American Society of Civil Engineers


Neural Networks for Estimation of Scour Downstream of a Ski-Jump Bucket


by H. Md. Azmathullah, (Res. Asst., Central Water and Power Res. Station, Khadakwasla, Pune 411 024, India; formerly, Res. Scholar, Civ. Engrg. Dept., I.I.T., Bombay, India. E-mail: rsazmat@civil.iitb.ac.in), M. C. Deo, (corresponding author), (Prof., Dept. of Civ. Engrg., Indian Inst. of TEchnol., BOmbay, Mumbai 400 076, India E-mail: mcdeo@civil.iitb.ac.in), and P. B. Deolalikar, (Joint Dir., Central Water and Power Res. Station, Khadakwasla, Pune 411 024, India. E-mail: cwprs3@vsnl.net)

Journal of Hydraulic Engineering, Vol. 131, No. 10, October 2005, pp. 898-908, (doi:  http://dx.doi.org/10.1061/(ASCE)0733-9429(2005)131:10(898))

     Access full text
     Purchase Subscription
     Permissions for Reuse  

Document type: Journal Paper
Abstract: The estimation of scour downstream of a ski-jump bucket has remained inconclusive, despite analysis of numerous prototypes as well as hydraulic model studies in the past. It is partly due to the complexity of the phenomenon involved and partly because of limitations of the traditional analytical tool of statistical regression. This paper addresses the latter part and presents an alternative to the regression in the form of neural networks. The depth of the scour hole developed along with its width and length is predicted using neural network models. A network architecture complete with trained values of connection weight and bias and requiring input of grouped parameters pertaining to discharge head, tail water channel depth, bucket radius, lip angle, and median sediment size is recommended in order to predict the depth, the location of maximum scour, as well as the width of scour hole. The neural network predictions have been compared with traditional statistical schemes. Although the common and simple feed forward back propagation network took a very long time to train as compared to some advanced schemes, it was found to impart equally reliable training as the latter. Use of causative variables in grouped forms was found to be more rewarding than that of their raw forms probably due to lesser scaling effect.


ASCE Subject Headings:
Neural networks
Scour
Spillways
Artificial intelligence
Floods