Vibration spectrum analysis using FFT in the microcontroller

Abstract

The present article proposes the realization and implementation of an algorithm for the selective extraction of a spectral component using fast FFT analysis, from the mechanical vibrations recorded with the help of a broadband acoustic sensor. The hardware device consists of a high-resolution data acquisition system and the storage of samples of the useful signal in a memory area. The selective processing of vibrations is done with the help of a FFT algorithm implemented with the help of the ATMEGA microcontroller based on the binary calculation made by weighted amounts with coefficients from the frequency band used. After obtaining the band spectrum of the signal, a software mask will be applied, with a frequency window that will be pre-set by the human user, and only the amplitude of the signal with the desired spectral component will be displayed, thus achieving an intelligent vibration tracking produced by a certain mechanical element or selected functional assembly. The vibration spectrum of a mechanical machine is composed of the sum of the individual spectral components of all interdependent mechanisms. Simulation performed for this paper and results processed using the proposed method lead to the conclusion that a unique harmonic spectrum, obtained using the Fourier transform algorithm with fixed frequency window and limited number of spectral markers, correspond in a unique way to a mechanical state of wear that the microcontroller can store using a variable magnitude vector. Obtaining a buffer of spectral markers stored together with information about the speed and engine rpm allow the development of an original method of fault detection. The article proposes the implementation of the software method with a hard architecture that includes, low pass filter, information processing and display units.