Research And Development Of Dynamic Monitoring System Based On The Fusion Of Near-field Dynamics And Machine Learning

In recent years,peridynamics theory become a new method of solid mechanics analysis.Compared with traditional mechanics,it has natural advantages in material failure and crack growth analysis.Therefore,this paper attempts to integrate peridynamics theory into dynamic monitoring system,so that it can give full play to its advantages in material structure health monitoring.In addition,in order to improve the real-time performance of the system and employ the peridynamics theory analysis more reasonably,the parallel processing of the peridynamics theory analysis is performed,and the machine learning methods are adopted to identify the abnormal state of the structure,all of these construct the dynamic monitoring system under the fusion of peridynamics theory and machine learning.In this paper,the framework of dynamic monitoring system based on the fusion of peridynamics theory and machine learning is proposed,which includes three modules:data collection,data storage and data analysis.The data collection module is composed of wireless sensor network for real-time data acquisition,data storage module is composed of database for historical data storage and data query,data analysis module is composed of peridynamics theory analysis and machine learning anomaly detection method,and data analysis module is also the main content of this paper.For peridynamics theory analysis,this paper uses the bond-based peridynamics theory of brittle materials,and proposes a damage function based on polynomial fitting which is more accurate for the damage estimation of the bond with small deformation and the results more consistent with the damage and failure of brittle materials.In addition,a thread level parallel scheme based on OpenMP is proposed for the peridynamics theory calculation process,parallelization of the material point cycle in the algorithm can effectively improve the calculation speed.In our experimental environment,the speedup of 20-25 can be obtained for general calculation tasks,these methods are also added to the existing peridynamics theory modeling and simulation software.Machine learning anomaly detection algorithm can quickly and accurately identify the abnormal state of the monitored object with the collected real-time data,which can make the peridynamics theory analysis more targeted.This paper analyzes the performance of various machine learning methods in two classification and multi classification anomaly detection experiments,and finds that the neural network algorithm can achieve better both in prediction performance and prediction speed.Furthermore,the transferable performance of neural network model is tested,we find that the fine-tune method of transfer learning can make the neural network achieve better prediction performance in a short retraining time.In this paper,the peridynamics theory analysis and machine learning anomaly detection are combined and integrated into the dynamic monitoring system.This not only makes the two methods give full play to their own advantages,but also makes the system have stronger analysis ability,which provides a new way for the application of peridynamics theory.