ARPA-E Archived Funding Opportunities

The Advanced Research Projects Agency – Energy (ARPA–E) intends to issue a new Funding Opportunity Announcement (FOA) that would seek to enhance the pace of energy innovation by accelerating the incorporation of machine learning into the engineering design processes for energy technologies and systems.

In order to organize the anticipated efforts, a simplified engineering design process framework has been adopted (Figure 1 - see attached document). Within the context of this framework, it is possible to conceptualize how machine learning tools would help engineers to execute and solve several general mathematical optimization problems, common to many (perhaps most) engineering design processes, in a manner that dramatically accelerates the pace of energy innovation.

The envisioned program would seek to enhance several aspects of the design process via machine-learning tools:

1) Hypothesis Generation Tools: Enhance the creativity of the hypothesis generation (i.e. conceptual design) process by helping engineers develop new concepts and by enabling the consideration of a larger and more diverse set of design options. Many of the design problems at this stage of the process can be characterized as Mixed Integer Non-Linear Optimization problems;

2) Hypothesis Evaluation Tools: Enhance the efficiency of the high-fidelity evaluation (i.e. detailed design) process by accelerating the high-fidelity analysis and optimization of the hypothesized solution concepts. Many of the design problems at this stage of the process can be characterized as Non-Linear Constrained Optimization problems; and

3) Inverse Design Tools:Reduce (ideally eliminate) design iteration by developing the capability to execute “inverse design” processes in which the product design is effectively expressed as an explicit function of the problem statement.

ARPA-E envisions posing several challenge problems to motivate the development of enhanced design processes/tools. These challenge problems are in areas that ARPA-E feels to be of significant importance and for which it feels that adequate data either are available or can be generated during the program. Note: the intended FOA evaluation criteria are expected be principally focused on the potential impact that the proposed ML-enhanced design tools might have on future general engineering design processes through their potential to reduce design cost, time and risk and/or increase design performance, robustness and novelty.

Potential design challenge problems include the following:

· Hypothesis Generation (i.e. Conceptual Design)

• Thermodynamic Cycles/Chemical Processes (e.g. Gas Separations)
• Electrical Circuits
• Materials/Molecules

· Hypothesis Evaluation (I.e. Detailed Design)

• Fuel/Electrolyzer Cells
• Gas Compressors
• Solar Cells

· Inverse Design

• Aerodynamic Surfaces
• Optical Devices

ARPA-E envisions projects that seek to develop machine learning enhanced tools that facilitate the solution to one of the above challenge problems. It is also envisioned that, for any of the above categories, there will be an option for applicant teams to propose their own, alternative challenge problem so long as it is sufficiently justified (i. e. that it is both highly impactful and especially appropriate/ripe for enhancement via machine learning). It is expected that each proposal would explicitly identify a selected challenge problem, an anticipated ML-enhanced solution approach, a data[1] acquisition/generation strategy, the major development risks, and an anticipated path to market for the design tool / software to be developed. It is important to note that ARPA-E does not envision developing prototypes of physical systems through this FOA – the focus is on developing the ML-enhanced design tools only.

As described in more detail below, the purpose of this teaming announcement is to facilitate the formation of new project teams to respond to the pending FOA. The FOA will provide specific program goals, technical metrics, selection criteria, and other terms and requirements. However, for purposes of the Teaming Partner List, a summary of the currently anticipated scope is provided below.

In order to realize the envisioned program goals, ARPA‐E aims to bring together diverse engineering and scientific communities. These communities include, but are not limited to machine learning, mathematics/optimization, computer science, software, and energy (e.g. mechanical, chemical, materials, or electrical) engineering.

As a general matter, ARPA-E strongly encourages outstanding scientists and engineers from different organizations, scientific disciplines, and technology sectors to form project teams. Interdisciplinary and cross-sector collaboration spanning organizational boundaries enables and accelerates the achievement of scientific and technological outcomes that were previously viewed as extremely difficult, if not impossible.

The Teaming Partner List is being compiled to facilitate the formation of new project teams. The Teaming Partner List will be available on ARPA-E eXCHANGE (http://arpa-e-foa.energy.gov), ARPA-E’s online application portal, starting January 15, 2019. The Teaming Partner List will be updated periodically, until the close of the Full Application period, to reflect new Teaming Partners who have provided their information.

Any organization that would like to be included on this list should complete all required fields in the following link: https://arpa-e-foa.energy.gov/Applicantprofile.aspx. Required information includes: Organization Name, Contact Name, Contact Address, Contact Email, Contact Phone, Organization Type, Area of Technical Expertise, and Brief Description of Capabilities.

By submitting a response to this Notice, you consent to the publication of the above-referenced information. By facilitating this Teaming Partner List, ARPA-E does not endorse or otherwise evaluate the qualifications of the entities that self-identify themselves for placement on the Teaming Partner List.  ARPA-E will not pay for the provision of any information, nor will it compensate any respondents for the development of such information. Responses submitted to other email addresses or by other means will not be considered.

[1] In the interest of minimizing the cost of acquiring/generating training data, it is anticipated that the vast majority of the “data” used in the development of the desired tools will be generated with physics-based models.