23rd EANN / EAAAI 2022, 17 - 20 June 2022, Greece

Autoregressive Deep Learning Models for Bridge Strain Prediction

Anastasios Panagiotis Psathas, Lazaros Iliadis, Dimitra V. Achillopoulou, Antonios Papaleonidas, Nikoleta K. Stamataki, Dimitris Bountas, Ioannis M. Dokas


  Infrastructures are often subjected to harsh loading scenarios and severe environmental conditions, not anticipated during design, which result in their long-term structural deterioration. Thus, monitoring and maintaining the urban infrastructure is critical for the resilience. In this research, the authors present a modeling approach, for the prediction of future strain values in a Dutch Highway Bridge, based on eleven Deep Learning (DL) algorithms. Previous strain measurements were used as input. The performance of the developed Machine Learning model was evaluated using the Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and R square (R2) indices. The Fast Fourier Transform (FFT) was employed for the pre-processing of the involved time series. The obtained results are extremely promising for predicting the performance under design loads. All algorithms have proven their capacity to successfully predict the fluctuations of strain values. The authorities responsible for the function and management of the bridge, can feel confident to rely on these models in order to schedule in time proper maintenance works.  

*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.