Autonomous parameter exploration in thermoplastic material-extrusion additive manufacturing using Bayesian optimization with intermediate features

Material-extrusion additive manufacturing (MEX-AM) involves rapid and microscopic physical phenomena, such as melting, ejection, and solidification of thermoplastic polymers that occur at the nozzle tip. These processes are highly sensitive to various process parameters, such as nozzle temperature and scan speed, rendering parameter exploration costly and time-consuming. To expedite the identification of optimal processing conditions, this study focuses on tensile strength as the target property and introduces neck width, derived from sidewall asperities, as an in situ intermediate feature (IFV). Neck width can be measured non-destructively and demonstrates a strong correlation with tensile strength. A Bayesian optimization with intermediate features (BOIF) approach is proposed and implemented for higher-throughput parameter exploration through real-time monitoring of neck width during modeling. By optimizing nozzle temperature, this method achieves a throughput approximately 400 times greater than that of exhaustive tensile testing. When the parameter space is expanded to include scan speed, throughput increases by up to 800 times. These results highlight the effectiveness of the BOIF framework in reducing both the experimental cost and development time in MEX-AM process optimization. Moreover, the framework suggests that incorporating additional IFVs can enhance predictive accuracy, and extending the optimization to higher-dimensional parameter spaces can further amplify the efficiency gains. • Novel BOIF parameter exploration in material-extrusion additive manufacturing • 800× faster parameter search achieved using BOIF • Nozzle temperature and scan speed optimized by sidewall asperity monitoring. • Intermediate features improve search efficiency in Bayesian optimization. • BOIF reduces experimental costs and development time.