About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Applications of Uncertainty Quantification (UQ) in Science and Engineering
|
| Presentation Title |
A case study of Bayesian parameter estimation for thermal property inference and uncertainty quantification |
| Author(s) |
Jeremy Drew, Shravan Godse, Yuxing Liang, Abhishek Pathak, Jonathan A Malen, Rachel C Kurchin |
| On-Site Speaker (Planned) |
Jeremy Drew |
| Abstract Scope |
Bayesian parameter estimation (BPE) provides a versatile framework for uncertainty quantification (UQ) when fitting experimental data to an analytical model. In this case study, we showcase the application of BPE to infer interfacial thermal properties from Frequency Domain Thermoreflectance (FDTR) measurements. We compare our approach to standard estimation and UQ-related techniques including least-squares regression, Monte-Carlo-based approaches, and Root-Sum-Squared (RSS) uncertainty. In a well-characterized test system, we find that the uncertainty estimates from BPE are somewhat lower than these standard approaches and attribute this to the likelihood function dynamically reducing the weight of parameter sets that exhibit a poorer fit to the measurements. BPE also offers interpretability advantages; in this work, unexpected behavior in initial results led to the experimental remeasurement of a film thickness, revealing that the original value was incorrect. This highlights BPE’s potential not only for parameter inference, but also for guiding the incorporation of externally known quantities. |