Search

• Non-destructive testing with impact acoustics and non-defective machine learning is useful to evaluate supporting conditions of track slabs. • PCA and autoencoder is useful for non-defective machine learning to evaluate supporting conditions of track slabs. • It is desirable to measure near the hammering point or from not more than around 3 times the thickness of a track slab. Slab tracks are main ballastless tracks used in the high-speed railway system in Japan (Shinkansen). The track slab is a reinforced or pre-stressed concrete component in the slab track. This component is important to support traffic loading stably. We need to maintain the supporting condition of track slabs sustainably in the future because it has been passed nearly 50 years since the earliest of slab tracks have laid. Therefore, we need non-destructive testing methods to evaluating this condition of track slabs. We have been focused on non-destructive testing with impact acoustics. In this study, we improve this non-destructive testing by experiments in the full-scale models, eigenvalue analysis and non-defective machine learning. We have obtained the conclusion in this study as below: (1) Characteristics of impact acoustics depend on hammering position. For example, In the case of hammering at near the corner, impact acoustics is larger and damping more slowly than the case of hammering the center of long side. (2) It is desirable to measure near the hammering point or distance from not more than around 3 times the thickness of a track slab. (3) Principal component analysis and Autoencoder are useful for evaluating supporting conditions of track slabs. Both methods can discriminate the supported area and the voided one with accuracy rates of 81.92 % and 79.51 %, respectively. [ABSTRACT FROM AUTHOR]