Condition Recognition Of Pantograph-catenary Contact Region Based On Image Processing And Deep Learning

With the rapid development of China high speed railway and higher requirements for operational quality and security,monitoring train operation status is becoming increasingly important for ensuring rail operation safety.Pantograph and catenary(PAC)are collecting systems of the electric current and ensuring their reliability has become one of the most important tasks.The detection of contact point and measurement of contact force are key points to evaluate the operating status of the system.Thus,this thesis proposes a novel fast contact point detection method based on multiple strategies and then studies on the identification of PAC key parameters.It mainly includes the following contents:a)Based on the PAC image acquisition system,the monitoring data of various test scenarios are collected.A set of algorithm flow of cross feature enhancement and contact point detection for PAC image is established based on traditional image processing.b)A new rail dataset PAC-TPL2020 for contact point coordinate detection is established,which contains interference information of various complex situations,mainly including arcing interference,terrible weather,abnormal light source,catenary structure and other interference.Its image richness and quantity basically meet the requirements of training and verification of PAC contact point identification program.c)In face of complex environment and various interference in the detection of PAC contact points,a real-time tracking-detection-optimization multiple strategies-based contact point detection method is proposed based on KCF and deep learning.A new contact point regression residual network(CPRR-Net)is proposed to return the contact point coordinates of the contact area.The method of optimizing the detection results by combining the motion equation of contact point and Kalman filter is proposed to realize the stable and accurate detection of the contact point position.This work additionally employed the dataset PACTPL2020 to prove the robustness and high accuracy of our model in real-world scenarios.d)The method of non-contact PAC contact force measurement studied based on the image processing,which helps for engineering application.In order to obtain accurate acceleration data from displacement data and solve the key problem of calculating pantograph head inertia force,a data processing method combining multi-step difference quotient strategy and low-pass filtering is proposed.Finally,the pantograph head balance equation is established to calculate the PAC contact force.Finally,the feasibility and accuracy of the noncontact measurement method are verified by bench test and simulation data.