Intelligent Detection And Diagnosis Of Defects With Acoustic Emission And Wheel-rail Transmitted Feature

High-speed railways are the backbones of the economic prosperties and lifeline of national transportation,which can also be viewed as a brand of advanced manufactured products.The development and maintenance of Chinese railway is beneficial to the national strength and people’s livelihood.However,with an increasing transporting volume and workloads,importance about maintaining the reliability of rail tracks has been brougt into attentions of the public.Among the emerging NDTs,acoustic emission(AE)technique is broadly applied in the engineering structural detections,owing to its dynamic property,high sensitivity an d capability to detect the internal cracks at the early stages.However,the stationary fixed-point monitoring scheme of the AE technique is not applicable in rail health monitoring considering that the coverage requirement for rail length is so enormous in practice.Based on the earlier investigation,a wheel-mounted crack-induced AE measurement scheme is further developed in this thesis,which significantly reduces the investment cost of AE in railway,promotes the efficiency of such an AE detection system and achieves a more effective surveillance performance in the rail maintenance.To discuss its feasibility,the thesis is orgnized according to the following aspects in this study:To quantitatively describe sound transmission across the wheel-rail contact interface,a theoretical model about the contact acoustic nonlinearity charateristic is first investigated.The analytical model of acoustic wave transmission across the contact interface in this thesis is based on the contact acoustic nonlinearity theo ry,integrating the fractal theory and Rayleigh integration to describe the source-dependent,media-dependent and contact-condition-dependent relationship.The accuracy of the model in estimating the acoustic transmission intensity was validated with static test data from the test rig of wheel-rail contact simulation.A data-driven nonlinear autoregressive model was proposed to fit and model the rolling-generated chaotic noise signal,which was used in the denoising task subsequently.An architecture which is composed of the long short-term memory and generative adversarial network is employed in the training of the chaotic AE noise.Key hyper-parameters of it are optimized by the chaos theory.The denoising results showed that the marjority in the noise are the same kind of predictable AE noise,which can be accurately predicted and hence eliminated by the proposed autoregressive model.Then the extraction of the crack components and the suppression of abnormal residual noise are studied on the strength of independent component analysis,constrained by the higher-order statistics.The improved algorithm can further suppress the unpredictable burst-type abnormal noise.Comparing with the pure autoregressive model,the application of independent component analysis helps to reduce the false-alarms caused by the abnormal noise interferences to a great extent.Finally,the robust unsupervised acoustic emission clustering was conducted for diagnosis of status in crack events.A generlized silhouette index was initia lly advanced for internal cluster validation,and then it was applied in the improved and self-optimized density-based clustering,together with stochastic neighbor embedding,alleviating its sensitivity to the key parameters and the curse of dimensionality.The proposed method can discover the intrinsic patterns and natural clusters of data in the deterioration of rail specimen in tensile tests,which is also robust under the complex geometry and outlier interference.To summarize,the thesis provides a solid theoretical basis and feasibility analysis for the wheel-mounted AE crack detection scheme,which can be a guidance to its implementation,and also removes the major barriers of the scheme.