To develop an algorithmic framework to predict properties of composites to aid in meta-modelling process.
Composites are made by combining two or more natural or artificial materials to maximize their useful properties and minimize their weaknesses. One of the oldest and best-known composites, glass-fiber reinforced plastic (GRP), combines glass fibers (which are strong but brittle) with plastic (which is flexible) to make a composite material that is tough but not brittle. Composites are typically used in place of metals because they are equally strong but much lighter.
More than 20 % of the A380 is made of composite materials, mainly plastic reinforced with carbon fibres. The design is the first large-scale use of glass-fibre-reinforced aluminium, a new composite that is 25 % stronger than conventional airframe aluminium but 20 % lighter. Concrete is a versatile and cheap material, with a vast range of applications around the home. Brick laying, constructing paths and driveways, foundations to buildings and walls, are some of the practical applications. Concrete has a similarly wide and varied range in industrial applications. Currently, carbon fiber is used primarily in the spar, or structural element, of wind blades longer than 45m/148 ft, both for land-based and offshore systems. The higher stiffness and lower density of CF allows a thinner blade profile while producing stiffer, lighter blades.
Since precise material properties are often needed to satisfy industrial needs and criteria, design of composites with tailored properties is of utmost importance.
The design and analysis of such composites faces following challenges -
- Computational Expense
- Heavy Simulations
- Rigorous Experimentation
- Time Expensive Process
Macroscopic mechanical behavior originates from mechanics of underlying microstructure.
Key mechanical performances of materials:
- Elastic moduli
- Toughness
- Ductility
- Fracture resistance
- Hardness
Microstructural parameters that influence the macroscopic behaviors:
- Shape (Assumed circular in our analysis)
- Size
- Spatial arrangement
- Volume fraction and properties of constituents of the microstructure
- Traditional methods like ‘Rule of Mixture’ and ‘Micromechanics’ Theories cannot be applied for extremely non-linear materials (like that in Tyres), and the distribution of particles.
- Simulations are relatively more complicated. The methodology developed requires more effort as compared to Machine Learning based approaches.
- Simulations are also computationally expensive, especially when being done for a huge range of materials with variations in all properties for each material.
- Data driven models are computationally much cheaper, and easier to develop.
- These models rely on data, which is already available in most laboratories sitting idle. We can make use of this past data to make accurate future predictions.
- These models attain nearly the same (if not higher) degree of accuracy as compared to direct FE simulations.
- FE Model development for simulation of mechanical behaviour of particulate composites.
- Create input database based on several mechanical properties, like Young’s Modulus, Poisson’s Ratio, Volume Fraction, spatial arrangements of particles, etc.
- Perform simulations and extract stress-strain behaviour for several cases.
- Create a meta-model using Machine Learning algorithms.
- Generate composites with unknown material properties.
- Predict with meta-model, and validate the results with direct FE simulations.