Long-term prediction of rheological effects in concrete structures using probabilistic neural networks

Concrete structures, such as long-span bridges, dams, nuclear containments, etc., display complex long-term behavior due to creep and shrinkage that must be accurately evaluated for safe service. Incorrect assessment of long-term rheological effects can reduce serviceability, adversely affect prestressing forces (if any), and require costly retrofitting measures. However, even under laboratory conditions, it is difficult to accurately predict long-term rheological effects, and only few decade-long experiments are available for thorough validation of proposed rheological models. Prediction in real-life structures is an even more challenging task due to the spatiotemporal variations in concrete properties, dependence on uncontrolled environment condition and load history, and complicated internal strain evolution in indeterminate structures. Numerous aging structures have been designed worldwide under codes that underestimate rheological effects, and, as a consequence, excessive deflections have been observed in these structures. The state-of-the-art methods employ FEM stochastic analysis and Bayesian approaches to reduce uncertainty by incorporating information stemming from laboratory specimen testing and scarce in-situ measurements throughout the structure's life. This requires the definition of a good numerical model, which entails specification of complex geometry and appropriate boundary conditions, and simplifying assumptions regarding environmental conditions. Relatively recently, data from structures that are equipped with embedded SHM systems since pouring of concrete are becoming available and they open doors for novel data for data-driven or hybrid methods for prediction of long-term rheological effects. In this work, a hybrid method employing a probabilistic neural network and reduced-order analytical model is proposed, and its performance assessed using data collected over multiple years from a pedestrian bridge equipped with strain and temperature fiber optic sensors.