Deep Learning-Based Acoustic Emission Scheme for Nondestructive Localization of Cracks in Train Rails under a Load

Abstract

This research proposes a nondestructive single-sensor acoustic emission (AE) scheme for the detection and localization of cracks in steel rail under loads. In the operation, AE signals were captured by the AE sensor and converted into digital signal data by AE data acquisition module. The digital data were denoised to remove ambient and wheel/rail contact noises, and the denoised data were processed and classified to localize cracks in the steel rail using a deep learning algorithmic model. The AE signals of pencil lead break at the head, web, and foot of steel rail were used to train and test the algorithmic model. In training and testing the algorithm, the AE signals were divided into two groupings (150 and 300 AE signals) and the classification accuracy compared. The deep learning-based AE scheme was also implemented onsite to detect cracks in the steel rail. The total accuracy (average F1 score) under the first and second groupings were 86.6% and 96.6%, and that of the onsite experiment was 77.33%. The novelty of this research lies in the use of a single AE sensor and AE signal-based deep learning algorithm to efficiently detect and localize cracks in the steel rail, unlike existing AE crack-localization technology that relies on two or more sensors and human interpretation.

Keywords: acoustic emission sensor; acoustic emission testing; deep learning; nondestructive testing (NDT).

Figure 1

The diagram of the nondestructive single-sensor acoustic emission (AE) scheme for crack detection in steel rail.

Figure 2

The AE signal of cracks in steel rail: (a) AE signal, (b) denoised AE signal using total variation denoising (TVD).

Figure 3

Majorization-minimization (MM) procedure: (a) cost function F(x) to be minimized and initial x_o; (b) iteration 1 where majorizer G₀(x) is tangent to F(x) at x₀ and minimize G₀(x) for x₁; (c) iteration 1 where majorizer G₁(x) is tangent to F(x) at x₁ and minimize G₁(x) for x₂.

Figure 4

The typical deep learning algorithm for classification.

Figure 5

AE data sensor and acquisition: (a) AE data acquisition module; (b) AE sensor, MAG4R magnetic holder, and Hsu–Nielsen source.

Figure 6

Peak amplitude of AE sensor relative to time.

Figure 7

Locations of pencil lead break (PLB): (a) rail head, (b) rail web, (c) rail foot.

Figure 8

Un-denoised AE signals of PLB at: (a) rail head, (b) rail web, (c) rail foot.

Figure 9

Conversion of denoised AE signals (after pre-processing) of PLB into dataset for training and testing the deep learning algorithmic model.

Figure 10

Feature and target datasets for training and testing the deep learning algorithmic model.

Figure 11

The proposed deep learning algorithm for classification.

Figure 12

Training and testing the proposed deep learning algorithm and the model validation.

Figure 13

Onsite experimental setup: (a) advanced phased array ultrasonic testing (PAUT) on welding joint of the steel rail; (b) PAUT detection with defect in the welding joint.

Figure 14

Steel rail temperature measurement.

Figure 15

Onsite experiment on steel rail using the AE scheme: (a) installation of the AE sensor on the steel rail; (b) AE signal data of the steel rail under a load.

Figure 16

Receiver operating characteristic (ROC) curve where classes 0, 1, and 2 denote rail head, rail web, and rail foot: (a) 150 AE signals (first grouping); (b) 300 AE signals (second grouping).

Figure 17

Confusion matrix of the deep learning algorithm for classification: (a) 150 AE signals (first grouping); (b) 300 AE signals (second grouping).

Figure 18

AE signals of cracks in the welding joint of the steel rail under a load using the single-sensor AE scheme.