Acceleration-Guided Acoustic Signal Denoising Framework Based on Learnable Wavelet Transform Applied to Slab Track Condition Monitoring

Acoustic monitoring has recently shown great potential in the diagnosis of infrastructure condition. However, due to severe noise interference acoustic signals, meaningful features tend be difficult infer. It creates a considerable obstacle for an extensive application monitoring. To tackle this problem, we propose acceleration-guided signal denoising framework (AG-ASDF) based on learnable wavelet transform automatically denoise and extract relevant acceleration signal. This requires only training stage. Therefore, sensors (non-intrusive) need installed during phase, which is convenient crucial condition safety-critical infrastructure. A comparative study conducted among proposed AG-ASDF other feature learning / extraction methods, by using multi-class support vector machine evaluate detection effectiveness slab track signals. Different healthy unhealthy states tracks are imitated with three types supporting conditions railway test line. The classification outperforms methods significant accuracy improvement.