Onderzoeksdata

Deze oefening is afkomstig uit RDM Rose (2015, no longer active), activity sheet 5.2.2. Het is een facultatieve oefening die je kunt doen als je meer gevoel wilt krijgen bij het begrip onderzoeksdata.   

Case studies

Op page 6-22 van een document van de Universiteit van Southampton (2016) vind je vijf case studies op het gebied van onderzoeksdata in: 

1. medisch onderzoek
2. materiaalkunde
3. aerodynamica
4. scheikunde
5. archeologie

Bekijk één case study in detail en beantwoord dan de volgende twee vragen:

• Herken je de vijf manieren om naar onderzoeksdata te kijken? Hoe?
• Benoem een aantal mogelijke vraagstukken die onderzoekers hebben (of kunnen krijgen) bij het opslaan, beheren, archiveren en delen van hun onderzoeksdata.

Hieronder zie je de uitwerking (in het Engels) van een van de cursisten van Essentials 4 Data Support, die naar case study 3 (Aerodynamica) keek.

5 ways of looking at the data:

• Collection: numerical model simulations
• Types: models, algorithms and scripts; software configuration, post-process files a.o. Figures
• Electronic storage: textual, software code, software specific (mesh), multimedia (figures)
• Size and complexity: large output files (hundreds of gigabytes) with corresponding additional files such as the input/configuration files and post-processing results (figures and aggregated results)
• Life cycle: this type of numerical modelling is typically done in the research phase where various wing shapes are “tested” with the model and the performance is compared. A subset of all the simulations carried out, with typical results to underpin the drawn conclusions, is usually described in the publication and therefore minimal required to be published.

Possible issues:

• Storage: With data volumes of 300GB per 1 sec of simulated flow, the total data volume easily exceeds the size of a regular laptop’s hard drive. Using network or cloud storage, that also has a good connection with the HPC to be used is recommended.
• Manage: For keeping track of a variety of simulations with sometimes minor differences in model input/configuration it is important to think before starting. A clear directory structure and sufficient description of modifications, and reasons for it, is crucial for good handling of the results. For reproducibility it is important to keep track of the used software version (even more important if it varies between different simulations). I recommend to use a version control system for model input/configuration and pre/post-processing scripts.
• Archive: For archiving again the data volume and the associated costs may play a role. Therefore archiving only the simulations results for the simulations that are actively discussed in the publication to draw conclusions from and archiving the input/configuration and software version (all necessary information for reproduction) of the remaining simulations might be wise.
• Share: For sharing of model results, it is crucial that others are able to interpret and reproduce the results. This means that the remarks made in the “manage” section are once more important. Basically, proper data management during the research phase makes you ready for sharing at any time.