Machine Learning and Optimal Experimental Design for Thermodynamic Property Modeling

对于化学和能源工程中的许多任务，热力学性质的准确知识（例如，压力和温度以及密度和声速），并且所涉及的流体的相行为起着关键作用。在科学中，这些性质是基本理解化学物理行为和开发预测模型所必需的。对于工业而言，热力学性质是设计安全和可持续的工艺和机械的基础。然而，使用状态方程（EOS）的性质计算的质量在很大程度上取决于实验数据的可用性和准确性。这些数据的测量通常是在密集的测量点网格框架内进行的，这提供了一个全面的数据集。然而，为了开发一个准确的EOS，这种方法是耗时的，而目前还不清楚是否所有的数据最终是实质性的模型开发。因此，所需的时间和财政支出使得可靠模型的生成相当有限。考虑到这一点，它是非常可取的，以显着减少模型的开发时间，通过限制实验数据量所需的程度，并涉及功能的形式，这使得短的计算时间的应用程序在过程模拟。因此，该研究项目的主要目标是通过实现（1）可解释的机器学习（ML）之间的特定相互作用来解决上述问题，以找到EOS的理想函数形式，（2）最佳实验设计以找到最合适的测量点和（3）实际实验。一个潜在的工作流程可以想象如下：从初始的热力学性质测量开始，基于ML的EOS建模用于创建第一个函数形式。此表格用于预测下一个信息量最大的测量值，然后可以将其用作进一步EOS建模的输入。何时终止此工作流程是项目研究计划的固有部分。该项目的一个重要产出是一个用于热力学测量规划和模型开发的现场软件工具，该工具考虑了测量工作、模型准确性和可解释性。

For many tasks in chemical and energy engineering, the accurate knowledge of thermodynamic properties (e.g., pressure and temperature with density and speed of sound) and the phase behavior of the involved fluids plays a key role. In science, such properties are required for the basic understanding of chemical-physical behavior and for the development of predictive models. For industry, thermodynamic properties are the basis for the design of safe and sustainable processes and machinery. However, the quality of property calculations using equations of state (EOS) depends largely on the availability and accuracy of experimental data. Measurements of such data are often carried out within the frame of a dense grid of measurement points, which delivers a comprehensive data set. Nevertheless, with the aim to develop an accurate EOS, this approach is time-consuming, while it is unclear whether all data are ultimately substantial to the model development. As a result, the required time and financial expenditure makes the generation of reliable models rather limited. Considering this, it is highly desirable to significantly reduce the model development time by limiting the amount of experimental data to the required extent and to involve functional forms, which enable short computing times for the application in process simulation. Therefore, the major goal of the research project is to tackle the aforementioned issues by realizing a specific interplay between (1) interpretable machine learning (ML) to find the ideal functional form of the EOS, (2) optimal experimental design to find the most appropriate measurement points and (3) the actual experiment. A potential workflow can be imagined as follows: Starting from initial thermodynamic property measurements, ML-based EOS modeling is used to create a first functional form. This form is used to predict the next most informative measurements, which can then be used as input for further EOS modeling. When to terminate this workflow is inherent part of the project's research schedule. One important output of the project is an in-situ software tool for thermodynamic measurement planning and model development, which considers the measurement effort, model accuracy and interpretability.