NEW DELHI: Researchers at Indian Institute of Technology (IIT), Guwahati have developed new modelling methods to assess the probability of failure of composite materials in the aerospace and automobile sector.
For this, the researchers have used a combination of Machine Learning tools and state-of-art sampling techniques, to model and predict the failure and other mechanical properties of composite materials used in the aerospace and automobile sector.
According to the team, combination of these tools is able to better predict the failure of these materials over the multiple probability technique such as Monte Carlo simulation.
The results of the work have been published in the journal “Composite Structures”.
“Composite materials are made of two or more components and are extensively used in all kinds of aerospace, automobile and construction applications because of favourable properties such as excellent corrosion resistance, high strength and stiffness to weight ratios, durability, increased fatigue life, affordable cost, etc. The simplest type of composite is Fiber-reinforced-Plastics or FRP, which are widely used today,” said Nelson Muthu, Assistant Professor, IIT Guwahati.
“Specialized fiber reinforced composites are used in aerospace, for making specialized aircraft structures. The fuselage, wings, tail, doors, and interior of aircraft, for example, are made of composite materials. The Boeing 787 Dreamliner aircraft is 80 per cent composite by volume, which reduces fuel use by 20-25 pc compared to its predecessors,” he added.
Muthu said that despite their wide use, composite materials can fail due to problems such as fibre matrix debonding, delamination, fiber misalignment, matrix cracking, density variation, broken fibres, impact damage, etc. We need to understand and predict such failures so the composite and the component can be designed accordingly and reduce the risk of failure.
“Although computers are getting more powerful, it is still unrealistic to depend only on single simulation codes to predict properties that are dependent on multiple factors,” he said.
The IIT Guwahati team developed a computationally efficient multiscale metamodel-based approach that combines machine learning tools like support vector machines, among others, and sampling tools, such as Latin hypercube, to assess the failure risk in composites with many sources of uncertainty.
“We performed experiments to determine the uncertainties in the microscale and established the variation of the properties in the mesoscale using the computational homogenization technique within the finite element modelling framework. The uncertainty in the mesoscale was also estimated, and this information was transported to the macroscale simulations,” he said. (PTI)