New Operational Load Monitoring Approach Using Digital Image Correlation and Image Classification Networks

Abstract

The following paper presents a novel approach that can be applied to Operational Load Monitoring and Structural Health Monitoring processes. The approach is based on artificial intelligence (AI) and digital image correlation (DIC) techniques. DIC is an optical method that allows measuring full-field structural displacements and strains. In the presented approach only a relatively small fragment of the material’s surface is monitored by DIC. The obtained partial image of strains or displacements is then processed by a carefully trained AI model, an image classification network, able to predict the state of whole structure (e.g. materials stresses, potential loss of material continuity). The assumption is that all possible load cases and states of the monitored structure can be identified and simulated, so the data obtained from simulations can then be used to train the image classification network. A numerical example is presented as proof of the presented concept. A modern lightweight aerostructure in the form of a hat-stiffened composite panel was used as monitored structure in the presented example and its Operational Load Monitoring was performed based on a relatively small fragment of normal strains map. The reference maps to train the network were simulated numerically. The prediction model estimates the Tsai-Wu failure criterion value for the whole composite material. The obtained accuracy of predictions proved the effectiveness and efficiency of the proposed approach.

Multiscale Analysis of Composite Structures with Artificial Neural Network Support for Micromodel Stress Determination

“…The output information of the neurons is passed through the layers until the output is obtained. The network can be trained and learn complex patterns and relationships from provided reference data, by adjusting the weight and bias values of all its neurons [ 58 , 59 , 60 ].…”

“…The output information of the neurons is passed through the layers until the output is obtained. The network can be trained and learn complex patterns and relationships from provided reference data, by adjusting the weight and bias values of all its neurons [58][59][60]. ANNs mimic the principles of operation of animal brains and consist of interconnected nodes, or artificial neurons, organized into layers.…”