Application of Recurrent Neural Network to Prediction of Structural Deterioration

Abstract

Degradation and obsolescence of structures shows limitations to preemptive forecasting and corresponding response because of various uncertainties. In order to overcome this challenge, this study proposes a methodology for combining Structural Health Monitoring (SHM) data obtained during operation, which is one of the deep learning algorithms, and deterioration progress models based on mechanistic knowledge by using Recurrent Neural Network (RNN). Recurrent neural network is one kind of deep learning methodology used to learn the input data accepted in time order, and it is currently actively applied the areas of language awareness and modelling. Accurate prediction and evaluation of the aging of the structure can be possible by recursively updating the monitoring data through the recurrent neural network. Accurate prediction and evaluation of a structure's obsolescence can be achieved by updating it recursively with monitoring data via a recurrent neural network. It can be confirmed by comparing with the results of the existing aging prediction model and additionally the accuracy of the recurrent neural network model is determined by checking the results regarding to the number of base training data sets. It also verifies its performance and enables better forecasting by proposing new algorithms that can be applied to the obsolescence of structures as well as applying existing recurrent neural network.