Real time frequency domain decomposition for structural health monitoring

Abstract

Frequency Domain Decomposition (FDD) is a powerful technique to identify the modal properties of a structure. FDD is accurate but computationally expensive and up to now it could not be performed in real time without a super computer. Real time is intended as the same frequency of the sampling rate of the input signals. In this study, we present a new method that performs real time FDD taking advantages of General Purpose Graphic Processor Unit (GPGPU). A GPU is equipped with hundreds of arithmetic logic unit and it is capable of massive parallel computing. The algorithm runs on a normal desktop computer equipped with a GPU card, which is rather inexpensive compared to super computers. This new method has many advantages. First, GPU consume less power than CPU because it runs at lower frequency. Second, on a real structure the output signals of the structural health monitoring system can be fed directly to the algorithm. Therefore, only the identified modal properties have to be stored on-site or transmitted off-site though saving storage space and transmission bandwidth. Third, because modal properties are monitored in real time damage detection algorithms based on them can also be run in real time.