Application of artificial neural networks for stress state analysis based on the photoelastic method

The present article proposes an evolutionary development of the photoelasticity method for measuring stresses based on annular photoelastic sensors application along with stress pattern recording with the aid of a digital camera and its recognition using artificial neural networks. The analysis of the modern application of the photo-elasticity method for various problems within the theory of strength is presented. The principle of operation of photoelastic sensors based on the photoelasticity effect is considered. Optical patterns in an annular photoelastic sensor are presented for various values of the horizontal stress. The calculation of the stress state of the sensor for the following full-scale experiment has been performed, the estimate of the threshold conditions under which the sensor can be applied has been performed. As a result of a laboratory experiment, a dataset of 1500 isochromatic images has been assembled. A subspecies of a neural network, namely a convolutional neural network, has been applied as a machine learning algorithm. Different combination of models and optimizers have been employed. The application of downhole sensors for continuous monitoring of alterations in the rock mass stress state and the integration of this data into a digital field model based on Internet of Things technologies has been proposed.