Prediction of scour depth at culvert outlets using neural networks

Scour at culvert outlets is a phenomenon encountered world-wide. Research into the problem has mainly been of an experimental nature, with equations being derived for particular circumstances. These traditional scour prediction equations, although offering the engineer some guidance on the likely magnitude of maximum scour depth, are applicable only to a limited range of situations. A model for the prediction of scouring that is generally applicable to all circumstances is not currently available. However, there is a substantial amount of data available from research over many years in this area. This paper compares current prediction equations with results obtained from two Artificial Neural Network models (ANN). The development of a basic feed forward artificial neural network trained by back-propagation to model scour downstream of culvert outlets is described. A supervised training algorithm is used with data collected from published studies and the authors' own experimental work. The results show that the ANN can successfully predict the depth of scour with a greater accuracy than existing empirical formulae and over a wider range of conditions.