Experimental Modal Analysis Involving the Use of Artificial Intelligence to Accurately Evaluate the Natural Frequencies of Engineering Structures

There are several well-known methods that can be used to improve the estimate of the frequency for short signals such as the response of a structure to an impulsive excitation. However, the accuracy of the estimates is insufficient, thus more advanced methods are requested. We propose a machine learning approach to estimate the frequency values between the spectral lines of the Discrete Fourier Transform (DFT). The method involves Artificial intelligence and uses the data obtained by applying the DFT to a sinusoidal signal with known frequency and amplitude and varying lengths in time. The peak amplitudes in the spectrum have a similar shape to that of the sinusoidal signal generated. To determine this amplitude, we preprocess the signal by zero padding. Then we find the frequency and amplitude for three points on the main lobe of the DFT, namely the maximizer and its two neighbors. These points, derived for each signal length, represent the INPUT data used to train an ANN. The OUTPUT data consists of the frequency of the generated signal. Tests performed with generated signals with a different frequency and with data achieved from measurements confirm the validity of the method.