SBIR Phase I: Neural Component Architecture to Accelerate Modeling & Simulation

这个小型企业创新研究（SBIR）第一阶段项目的更广泛影响将来自于在工程和医疗保健等广泛领域开发新的，更有效的设计方面节省大量成本和时间。如果成功，该项目将通过用先进的数字模拟来代表系统的物理组件，使从汽车到航空航天部件和制药的所有模拟运行速度提高100倍。 到目前为止，软件不兼容性限制了这种建模的发展。本项目将通过先进的计算和编译技术解决这个问题，从而证明一种新的设计过程的可行性，显着降低成本。 这个小企业创新研究第一阶段项目将展示在模块化系统中使用神经组件的可行性。 我们将结合联合收割机的成功代理模型优化和神经常微分方程，允许基于组件的微分代数方程模型与自动模型降阶通过潜在的神经网络。我们的想法是使用我们在差分编程和学习软件方面的最新进展，将复杂模型构建为模块化预设计仿真组件的组装，以允许自动训练神经模型降阶，从而加速大型因果模型的解决方案。两种机器学习方法在加速传统的机械建模工作流方面具有很好的前景：代理优化和神经微分方程。 在这个项目中，我们将把这些组件集成到一个原型系统中。这个奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact of this Small Business Innovation Research (SBIR) Phase I project will result from enabling significant cost and time savings in developing new, more efficient designs in broad fields such as engineering and healthcare. If successful, the project will enable simulations of everything from automobiles to aerospace components and pharmaceuticals to run up to 100 times faster by representing a physical component of a system with an advanced digital analogue. To date, software incompatibilities have limited the development of this kind of modeling. This project will solve this problem through advanced computational and compiler techniques, and thereby demonstrate the feasibility of a new kind of design process with significant cost reductions. This Small Business Innovation Research Phase I project will demonstrate the feasibility of using neural components in a modular system. We will combine the successes of surrogate model optimization and neural ODEs to allow for component-based differential-algebraic equation models with automated model order reduction through a latent diffeq. The idea is to build complex models as an assembly of modular pre-designed simulation components using our recent advances in differential programming and learning software to allow for automated training of neural model order reduction for accelerating the solution of large acausal models. Two machine learning methods have promising prospects for accelerating traditional mechanistic modeling workflows: surrogate optimization and neural differential equations. In this project, we will integrate these components into a prototype system.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.