FEIT Research Project Database

Materials informatics: an intelligent multiphysics multiscale computational tool for autonomous in-service life management of fibre-reinforced polymer composite aircraft structures

Project Leader: Ellie Hajizadeh
Primary Contact: Ellie Hajizadeh (ellie.hajizadeh@unimelb.edu.au)
Keywords: computational materials science; Discrete Element Modelling; machine learning; polymer nanocomposites
Disciplines: Chemical & Biomolecular Engineering,Mechanical Engineering

Irrespective of the superior advantages of fibre-reinforced polymer composites (FRPCs) as a material of choice for aerospace structures, they are not being used at their full potential in critical load-bearing structural applications. This is due to the difficulty of tracking the initiation and propagation of the so-called barely visible impact damage (BVID) during the structure’s lifetime prior to failure. As a consequence, composites are not designed optimally but with a high safety factor.

FRPC structures may exhibit certain failures, such as matrix cracking, fiber-matrix debonding, fibre rupture and delamination, resulting from a combination of defects inherent to the existing composite manufacturing process, as well as a range of in-service conditions including anisotropic and interlaminar shear stresses. Additionally, these failures are further driven by and exacerbated due to unexpected static overload, impact events, thermomechanical fatigue, design errors, and overheating. Essentially, delamination is considered the greatest ‘weakness’ of laminated composite materials, leading to loss of structural integrity. Delamination occurs mostly under the surface, which makes it difficult or even impossible to detect visually.

The failure mechanisms in FRPCs are multiscale in nature, usually initiated by formation of nano- and micro-voids in the matrix, fine structural and intra/inter tow irregularities of the fibres, or at the fibre-polymer interphase, where damage evolution depends on the specific matrix and fibre characteristics, the strength and physiochemical properties of interphase, static and dynamics loading conditions and hygrothermal history. Also, of particular concern for polymer matrices of composites in service are environmental conditions such as UV, temperature and humidity fluctuations that can degrade the matrix and promote greater water absorption and plasticisation of the polymer matrix. Therefore, the effect of temperature and cyclic hygrothermal aging on the interlaminar shear strength needs to be investigated. Additionally, the viscoelastic nature of the polymer matrix demands accounting for deformation-rate and -duration and temperature dependencies of the mechanical behaviour of the polymer matrix, which adds extra complexities to the analysis of the fatigue life of the FRPCs compared to metals.

Therefore, the multiscale nature of these failure mechanisms and their strong dependence on the network microstructure of the polymer matrix and the mechanical and environmental conditions, demand a multiscale and multiphysics approach to allow for the accurate and efficient prediction of the state of the material and lifetime of the component before their catastrophic failure.

In practice, to address safety concerns and reduce the costs associated with early component replacements, and to provide a reliable prediction of the remaining life of the damaged components, it is critical to develop a robust structural health monitoring (SHM)method and an effective non-destructive testing (NDT) methodology to accurately ‘sense’ and ‘assess’ the state of the composite material, particularly when exposed to harsh environmental and operational circumstances.

There are two established approaches to SHM: physics-based models and data-driven models. The physics-based approach uses the inverse problem technique to calibrate numerical models (eg, finite element models) and attempts to identify damage by relating the measured data from the structures to the estimated data from the models (such as FEM). In contrast, the data-driven approach strives to implement different machine learning algorithms to learn from existing available data (e.g. experiments, model data and material simulation) to then conduct pattern recognition, classification, clustering, and regression analysis. These machine-learned algorithms can be adopted to depict the structural behaviour from ‘measured NDT data’, and to perform statistical analysis such as prediction, regression analysis, and pattern recognition for damage identification. Therefore, we emphasize that the establishment of a precise and efficient dynamic model for a structure is an important precondition.

We propose to develop a software application for prediction of the state of the material and SHM tool based on a hybrid approach — hybrid pattern recognition paradigm — which takes into account physical modelling (materials simulation), structural monitoring (NDI), and information from visual inspections, as represented in Figure 2. In this data fusion approach, the machine learning algorithms can learn from data collected from physical models, experimental measurements, and visual-based sources ie, ‘measured NDI data’ and ‘material simulation data’, which potentially improves their knowledge from the structures to better identify damage at early stages with high statistical certainties.

Therefore, we aim to develop an intelligent, portable, and modular SHM tool – to accommodate future capability extension – by integrating the physical models of impact and fatigue behaviour of undamaged and damaged FRPCs validated by NDT and NDI data with advanced machine learning algorithms. In order to achieve this, we need to solve two problems: (1) ‘forward problem’ and (2) ‘inverse problem’.

Related articles on methodologies

[1] Elnaz Hajizadeh, Shi Yu, Shihu Wang, and Ronald G. Larson, “A novel hybrid population balance—Brownian dynamics method for simulating the dynamics of polymer-bridged colloidal latex particle suspensions”, J. Rheol. 62, 235–247 (2018).

[2] Elnaz Hajizadeh, Billy Todd, Peter Daivis, Journal of Rheology, 58, 281–305 (2014)

[3] Elnaz Hajizadeh, Billy Todd, Peter Daivis, J. Chem. Phys, 142, 174911 (2015)

[4] Elnaz Hajizadeh, Billy Todd, Peter Daivis, J. Chem. Phys, 141, 194904 (2014)

[5] Guorui Zhu, Hossein Rezvantalab, Elnaz Hajizadeh, Xiaoyi Wang, Ronald G Larson, Journal of Rheology, 60, 327–343 (2016)

[6] Elnaz Hajizadeh, Ronald G Larson, Soft Matter, 13, 5942–5949 (2017)