UCLA

Spectroscopy methods are widely used for system characterization. Typically a perturbationsignal is introduced to the system, and corresponding outputs are detected. The underlying properties of a system can be determined by comparing the output to the input. This work employs advanced statistical methods to improve overall performance of such methods applied to energy transport and storage, in terms of accuracy, time efficiency and cost for spectroscopy methods to characterize thermal and electrochemical systems.

The rest part of this work reports a custom spectroscopy system that employs periodicheating to characterize thermal diffusivity under ambient conditions; the technique is also known as Angstrom's method. We employ forced convection to reduce variation in convective heat transfer losses along the heat propagation direction to enable the experiment to be conducted outside vacuum. We employ IR thermography to for data-rich temperature detection and further introduce a Bayesian framework for uncertainty quantification and uncertainty reduction with increased data. We demonstrate accurate results (< 5% error) for multiple short metal strips.

In the second part we extend the room temperature spectroscopy approach to high temperatures.For high-temperature characterization, the testing environment is difficult to control precisely; therefore, additional unknowns exist in the physical model as compared to low-temperature measurements. In addition, nonlinear radiation loss are present, and the transient heat transfer process must be modeled numerically. Solving the inverse problem using conventional regression approaches is challenging because the solutions are not unique and lack uncertainty estimates. Therefore, we develop a Bayesian framework to solve the inverse problem. The main challenges overcome are: (1) probing the posterior distribution given a computationally expensive forward model and (2) achieving convergence in the sampling process for a model with a relatively high number of unknowns. In this study we report a computationally efficient parametric surrogate model to accelerate the Bayesian analysis and employ a No-U-Turn sampler to achieve good convergence in the sampling process. The custom instrument exhibits high accuracy (approx. 5% error) and requires only 10 min. obtain steady high-temperature results, compared to several hours using conventional methods and commercial instruments.

Lastly we demonstrate a broadband spectroscopy instrument for electrochemical impedancespectroscopy (EIS) measurements for supercapacitors. Measuring EIS is challenging because (1) low-frequency scans are time consuming and (2) impedance of supercapacitors typically changes by several orders of magnitude across a wide frequency band. The custom instrument employs a broadband exponential chirp signal for rapid frequency scanning. To account for large impedance variations, we report a custom circuit for automatically adaptable impedance matching during measurements. The custom instrument is accurate and exhibits four times measurement time reduction and ten times cost reduction compared to leading commercial instruments.