Long short-term memory model for quantitative analysis of corrosion products in concrete

A quantitative analysis model is proposed to describe the content evolution of corrosion products in concrete structures subjected to sulfate or chloride attack. The model is inspired by the idea of the long short-term memory (LSTM) algorithm in the machine learning field. To establish a relationship between engineering data and theoretical computations, training samples of the model were generated by solving the rate equations of sulfate corrosion reactions in concrete with wide-range initial conditions and validated by the existing experimental data. The Pearson correlation coefficient was used to determine the model features. The model performance was comprehensively evaluated and the results show that the proposed model has sufficient accuracy and feasibility. It could effectively predict the contents of corrosion products during sulfate attack using several input values rather than the initial contents of all chemical constituents. The proposed model builds a bridge between experimental methods and theoretical predictions, adequately inheriting the advantages of each.