Fast and robust identification of railway track stiffness from simple field measurement

We propose to combine a physics-based finite element (FE) track model and data-driven Gaussian process regression (GPR) directly infer railpad ballast stiffness from measured frequency response functions (FRF) by field hammer tests. Conventionally, only the rail resonance full are used as FRF features identify stiffness. In this paper, eleven features, including sleeper resonances, single curve selected predictors of GPR. To deal with incomplete measurements uncertainties in we train multiple candidate GPR models different kernels training sets. Predictions fused using weighted Product Experts method that automatically filters out unreliable predictions. compare performance proposed updating particle swam optimization (PSO) on two synthesis datasets wide range scenarios. The results show enriched fusion strategy can effectively reduce prediction errors. worst-case scenario three 5% injected noise, average errors for approximately 12% 6%, outperforming PSO about 6% 3%, respectively. Moreover, enables fast predictions large datasets. 400 samples takes 10 s compared 40 min PSO. Finally, application example shows is capable extracting values simple setup, i.e., one accelerometer impact location.