Predicting dimensional fidelity of FDM-fabricated PETG and TPU microchannels using Takagi-Sugeno fuzzy neural networks

Achieving reliable dimensional fidelity in microscale features remains a critical bottleneck in fused deposition modelling (FDM) due to complex thermal–viscoelastic interactions at small length scales. The novelty of this work lies in the development of an interpretable predictive modelling framework based on Takagi–Sugeno fuzzy neural networks (TS-FNN), which overcomes the "black-box" limitations of traditional artificial intelligence by providing both high-precision prediction and physical transparency for microscale FDM. We investigated five key process parameters using a Taguchi experimental design to model four dimensional responses for PETG and TPU microchannels. Quantitative results demonstrate exceptional predictive performance, with R² values reaching 0.997 for width accuracy and remaining above 0.897 across all robustness metrics. The optimal TS-FNN configurations, utilizing a structured architecture of 27 fuzzy rules, achieved remarkably low root mean square errors (RMSE as low as 0.00178 mm), while validation on previously unseen parameter combinations yielded mean prediction errors as low as 3–4 µm ensuring high-fidelity control relative to the microscale feature dimensions. Beyond numerical accuracy, the fuzzy rule-based structure explicitly identifies material-dependent sensitivities, revealing how rheological differences between rigid (PETG) and compliant (TPU) thermoplastics dictate dimensional stability. This research provides a rational methodology for identifying material-specific processing windows and represents a significant advancement in the predictable, interpretable control of additive manufacturing for functional microscale devices.