An Automatic Analysis Approach of Bridge Life Based on Engineering Mathematics and LSTM

With the advancement of sensor networks, a variety of methods have been proposed in the field of bridge health monitoring to predict the lifespan of bridges. These methods can be divided into two categories: statistical methods and machine learning methods. statistical methods are static techniques so that fail to capture real-time changes in bridge health. machine learning methods are better to take into account the temporal variations but require high-quality datasets. In practical engineering applications, missing or excessive data often result in over-fitting or under-fitting issues. This study propose a new approach that modifies the parameter structure of the Long Short-Term Memory (LSTM) model. By changing the parameter structure from vectors to matrices, the author enables shared learning of both local and global bridge lifespans. To evaluate the effectiveness of this approach, a series of experiments were conducted on several bridges built with the ANSYS platform. This approach has several advantages over existing methods. (i) it provides a set of physical formulas to reduce data dimensionality, making it applicable to bridges with different sensor data; (ii) by using LSTM to consider the temporal characteristics of bridge health, our approach enhances the accuracy of lifespan prediction; (iii) the authors introduce a method that partitions bridges into multiple local regions for sensor data processing, enabling the acquisition of both local and global lifespan information. These features greatly improve the applicability of our approach in practical engineering scenarios.