ANALYSIS AND OPTIMIZATION OF SCOUR DEPTH AROUND T-SHAPE SPUR DIKES

Abstract

Alluvial rivers are prone to bank erosion and channel instability, a problem that worsens during high flows and floods, when rising velocity and bed shear stress can rapidly erode banks, shift the channel, and damage surrounding land and infrastructure. Spur dikes are a common river-training response to this, built to redirect the flow, protect the bank, and encourage sediment deposition. Their own use, however, brings a problem of its own: local scour at the structure, which if left unchecked can undermine the foundation and cause the spur dike to fail. Designing safer and more effective spur dikes therefore depends on a proper understanding of the flow and the scour it produces around them. This thesis examines how scour around a spur dike is shaped by its structural configuration, with a particular focus on improving traditional designs to reduce scour and give better bank protection. Its main contribution is a new configuration, the curved T-shape spur dike, obtained by curving the flange of the conventional T-shape, with the aim of spreading the flow around the structure more evenly, weakening the vortices it generates, and so reducing local scour at the head. These objectives were pursued through an experimental programme in a laboratory flume under clear-water conditions, comparing scour around the conventional I- and T-shape spur dikes with the proposed curved T-shape across a range of hydraulic conditions. The spacing between spur dikes placed in series was examined too, since it governs how scour develops between adjacent structures. Maximum scour depth, its development over time, and the resulting bed morphology were recorded throughout. Scour proved to be a progressive process, reaching equilibrium in around 8 hours in every run. The spur dike length (b) was fixed at 0.22 m, or 20% of the channel width, and of the variables examined, the flow condition U/Uc (the ratio of approach velocity to critical velocity) proved to be among the most influential: a higher flow intensity generates stronger vortices and moves more sediment, and so erodes the spur head more severely. Sixteen configurations of the curved T-shape spur dike were tested to find the best geometry, combining four flange v lengths, X = b/2 (0.11 m), 2b/3 (0.147 m), 5b/6 (0.183 m), and b (0.22 m), with four crest widths, Y = b/10 (0.022 m), 3b/20 (0.033 m), b/5 (0.044 m), and b/4 (0.055 m). X = b (0.22 m) with Y = b/5 (0.044 m) gave the best performance, with equilibrium scour depths of 0.142 m, 0.138 m, and 0.133 m at U/Uc of 0.95, 0.89, and 0.83, respectively. Spacing between spur dikes in series was tested at 2b, 2.5b, 3b, 3.5b, and 4b, and 3b proved optimum, cutting scour at the second spur dike by 50% relative to the first. The curved T-shape spur dike was then set against the conventional I- and T-shape spur dikes, and gave a lower scour depth under every hydraulic condition tested, an improvement traced to its curved flange, which redirects the approaching flow more smoothly toward the centre of the channel. CFD simulations reproduced this same behaviour, showing lower near-bed velocities, a weaker horseshoe vortex, and less flow separation around the curved T-shape than around the I- or T-shape spur dike. That the experiments and the simulations agree confirms the proposed design's ability to limit local scour and improve hydraulic performance, and points to the curved T shape spur dike as a way of strengthening bank protection and the long-term stability of alluvial channels. Machine learning was also used to estimate scour depth around spur dikes. A standalone Random Tree (RT) model overfit the data and generalised poorly, so it was hybridised with additive regression (AR), bagging (B), and random subspace (RSS) to give seven combinations – AR_RT, B_RT, RSS_RT, RSS_AR_RT, RSS_B_RT, B_AR_RT, and RSS_B_AR_RT – built on the key hydraulic and geometric parameters as input variables and evaluated for predictive performance. All seven performed well in training, and two carried this through to testing: B_AR_RT (R-Square = 0.9702, RMSE = 0.1305, NSE = 0.9692, and COC = 0.9845) and RSS_RT (R-Square = 0.9699, RMSE = 0.1321, NSE = 0.9685, and COC = 0.9841). Of these, only B_AR_RT held up as a reliable predictor of scour depth, with an R-Square of 0.9007, an RMSE of 0.1801, and a COC of 0.9491, suggesting that this hybrid approach can give dependable scour-depth estimates and stand in for extensive laboratory investigations at the preliminary design and analysis stage. Three regression equations, linear, non vi linear, and second-order polynomial, were also developed to predict scour depth around I- and curved T-shape spur dikes: the second-order polynomial gave the most accurate predictions for the I-shape, while for the T-shape, all three regression equations performed satisfactorily. Taken together, these results give a rounded picture of the hydraulic and scour processes at work around spur dikes, and show the curved T-shape spur dike to be an effective alternative to conventional configurations for reducing local scour.