Vibration-Based Railway Suspension Monitoring Under Varying Operating Conditions via Robust Statistical Time Series Type Methods

The on-board condition monitoring of railway suspension components is important for ensuring passenger comfort, maintaining proper operation, enhancing safety, and reducing operational costs. In this context statistical time series type random-vibration-based methods constitute a potentially efficient and cost-effective framework. Yet a significant challenge emerges from the competing requirements for high sensitivity to suspension component degradation and robustness to varying Operating Conditions (OCs) and uncertainty. Varying Operating Conditions and uncertainty are related to factors such as varying payload, travelling speed, and rail profile. The problem is further aggravated by the complexity of the vehicle dynamics, including the wheelbase filtering effect. Despite the significant progress made so far and the development of various methods, the challenges related to varying OCs and uncertainty largely remain as a technical barrier. The aim of the present study is the introduction of a data-based methodology aspiring to overcome them. The methodology is based on Multiple-Input Single-Output Transmittance Function (MISO-TF) time series models of a special form allowing for eliminating the effects of varying rail profile while properly treating the wheelbase filtering effect. Monitoring is based on a MISO-TF feature space, using either a Random Coefficient Gaussian Mixture or a Multiple Model based scheme. The former, although more elaborate, allows for the introduction of a complete probabilistic framework. An interesting advantage of both versions is that operation can be based on a very limited number of sensors and only a partial data-driven model of the dynamics. The methodology is comprehensively assessed via hundreds of SIMPACK based Monte Carlo runs under variable payload and rail profiles. The achievable performance is shown to be very good for both versions and for 10%, 15%, and 20% of condition degradation in either primary or secondary suspension components.