Deep Learning-based delamination assessment of complex composite structures from UGW responses under varying environment

Projektinformationen

SafeCom

ID Finanzhilfevereinbarung: 101102294

DOI

10.3030/101102294

Projekt abgeschlossen

EK-Unterschriftsdatum 30 Juni 2023

Startdatum 1 Februar 2024

Enddatum 28 Februar 2026

Finanziert unter

• Marie Skłodowska-Curie Actions (MSCA)

Gesamtkosten

Keine Daten

EU-Beitrag

€ 175 920,00

Koordiniert durch

KATHOLIEKE UNIVERSITEIT LEUVEN
Belgium

Periodic Reporting for period 1 - SafeCom (Deep Learning-based delamination assessment of complex composite structures from UGW responses under varying environment)

Berichtszeitraum: 2024-02-01 bis 2026-02-28

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

Structural components made from composite materials are increasingly used in transport, aerospace, renewable energy, marine structures and other safety-critical engineering systems because they are lightweight, strong and corrosion resistant. However, these structures can suffer from hidden internal damage during manufacturing or service. One of the most critical forms of damage is delamination, where layers of the composite separate from each other. Delamination is difficult to detect visually, but it can reduce stiffness, change vibration behaviour and, if not identified early, may lead to serious structural failure.

The SafeCom project addressed this challenge by developing computational and data-driven methods for assessing delamination damage in complex composite structures. The project focused on structural health monitoring, where sensors and numerical models are used to evaluate the condition of structures without stopping their operation or causing damage. In particular, SafeCom investigated how vibration responses and ultrasonic guided waves can be used to detect and quantify hidden damage.

The overall objective was to support the development of faster, more reliable and more intelligent damage assessment tools for composite structures operating under realistic service conditions. The project aimed to identify important damage characteristics such as the location, size and interface of delamination. To achieve this, the work combined numerical modelling of damaged composites, contact-based dynamic analysis, vibration-based inverse identification, guided-wave signal processing and Deep Learning methods.

A key motivation of the project was to move towards monitoring approaches that can reduce inspection time, limit unnecessary maintenance, and improve the safety and sustainability of composite structures. By enabling earlier and more accurate detection of damage, the project contributes to safer transport and energy systems, more efficient maintenance planning, and longer service life of high-performance engineering structures.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

SafeCom developed a computational and data-driven framework for assessing hidden delamination damage in composite structures. The work combined physics-based modelling, vibration and wave-response analysis, inverse identification, and Deep Learning.

The first activity was the development of numerical models for delaminated composite beams and plates. The project used advanced finite element approaches, including degenerated shell formulations, sub-laminate modelling and contact constraints, to represent how separated composite layers open, close and interact during vibration. These models were used to study free vibration, constrained vibration and transient dynamic response of damaged laminates.

The second activity focused on damage identification from structural responses. A hybrid inverse-identification method was developed to estimate the location, size and interface of delamination using vibration features such as natural frequencies and mode shapes. This provided an interpretable and efficient route for quantifying damage.

The third activity addressed guided-wave-based monitoring using machine learning. A physics-informed method was developed to select the most damage-sensitive time windows from raw ultrasonic guided-wave signals. These selected signal segments were then processed using a Temporal Convolutional Network to identify damage location and severity. This approach reduced redundant signal content and improved learning efficiency.

The project also positioned these developments within broader structural health monitoring and digital-twin workflows through work on sensing technologies and reduced-order modelling. The main outcomes include validated modelling tools, damage-identification algorithms, guided-wave Deep Learning workflows, and several scientific publications and submitted manuscripts.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

SafeCom advanced the state of the art by connecting several areas that are often treated separately: delamination modelling, contact dynamics, vibration-based inverse identification, guided-wave signal processing and Deep Learning.

A first key result is the development of contact-aware models for delaminated composite laminates. Conventional models often ignore the physical interaction between separated layers, which can lead to unrealistic interpenetration or misleading dynamic responses. SafeCom introduced modelling strategies that enforce contact between delaminated layers and capture breathing-type behaviour more realistically.

A second result is an efficient inverse-identification framework for delamination detection. The project introduced a strategy to handle both continuous damage parameters, such as size and location, and discrete parameters, such as damaged interface, within one optimization framework. This improves the practicality of identifying complex delamination scenarios.

A third result is the physics-informed guided-wave Deep Learning workflow. Instead of feeding entire long wave signals into a neural network, the method first identifies the most informative time windows where damage-sensitive wave interactions occur. This improves interpretability, reduces computational burden and supports more reliable damage localization and severity classification.

These results provide a basis for future digital twins and predictive maintenance tools for composite structures. Further uptake will require larger experimental datasets, testing under real operating and environmental conditions, integration with sensor networks, and demonstration on industrial-scale structures such as aircraft components, wind turbine blades, marine structures or transport systems.