Development of visual odometry based machinery motion assessment system / Low Shee Teng

Low, Shee Teng (2022) Development of visual odometry based machinery motion assessment system / Low Shee Teng. Masters thesis, Universiti Malaya.

	PDF (The Candidate’s Agreement) Restricted to Repository staff only Download (638Kb)
Preview	PDF (Thesis M.A) Download (1774Kb) | Preview

Abstract

Monitoring the vibrations of a machine's mechanical components is critical to its proper operation as for performing preventive maintenance. Recently, a sizable number of the study approaches in vibration analysis are based on non-contacting vibration measuring equipment that offering various advantages than the conventional sensors. New methods for gathering information about the vibration of the machine have evolved simultaneously with the constant improvement of the visual odometry (VO) systems due to the rapid development of computer vision (CV) field. Digital image analysis, video analysis, and other visual inputs are all examples of CV, which is a branch of artificial intelligence (AI) that empowers computers or systems to obtain significant information from digital images, videos, as well as other visual inputs and to take actions or make recommendations based on this information. Research laboratories to actual industrial installations were made possible because of their actual effectiveness. The use of visualization tools can often be a useful addition to vibration analysis or even a complete replacement for more traditional approaches. The non-contacting attributes and the ability to simultaneously observe several spots in the defined region are the most important factors. Motion magnification (MM), an image processing technique that provides the visual observation of vibration processes that are not visible in their native state, is an image processing technology. Four types of methodologies involving optical flow (OF), motion amplification and MM have been implemented and linear based Eulerian Video Magnification (EVM) have been implemented as a benchmark. Method 1 include the calculation of OF follow by motion amplification on the video. Method 2 would be the same as Method 1 but including the insertion of the cut-off frequency. Method 3 would be combining Method 2 with linear based EVM, and Method 4 would be purely linear based EVM. These algorithms are implemented in terms of their computational complexity and visual quality as well as how they provide the amplified motion of video output. Machine diagnostics can be improved by using visual methods that magnify motion. Motion amplification aids in the visualization of complex vibration problems that are otherwise inaccessible to the human eye. When used in conjunction with other tools, this instrument can save time and money in the areas of routine condition monitoring programs, troubleshooting, vibration analysis and root cause analysis. In this research, the output of the video amplifying and magnifying algorithm have been compared. According to the findings, EVM is the most appropriate VO for a machinery motion assessment system because it has performed the best magnification work in this project. The EVM technique has the best magnification when comparing the data acquired from these approaches; nonetheless, the EVM method exhibits superior noise characteristics than Method 3. Method 2 outperforms Method 1 in terms of edge distortions, but the results are foggy at the end of the system because of the blurring that occurs at the end of the system.

Actions (For repository staff only : Login required)