Optimization of Microfluidic Biosensor Enhanced by Electrothermal Effect Using Taguchi's Method Coupled with MLP-ANN Models

This study focused on optimizing the performance of a microfluidic biosensor by analyzing the effect of four control parameters on the detection time
$(\mathrm{T}_{\mathrm{R}})$
of the device. The control parameters were the applied voltage
$(\mathbf{V}_{\mathbf{rms}})$
, the channel height (H), the biosensor length
$(\mathrm{l}_{\mathrm{S}})$
, and the biosensor position
$(\mathrm{X}_{\mathrm{S}})$
. The Taguchi method was used to design an
$\mathbf{L}\mathbf{16}(\mathbf{4}^{\mathbf{4}})$
orthogonal network to reduce the variance in the experiment and identify the optimal setting of the control parameters. The signal-to-noise (S/N) ratio curve was used to optimize the individual performance parameters and developed numerical predictive model using artificial neural network (ANN). The results showed that the optimal response of the design parameters was achieved with
$\mathbf{V}_{\mathbf{rms}}=\mathbf{20}\mathbf{V}, \mathbf{H}=\mathbf{20} \boldsymbol{\mu} \mathbf{m}, \mathbf{l}_{\mathbf{s}} =\mathbf{40}\ \boldsymbol{\mu} \mathbf{m}$
and,
$\mathbf{X}_{\mathbf{S}} =\mathbf{100} \boldsymbol{\mu} \mathbf{m}$
, which resulted in a minimum response time of 193.38 min (3.2h). The analysis of variance (ANDVA) was also performed to determine the contribution of each input parameter to reducing the response time. The results indicated that Vrms made the most significant contribution at 35%, while Xs had the least impact at 3%. The ANN model demonstrated strong predictive capabilities, as reflected in the mean absolute percentage error (MAPE), root mean square error (RMSE), values for explained variance (VAF), and coefficient of determination (R2). In summary, this research offers valuable insights into enhancing microfluidic biosensor optimization and Highlights the importance of selecting the right control parameters for achieving peak performance