Accelerated Testing with Digital Engineering

Challenge

Aircraft structures require significant testing to ensure their airworthiness. These testing techniques require high-quality data to understand how
components behave and how best to identify defects. Collecting data through physical experiments is both expensive and time intensive and often doesn’t yield enough data. Reviewing and analysing data is resource intensive, relying on an individual's ability to recognise small changes that might indicate emerging defects.

There’s a growing need to speed up the testing process. With the emergence of new manufacturing techniques and the development of the next generation of single aisle aircraft on the horizon, being able to test effectively, accurately and more quickly is critical.

Solution

Using MOOSE, an open source multiphysics simulation framework, we developed mathematical models of the physical testing process. These models enabled the exploration of variation in test conditions and how that affected structural response. This produced data that could be used alongside physical tests to understand behaviour and provide further insight to make decisions. Sensitivity analysis was also applied to understand how uncertainty within test setup conditions affected measured outputs. By varying parameters, their influence on outputs could be quantified and insight gained into which parameters had the most impact and therefore needed the most control when setting up.

A cohesive zone model was also developed. This is a simulation technique used to show how damage can initiate and develop within a structure. By simulating strain, a measurement taken during physical bending tests, synthetic strain data was generated. This could be used to train machine learning models, supporting further analysis without needing to rely on physical testing alone.

Impact

Through modelling and simulation, deeper insight into the structure testing process was gained. Analysis showed how the test set up affected results and where uncertainty had the greatest impact, enabling further investigation into these areas. By identifying sources of influence within the testing set up, parameters with stronger influence could benefit from smaller tolerances whilst less influential parameters could be given more flexibility. This can help to reduce testing complexity, setup time and the associated costs with both of these.

The machine learning analysis also highlighted the potential of this kind of technology. This technique could complement an engineer’s toolkit enabling them to gain a deeper insight into potential behaviours and help shape further modelling, simulation and testing set ups.