Research On Curve Negotiation Safety Of Train Using Digital Twin

In recent years,the dynamic simulation model for train curve negotiation has evolved from a simple model to a more complex one that reflects actual operating conditions well,thus demonstrating a positive development.There are,however,several issues,such as an inability to achieve both accuracy and efficiency,the separation of the real data from the simulation data,and insufficient matching and fusion between the two data sets,which results in low timeliness.Digital twins,an important part of high-speed railway digitization,can not only mirror the physical world,but also receive physical information and explore the prophets of the physical world,reducing the losses caused by unpredictable and dangerous emergency situations.By applying digital twins to train curve passage safety analysis,we not only provide a new way to improve train curving performance but also explore a new approach to realize the digitalization of high-speed rail.Accordingly,based on fully summarizing related research at home and abroad,this thesis undertakes the following tasks:(1)Research on the train curve negotiation model for digital twins;(2)Research on model data fusion;(3)Research on fast solution methods for simulation models;(4)Research on real-time prediction algorithm;(5)Research on digital twin prototype system of train curve negotiation safety.During this research work,four key processes of digitization,interaction,prophetic perception,and foresight have been achieved.The main research work and main research results are summarized as follows.An application framework for the digital twin is developed based on the theoretical basis of the digital twin.Then,the physical model of the vehicle,the physical model of the curve line,and the physical model of the suspension devices such as the coupler and buffer device are constructed.By digitizing virtual models,we can produce train curve models that accurately describe the physical world.Additionally,the topography and curved lines of Chaoyang Station to Beipiao Station from the Beijing-Shengyang high-speed railway are reconstructed.For realtime control of the marshalling train’s position when the train negotiates a curve,an algorithm for automatically generating the curve line and a control algorithm for controlling its position is used.It is possible to create high-fidelity digital models of train curve negotiation to better describe the physical world,with the goal of fusing model data in the future.Based on the study of the train curve passing model,two data acquisition schemes are proposed,i.e.the monitoring and acquisition system and the measurement and control acquisition system,which are intended to monitor the running state during curve passing by monitoring speed,displacement,attitude,and other related data.Following this,a hybrid database including a REDIS non-relational database and a MYSQL relational database is designed for the collected twin data.Data transmission between twin data and twin models is achieved using three network protocols,i.e.UDP,HTTP,and WEBSOCKET,to ensure realtime and reliability.Last but not least,an extended Kalman filter method is used to integrate the optimization algorithm to provide a more accurate estimate of the running state.Using collected data,the acquired virtual model is updated in real-time at a fixed time interval to realize data fusion and reflect the running status of the train as it passes through the curve.An efficient solution method based on parallel physical information neural networks is proposed for multi-body dynamics problems with many degrees of freedom and strong nonlinearity.This method integrates the mechanical differential equation into the loss of the neural network to solve the differential equation and improves the training efficiency via parallel computing and the adaptive differential method.A comparison of the calculation results with traditional numerical methods reveals that this method is computationally stable while maintaining accuracy.Because of this,the influence of different curve radiuses and different running speeds is considered when analyzing the safety of vehicle curve passing.Moreover,the analysis of the train curve passing performance provides insight into the mechanical performance law of the train curve negotiation and then offers suggestions for enhancing the safety of the train curve negotiation.Real-time prediction of dynamic responses is required in the digital twin system,and two machine learning algorithms of MQRNN and OSELM,are proposed to accomplish this task.In the online learning process of dynamic data flows,the OSELM algorithm does not need to save and relearn previously learned sample data,and can quickly update itself according to the latest data collected;Using MQRNN,a multilayer perceptron-based encoder-decoder structure can be built,quickly enabling quantile regression predictions that are nonlinear and multi-level.By comparing the prediction results with the test results,it can be seen that both algorithms are capable of maintaining predictions.Additionally,OSELM has the benefit of fast learning speed and strong generalization ability,whereas MQRNN has the benefit of robustness and stability.Furthermore,it provides a reference for overcoming the limitations caused by an incomplete understanding of the physical world currently.Since these above researches,a digital twin prototype system for train curve passing safety is developed.This system allows not only the creation of twin models and twin scenes but also the motion simulation of the train within the scene,as well as the simulation of natural landscapes such as rain and snow.Moreover,it incorporates many functions such as parameter selection,real-time status monitoring,and performance prediction,so that it can be expanded in the future.Finally,by testing the twin system,we can verify that the key technologies in the system are feasible.