important parameter of machined components namely surface roughness R t is estimated with greater degree of accuracy compared to the existing methods. The main focus of this work is to find more efficient and more general algorithms and implement them on more technologically sophisticated equipment. In particular, parallel architectures are being used to ease the enormous computational load of operations conducted on image data sets. The advantage of the proposed structure is that the image enhancement operation is evolved from primitives and does not require any apriori information about the type of noise. The developed system employs a genetic image interpretation based on a hypothesis and verifies principle to evolve image operators and perform adaptive image processing. Also, the evolution is performed at the simplified functional level and uses simple logical functions and adders as primitive components. This research work solves partially the unsolved problem of automating the sequence of relevant operations required to solve a specific task. The developed work is tested in online on images of specimens grabbed by the computer vision system with linearly decreasing intensity. The grabbed image is enhanced using virtual reconfigurable circuit configured using evolutionary algorithms and features are extracted using two different schemes, one using Fourier transform (FT) and the other using wavelet decomposition. The FT method is used to extract the features of image texture namely the major peak frequency F 1, and the principal component magnitude squared value F 2. Using the wavelet (Db4) multiresolution decomposition algorithm the energy details of the sub band images namely Energy total (Et), Energy horizontal (Eh), Energy vertical (Ev) and Energy diagonal (Ed) are extracted. These extracted features of the enhanced image are given as input to a trained neural network and the surface roughness parameter R t is estimated. For comparative study image enhancement schemes using conventional filtering technique are considered and it is established that evolved solutions outperform the conventional one both in terms of performance and implementation cost. The evolved image filter is implemented in realtime on a Xilinx FPGA based image processing board, which is available as a commercial off the shelf hardware device. This allows complex and fast computation to be performed by dedicated hardware instead of software in digital computer, as hardware units can operate in parallel. The objectives of this research work include; To develop a novel Evolvable hardware based image enhancement technique, to improve the quality of the images of surfaces by adapting to changes to process variations and adverse conditions in a CIM environment such as contrast reversal and intensity gradients, angular uncertainties, blur caused by changes in depth field, scale changes, partial obliteration or missing features. To implement the evolvable hardware unit in real-time on a Xilinx FPGA board. To apply wavelet transforms and 32 ISSN (Online): (Print) - Rising Research Journal Publication

3 Fourier Transforms independently and extract the relevant features of image texture. To train a neural network and use it for estimating the surface roughness Rt of components manufactured using processes such as grinding and milling. To compare the surface finish obtained using each network with that using stylus approach. METHODOLOGY An attempt is made in this paper to describe an exhaustive survey of the works done so far in the area relevant to this work. The survey covers the direct and indirect methods of measuring schemes for surface roughness. Literature study on structural and statistical approaches of surface texture analysis performed on various machining operations is also presented. Different measurement parameters used in evaluating the surface roughness of a machined component are discussed. The experimental setup used for grabbing the images is also discussed. The detail of the high speed image processing card on which the image processing algorithm is implemented is also presented. A brief reference on the Fourier transform and wavelet transform is made. The extraction of image texture features using the above transforms is discussed elsewhere. RESULT AND DISCUSSION The performance of the proposed Evolvable HardWare (EHW) based image enhancement technique is reviewed. To evaluate the effectiveness of the proposed EHW technique two performance measures namely (i) the variation of optical surface roughness parameter Ga with respect to Rt and (ii) Peak Signal to Noise Ratio (PSNR) value of the images are obtained. For comparison studies the enhancement schemes reported by Goch et al (1999) is considered. The results are presented separately for ground and milled surfaces. Trials on measuring the surface roughness of ground surfaces were performed on mild steel specimens. The raw images were captured using high resolution vision system (Table 1 and 2). CONCLUSION It can be inferred from results of above section that among all the approaches used to enhance image quality in computer vision, application of EHW techniques to the grabbed image was found to be the most effective method. Hence, this technique can be efficiently used for non-contact inspection of components, which has assumed considerable significance. The parameter (optical roughness value) after applying EHW technique had a better correlation (i.e. higher correlation coefficient) with the average surface roughness (R t) measured using a conventional and widely accepted stylus type instrument for the components manufactured particularly using milling and grinding processes. Also, the PSNR value is found to be best in the EHW based scheme. The higher PSNR indicates a better resolution and degradation and blurring of edges gets removed to a large extent. It is worth mentioning that this improvement is significant for the milled surfaces compared to the ground surfaces. This trend can be explained from the fact that there is a large local variation in the characteristic of ground surface, as compared to the more uniform and fine milled surfaces resulting in better correlation between stylus R t and vision roughness R t for milled surfaces. From the real-time implementation point of view also, it has already been established that the proposed EHW based approach is the best suited. Conventional filtering schemes may employ various image filters for different scenarios but it is very difficult and time consuming to find a set of appropriate filters and as a result most of them are suited for off-line implementation only. On the contrary, the proposed structure has the advantage that it does not require any expert knowledge to find the type and order of filters for a given domain. This work has described the use of computer vision techniques to inspect the surface roughness of a work piece under various milling and grinding operations. The machined surfaces using milling and grinding for mild steel with a surface finish range of R t 1.57 µm to µm for grinding and 2.5µm to µm for milling are evaluated in this work for their surface quality using a computer vision system. The advantages of the present approach are the non-contact measurement and ease of automation. The results obtained in the present work clearly indicate that the machine vision approach can be effectively used to evaluate the surface finish of machined components (Bennett, 1978 and Majumdar and Bhushan, 1990). A novel image enhancement structure using evolvable hardware is used. The VRC unit and the Genetic Algorithm processor are implemented using VHDL language. The image enhancement algorithm presented in this work is unique, and is ideally suited for a real time environment, where no apriori information is available about the noise mixed with the acquired image. Conventional nonadaptive filters perform well only as long as the spatial density of the impulse noise is less. On the contrary an adaptive filter can handle impulse noise even with higher probabilities and also can preserve sharpness and detail while smoothing non-impulse noise. Testing the algorithm with simulation alone may not always reflect the actual system performance and can neglect the effects of physical uncertainties. In this work, the evolved image filter is implemented and tested in a hardware unit. In all the previous analogous works surveyed so far, it is observed that the image enhancement algorithms are suited for offline implementation only. This is because of the reason that the algorithms have been implemented in a 33 ISSN (Online): (Print) - Rising Research Journal Publication

4 sequential processor and the time it takes to process an image is large. However, in this work the FPGA hardware is preferred over a processor based implementation and due to this the evolution time is drastically reduced and is highly suitable for use in on-line (Adkins, 1969; Thwaite, 1979: Hisayoshi Sato and Mansonori O-Hari, 1982 and Hasewage et al., 1996). The optical roughness parameters and image features using Fourier Transformation (, F 1 and F 2) and wavelet Transformation (E t, E h, E v and E d) are calculated for the milling and ground surfaces after applying the EHW based image processing algorithm. An ANN is used for predicting the roughness values of the same components using the above roughness parameters as input. The predicted surface finish values using ANN are found to correlate well with the conventional stylus surface finish (Rt) values. In the case of WT based image feature extraction this correlation is more (i.e correlation coefficient is higher) in milling compared to ground components. Several verification tests have shown that the maximum absolute error between the surface roughness measured by vision system and that measured by the stylus instrument is less than 10% for milling and limited to 15% for grinding. Thus, the developed computer vision based surface roughness evaluation can be effectively used to measure the surface roughness over a wide range of cutting conditions in milling and grinding. The processing speed (i.e. speed with which features of captured image enhanced), increases significantly when implemented in such hardware units and this method is found to be several times faster than conventional ones. Therefore this concept could be used for milled and ground surfaces and easily be extended for estimation of roughness of surfaces manufactured by other processes such as shaping, broaching, etc. REFERENCES Rainer Brodmann and Gerd Thus Optical roughness measuring instrument for fine machined surfaces, Opt. Eng., 24 (3): Serope Kalpakjian Manufacturing Engineering and Technology, Addison Wesley, Reading, MA, pp Thomas T.R Rough Surfaces, Longman, New York. Wick C. and R.F. Veilleux Tools and manufacturing Engineers handbook, Quality Control and Assembly, 4 th edn., SME. Thwaite E.G Wear, 57:71. Adkins, H A look at surface finish, Am. Mach., 113: Hisayoshi Sato and Mansonori O-Hari Surface roughness measurement by scanning electron microscope, Annals of CIRP, 31: Hasewage M., Okunda K., Liu K. and M. Nunobiki A new approach to modeling machined surface profiles, Proceedings: Institute of Mechanical Engineering, 240: Bennett H.E Scattering characteristic of optical materials, Optical Engineering, 17: Majumdar A. and B. Bhushan Role of fractal geometry in roughness characteristic and contact mechanics of surface, Journal of tribology, 112: 205. Goch G., Peters J., Lehmann P. and H. Liu Requirements for the application of speckle correlation techniques to on-line inspection of surface roughness, Annals of CIRP, 48: Table 1. Machining Parameters used in Grinding and the corresponding surface roughness values of the components Optical Parameters Test no Speed (m/s) Feed m/min Depth of cut ( m) G 1 a G 2 a G 3 a G 4 a Rt (µm) Ga 1 = Grey level arithmetic average for raw image 2 = Grey level arithmetic average after applying Keren approach 3 = Grey level arithmetic average after applying EHW Algorithm 4 = Grey level arithmetic average after applying Vandewalle approach. 34 ISSN (Online): (Print) - Rising Research Journal Publication

5 Table 2. Results of the different super-resolution algorithms on the surface images (Ground surfaces) PSNR (db) Sl.No Speed (m/s) Depth of cut (mm) Feed (mm/rev) Rt ( m) Fig 1. Surface texture *************** 35 ISSN (Online): (Print) - Rising Research Journal Publication