Automatic Defect Detection in Mechanical Members using Computer Vision Avani S Rao, Akshay Kumar C, Ankit Kumar Begwani UG Student, PES Institute of Technology, Bangalore, India avani_rao@rediffmail.com, akkupesit28@gmail.com, ankit.begwani@gmail.com Shylaja S S Professor PES Institute of Technology, Bangalore, India shylaja.sharath@pes.edu Abstract— The Mechanical members are subjected to static and dynamic loads during their service period. Sometimes these members fail before their expected service life. Some of the major causes are "crack development" and "misalignments of the members". The "improper assembly" and "missing components" of the members also affect the structural integrity and may cause failures. This fosters the need for a meticulously designed quality control protocol before making the product available to the customers. Visual inspection is a prominent Non-Destructive Testing(NDT) method in the world. The Industry employs highly skilled and trained professionals for the visual examination. The credibility of this method is completely dependent on the skills of these professionals, which might sometimes be imprecise. To address the above indicated defects, this paper proposes an Automated system for the defect detection using Computer Vision. SIFT(Scale Invariant Feature Transform) and Centroid Matching techniques are employed to detect the cracks and misalignments whereas, Template Matching, a proficient algorithm in Image Processing, is used to detect the improper assembly and missing components. The techniques used produce satisfactorily good results. Keywords—SIFT, cracks, misalignment, improper assembly, missing components, NDT, Morphology, Centroid, Moments, template. Abbreviations— SIFT- Scale Invariant Feature Transforms, NDT- Non-Destructive Testing. I. INTRODUCTION Visual inspection is the most widely used NDT method in the world. Nearly everyone in the industry uses this method on a daily basis prior to performing the other NDT methods. Automation of the visual inspection process is necessary to enhance the quality control process on the assembly line. An automated visual examination can eliminate the probable use of the other conventional NDT methods, saving the production costs. Defects may be produced during the processing, fabrication and use during the service period of a material or product. They are introduced either early in the processing chain or later during its service period due to stress concentrations and effects of the external agents. The carryover of defects from earlier fabrication process may cause failures. Defects arising during service period may result from the presence of defects introduced during processing and fabrication, the inadequate specification of materials, or sometimes even from operation outside the intended design criteria. The failures that occur during fabrication process can be minimized by careful examination of the products before the dispatch. The industries have very sound quality control protocols at the assembly lines to make sure that the product meets the standards and specifications for the intended use. II. PREVIOUS WORK The paper in [1] deals with the analysis the presence of a fault based on the changes in modal frequency and flexibility coefficients that depend on the type of structure. The paper in [2] draws conclusions from the local stiffness matrix at the cracked section of a shaft as well as un-cracked shaft. The author of [3] has devised a unique technique for identifying defects which comprises of the digital processing and Fast Fourier Transform (FFT) from which transfer function data is obtained and decision is taken based on modal properties by a least square error estimator. The structural damage has been detected and identified in [4] by studying the changes in the random vibrational response of the damaged structures which are obtained by various signal processing techniques. Kong and Chen [5] proposed a methodology using wavelet transform, fuzzy logic and neural network technique to identify the faults in the mechanical members. Chandrashekhar and Ganguli [6] used Monte Carlo simulation to study the changes due to uncertainty in the geometric properties of the member. G. M. Atiqur Rahaman and Md. Mobarak Hossain in [7] have employed the image processing techniques for defect detection in ceramic tiles that are quite simple and efficient. 978-1-4799-8619-4/15/$31.00 ©2015 IEEE IEEE 2nd International Conference on Knowledge Collaboration in Engineering March 27- 28, 2015