Non-Contact Diagnostics of Oscillatory Systems Using Computer Vision Methods

This paper proposes an innovative non-contact approach to the determination of vibration parameters of technical objects — frequency, amplitude, and phase — based on computer vision methods. The relevance of this work strongly stems from the fact that conventional vibration diagnostics using contact accelerometers face significant limitations in high-temperature environments, on rapidly rotating elements, and in chemically aggressive atmospheres. In such conditions, mechanical fixation of sensors is impossible or compromises measurement reliability. As an alternative, the authors propose a method based on dense optical flow estimation between high-speed video frames combined with Fast Fourier Transform (FFT) spectral analysis. An algorithm has been developed and implemented using the OpenCV library, employing the Farneback method with optimisation of its key parameters: analysis window size, approximating polynomial degree, number of Gaussian pyramid levels, and smoothing parameter. Experimental validation was performed on a test rig equipped with an electromagnetic shaker generating a frequency of 142.3948 Hz and a peak displacement amplitude of 3.0 3.2 mm; video registration was carried out at 2000 fps. Four typical parameter configurations were investigated, each independently repeated ten times. High stability of frequency estimation has been explicitly demonstrated: absolute error did not exceed 0.004 Hz, relative error did not exceed 0.0025 % the amplitude coefficient of variation lay in the target range of 2.9 5.0 %. The obtained results are fully comparable with those of professional accelerometric systems in accuracy and substantially superior in deployment simplicity, absence of mechanical fixation requirements, and overall cost. The method is perfectly applicable to equipment diagnostics in inaccessible and hazardous environments.