Abstract: This paper introduces an automatic positioning and machining method for contoured parts based on machine vision. This research utilizes a relaxation-restored image and edge feature extraction algorithm on a visual operating platform to automatically generate CNC machining programs for arbitrarily positioned contoured parts. Simulation experiments show that this research can meet the requirements of fixtureless CNC machining. Keywords: machine vision, edge feature, process path, relaxation recovery, outline work parts [b][align=center]Research of Numerical Control System for Machining without Special Fixture Based on Machine Vision[/align][/b] Deng Shaping, Wu Xiaojun, Ding Lijun, Zhang yun Abstract: This paper introduces a method that the position of outline work parts can be discerned and defined automatically based on machine vision. On the visible operating platform, the image edge of the outline work parts placed any-angle is restarted and its characteristic is abstracted by relaxes recovers algorithm. The machining program of the work parts for numerical control machine can be produced automatically based on the edge characteristic. The simulation experiment shows that the research can satisfy the demand of NC machining without special fixture. Keywords: machine vision, edge characteristic, process path, relaxes recovers, outline work parts. 1 Introduction Developed industrial countries have regarded flexible manufacturing systems (FMS) as the main development direction of the machinery manufacturing industry, and modular fixtures are one of the cutting-edge research topics in this field[1]. This system targets contoured parts, employing machine vision image acquisition and contour recognition to directly generate CNC machining source programs for workpieces in arbitrary poses, achieving fixtureless, automatic positioning and machining. Developed on the Visual C++ 6.0 platform, the system implements the entire process of image acquisition, self-positioning, source code generation, and simulated toolpath. The system can also anticipate problems such as tool-workpiece collisions and overcuts, and can automatically set parameters, perform tool analysis, dynamically process and display tool principle data, and provide a visual simulation demonstration of the machining process. 2. System Structure This system uses a 600-line WAT-902 series monochrome CCD and a manual aperture CS interface T2314FICS-3 megapixel industrial automation lens to capture images of workpieces in arbitrary poses on the worktable. The analog video signal is then input into an MV-20 industrial image acquisition card (which features unique video filtering technology, high definition and display speed, and 4-channel switching for high-quality monochrome and color images). The analog signal is converted into a digital signal, input into a general-purpose PC for image processing, extracting workpiece contour edge feature points and storing them in text format. This system can automatically acquire the workpiece pose on the worktable using a machine vision system; extract the workpiece contour edge feature information by combining image processing algorithms; simulate CNC machine tool machining positioning and tool path animation on the main interface of the Visual C++ application using two-dimensional planar graphics; and convert the equipment coordinates into NC language that the CNC machine tool can recognize according to the extracted workpiece contour. 3. Workpiece Pose Acquisition and Recognition Process under Machine Vision The machine vision system is computer-centric and consists of modules such as a vision sensor, a high-speed image acquisition system, and a dedicated system or general-purpose PC with image processing capabilities. It can automatically acquire object images and process, analyze, measure, qualitatively analyze, and quantitatively interpret their features to obtain a certain understanding of the target object and make corresponding decisions. 3.1 Information Source of Machine Vision System The vision sensor (CCD) is the main source of machine vision information. The CCD surface is imaged using a T2314FICS-3 megapixel industrial automation lens (manual aperture CS interface), converting light energy into voltage. When a photon with an appropriate wavelength strikes a specific material in the device, a charged quantum is generated and remains in the generated region. At the horizontal blanking moment of the grating scan, the computer obtains brightness information from the voltage signal encoding of the new line. The effective video is generated during the retrace of the grating scan. 3.2 Sampling and Quantization Process: The charge on the detector is converted into voltage through a resistor and amplified. An analog-to-digital converter converts this signal into a digital representation, simultaneously performing sampling and quantization functions. Since digital images are acquired using grating scanning and sampling, they are one-to-one in time and space. The sampled signal must be represented relative to the vertical blanking moment at the top of the image signal; this time relationship corresponds one-to-one with a specific position on the screen. The quantization process transforms the grayscale of a pixel into discrete integer values. The number of quantization bits is directly related to image quality. Considering the convenience of computer operation, this system uses 8-bit quantization, with the grayscale value of the image pixel between 0 and 255. 3.3 Image Generation Mechanism First, the brightness at point (x,y) is defined as an ideal function f(x,y) of two spatial variables. After image quality degradation, the observed image g(x,y) is obtained, g(x,y) = D(f(x,y)). When photographing the same stationary object, noise will appear randomly at any time, and the location and magnitude of the noise will be different at each instant. Image noise includes random noise and coherent noise. The former manifests as speckle interference, and the latter manifests as mesh interference. This system is a linear displacement-invariant system. The original image f(x,y) and the noisy image n(x,y) are both stationary random fields. The noise n(x,y) is additive noise with a mean of 0 and is uncorrelated with f(x,y). In addition to the deterministic distortion operator D, there is also the interference of additive noise n. Therefore, the general model of linear displacement-invariant image degradation is a superposition of the distortion operator D and random noise g = D(f) + n. F, n, and g are sampled at the same interval to produce the same array. Let f, n, and g be defined as a set of pixels, written in the form of a pixel column vector such as [f[sub]1[/sub],f[sub]2[/sub],…,f[sub]N[/sub]][sup]T[/sup]. Then, the observed image g is obtained, and f is optimally estimated, i.e., restoration (the inverse problem of blur). The main purpose of image restoration processing is to remove interference and blur, restoring the original appearance of the image. A criterion function is selected: This function is the sum of noise terms related to the measurement and prior terms related to the real image, depending on the observed data and the unknown image, respectively. f is considered a conceptual function, assigned values ​​between sampling points, and f is solved by interpolation. Next, image feature extraction is performed. Edges are the boundaries between the workpiece and the background in the image, and also the places where the image grayscale and color change sharply. Due to the low-frequency characteristics of the sensing element, there are blurred parts at the edges between the workpiece and the background, with small grayscale changes. The gradually non-convex (GNC) method is used to handle the weak diaphragm case by minimizing the objective function for the existence of edges, i.e., finding images with sharp edges. Here, f is the gradient of the image at point i, li represents the discontinuity of f at the i-th pixel, and its value can range from 0 to 1. When li = 1, the pixel at point i is an edge point, and fi represents the brightness of the i-th pixel. By removing the discontinuity factor, the problem is simplified to an operator containing only continuous variables, returning a scalar metric for the local "edges" of the image. The V function is a truncated parabola in the prior energy map of the GNC algorithm, penalizing gradient occurrences in the image in an incremental manner. Since the function is usually non-convex and can have multiple minimum values, it cannot be minimized by techniques such as gradient descent. Therefore, the following piecewise smoothing function is used to approximate V: where t is equivalent to the magnitude of the edge gradient, and C* is a scalar constant. Combined with the "annealing" algorithm, a continuous stepwise approximation of the prior shape is achieved. In the GNC method used, the original shape of the prior term is preserved. The smaller p is, the closer the approximate result is to the ideal prior energy. As p decreases from 1 to 0, it continuously tracks the local minimum, smoothly changing V* until it precisely approximates V, while generating a prior energy family. During this stage, the object is measured, and multiple measurement values ​​are extracted to form a feature vector. This simplified information expresses the knowledge for subsequent understanding and matching. Finally, in subsequent processing, salt-and-pepper noise is eliminated through erosion followed by dilation. After processing, white isolated point noise is eliminated during erosion. See Figures 1, 2, and 3: 4 Simulation Software Architecture 4.1 VC++ Program Structure and Project File Establishment 4.1.1 Interface Design Most of the Visual C++ programs in this research project are built on this class library. To simulate tool behavior within a fixed-size area and simultaneously view the corresponding NC language in another view window, it is desirable to keep the view separator lines fixed while selecting the split window. A class CFixSPlitterWnd is derived from CSplitterWnd to implement form splitting. Then, by responding to mouse messages such as WM_MOUSEMOVE, WM_LBUTTONDOWN, and WM_SETCURSOR, the mouse behavior is changed so that the double-arrow shape is no longer displayed within the split form, and dragging functionality is disabled. From this point, the client area of ​​the form is divided into two fixed parts: one part corresponds to the default view class of AppWizard, and the other part corresponds to a newly added form view class derived from CFormView. 4.1.2 Command and Data Input Menu Functionality Command input is implemented by adding menu commands to the main menu to open data input dialog boxes and perform other operations. The menu editor is opened, an "Options" menu is inserted, and menu options "Text," "Shaping," "Tool Flow," "Command," and "Refresh" are added to the "Options" menu. Simultaneously, corresponding accelerator key table entries are created in the accelerator key editor and assigned the same ID number as the menu options. During program execution, the ENABLE and GRAYED attributes of the "Text," "Shaping," "Tool Path," "Instruction," and "Refresh" menu options are interlocked to ensure that instructions are generated immediately after data is read. Workpiece positioning can be changed, tool path simulation cannot be interrupted to avoid resource waste, and refreshing after tool path completion allows for cyclical implementation. A common file dialog box embedded in the program using the CFileDialog class is used to implement the standard Windows "Open" dialog box. The storage path of the part's vertices after contour recognition based on machine vision image acquisition is selected and then read out in the same mode. 4.1.3 Tool Path Mode Selection: Add a tool path mode dialog box resource and edit the dialog box. Add grouped box controls titled "Tool Path" and "Tool Speed," and add radio button groups for "Internal Circle Machining" and "External Circle Machining" tool paths and "Fast" and "Slow" tool speeds respectively, roughly simulating the requirements of CNC machine tool machining. Select the machining path, and a dialog box for setting the depth of cut pops up. After setting, click the "Start" button to begin tool path machining. 4.2 Effect timers are a method of periodically calling functions at specified intervals, similar to timer interrupts in DOS. There are two ways to create a timer using the CWin::SetTime() function: first, specify a static function that is called when the timer expires; second, send a WM_TIMER window message to the window that calls SetTime() when the timer expires, and call the message handling function OnTimer() for appropriate processing. The program calls the timer setting function in the OnOptionsGo() function of the window class. Periodic calls generate an animation effect of continuous tool rotation and tool movement. Due to space limitations, the implementation process of the message response functions is not detailed here; relevant code can be found in the source code. 5 Conclusion This system uses image restoration and image feature extraction methods that perform relaxation operations, improving edge detection performance, achieving smooth denoising and degradation suppression, and extracting clearer features for image segmentation. This image processing software was developed using the Visual C++ 6.0 development platform through an image access subroutine in IFS, and the identified workpiece vertices are written to a .txt file in a fixed pattern. The vertices are then read from the text using the same pattern, and the behavior of the hardware is simulated on a PC using software. The provided CNC machine tool machining process is basically consistent with the actual machining process, which can help predict various problems that may arise during machining. In the noise-removable GNC algorithm used, nonlinear operation is an absolute element for success. A linear operation plus appropriate subsequent nonlinearization can calculate any image, which is of great significance for the implementation of the GNC algorithm using a simple and direct neural network. As CNC machine tools develop towards flexibility and unmanned operation, the level of functional integration is more reflected in functions such as automatic workpiece positioning. 6. Innovations of the Authors 6.1 This system adopts the relaxation algorithm, which has been proven in practice to be an optimized algorithm for image edge detection, extracting clearer image features and meeting the accuracy requirements of CNC machining. 6.2 As a vision system for CNC systems, this system can realize fixtureless positioning, tool self-positioning machining, and unmanned diagnosis. Its application in flexible manufacturing systems can greatly improve the system's real-time performance, accuracy, and efficiency. References [1]. Li Boru et al. Research on combined fixtures in flexible manufacturing systems [D], Advanced Manufacturing Technology Papers, www.e-work.bj.cn; [2]. Sun Jixiang, Image Processing [M], Science Press, 2004; [3]. Wang Junxiu, Restoration of degraded images using VC++.Net [J], Microcomputer Information, (2004) 04, p. 82.