Predictive modeling and optimization of surface roughness in Reverse-µEDM fabricated microeletrode arrays using ML...

Area of Science:

• Materials Science and Engineering
• Biomedical Engineering
• Manufacturing Technology

Background:

• Microelectrode arrays (MEAs) are vital for neural signal acquisition in healthcare.
• Achieving optimal surface roughness in MEAs is critical for signal accuracy.
• Reverse-Micro-Electrical-Discharge Machining (Reverse-µEDM) offers high precision for MEA fabrication.

Purpose of the Study:

• To fabricate MEAs using Reverse-µEDM.
• To investigate and optimize surface roughness parameters.
• To predict surface roughness using machine learning models.

Main Methods:

• Fabrication of MEAs via Reverse-µEDM.
• Taguchi's L18 experimental design to analyze voltage, capacitance, and feed rate effects.
• Surface roughness analysis using non-contact profilometry and Scanning Electron Microscopy (SEM).
• Machine learning models (ANN, SVR, Random Forest, Gradient Boosting, GPR) for surface roughness prediction.
• Leave-One-Out Cross-Validation (LOOCV) for performance evaluation.

Main Results:

• Capacitance was the dominant factor (86%) influencing surface roughness, followed by voltage (11%).
• Random Forest Regression achieved the highest prediction accuracy for surface roughness (R² value, MAE = 0.18 μm).
• Artificial Neural Network (ANN) showed comparable accuracy but required longer training times.

Conclusions:

• Reverse-µEDM is a suitable method for fabricating high-precision MEAs.
• Machine learning, particularly Random Forest, effectively predicts surface roughness for MEAs.
• Optimized MEA surface roughness is achievable, enhancing neural signal recording capabilities.