Ke Wang

Chemical and Biomolecular Engineering

Bayesian Optimization of Additive Manufacturing for Thermoelectric Materials

Motivation: Solid-state thermoelectric (TE) generators (TEGs) are a promising technology for powering distributed sensor networks, biomedical devices, wearable electronics, and beyond. [1-3] However, the efficiency of TEGs is limited by the TE materials and their manufacturing methods. [4, 5] This research project, in collaboration with Profs. Tengfei Luo, Yanliang Zhang, and David Go, seek to establish a paradigm shift from heuristic time- and resource-intensive Edisonian search to systematical data-driven Bayesian optimization (BO) for TE materials and devices [8]. Figure 1 summarizes the necessary manufacturing steps for the desired high- functionality TE material and the research objectives.

Prior results: As part of a now-complete ND Energy-affiliated DOE-funded project (AMO, DE-EE000910), our team successfully demonstrated key elements of the data-driven optimization framework, including:

1. Machine learning enables accurate predictive modeling of sparse and high-dimensional flash sintering conditions [6]

2. Bayesian optimization assists experts in efficiently identifying the state-of-the-art flash sintering experimental condition and TE material composition [7, 8, 10]

3. Physics-informed ML enables precise control of thickness with less than five experiments [9]

Research Objectives

We propose building on these prior results to synthesize record-setting TE devices through co-optimizing materials and manufacturing processes, organized into three research objectives.

Objective 1: Accelerate low-temperature processing of printed indium tin oxide (ITO) nanoinks using Bayesian optimization of non-thermal plasma jet sintering. This objective uses multi-objective Bayesian optimization to systematically explore the trade-off between electrical conductivity and maximum substrate temperature for non- thermal plasma sintering. Expected outcome: Identify the optimal plasma sintering conditions for record-setting TE materials, e.g., electrical conductivity over 7 S m-1 and maximum temperature less than 35 °C, resulting in one high-impact peer-reviewed journal publication [12, in preparation]. Moreover, we anticipate this framework is transferable to other plasma systems (e.g., sintering, catalysis).

Objective 2: Machine learning-assisted direct ink writing of three-dimensional P-type BiSbTe thermoelectric devices. Our prior work demonstrates that controlling film thickness is critical in optimizing material functionality before sintering. [9] This objective will build upon prior work in Bayesian optimization of flash sintering [6] and predictive thickness control of jet printing [7] to optimize the material functionality for a device. Expected outcome: Fabricate an optimized P-type TE material with record-setting P-type power factor over 4000 W m-1K-2, using four or fewer rounds of experiments, resulting in one high-impact peer-reviewed publication [13, in preparation].

Objective 3: Machine learning-enabled co-optimization for N-type AgSe-based material composition and flash sintering. Our prior work individually optimized sintering [7] and material composition [10]. In this objective, we will develop an ML framework to co-optimize material composition and manufacturing conditions simultaneously. Expected outcome: Fabricate an optimized N-type TE material with record-setting N-type power factor over 3000 W m-1K-2, using ten or fewer rounds of experiments, resulting in one high-impact peer-reviewed publication [14, in preparation].

Long-term Impact: This project will develop and demonstrate new ML-guided optimization capabilities, producing more record-setting TE material and devices. Moreover, we emphasize that these ML approaches are transferable to a broad range of material and additive manufacturing systems. The high-impact publications from objectives 1 to 3 will help ND Energy researchers remain competitive for future collaborative projects in the ML + materials + additive manufacturing area.

References

[1] Zeng, Minxiang, Duncan Zavanelli, Jiahao Chen, Mortaza Saeidi-Javash, Yipu Du, Saniya LeBlanc, G. Jeffrey Snyder, and Yanliang Zhang. "Printing thermoelectric inks toward next-generation energy and thermal devices." Chemical Society Reviews 51, no. 2 (2022): 485-512.

[2] Pourkiaei, Seyed Mohsen, Mohammad Hossein Ahmadi, Milad Sadeghzadeh, Soroush Moosavi, Fathollah Pourfayaz, Lingen Chen, Mohammad Arab Pour Yazdi, and Ravinder Kumar. "Thermoelectric cooler and thermoelectric generator devices: A review of present and potential applications, modeling and materials." Energy 186 (2019): 115849.

[3] Jouhara, Hussam, Alina Żabnieńska-Góra, Navid Khordehgah, Qusay Doraghi, Lujean Ahmad, Les Norman, Brian Axcell, Luiz Wrobel, and Sheng Dai. "Thermoelectric generator (TEG) technologies and applications." International Journal of Thermofluids 9 (2021): 100063.

[4] Tohidi, Farzad, Shahriyar Ghazanfari Holagh, and Ata Chitsaz. "Thermoelectric Generators: A comprehensive review of characteristics and applications." Applied Thermal Engineering 201 (2022): 117793.

[5] Masoumi, Saeed, Seamus O'Shaughnessy, and Amir Pakdel. "Organic-based flexible thermoelectric generators: From materials to devices." Nano Energy 92 (2022): 106774.

[6] Wang, Ke, Mortaza Saeidi-Javash, Minxiang Zeng, Zeyu Liu, Yanliang Zhang, Tengfei Luo, and Alexander W. Dowling. "Gaussian process regression machine learning models for photonic sintering." In Computer Aided Chemical Engineering, vol. 49, pp. 1819-1824. Elsevier, 2022.

[7] Mortaza, Saeidi‐Javash, Wang, Ke, Minxiang Zeng, Tengfei Luo, Alexander W. Dowling, and Yanliang Zhang. "Machine learning-assisted ultrafast flash sintering of high-performance and flexible silver–selenide thermoelectric devices." Energy & Environmental Science 15, no. 12 (2022): 5093-5104.

[8] Wang, Ke, and Alexander W. Dowling. "Bayesian optimization for chemical products and functional materials." Current Opinion in Chemical Engineering 36 (2022): 100728.

[9] Wang, Ke, Minxiang Zeng, Jialu Wang, Wenjie Shang, Yanliang Zhang, Tengfei Luo, and Alexander W. Dowling. "When physics-informed data analytics outperforms black-box machine learning: A case study in thickness control for additive manufacturing." Digital Chemical Engineering 6 (2023): 100076.

[10] Shang, Wenjie, Minxiang Zeng, Wang, Ke, A. N. M. Tanvir, Mortaza Saeidi‐Javash, Alexander Dowling, Tengfei Luo, and Yanliang Zhang. "Hybrid data‐driven discovery of high‐performance silver selenide‐based thermoelectric composites." Advanced Materials (2023): 2212230.

[11] Cheng, Zhongyu, Wang, Ke, Wenjie Shang, Ali Tanvir, David Go, Alexander Dowling, Tengfei Luo, and Yanliang Zhang. "Non-thermal plasma jet sintering of indium tin oxide (ITO) thin films based on Bayesian optimization." Bulletin of the American Physical Society (2023).

[12] Wang, Ke, Cheng Zhongyu, Wenjie Shang, AMN Tanvir, Tengfei Luo, Yanliang Zhang, Alexander W. Dowling, David B. Go. "Accelerating low-temperature processing of printed nanoinks using Bayesian optimization of non- thermal plasma jet sintering. " (in preparation)

[13] Kaidong Song, Guoyue Xu, Wang, Ke, Wenjie Shang, AMN Tanvir, Tengfei Luo, Alexander W. Dowling, Yanliang Zhang. "Machine learning-assisted direct ink writing of three-dimensional BiSbTe thermoelectric devices. " (in preparation)

[14] Wang, Ke, ANM Tanvir, Wenjie Shang, Md Omarsany Bappy, Qiang Jiang, Alexander W. Dowling, Tengfei Luo, Yanliang Zhang. "Machine learning-assisted co-optimization of BiSbTe composition and flash sintering. " (in preparation)