Periodic Reporting for period 1 - AI2AM (Artificial Intelligence driven topology optimisation of Additively Manufactured Composite Components)
Reporting period: 2021-09-01 to 2023-08-31
The EU is a world leader in the development and manufacturing of structural products for the transport, energy, and construction industries that generate an annual income of €1tn (10.8% in gross value) for the European economy. The European “Green Deal” sets an ambitious target of rendering Europe climate neutral by 2050. The recently approved Circular Economy Action Plan brings forward the requirement for delivering sustainable and resilient end-products across the entire industrial ecosystem. Hence the question on how to maximize the operational life-cycle of critical complex structures, e.g. airplanes, wind-turbines, bridges while drastically reducing manufacturing and and maintenance costs becomes seminal. Additively Manufactured fibre reinforced composite (AMC) components built via fused deposition modelling rapidly find applications within the European aerospace and transport industry as they are prone to, less machine, material, and labour costs, less manufacturing waste, and usage of more efficient materials. A drawback of AMC components is their typically complex and in cases tessellated geometry; this gives rise to combined damage mechanisms that deviate from the usual “high strength and ductile metal” paradigm. This “complexity”, if controlled, can result in components of tailored properties, e.g. of increased fracture toughness and pseudo-ductile post fracture response. Unfortunately, current analysis and design methods lack the necessary level of refinement, or the underlying theoretical framework indeed, to address this critical issue.
The overarching aim of AI2AM is to provide an integrated design framework to address this challenge by greatly advancing the current state-of-the-art by employing a truly cross-disciplinary research methodology. On the modelling side, this project is designed to develop a novel, rapid and high-fidelity physics-based method custom fit for the analysis of such complex structures. On the design side, this research will deliver the first ever AI driven topology optimisation framework for AMC components of increased fractured toughness.
To this end, over a period of two years (2021-2023), AI2AM delivered novel numerical methods and tools for the damage analysis and topology optimisation analysis of fibre reinforced 3D printed domains. The main conclusions of the project are:
1) Cohesive phase field models provide a versatile paradigm for the damage analysis of 3D printed composites. Our results demonstrate that the numerical predictions match experimental results when the pertinent material parameters, i.e. fracture toughness per material orientation and critical stress per material orientation are well-defined.
2) Both Scaled Boundary Finite Element and Virtual Element methods provide well-behaved fracture simulations of increased computational efficiency when compared to standard finite element solvers. This is mainly due to the ability of the methods to solve over complex discretization topologies without the need for local mesh refinement.
3) Fusing topology optimisation with machine learning surrogate models succeeds in providing designs of increased fracture toughness compared to conventional designs. However, further research is required for the results to be generalizable.
4) Deploying highly-parallelized solvers for coupled field problems along-side with discrete methods that can operate over quad-tree meshes is necessary for rendering phase field methods applicable to optimisation problems. As a side-project we were also able to apply similar algorithms to fluid-structure interaction problems achieving 50x speed-ups when operating in multi-core servers.
The overarching aim of AI2AM is to provide an integrated design framework to address this challenge by greatly advancing the current state-of-the-art by employing a truly cross-disciplinary research methodology. On the modelling side, this project is designed to develop a novel, rapid and high-fidelity physics-based method custom fit for the analysis of such complex structures. On the design side, this research will deliver the first ever AI driven topology optimisation framework for AMC components of increased fractured toughness.
To this end, over a period of two years (2021-2023), AI2AM delivered novel numerical methods and tools for the damage analysis and topology optimisation analysis of fibre reinforced 3D printed domains. The main conclusions of the project are:
1) Cohesive phase field models provide a versatile paradigm for the damage analysis of 3D printed composites. Our results demonstrate that the numerical predictions match experimental results when the pertinent material parameters, i.e. fracture toughness per material orientation and critical stress per material orientation are well-defined.
2) Both Scaled Boundary Finite Element and Virtual Element methods provide well-behaved fracture simulations of increased computational efficiency when compared to standard finite element solvers. This is mainly due to the ability of the methods to solve over complex discretization topologies without the need for local mesh refinement.
3) Fusing topology optimisation with machine learning surrogate models succeeds in providing designs of increased fracture toughness compared to conventional designs. However, further research is required for the results to be generalizable.
4) Deploying highly-parallelized solvers for coupled field problems along-side with discrete methods that can operate over quad-tree meshes is necessary for rendering phase field methods applicable to optimisation problems. As a side-project we were also able to apply similar algorithms to fluid-structure interaction problems achieving 50x speed-ups when operating in multi-core servers.
There have been five key developments within the scope of the AI2AM project. In particular
1) A novel accurate and validated phase field theory was developed to simulate the exact damage patterns and the quasi-brittle post-fracture response of anisotropic material domains. This provided a niche physic's canvas for the simulation of additively manufactured long fibre composites.
2) A novel and rapid forward simulation tool for damage propagation in anisotropic materials was developed harnessing the computational merits of the scaled boundary finite element method. The source code implementation was verified against the standard finite element method. In addition, a Virtual Element variant was developed that demonstrated additional computational merits.
3) Using the aforementioned computational framework, we generated 20TB of training data pertinent to material coupon tests typically employed in the industry to grade additively manufactured materials.
4) We subsequently developed machine learning algorithms trained on the aforementioned dataset. Two pathways were explored, i.e Physics Informed Neural Networks using multilayer perceptrons and graph neural networks.
5) Finally, a novel topology optimization framework has been developed using the aforementioned physics canvas. Within this setting, a polymaterial SIMP algorithm was implemented that can account for fibre-reinforced, matrix only, and void subdomains. Two industrial cases were examined using this methodology, i.e. a wing rib for a solar HALE and a bicycle rocking joint.
The project results have already been disseminated via presentations in international conferences, technical workshops and peer-reviewed publications in scientific journals. All publications, generated data and codes are openly available. The Fellow has further participated in several outreach activities, including 5 presentations within the remit of the Science is Wonderful 2021 action, one presentation in an Elementary school at Athens Greece and the Researcher's Night at NTUA Greece.
1) A novel accurate and validated phase field theory was developed to simulate the exact damage patterns and the quasi-brittle post-fracture response of anisotropic material domains. This provided a niche physic's canvas for the simulation of additively manufactured long fibre composites.
2) A novel and rapid forward simulation tool for damage propagation in anisotropic materials was developed harnessing the computational merits of the scaled boundary finite element method. The source code implementation was verified against the standard finite element method. In addition, a Virtual Element variant was developed that demonstrated additional computational merits.
3) Using the aforementioned computational framework, we generated 20TB of training data pertinent to material coupon tests typically employed in the industry to grade additively manufactured materials.
4) We subsequently developed machine learning algorithms trained on the aforementioned dataset. Two pathways were explored, i.e Physics Informed Neural Networks using multilayer perceptrons and graph neural networks.
5) Finally, a novel topology optimization framework has been developed using the aforementioned physics canvas. Within this setting, a polymaterial SIMP algorithm was implemented that can account for fibre-reinforced, matrix only, and void subdomains. Two industrial cases were examined using this methodology, i.e. a wing rib for a solar HALE and a bicycle rocking joint.
The project results have already been disseminated via presentations in international conferences, technical workshops and peer-reviewed publications in scientific journals. All publications, generated data and codes are openly available. The Fellow has further participated in several outreach activities, including 5 presentations within the remit of the Science is Wonderful 2021 action, one presentation in an Elementary school at Athens Greece and the Researcher's Night at NTUA Greece.
Research-wise, AI2AM delivered progress across two main fields. On the computational mechanics front, a new phase field model has been developed that accurately predicts the damage patterns of long-fibre reinforced composites. The resulting governing equations were solved using the scaled boundary finite element method and the virtual element method with the latter proving advantageous in terms of computational efficiency.
Further to the initial objectives of the project, the experience gained from implementing coupled solvers facilitated the interactions of the researcher with colleagues working in fluid-structure interactions and eventually leading to additional research outputs pertinent to wave induced deformations in deformable media.
On the optimization front, a novel SIMP driven topology optimization algorithm was developed. The novel traits of the algorithm are the fact that it results in multi-material configurations of increased fracture toughness or minimum compliance depending on the end-user requirements.
These developments will accelerate the uptake of additively manufactured composites by the aero-space and construction industries for structural applications leading to a significant reduction in material use. These exploitations routes are already been explored via ongoing projects that the Fellow has mananed to secure during the period of this Individual Fellowship.
Further to the initial objectives of the project, the experience gained from implementing coupled solvers facilitated the interactions of the researcher with colleagues working in fluid-structure interactions and eventually leading to additional research outputs pertinent to wave induced deformations in deformable media.
On the optimization front, a novel SIMP driven topology optimization algorithm was developed. The novel traits of the algorithm are the fact that it results in multi-material configurations of increased fracture toughness or minimum compliance depending on the end-user requirements.
These developments will accelerate the uptake of additively manufactured composites by the aero-space and construction industries for structural applications leading to a significant reduction in material use. These exploitations routes are already been explored via ongoing projects that the Fellow has mananed to secure during the period of this Individual Fellowship.