Dynamic versus static artificial neural network model for masonry creep deformation

Abstract

One of the inherent modelling problems in structural engineering is creep of quasi-brittle materials such as concrete and masonry. The creep strain represents the non-instantaneous strain that occurs with time when stress is sustained. It has long been known that creep in brickwork results in deformations that increase gradually over time, and dependable and accurate prediction models for the long-term, time-dependent creep deformation of brickwork structures are, therefore, required. Several models, with limited accuracy, have been developed over recent decades to predict creep in concrete and masonry structures. The stochastic nature of creep deformation and its reliance on a large number of uncontrolled parameters (e.g. relative humidity, time of load application, stress level) make the process of prediction difficult and the development of accurate mathematical models almost impossible. Artificial neural networks have been introduced as an efficient modelling technique for applications incorporating a large number of variables and have proven successful in many cases, especially in problems for which the characteristics of the process are difficult to describe using mathematical models. This study introduces a creep prediction model, based on non-linear autoregression with exogenous inputs (Narx), which is able to detect and consider time dependency (which is the major factor in creep deformation of brickwork structures) within its architecture. The performance of the proposed Narx model was verified with experimental creep data from brickwork assemblages collected over the last 15 years. The results show that the accuracy of the Narx model outperforms existing artificial neural network models and is able to achieve a prediction error of less than 12%, strongly suggesting that the Narx model is more suitable for modelling a time-varying process such as creep prediction.