Multiparameter estimation and LSTM-based prediction method for health state of single-layer reticulated shells

Abstract

Single-layer reticulated shells (SLRSs) are typically used in large public buildings. Accurately assessing and monitoring the health status of SLRSs can detect potential structural issues and facilitate the adoption of early preventive measures to avert accidents. However, it is difficult to determine the state parameters that reflect the health status of SLRS. At the same time, there are numerous potential performance degradation patterns of SLRS. Therefore, evaluating and predicting the SLRS health status is challenging. Under the condition of damage changes uniformly and continuously, this study proposes a multiparameter estimation and long short-term memory (LSTM)-based prediction method for the SLRS health state. First, three state parameters related to frequency, vibration mode, and energy are constructed, which can be obtained through structural response. Thereafter, a state estimation vector H is proposed based on the above parameters. The changing trend of H reflects the changing of SLRS health state and can provide maintenance suggestions. Second, different state estimation vectors under different damage patterns are constructed to train the LSTM neural network for predicting the future health state of the structure and providing predictive maintenance suggestions. Finally, three numerical examples based on a spherical SLRS are presented. For the numerical example of a spherical SLRS, there are two types of inflection points in the spherical SLRS: the first inflection point corresponds to a slight amount of damage to the structure, and the second inflection point corresponds to a decrease in the overall bearing capacity of the structure. The trained LSTM network can predict H accurately in three different damage patterns. Predictive maintenance suggestions can be made based on H and the prediction results of the trained LSTM network: the spherical SLRS should be repaired when the first inflection point appears and before the second inflection point appears.