Feature Engineering of Time-Domain Signals Based on Principal Component Analysis for Rebar Corrosion Assessment Using Pulse Eddy Current

Corrosion of reinforcing steel bars in concrete structures is a leading cause of durability loss in civil infrastructure. This work introduces a low-cost pulsed eddy current (PEC) sensing system that employs a novel probe combining an excitation coil with a high-resolution anisotropic magnetoresistive (AMR) sensor. The probe achieves high sensitivity to corrosion at lift-off distances up to 55 mm while using only 0.2 A excitation current, making it suitable for reinforced concrete with significant cover depth.

A key contribution is the feature engineering of time-domain PEC signals. Instead of relying on raw signals, the study extracts area-under-the-curve (AUC) features from selected time segments and applies principal component analysis (PCA) to enhance discrimination between corroded and uncorroded regions. PCA not only improves detection at larger lift-offs but also enables quantitative estimation of rebar diameter loss, validated through accelerated corrosion tests and chamfered rebar samples.

The approach demonstrates that PCA-based feature engineering substantially boosts the robustness and resolution of PEC sensing, paving the way for portable, field-deployable NDT systems for real-time rebar corrosion monitoring.

Key Findings

• A pulsed eddy current (PEC) probe with an AMR sensor detects rebar corrosion at lift-offs up to 55 mm with low excitation current.
• Area-under-the-curve (AUC) features from PEC signals improve sensitivity compared to raw signals.
• Principal component analysis (PCA) enhances corrosion discrimination and robustness at larger lift-offs.
• PCA-based features enable quantitative estimation of rebar diameter loss, validated with accelerated corrosion and chamfered samples.

Recommended citation: Tamhane D., Patil J., Banerjee S., Tallur S. Feature Engineering of Time-Domain Signals Based on Principal Component Analysis for Rebar Corrosion Assessment Using Pulse Eddy Current. IEEE Sensors Journal. 2021;21(19):22086–22094. doi:10.1109/JSEN.2021.3103545.
Download Paper