Reinforcement Learning-based Algorithm For Solving Structural Damage Inversion Equation

While the large-scale infrastructure structure promotes social and economic development and improves people’s living standards,the safety and integrity of its structure are constantly threatened by various factors.Once the structure fails,it will often cause catastrophes to human society and the natural environment.Therefore,it is of great significance to be able to identify the location and degree of early damage of structures in time.With the development of the artificial intelligence and the big data technology,structural damage identification is gradually moving towards automation and intelligence.In this thesis,the reinforcement learning-based algorithm for structural damage identification is studied.The main contents are as follows:First,a reinforcement learning-based algorithm for solving the sensitivity-based structural damage identification inversion equation is proposed.Based on the structural sensitivity model,the target equations of the structural damage identification are proposed.According to the physical properties of the structural damage,the appropriate activation function is selected for the neural network in the reinforcement learning algorithm.Combing with the sparse nature of the damage,a well-oriented profit function and the interaction process between the agent and the environment are designed for the algorithm.A plane truss simulation model are employed to illustrate the ability of the proposed method.Compared with the damage identification results of reinforcement learning algorithm and basis pursuit noise reduction algorithm,the effectiveness and noise resistance of the proposed reinforcement learning algorithm in structural damage location and degree quantification are verified.Secondly,considering the uncertainties in actual structures,a reinforcement learning algorithm for structural damage identification based on Bayesian probability model is proposed.By solving the damage parameters in the inversion problem by reinforcement learning algorithm,the identification results of damage conditions simulated by the numerical model illustrates the accuracy and robustness to noise of the proposed algorithm.Finally,using the Benchmark model of the actual bridge,the simulated damage cases and the actual damage cases of the Benchmark model are identified respectively.The damage identification results show that the results of RL algorithm can accurately locate and quantify the real damage,which has good potential in actual engineering applications.