An optimization method for failure analysis in beam like structures

Abstract

Abstract Industrial revolution plays a major role in the economy of a nation. To have an uninterrupted support from any of the industries, a regular structural health monitoring is very much essential. Again, the online monitoring is preferred over any of the conventional approach. In the recent past many researchers work on the failure analysis of online structural health monitoring. Though visual inspection can’t give better results in many aspects, so it is better to work on a method which can give better results with better accuracy and should be with less time consuming. In this present work we focus on inverse problems for the determination of crack parameters in the form of its location and severity. Basically, the present work is limited to, the failure analysis of a beam like structure. We have considered a cantilever beam for the analysis. The research work has been conducted in two phases. In the first phase the theoretical investigation of the structure has been performed through a numerical method called finite element analysis (FEA), which is comprised of different matrices such as mass matrix and the element stiffness matrix based on the material properties as well as loading conditions. The analysis has been performed with and without the presence of the crack parameters, which leads to give a result of different natural frequencies at different modes of free vibration. The results of natural frequency create a platform to identify unhealthy structures. In the second phase, the main objective of the findings in which the theoretical results have been trained to get the crack parameters as crack depth and crack location. Out of the several inverse techniques we choose the Extreme Learning Machine (ELM) as the best method for this proposed work. ELM is a non-iterative single layer feed forward neural network (NN) system in which the weights and biases (parameters of the hubs) are set haphazardly and require not to be tuned amid the fair treatment. The weights (output) of the network are determined analytically through simple generalized Moore Penrose pseudo inverse operation. It is observed that the ELM gives a better result with better accuracy with less computation time. The robustness of the proposed work has been verified through a comparison study