Nghiên cứu đề xuất giải pháp và công cụ phát hiện tự động khuyết tật hàn qua phim chụp ảnh bức xạ

BỘ GIÁO DỤC VÀ ĐÀO TẠO TRƯỜNG ĐẠI HỌC SƯ PHẠM KỸ THUẬT THÀNH PHỐ HỒ CHÍ MINH LUẬN VĂN THẠC SĨ NGUYỄN NGỌC HÀ NGHIÊN CỨU ĐỀ XUẤT GIẢI PHÁP VÀ CÔNG CỤ PHÁT HIỆN TỰ ĐỘNG KHUYẾT TẬT HÀN QUA PHIM CHỤP ẢNH BỨC XẠ NGÀNH: KỸ THUẬT CƠ KHÍ - 60520103 SKC007558 Tp Hồ Chí Minh, tháng 10/2017 BỘ GIÁO DỤC VÀ ĐÀO TẠO TRƯỜNG ĐẠI HỌC SƯ PHẠM KỸ THUẬT THÀNH PHỐ HỒ CHÍ MINH LUẬN VĂN THẠC SĨ NGUYỄN NGỌC HÀ NGHIÊN CỨU ĐỀ XUẤT GIẢI PHÁP VÀ CÔNG CỤ PHÁT HIỆN TỰ ĐỘNG KHUYẾT TẬT HÀN QUA PHIM CHỤP ẢNH BỨC XẠ NGÀNH: KỸ THUẬT CƠ KHÍ – 60520103 Hướng dẫn khoa học: PGS TS ĐẶNG THIỆN NGƠN Tp Hồ Chí Minh, tháng 10 năm 2017 i ii iii iv v vi vii viii PHỤ LỤC CODE MATLAB SỬ DỤNG CHO PHẦN MỀM NHẬN DIỆN KHUYẾT TẬT HÀN function varargout = Main(varargin) % MAIN MATLAB code for Main.fig % MAIN, by itself, creates a new MAIN or raises the existing % singleton* % % H = MAIN returns the handle to a new MAIN or the handle to % the existing singleton* % % MAIN('CALLBACK',hObject,eventData,handles, ) calls the local % function named CALLBACK in MAIN.M with the given input arguments % % MAIN('Property','Value', ) creates a new MAIN or raises the % existing singleton* Starting from the left, property value pairs are % applied to the GUI before Main_OpeningFcn gets called An % unrecognized property name or invalid value makes property application % stop All inputs are passed to Main_OpeningFcn via varargin % % *See GUI Options on GUIDE's Tools menu Choose "GUI allows only one % instance to run (singleton)" % % See also: GUIDE, GUIDATA, GUIHANDLES % Edit the above text to modify the response to help Main % Last Modified by GUIDE v2.5 06-Sep-2017 17:13:57 % Begin initialization code - DO NOT EDIT gui_Singleton = 1; gui_State = struct('gui_Name', mfilename, 'gui_Singleton', gui_Singleton, 'gui_OpeningFcn', @Main_OpeningFcn, 'gui_OutputFcn', @Main_OutputFcn, 'gui_LayoutFcn', [] , 'gui_Callback', []); if nargin && ischar(varargin{1}) gui_State.gui_Callback = str2func(varargin{1}); end if nargout [varargout{1:nargout}] = gui_mainfcn(gui_State, varargin{:}); else gui_mainfcn(gui_State, varargin{:}); end % End initialization code - DO NOT EDIT 67 % - Executes just before Main is made visible function Main_OpeningFcn(hObject, eventdata, handles, varargin) % This function has no output args, see OutputFcn % hObject handle to figure % eventdata reserved - to be defined in a future version of MATLAB % handles structure with handles and user data (see GUIDATA) % varargin command line arguments to Main (see VARARGIN) % Choose default command line output for Main handles.output = hObject; % Update handles structure guidata(hObject, handles); end % UIWAIT makes Main wait for user response (see UIRESUME) % uiwait(handles.figure1); end % - Outputs from this function are returned to the command line function varargout = Main_OutputFcn(hObject, eventdata, handles) % varargout cell array for returning output args (see VARARGOUT); % hObject handle to figure % eventdata reserved - to be defined in a future version of MATLAB % handles structure with handles and user data (see GUIDATA) % Get default command line output from handles structure varargout{1} = handles.output; end % - Executes on button press in pushbutton1 function pushbutton1_Callback(hObject, eventdata, handles) % hObject handle to pushbutton1 (see GCBO) % eventdata reserved - to be defined in a future version of MATLAB % handles structure with handles and user data (see GUIDATA) [filename, pathname] = uigetfile({'*.bmp;*.jpg;*.gif;*.png';'*.*'}, 'Pick an Image File'); if pathname ~= S = imread([pathname,filename]); axes(handles.axes1); imshow(S); handles.S = S; guidata(hObject, handles); end end % - Executes on button press in pushbutton2 function pushbutton2_Callback(hObject, eventdata, handles) %a3Test gets the name of the image as an input and displays the %probabilities of each number %imageName name of the image to load %close all; %close all figures number = handles.S; %Load the model load model; %load the model of the network %Load the image and transform it % number = imread(imageName); %read the image file number = rgb2gray(number); imshow(number); %show the image 68 movegui('northwest'); number = im2double(number); %transform the format from 8bit to double number = number'; %this is just to adapt to USPS database data = number(:); %finally transform the matrix into a vector %Compute the probabilities probabilities = computeProbabilities(model, data); %Show the probabilities in a graph figure; numbers = [0 10 11 12]; bar(numbers,probabilities); movegui('northeast'); end function ret = logistic(input) ret = / (1 + exp(-input)); end function ret = log_sum_exp_over_rows(a) % This computes log(sum(exp(a), 1)) in a numerically stable way maxs_small = max(a, [], 1); maxs_big = repmat(maxs_small, [size(a, 1), 1]); ret = log(sum(exp(a - maxs_big), 1)) + maxs_small; end function ret = computeProbabilities(model, data, wd_coefficient) % model.input_to_hid is a matrix of size (n_hid,256) % model.hid_to_class is a matrix of size (13,256) % data.inputs is a matrix of size (256,) % data.outputs is a matrix of size (13,) % first, the forward pass, i.e calculate a variety of relevant values hid_in = model.input_to_hid * data; % input to the hidden units, i.e before the logistic size: (n_hid,) hid_out = logistic(hid_in); % output of the hidden units, i.e after the logistic size: (n_hid,) class_in = model.hid_to_class * hid_out; % input to the components of the softmax size: (13, ) class_normalizer = log_sum_exp_over_rows(class_in); % log(sum(exp)) is what we subtract to get normalized log class probabilities size: (1,) log_class_prob = class_in - repmat(class_normalizer, [size(class_in, 1), 1]); % log of probability of each class size: (10, ) class_out = exp(log_class_prob); % probability of each class Each column (i.e each case) sums to size: (10, ) ret = class_out; end function ret = model_to_theta(model) input_to_hid_transpose = transpose(model.input_to_hid); hid_to_class_transpose = transpose(model.hid_to_class); ret = [input_to_hid_transpose(:); hid_to_class_transpose(:)]; end function ret = theta_to_model(theta) n_hid = size(theta, 1)/(256 + 13); ret.input_to_hid = transpose(reshape(theta(1: 256*n_hid), 256, n_hid)); ret.hid_to_class = reshape(theta(256 * n_hid + : size(theta,1)), n_hid, 13).'; 69 end % - Executes on button press in pushbutton3 function pushbutton3_Callback(hObject, eventdata, handles) % hObject handle to pushbutton3 (see GCBO) % eventdata reserved - to be defined in a future version of MATLAB % handles structure with handles and user data (see GUIDATA) end 70 PHỤ LỤC Nội dung báo “Nghiên cứu đề xuất giải pháp và công cụ phát hiện tự động khuyết tật hàn qua phim chụp ảnh xạ” trình bày hội thảo ICSSE 2017 ngày 21/07/2017 đăng tải IEEEXplore (DOI: 10.1109/ICSSE.2017.8030899) ngày 11/09/2017 71 Organized by Ho Chi Minh City University of Technology and Education (HCMUTE), Vietnam Taiwan Association of System Science and Engineering (TASSE) Sponsored by IEEE - Institute of Electrical and Electronics Engineers IEEE Systems, Man, and Cybernetics Society, Taiwan Ho Chi Minh City Department of Science and Technology Daikin Air Conditioning Vietnam Joint Stock Company ICSSE 2017 IEEE International Conference on System Science and Engineering July 21-23, 2017 • Ho Chi Minh City, Vietnam COPYRIGHT BMS Part Number: CFP1746J-CDR ISBN: 978-1-5386-3421-9 Copyright and Reprint Permission: Abstracting is permitted with credit to the source Libraries are permitted to photocopy beyond the limit of U.S copyright law for private use of patrons those articles in this volume that carry a code at the bottom of the first page, provided the per-copy fee indicated in the code is paid through Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923 For reprint or republication permission, email to IEEE Copyrights Manager at pubs-permissions@ieee.org All rights reserved Copyright ©2017 by IEEE CONTENTS SS1 INTELLIGENT HEALTHCARE SYSTEMS (14 papers)  SS01-Design Strategies to Improve Self-feeding Device - FeedBot for Parkinson Patients Nguyen Truong Thinh, Tran Tan Thanh and La Hoang Thang  SS04-Camera-based Heart Rate Measurement Using Continuous Wavelet Transform Bing-Fei Wu, Po-Wei Huang, Tsong-Yang Tsou, Tzu-Min Lin and Meng-Liang Chung  SS15-Edge Detection Based On Fuzzy C Means In Medical Image Processing System 12 Mong Hien Nguyen, Thanh Binh Nguyen and Quoc Viet Ngo  SS06-Adaptive Exploration Strategies for Reinforcement Learning 16 Kao-Shing Hwang, Chi-Wei Hsieh and Wei-Cheng Jiang  SS07-Using Voice and Gesture to Control Living Space for the Elderly People 20 Haobijam Basanta, Yo-Ping Huang and Tsu-Tian Lee  SS09-Kinematic Measurement of the Ankle Joint using a d.o.f electro-goniometer 24 V Thang Nguyen, L T Giang Huynh , V A Dung Cai, M Tam Nguyen  SS12-Hands-Free Control of an Electric Wheelchair Using Face Behaviors 29 Viet Ngo Ba, Hai Nguyen Thanh and Thuyen Ngo Van  SS13-Long-tail Effect on ECG Classification 34 Thai Nguyen Huu, Nghia Nguyen Thanh, Binh Duong Van, Hai Nguyen Thanh and Hung Nguyen Manh  SS14-Error-Rate Analysis for ECG Classification in Diversity Scenario 39 Thuy Le Thi Minh, Nghia Nguyen Thanh, Binh Duong Van, Hai Nguyen Thanh and Hung Nguyen Manh  09-Medical Classification using Interval Intuitionistic Fuzzy Sets 44 Chih-Min Lin and Jing Zhao  36-A study of Aloe Vera Peeling and Dicing system 48 Tuong Phuoc Tho and Nguyen Truong Thinh  126- Analysis and Design of Transtibial Prosthesis 52 Jhon Hernandez Martin, Luis Alberto Parra Piñeros, German Antonio Mendieta  177- An Effective Video-based Model for Fall Monitoring of the Elderly 57 L.U Thuc Hoang, Van Tuan Pham and Jenq-Neng Hwang  192- Kinect-based Virtual Training System for Rehabilitation 62 The Chien Hoang, Ha Trang Dang, Viet Dung Nguyen i  85- Dynamic Programming based Control for Perturbed Discrete Time Nonlinear Systems .339 Van Hung Pham, Hoai Nam Nguyen and Doan Phuoc Nguyen  91- Design of a Compliant Bio-Inspired Camera-Positioning Mechanism for Autonomous Mobile Robots 345 Ngoc-Phuong Hoang and Huy-Tuan Pham  101- Experimental Investigation of Speed Control of Hydraulic Motor Using Proportional Valve 350 Hai Tran Ngoc, Cung Le and Anh Dung Ngo  125- An Extended Navigation Framework for Autonomous Mobile Robot in Dynamic Environments using Reinforcement Learning Algorithm 356 Nguyen Van Dinh, Nguyen Hong Viet, Lan Anh Nguyen, Hong Toan Dinh, Nguyen Tran Hiep, Pham Trung Dung, Trung-Dung Ngo and Xuan-Tung Truong  140- Fault Detection and Isolation for Robot Manipulators Using Statistics 360 Cao Thanh Trung, Hoang Minh Son, Dao Phuong Nam, Tran Nhat Long, Do Tien Toi, Phan Anh Viet  208- Advanced Control Methods for Two-Wheeled Mobile Robots 364 Khanh G Tran, Phuoc D Nguyen and Nam H Nguyen MECHANICAL AND AUTOMOTIVE ENGINEERING (14 papers)  24-Effects of Resistance, Capacitance and Self-Inductance on Accumulated Energy in the Hybrid Ignition System 369 Do Van Dung, Do Quoc Am, Nguyen Tan Ngoc  56-Computational Analysis on Hybrid Electric Motorcycle with Front Wheel Electric Motor using Lithium Ion Battery 375 Van Trang Nguyen, Pyung Hwang and Thinh Huynh  59- Active Fault Tolorant Torque Distribution Control of In-Whel Motors Electric Vehicles Based on Karmal Filter Approach 380 Trung-Hieu Nguyen, Bo-Chiuan Chen, Dejun Yin, and Phuoc-Son Huynh  73- Experimental Research on Fuel Control System of Internal Combustion Engine using Dual Fuel LPG – Diesel 385 Long Giang Nguyen Van, Dzung Do Van and Phuoc-Son Huynh  75-An Approach to the Automatic Detection of Weld Defects in Radiography Films Using Digital Image Processing 391 Dang Thien Ngon, Le Chi Cuong and Ha Nguyen Ngoc  76- Research on Using PID Algorithm to Control Simulation Model of Regenerative Braking System Based on Driving Cycles 395 Duong Tuan Tung, Do Van Dung and Nguyen Truong Thinh vi Proceedings of the International Conference on System Science and Engineering 2017 (ICSSE 2017) An Approach to the Automatic Detection of Weld Defects in Radiography Films Using Digital Image Processing Ngon Dang Thien*, Cuong Le Chi, Ha Nguyen Ngoc  Actual survey with local NDT companies, including Alpha NDT, CA NDT, TTA, QIS, APAVE, OCD VSP, radiography films interpretation has still being manually performed by technicians Abstract— Oil refinery industry is using large amount of radiographic films of weld defects, requiring human work by technicians to interpretation This paper presents a method and automated system for determination of weld defect from radiographic films by using the image processing technique The hardware and software system rapidly and automatically convert the radiographic films into digital images, then analyzes and compares with the digitalized radiographic images database The probable defects and its type recognized defect are determined This approach will be significant in developing a professional system for partially replacing technician work in image interpretation, ensuring quality and productivity in welding manufacturing technology Keywords— weld defect; radiographic films; interpretation; image processing; non-destructive testing To improve the quality and productivity of weld assessment, a tool that automatically assess the weld defect is highly demanded and should be developed This tool helps and replaces human work of film interpretation, ensure objectivity of the quality assessment, thus reduce inspection cost and increase competitiveness for company II THEORITICAL BACKGROUND Reza Nejatpour et al has used image processing, CAD method for welding defect detection [1] This method is only applied for surface defects of weld joint Nacereddine et al recommended for the digitized radiographic image processing with threshold methods, however, the detection system to identify weld defects has not been developed [2] I INTRODUCTION Oil and gas refinery industry is a significant area in Vietnam nowadays, where a large amount of weld and the inspection of weld defects using analog X-ray radiographic films The films producing generates environment pollution, human health risk and, especially, human work by technicians to exactly interpret the defect for weld repair and maintenance Rapid defect detection will avoid disaster risk and reduce cost for repairing or replace these products, thus increase enterprise competitiveness Saba Madani et al propose image processing by using KMeans algorithm to put the similar colors of the image up to a certain distance [3] The results include the general and basic concepts related to image processing, but limiting in research and development Kasban et al present a new cepstral approach using neutral network for defect detection from radiographic images [4] The proposed approach has two phases: a training phase and a testing phase The result is capable of recognizing excessive penetration, lack of side-fusion undercut, longitudinal crack inclusion, tungsten inclusion, localized porosity, burn-through, lack of side-fusion, and crater pipe at low degradation cases Radiographic Testing (RT), which is based on differential radiation absorption by variation of part thickness or differential material density, is a widely used Non-Destructive Testing (NDT) technique to evaluate the structural continuity of welded components The defects are detected by the brightness (darkness) of analog radiographic films When completed film processing, technician will proceed interpretation for image on these films and give evaluation (acceptable or rejected) basing on approval criteria The evaluation of weld quality must meet the following requirements: sufficient qualified technicians and rapid evaluation of with a large amount of radiographic films However, almost mechanical manufacture companies has limited number of qualified interpretation technicians, with low processing speed and the expensive commercial equipment In general, the implementation of welding defects assessment by image processing methods are being more interested in world research, however the proposed methods are applied for surface defects of product or propose image processing method High-tech applications using neutral network seems to be inapplicable universally for commercialization to replace human work of radiographic film interpretation III MATERIAL AND METHOD A Radiographic film interpretation requirements Ensure interpretation reliability; Rapid processing progress; Applicable for all kinds of welds with various sizes *Thien Ngon Dang is with the Department of Machinery Manufacturing Technology, Ho Chi Minh City University of Technology and Education, 708000 Vietnam (mobile phone: (84) 913 804803, fax: (848) 38964922, email: ngondt@hcmute.edu.vn) Chi Cuong Le is with the Department of Welding & Metal Technology, Ho Chi Minh City University of Technology and Education, 708000 Vietnam (e-mail: lccuong@hcmute.edu.vn) Ngoc Ha Nguyen (Mr.) is master's student at the Ho Chi Minh City University of Technology and Education, school of mechanical engineering (e-mail: hann@pvep.com.vn) 978-1-5386-3422-9/17/$31.00 ©2017 IEEE B Propose method The proposed method for automated weld defect detection system using radiographic image processing is shown in Fig 391 Proceedings of the International Conference on System Science and Engineering 2017 (ICSSE 2017) Step 1: Digitalizing radiographic films using digital camera system, then converting image to black and white images where AAAA is the name of main weld defect, BBBB is the classification of main weld defect, and XXXX is the serial number, starting from 0001 Step 2: Applying propose process of detect defection based on database and software E Structure of defect detection system Structure of defect detection system is as following Fig The RT films are converted into the digital image files and stored in a personal computer and then is read by a specially programmed software Step 3: Classifying defects and represent results Radiographic film RT films - Digital Camera - Image Files Digitalizing PC Digitalize system - Defect Database - Software Defect Detection F Algorithm of image processing and weld defect detection Algorithm of image processing and weld defect detection is shown in Fig Figure Propose method for automated weld defect detection system using radiographic image processing C Digitalizing method RT film image files to be interpreted Fig shows the digitalizing method for radiographic film The light system is to indicate radiographic film for capturing The digital camera used is Canon PowerShot G9 12.1MP with 6x Optical The image processing uses Contrast stretching, Histogram equalization, Contouring, Brightness by Photoshop software or equivalent software in order to make black and white images The image files are the black & white images, and is stored to database Digital Camera Program Figure Structure of defect detection system Interpretation of Result Light system Weld defect images Database Image Processing Detected Image files Compare with Database Image detection Not detected Classify defect Figure Flow chart for digitize method Documentation Fig below show digitalize equipment Figure Algorithm of image processing and weld defect detection IV PERFORMING IMAGE PROCESSING A Image Acquisition Image acquisition is the process of obtaining a digitized image from a real world source Each step in the acquisition process may introduce random changes into the values of pixels in the image which is referred to as noise A RT film image is captured and stored into the computer for further processing This may be achieved by taking a photograph with a conventional camera In our method, we used Canon PowerShot G9 digital camera for image capturing Then, this image is trimmed with m × n (width and height) to make all the images to fix size B Image enhancement Actually, image enhancement technique is to make the image clearer so that various operations can be performed easily on the image To that, at first the captured RGB image is converted to the gray level image Contrast stretching (often called normalization) is a simple image enhancement technique that attempts to improve the contrast in an image by stretching the range of intensity 1-Sliding mechanism; 2-Digital camera; 3-Tray to position radiographic films; 4-Light source; 5-Computer Figure Digitize equipment D Database classification The coding of radiographic image files shall be as below AAAA BBBB XXXX 392 Proceedings of the International Conference on System Science and Engineering 2017 (ICSSE 2017) values it contains to span a desired range of values, e.g the full range of pixel values that the image type concerned allows to a file, named m1 and m2 Then we count the total number of black pixels (in binary, it is represented as 1) in m1 and that in m2 The two are then compared If the number of black pixels in m1 is greater than those of pixels in m2, a defect is diagnosed to be presented in the considering image, otherwise we can say that no defect is presented To understand this concept clearly, let us consider the example below The general process of the contrast stretching operation on grayscale image is to apply the following equation on each of the pixels in the input image to form the corresponding output image pixel: O( x, y)  ( I ( x, y)  min)( ni )i max  Let, we have a test image and a reference image of equal size (5×5) Now applying preprocessing steps on the test image we find matrix m1 as: where, O(x,y) represents the output image, I(x,y) represents the xth pixel in the yth column in the input image In this equation, ni represents the number of intensity levels, i represents the initial intensity level, "min" and "max" represent the minimum intensity value and the maximum intensity value in the current image respectively 1 0  0  0 0 C Noise Reduction Noise reduction is a process of removing noise from a captured image To remove noise some filtering techniques can be proposed as follows: 0 0 0 1 0 0  0  1  Again, applying the preprocessing operations on the reference image another matrix m2 is found: 1 0  0  0 0 One method to remove noise is convolving the original image with a mask that represent a low-pass filter or smoothing operation For example, the Gaussian mask comprises the elements, determined by a Gaussian function This convolution brings the value of each pixel into closer harmony with the surrounding values In general, a smoothing filter sets each pixel to the average value, or a weighted average, of itself and its surrounding pixels; the Gaussian filter is just one possible set of weights But smoothing filters tend to blur an image, since the pixel intensity values that are significantly higher or lower than the surrounding neighborhood would "smear" across the area Because of this blurring, linear filters are seldom used in practice for noise reduction 0 0 0 0 0 0 0 0  0  0  Here, the number of black pixels for the reference image is and for the test image this number is So, here obviously 6>2 and we can make decision that defect is presented in the test image E Classifying the defect using software After applying preprocessing operations, we have database of reference images By using Python software, which is a multi-paradigm programming language, defects will be classified some common welding defects and recognize with test images V RESULTS AND DISCUSSION The proposed system detects defect successfully In this section, we show the result, using Python software, by comparing the primary RT film image with reference image database We also classify different types of defects found through defect detection process Now, RT films with common specific defects can be listed as below A Lack of fusion Figure Image processing D Applying the proposed defect detection process All pre-processing operations are applied to the reference image, stored in the computer database to compare with the considering image of the weld Let, the resulting image is I2 Now we consider I1 as the resulting image found from the test image after applying all preprocessing operations The proposed technique is represented hereunder I1 and I2 are stored as matrices form Figure “Lack of fusion” defect 393 Proceedings of the International Conference on System Science and Engineering 2017 (ICSSE 2017) B Burn through By developing this technology, we should find solutions for shorten processing time with assure the accuracy It is obvious that the conclusion image processing can have very wide application in welding defects inspection In order to achieve this, digital image processing is definitely one of the most critical technologies for solving numerous problems for human at the present time and in the future REFERENCES [1] Reza Nejatpour, Ali Asmari Sadabad, Ali Akbar Akbari Automated weld defects detection using image processing and cad methods 2008 ASME International Mechanical Engineering Congress and Exposition, October 31-November 6, 2008, Boston, Massachusetts, USA [2] N Nacereddine, M Zelmat, S S Belaïfa and M Tridi Weld defect detection in industrial radiography based digital image processing 2005 ISSN 1307-6884 [3] Saba Madani, Mortaza Azizi Detection of Weld Defects in Radiography Films Using Image Processing 2015 Faculty of Science, Cumhuriyet University [4] H Kasban, O Zahran, H Arafa, M El-Kordy, S.M.S Elaraby, F.E Abd El-Samie Welding defect detection from radiography images with a cepstral approach 2010 Elsevier Ltd [5] ASM Handbook Volume 17 Nondestructive Evaluation and Quality control, 9th Edition [6] Maria Petrou, Costas Petrou Image Processing: The Fundamentals 2010 John Wiley & Sons, Ltd ISBN: 978-0-470-74586-1 [7] ASME Boiler & Pressure Vessel Code Section V Nondestructive Examination 2015 Edition [8] ASME Boiler & Pressure Vessel Code Section VIII Div I Rule for Construction of Pressure Vessels 2015 Edition [9] ASME Code for Pressure piping Process Piping Section B31 2014 Edition [10] AWS D1.1 Structural Welding Code – Steel 2015 Edition [11] API 1104 Welding for Pipelines and Related Facilities 2010 Edition [12] M S Rogers Radiographs Course Reference WIS 10 TWI 2002 Figure Burn through defect C Slag inclusion Figure Slag inclusion defect D Porosity Figure 10 Porosity defect VI CONCLUSIONS This paper presents the methodology for weld defect from radiographic films by using the image processing technique and results obtained as follows:  A method as well as a system for digitizing RT films was proposed and fabricated;  A database of common welding defects, such as lack of penetration, burn through, slag inclusion and porosity, was built up  Software to automatically recognize and evaluate welding defects from the RT films by image processing  The ability to identify common welding defects is over 95% by using this method This method needs to be optimized and developed to identify other welding defects which are difficult to diagnose, such as small crack, excess penetration, undercut 394 S K L 0 ... bày giải pháp công cụ phát tự động khuyết tật hàn qua phim chụp ảnh xạ bằng kỹ thuật xử lý ảnh, nhằm thay người cơng việc giải đốn ảnh chụp xạ Cơng cụ phát tự động khuyết tật hàn qua phim. .. chụp ảnh xạ giải vấn đề sau: - Đề xuất giải pháp tự động đánh giá khuyết tật mối hàn qua phim chụp ảnh xạ - Chế tạo thiết bị số hoá phim chụp ảnh xạ - Xây dựng sở dữ liệu ảnh số theo nhóm khuyết. .. quyền chủ động Ý nghĩa khoa học thực tiễn đề tài 3.1 Ý nghĩa khoa học - Đề xuất giải pháp, cấu trúc hệ thống, giải thuật phần mềm tự động phân tích, đánh giá phát tự động khuyết tật hàn qua phim chụp