Numerical Uncertainty in Fluid Flow Calculations: Needs for Future Research

Abstract

Computational fluid dynamics (CFD) has established itself as a viable technique for performing research and solving engineering problems, and when used correctly, can give accurate results for many fairly complex problems. This success has led to an ever increasing number of journal publications, many code developers, and surprisingly many users in the industry. Commercial CFD packages are often marketed by claiming that a particular code can solve almost every fluid flow problem, while many users, both in industry and academia, stand aloof from quantitative error measures, instead being dazzled by colorful computer generated output. This is mostly due to insufficient education in the scientific computing discipline which often leads (intentional or not) to misuse and wrong conclusions. Every year, hundreds of papers are published in conference proceedings, and journals, on the advancement and application of CFD techniques. Whenever something is spawned in such large quantities it is very easy to lose sense of quality control. To assert quality, papers often end with a conclusion such as “good agreement is found between experiments and predictions” to which the readers have become so immune that it no longer has meaning. Unfortunately, very little information is provided about the numerical uncertainty and the experimental data are often treated as if they are 100 percent accurate.