Chair: Prof. Daeil Kwon (Ulsan National Institute of Science and Technology)
Crack Detection on Pressed Panel Products using Image Processing Techniques with Camera System Mr. Hoyeon MOON, Hweekwon JUNG, Changwon LEE and Gyuhae PARK (Chonnam National University)
The detection of cracks on panel products is a vital step for ensuring the quality of panel products. General crack detection technique has been performed by human inspectors who are good at detecting crack, which is spend many times and much money. Therefore, it is necessary to detect efficiently crack on panel products by machine vision system during the press forming process. In this study, we performed an automated crack detection using two image processing techniques with camera system. The first technique is evaluating the panel edge lines which are extracted from a percolated object panel image. This technique does not require a reference image for crack detection. Another technique is based on the comparison between a reference and a test image using the local amplitude mapping on the edge line image. The reference image of panel product is automatically acquired by camera system using image difference and time interval technique. As a result, cracks are efficiently detected using two crack detection techniques based on image processing. For demonstrating the proposed techniques, experiments were performed in the laboratory and the actual manufacturing lines. Experimental results show that the proposed techniques could effectively improve the crack detection rate with improved speed.
A PHM Testbed for Prognostics of the Machine Tools Mr. Kyusung JUNG, Hyungjun PARK, Seokgoo KIM, Dawn AN and Joo-Ho CHOI (Korea Aerospace University, Korea Institute of Industrial Technology)
In manufacturing, a machine tool needs to maintain in good condition to prevent degradation in accuracy and disruption in production. All machine tools degrade as it operates, but it is difficult to discern degradation before visual identification, especially detecting degradation in real time. As failure of main elements in the machine result in the significant loss in cost and time, manufacturers need automated and efficient method to diagnose and predict the condition of its elements while in its operation. This paper addresses the Prognostics and Health Management (PHM) architecture for the machine tool, in which the functional model is created from the target system, critical failure modes are identified, sensor units are design to measure failure cause, symptom and the effect on the quality. The proper sensors, features and PHM algorithms are suggested for each of the failure modes as well. To demonstrate and validate this approach, a testbed is designed and operated for machine tools equipment that can implement PHM technique by detecting the faults of critical components, monitoring their degradation and predicting the remaining useful life. After the implementation, cost benefit analysis of the PHM application is conducted. Final goal is to apply and validate the system in the real field.
Mechanical Property Estimation for FDM 3D Printed Parts using Gaussian Process Regression Mr. Heechang KIM, Seungtae PARK, Eunju PARK, Seungchul LEE and Namhun KIM (UNIST)
Recently, as the application of additive manufacturing products has increased, new demands for the manufacturing industry are increasing. FDM is one of the most popular methods of the additive manufacturing. The products which is printed by the FDM method can be used in various fields, but the mechanical property of the products is considered to be weaker than the conventional cutting or casting processing. Therefore, improvement of the mechanical properties of the FDM printing product is a factor that can contribute to the actual manufacturing industry. In this study, we propose an analysis on the improvement of mechanical properties of output products due to various experimental variables. The experimental variables include orientation, infill rate, and material. Also, we propose a statistical method to estimate the maximum tensile strength of the product which changes according to the variables before the printing of the product, and experimentally verify it in order to optimize product manufacturing, manufacturing time and material waste saving in the actual industrial field. We use the Gaussian process to estimate nonstationary mechanical properties with respect to infill rate which pioneers the estimation approach in specimen-related experiments. By using GP, we come up with estimation values with its uncertainty which guarantees some extent of confidence interval for our estimation values.
Detection of localized faults in bearings using 2D envelope signal analysis Mr. Sheraz Ali KHAN, Jaeyoung KIM and Jongmyon KIM (Univesity of Ulsan, University of Ulsan)
Bearing faults are the leading cause of failure in induction motors and result in the longest downtime per failure in wind turbines. The accurate and timely detection of these faults is essential in avoiding unexpected shutdowns and the consequent economic losses. State of the art bearing fault detection is mostly based on envelope analysis. It generally involves band pass filtering the raw signal, demodulating the filtered signal to construct the envelope and then the Fourier analysis of the envelope signal to determine the presence of peaks at characteristic defect frequencies and their sidebands. Recent work on envelope analysis is mostly concerned with improving the visibility of these defect frequencies in the envelope power spectrum by selecting an optimal band for filtering. This is done using techniques like sub-band analysis and spectral kurtosis, thereby making the process tedious and computationally complex. This paper presents a 2D imaging based approach to envelope signal analysis, which is simple yet highly effective. The envelope signal, when projected onto a two dimensional grayscale intensity space, shows a unique texture for each fault. This study shows that these unique textures can be used for automated detection of localized defects in bearings.