Gradient Descent Based Restoration Method of Track Irregularity in Asymmetric Chord Reference Method

This paper proposes a gradient descent based restoration method of track irregularity. Based on the theory of asymmetric chord-reference method (CRM), the restoration of track irregularity is described as an optimization problem for an underdetermined linear system. Gradient descent method is employed to solve this optimization problem, where a quadratic cost function considering penalization is used. To evaluate the performance of the proposed method, an inspection trolley was setup and used in a field test on a scaled bridge model. Comparison between the proposed method and level measurement validates a good accuracy of gradient descent based restoration method. Compared with traditional method which needs a specially designed inverse filter, the proposed method has a clear physical meaning, which only needs configuration of asymmetric CRM and measured chord reference value to establish the optimization model. This suggests that gradient descent method has good operability in the field test. And the repeatability assessment reveals that the proposed method has a good track irregularity restoration reproduction capacity.