Research On Optimal Sensor Placement For Structural Health Monitoring Based On Deep Reinforcement Learning

The scale of infrastructure construction in my country is the largest in the world.Environmental erosion,natural disasters,material aging,etc.will inevitably cause engineering structural damage and even catastrophic accidents.Structural health monitoring has become an important way to ensure structural safety.The optimal placement of sensors is a key step.This paper proposes an optimal sensor placement method based on deep reinforcement learning.Using the powerful feature extraction capabilities of deep neural networks and the optimization strategy of reinforcement learning,it effectively solves the non-convex,high-dimensional and discrete decision-making(0 or 1)sensor placement objective functions of the more complex structure sensor optimal placement objective function.The details are as follows:The sensor placement criteria and the sensor placement optimization algorithm are analyzed in detail.The sensor placement criteria based on structural modal reconstruction and uncertain structural parameter identification are respectively introduced and analyzed.On this basis,the sensor placement optimization algorithm is further studied.It reveals that these algorithms cannot automatically realize the discrete representation of the results and the accuracy of the optimization results needs to be improved.Using the optimization strategy of reinforcement learning and the powerful feature extraction capabilities of deep neural networks.Use deep neural networks instead of Q tables for reinforcement learning,designed a double deep neural network,and completed the calculation process of the deep reinforcement learning optimization algorithm.It is found that the traditional deep reinforcement learning optimization algorithm has an overestimation problem,Relevant research on the improvement of the algorithm is carried out to solve this problem.In view of the optimal configuration of sensors,the system designs relevant elements of deep reinforcement learning such as state space,action space and reward function.Finally,the algorithm is implemented through the python 3.6 platform.Based on the calculation examples of different structure types,this paper verifies and analyzes the sensor layout optimization algorithm based on deep reinforcement learning.Calculated the optimal layout of sensors for a shear structure,a frame structure,and a cross-sea bridge,compared with the results of traditional sensor placement optimization algorithms to verify the effectiveness of the deep reinforcement learning method.More importantly,the method automatically realizes the discrete representation result of 0 or 1,and provides clearer indication information for sensor placement decisions at different positions of the actual engineering structure.