Optimization of annular microfluidic biosensor enhanced by active and passive effects using Taguchi’s method coupled with multi-layer perceptron neural networks (MLP-NN) models

Abstract

Optimizing the performance parameters, including the detection time, of microfluidic biosensors is crucial for enhancing their efficiency and accuracy. In this study, the detection time $({T}_{R})$ was optimized by considering control parameters, such as the width of the annular reaction surface ($\delta R$), the applied voltage (${V}_{\mathrm{rms}}$), the inlet concentration (${C}_{0}$), the inlet average velocity (${U}_{\mathrm{ave}}$), and the presence of an obstacle. Taguchi’s method was employed to design an L8(2⁵) orthogonal network, enabling optimal parameter settings. The performance parameters were optimized using the signal-to-noise (S/N) ratio curve, and numerical predictive models were developed using multiple linear regression (MLR), quadratic regression, and multi-layer perceptron artificial neural network (MLP-ANN). The theoretical analysis resulted in optimized design parameters, with $\delta R=5\mu m$, ${V}_{\mathrm{rms}}=5V$, ${C}_{0}=10\mathrm{pmol}/{\mathrm{m}}^{3}$, ${U}_{\mathrm{ave}}=0.5\mathrm{mm}/\mathrm{s}$, and the presence of an obstacle, leading to a minimum response time of 10325 s (2.87 h). Among the key parameters, ${U}_{\mathrm{ave}}$ had the highest contribution (62%) in reducing the response time, while C₀ had the lowest contribution (0.1%). The MLP-ANN model exhibited high prediction performance for the response time of the new microfluidic biosensor design, as demonstrated by MAPE, RMSE, VAF, and R² values. Overall, this study underscores the significance of optimizing performance parameters, particularly the detection time, in microfluidic biosensors. It utilizes Taguchi's method, develops numerical predictive models, and provides optimized parameter settings. The results highlight the substantial impact of ${U}_{\mathrm{ave}}$ and the minimal contribution of C₀ in reducing the response time. Additionally, the MLP-ANN model demonstrates outstanding prediction accuracy for the response time of the new microfluidic biosensor design.