Deterioration prediction of existing concrete bridges using a LSTM recurrent neural network

Bridge censored databases can be used to analyze and assess structural deterioration conditions, but conducting the analysis is difficult. This difficulty occurs because many factors affect deterioration, time span of data for these depends on years in service respective bridge. In addition, values some are not regularly observed. The present study uses long short-term memory (LSTM) consider twelve potentially influencing recognize relationships between grades. Testing model an inspection database 3,368 bridges indicates that LSTM obtained accuracy exceeding 80%, i.e., outperforms performance a multilayer perceptron established using same database. For four types bridges, shows equivalent performance. predictive ability coastal slightly superior non-coastal bridges. No significant differences determined different deck areas. Practically, predict bridge paths, could help decision-makers formulate intervention strategies improving quality maintenance management.