Article: Deep learning-based recovery of cutting tool vibrations from spatiotemporally aliased video Journal: International Journal of Mechatronics and Manufacturing Systems (IJMMS) 2024 Vol.17 No.3 pp.276 - 294 Abstract: Visual vibrometry involves motion registration of vibrating objects from their video using image processing and computer vision methods. Its advantages include being full-field, contactless and not requiring sophisticated data acquisition devices. To leverage its use to record motion of cutting tools that vibrate at high frequencies and with small motion requires video to be recorded at high speeds and resolutions. However, since cameras trade speed for resolution, motion can become spatiotemporally aliased. This paper shows that it is possible to recover motion by upsampling the spatiotemporally aliased video using convolution neural networks (CNN)-based deep learning methods. Since the upsampled motion's spectra has at least as many peaks as the upsampled rate, we also present a simple method to deduce true modes from observed ones. Methods presented are generalised and can measure high frequency and small vibratory motion using even low-speed and resolution cameras. Inderscience Publishers - linking academia, business and industry through research

Title: Deep learning-based recovery of cutting tool vibrations from spatiotemporally aliased video

Authors: Harsh Singh Rajput; Varun Raizada; Mohit Law

Addresses: Department of Mechanical Engineering, Machine Tool Dynamics Laboratory, Indian Institute of Technology Kanpur, Kanpur, 208016, Uttar Pradesh, India ' Department of Mechanical Engineering, Machine Tool Dynamics Laboratory, Indian Institute of Technology Kanpur, Kanpur, 208016, Uttar Pradesh, India ' Department of Mechanical Engineering, Machine Tool Dynamics Laboratory, Indian Institute of Technology Kanpur, Kanpur, 208016, Uttar Pradesh, India

Abstract: Visual vibrometry involves motion registration of vibrating objects from their video using image processing and computer vision methods. Its advantages include being full-field, contactless and not requiring sophisticated data acquisition devices. To leverage its use to record motion of cutting tools that vibrate at high frequencies and with small motion requires video to be recorded at high speeds and resolutions. However, since cameras trade speed for resolution, motion can become spatiotemporally aliased. This paper shows that it is possible to recover motion by upsampling the spatiotemporally aliased video using convolution neural networks (CNN)-based deep learning methods. Since the upsampled motion's spectra has at least as many peaks as the upsampled rate, we also present a simple method to deduce true modes from observed ones. Methods presented are generalised and can measure high frequency and small vibratory motion using even low-speed and resolution cameras.

Keywords: deep learning; CNN; convolution neural networks; convolutional neural networks; aliasing; upsampling; visual vibrometry; cutting tools.

DOI: 10.1504/IJMMS.2024.143317

International Journal of Mechatronics and Manufacturing Systems, 2024 Vol.17 No.3, pp.276 - 294

Received: 23 Apr 2024
Accepted: 23 Jul 2024
Published online: 13 Dec 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Deep learning-based recovery of cutting tool vibrations from spatiotemporally aliased video

Keep up-to-date