Session: 03-03-02: Damage, Fatigue, and Fracture

Paper Number: 107641

107641 - Bayesian Identification of Input Parameters to Simulate Progressive Damage in Composites 

Despite many superior properties of Fibre Reinforced Polymers (FRPs) such as high strength to weight ratio or damage resistance, these materials typically suffer from a wide range of variability in their mechanical behaviour. These variations mainly arise from manufacturing-induced defects and imperfections such as fibre misalignment, inconsistent fibre volume fraction and wrinkling among others.

It is important to acknowledge these variations in properties when designing FRP composites. Therefore, statistically-based material data is required beyond simple average values.

For the certification of FRP composites for aerospace structural applications, industry must rely on the building-block approach for which comprehensive and costly testing programs must be undertaken in order to guarantee safe and reliable use of FRP composite structures and to obtain statistically meaningful properties.

Finite element (FE) simulations promise a significant reduction in cost and time to develop and deploy lightweight FRP materials by predicting their mechanical behaviour before manufacturing and physical testing. When efficient FE models are considered, it also enables the simulation of large-scale structures at higher levels of the building block.

Progressive damage simulation of FRP composites by means of Finite Element Analysis (FEA) has matured over the past 2-3 decades. Various damage modelling techniques are nowadays available as built-in tools in commercial FE software packages ranging from continuum damage mechanistic material models to discrete techniques such as the cohesive zone method or X-FEM.

Irrespective of the preferred damage modelling technique, one key question is how to consider the inherent uncertainties and variations in mechanical properties of FRP composites.

This paper combines three computational methods - FEA, machine learning and Markov Chain Monte Carlo - to find the range and distribution of various FE input parameter to simulate the statistical response of FRP composites subjected to open-hole tension tests.

First, 15,000 finite element simulations of open-hole tension tests are generated with randomly varying input parameters by applying continuum damage mechanics material models. The data is then used to train and validate a neural network consisting of 5 hidden layers and 32 nodes per layer to generate a highly efficient surrogate model. With the help of such surrogate model, the application of Markov Chain Monte Carlo algorithms enables Bayesian parameter estimation to learn input parameters and their distribution.

The presented data-driven approach is a promising method to reduce experimental efforts required for effective and realistic calibration of computational damage models. The implementation of such a method in the building-block approach can potentially reduce the cost and time associated with certification of aerospace composite structures, hence expediting the development and implementation of next generation high-performance composites.

Presenting Author: Johannes Reiner Deakin University

Presenting Author Biography: Dr Reiner is a Senior Lecturer in Mechanical Engineering within the School of Engineering at Deakin University. His research is centred around the efficient simulation of advanced composite and hybrid composite materials including the evolution of progressive damage and energy absorption. He received his Diploma degree in Applied Mathematics from the Karlsruhe Institute of Technology (KIT) in 2012. He joined the Composites group at the University of Queensland (UQ) in Australia shortly afterwards to pursue his PhD project on the computational failure modelling of composites and hybrid titanium composite laminates. From 2017 to 2019, he was a Postdoctoral fellow at the University of British Colombia (UBC) under the supervision of Prof. Reza Vaziri to study the simulation of damage and failure in dynamically loaded composite structures as well as manufacturing-induced defects.

Authors:

Johannes Reiner Deakin University