A novel integrated framework for reproducible formability predictions using virtual materials testing

Authors: Adam Plowman, Patryk Jedrasiak, Thomas Jailin, Peter Crowther, Sumeet Mishra, Pratheek Shanthraj, Joao Quinta da Fonseca

Aluminium alloy sheet material is starting to replace steel alloys in automotive components due to its low density and improved recyclability. During production, these materials undergo forming processes, where the sheet is formed into some desired final shape (such as a vehicle pillar). Forming materials can greatly affect their mechanical properties, such as strength. Thus, to ensure these lightweight alloys can be used to maximum effect, it is important to understand the link between formability (the ease with which a material can be formed without damaging it) and the material’s structure at the microscopic scale, which determines its mechanical properties. Formability can be characterised accurately by performing many complex (and expensive!) experiments that probe how a material responds to being loaded in different directions. However, virtual materials testing, in which we combine a small set of experiments with many computational simulations, is a cheaper and more accessible approach. In our work, we have developed an open-source, configurable framework for generating and sharing reproducible hybrid workflows that combine experimental and computational simulations to generate formability predictions. We applied this framework to a particular alloy (Surfalex HF AA6016A), and demonstrated agreement with experimental testing. 

Applying open research practices

• All experimental data collected during the work (for comparison with and calibration of the modelling predictions) is available on the public data repository Zenodo.
• The framework we developed, MatFlow, is an open-source Python package that is hosted publicly on GitHub.
• The workflows we developed that use MatFlow to generate formability predictions are available on Zenodo. This means that anyone can download any of the workflows and run them using MatFlow to reproduce our results. Alternatively, the workflows could be run using different experimental input data to predict formability of a different material.
• We have also made available (in a public GitHub repository) additional post-processing code in the form of Jupyter notebooks, which can, for example, generate the figures we have included in our publication.
• We have submitted this work to the open-access Materials Open Research journal, which adopts an open peer review process.

Overcoming challenges

• Managing a large quantity of computational data required good organisation. We wrote and used existing code to automate parts of the process. For example, we developed a tool called DataLight for programmatically uploading datasets to the Zenodo data repository.
• Ensuring full transparency and reproducibility of the work; the computational work includes over 150 simulations which are interdependent (for example, the initial model calibration required running simulations in a loop until the results matched those from experiment; the calibrated model is then used in other simulations). This complexity was mitigated by our development of MatFlow, which manages the data dependency relationships.

Benefits of using these open research practices

• Since our workflows are highly configurable and publicly available, an interested materials researcher can apply them to their own materials input data with relatively few changes.
• Community oversight of analysis methods; our methodology is completely transparent and available for inspection both by the peer reviewers and by the materials community in general. Thus, in contrast to the more typical approach where analysis is described incompletely within the publication, any issues with our approach are more likely to be spotted and so addressed.

Top tip

Make use of research software engineers who can help to make your analysis code more readable, reusable, and citable!

Related links