Machine Learning-Enhanced Acoustic Emission Technique for Impact Source Identification and Differentiation in Steel Pipes - Presented at the 12th Annual Conference (DIVK12)

Q: Why is acoustic emission monitoring needed for steel pipes?

External impacts of varying magnitudes can generate defects, starting as micro-cracks that may progress toward pipeline failure. SHM is required to support sustainable reliability and integrity.

Q: How do the pencil lead break (PLB) and drop-ball experiments contribute to the study?

The PLB experiment calibrates the AE setup and provides baseline signals across source distances for open-ended and damped pipes. The drop-ball experiment generates distinct burst-type AE signatures, allowing feature extraction for ML training.

Q: Which acoustic emission features are used to train the supervised machine learning models?

The extracted time-series features include AE energy, peak amplitude, and rise time from both experiment signals.

Aug 4, 2026

Upload by Aria

Machine learning-enhanced acoustic emission (AE) testing is developed to identify and differentiate external impact sources in carbon steel pipes used for oil and gas transportation. The work targets a key SHM need, as impacts can initiate micro-cracks and progress toward pipeline failure. Two calibration and validation experiments—pencil lead break and drop-ball impacts—generate time-domain and time-frequency signals. Extracted features (AE energy, peak amplitude, rise time) train supervised ML models to improve differentiation accuracy under varying source distances, ball masses, and heights.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Introduction

Aim

Background

Acoustic Emission Testing (AET) at a Glance

Supervised Machine Learning

Experiments and Results

Pencil Lead Break (PLB) Experiment

Time Domain and Time-Frequency Analysis

Drop ball Experiment

Feature Extraction and Model Training

Extracted AE time-series features

FATUKASI O.S., ABOLLE-OKOYEAGU, J., PRATHURU, A. and SOLADEMI, O. 2024. Machine learning-enhanced acoustic emission technique for impact source identification and classification in steel pipes. Presented at the 12th Annual conference of Society of structural integrity and life (DIVK12), 17-19 November 2024, Belgrade, Serbia. Machine learning-enhanced acoustic emission technique for impact source identification and classification in steel pipes. FATUKASI O.S., ABOLLE-OKOYEAGU, J., PRATHURU, A. and SOLADEMI, O. 2024 This document was downloaded from [https://openair.rgu.ac.uk](https://openair.rgu.ac.uk) Machine Learning-Enhanced Acoustic Emission Technique for Impact Source Identification and Differentiation in Steel Pipes Presented by; Oluseyi Samuel Fatukasi Dr. Judith Abolle-Okoyeagu Dr. Anil Prathuru Oludare Solademi School of Computing, Engineering & Technology Robert Gordon University United Kingdom Introduction • Carbon steel pipes are used for oil and gas transportation. • External impacts of various magnitudes are sources of defects in steel pipes. • Pipe defects start as micro-cracks and may progress to cause failure of pipelines. • Failures of pipelines are associated with disasters. • 2023 Pipelines’ significant incident consequences is $216,784,416 . • External Impacts caused about 47% of subsea pipeline failure-IAGA (Zhang et al., 2023) . • There is need for SHM of pipes to ensure sustainable reliability and integrity. Aim • To develop a machine learningenhanced acoustic emission technique for impact source identification and differentiation in steel pipes. • Acoustic Emission Testing (AET) is a Passive NDT. • Materials crack initiation and growth degradation release the elastic stress waves. Background AET at a glance • Conventional AE analysis often struggle to differentiate between closely related damage mechanisms. • The integration of ML models will enhance the accuracy of AET’s differentiation of external impacts sources in steel pipe. Supervised Machine Learning Pencil Lead Break (PLB) Experiment • The PLB experiment aims to calibrate AE set-up. • The test object is 100cm carbon steel pipe, 15cm internal diameter, 1cm thickness . • 20 PLBs were broken at 25cm, 40cm & 55cm distance to the sensor. • The experiment was repeated on damped pipe. • Data sampled at 2.5M/s . PLB test illustration (ASTM E 976 – 99) Ø 2 mm Guide Ring Dimension Pencil lead break experiment set-up Pencil Lead Break Results • Open-ended pipe exhibited higher Energy and peak amplitude across 3 source points. • Open pipe exhibited highest AE Energy (0.0000019V2s). • Open pipe recorded highest peak amplitude (0.0007Volts) . • AE energy and amplitude decrease with longer distances due to wave attenuation. • The number of rise decreases as the PLB source is farther away from sensor. Time Domain Analysis Raw time domain signal from open pipe Raw time domain signal from damped pipe Time-frequency Analysis • Digital filtering is applied to frequencies below and above 100KHz. • Both pipe set-ups recorded higher AE energy in Low pass Cutoff frequency (< 100KHz) . • Both set-ups recorded the highest AE energy atthe 25cm source point. • Open pipe exhibited peak energy at 49 kHz and 2.46µs time at low pass band. • Damped pipe recorded highest AE energy at 48KHz frequency and time 1.6µs. Open Pipe Damped Pipe 3D spectrograms for: (a) low pass (b) high pass for open pipe (c) low pass and (d) high pass for damped pipe Drop ball Experiment • 9g and 17g steel balls dropped from 20cm and 30cm heights. • AE source points are located at 25cm, 40cm, and 55cm source to sensor distances. • The experiment consists of 12 variables of 100 tests each (1200 tests) • FFT, STFT were performed on the AE wave signals. • The extracted time series features (AE energy, peak amplitude, and rise time) were used to train supervised ML models. S

FAQs

Why is acoustic emission monitoring needed for steel pipes?

External impacts of varying magnitudes can generate defects, starting as micro-cracks that may progress toward pipeline failure. SHM is required to support sustainable reliability and integrity.

How do the pencil lead break (PLB) and drop-ball experiments contribute to the study?

The PLB experiment calibrates the AE setup and provides baseline signals across source distances for open-ended and damped pipes. The drop-ball experiment generates distinct burst-type AE signatures, allowing feature extraction for ML training.

Which acoustic emission features are used to train the supervised machine learning models?

The extracted time-series features include AE energy, peak amplitude, and rise time from both experiment signals.

Aug 4, 2026

Upload by Aria