A novel hybrid computational fluid dynamics-artificial neural network approach is implemented to predict the mixing efficiency of a T-shaped induced charge electrokinetic micromixer for non-Newtonian fluids. 12,500 data observations produced from computational fluid dynamics—benchmarked against experimental results—are used to develop an optimized deep neural network model for the prediction of mixing efficiency. The optimized neural network model with transfer function in hidden layers has an architecture of 7-85-85-1 and it predicts the mixing efficiency of the induced charge electrokinetic micromixer with the maximum deviation of 2.74 %. Global sensitivity of the artificial neural network model is assessed using Shapley values and it is found that length of the conducting link is the most influencing parameter for designing induced charge electrokinetic micromixer. If more than one conducting links are employed, the pitch transverse to fluid flow is more critical than pitch along the fluid flow direction in mixing zone. Pseudoplastic fluids, marked by pronounced micro-vortices, exhibit superior mixing efficiency, and accelerated mixing at higher electric field strengths compared to dilatant fluids, achieving a mixing efficiency exceeding 99 %. The optimized artificial neural network model predicts mixing efficiency significantly faster compared to computational fluid dynamics and conclusively demonstrates its ability to expedite the design process for electrokinetic micromixers.

中文翻译：

---

使用混合计算流体动力学-人工神经网络方法预测非牛顿流体感应电荷动电微混合器的混合效率

采用一种新颖的混合计算流体动力学-人工神经网络方法来预测非牛顿流体的 T 形感应电荷动电微混合器的混合效率。根据计算流体动力学产生的 12,500 个数据观察结果（以实验结果为基准）用于开发优化的深度神经网络模型，用于预测混合效率。优化后的隐含层传递函数神经网络模型的结构为7-85-85-1，预测感应电荷动电微混合器的混合效率最大偏差为2.74%。使用Shapley值评估人工神经网络模型的全局灵敏度，发现导电链路的长度是设计感应电荷动电微混合器的最有影响的参数。如果采用多于一个的导电连杆，则在混合区中，横向于流体流动的节距比沿流体流动方向的节距更为关键。假塑性流体具有明显的微涡流，表现出优异的混合效率，与胀流流体相比，可以在更高的电场强度下加速混合，实现超过 99% 的混合效率。与计算流体动力学相比，优化的人工神经网络模型预测混合效率的速度明显更快，并最终证明了其加快电动微混合器设计过程的能力。