Real structures have pier and abutment widths of the order of 1-3 metres and flood flow velocities upwards of 2m/s. Thus Reynolds Number will be significantly above 10^6, which is above the critical flow transition for circular piers. This means that pier drag factor will be lower than in models with RE< 5x10^5 and there will be no coherent vortex shedding, which in scale models produces suction, lifting disturbed sediment out of the scour hole. This epistemical source of error was missed by Bateni in 2007 and does not appear to have been considered by subsequent machine learning researchers.
Hello here are some links you might want to take a look at, it may not answer the question directly but might help you understand the scaling and machine learning further.
Article Machine Learning at Scale
https://towardsdatascience.com/the-influence-of-data-scaling-on-machine-learning-algorithms-fbee9181d497
https://medium.com/analytics-vidhya/why-scaling-is-important-in-machine-learning-aee5781d161a
This answer deals with the issue of scaling within a machine learning context rather than a civil engineering context. Rephrasing the question in machine learning terms: Can you rely on AI predictions for cases that are significantly outside the range of the training data, and could increasing the volume of training data give any improvement in such a prediction. To coin a phrase, AI surely stands for artificial interpolation, not artificial extrapolation.
- Badges
- Science topic