Machine Learning-Assisted Optimization of CNC-Micromilled PMMA Microchannels for Predictable Hydraulic Performance and Low Protein Fouling Resistance

Polymethyl methacrylate (PMMA) microfluidic devices are widely used because of their optical transparency and biocompatibility; however, surface roughness variations introduced during micromilling influences the hydraulic performance of the channels. In this study, a data-driven approach was developed to clarify the relationships between computer numerical control (CNC) micromilling parameters, surface roughness, and flow resistance in PMMA microfluidic channels. A total of 264 machining instances were used to train machine learning (ML) models for surface roughness prediction, with the best-performing models achieving a test root-mean-square error (RMSE) of 0.168 to 0.186 μm. These models were then combined using an inverse-RMSE ensemble to improve predictive robustness. To link the surface morphology of the channels to the hydraulic performance, an 84-point friction-factor dataset (10 ≤ Re ≤ 60; relative roughness 5–7.5%) was modeled (R 2 = 0.937), and used to estimate pressure drop based on predicted roughness. The micromilled PMMA channels with friction factors ranging from 10.29 to 4.04 for Reynolds numbers (Re) ranging from 10 to 40, showed smooth-pipe behavior at relative roughness of <1%. In addition, preliminary protein fouling experiments were carried out on micromilled PMMA channels with micro-scale surface roughness ranges. The results revealed a weak positive correlation between channel roughness and residual protein concentration (R 2 ≈ 0.29), suggesting that the morphology of micromilled channels exerts a potential influence on biomolecular adsorption during protein fouling. The proposed framework enables pre-fabrication selection of machining parameters to achieve targeted hydraulic performance and provides a pathway for integrating manufacturing data with microfluidic design.