Detecção de falhas em vigas em balanço através da análise de sinais de vibração e estatísticas de ordem superior

Abstract

Cantilever Beams are great allies in the development of new technologies for Predictive Maintenance, given that they efficiently simulate common structures present in the daily life of engineering. The construction/composition of cantilever beams resembles many structures present in the daily life of industrial equipment and construction and, therefore, are widely employed in the development and testing of new analysis methods for detecting flaws in structures. Motivated by the good performance of beams in aiding the development of these methods, this paper proposes a new technique for the detection of flaws in cantilever beams by means of vibration signal analysis and Higher Order Statistics (HOS). The innovation of the work is in the use of second, third and fourth order cumulants, extracted from the vibration signals of the cantilever beam in order to detect flaws, if any, as well as to identify the level of the flaws. For this, we developed an experimental bench to collect vibration signals, which were subsequently processed via MatLab® software. The most representative HOS, in terms of flaw detection/classification capability, were selected using the Fisher's Discriminant Ratio and, later, presented to two classifiers proposed for different approaches. We used the unsupervised SVM (support vector machines) classifier for flaw detection in a scenario in which flaws are unknown. For the approach in which flaws are known, we used a multilayer perceptron (MLP), which detects and identifies the level of the flaws. The results showed that the HOS are capable of forming compact vector signatures (with only two or three parameters) of the types of flaws and of the vibration signals from the cantilever beam without flaw, leading to simple architecture classifiers that reach performances of 100% for small processed signal windows.