Coupling of Real-World Data and Fast Response Algorithms to Improve Simulation Correlations and Optimise Construction

该项目将在卡特彼勒英国发动机有限公司（CAT）内部开发和部署MODSUITE。MODSUITE是一种新颖的数据分析和优化工具，将用于分析发动机和机器测试数据，以应对发动机内部发生的过程的基于预测物理的模型。该工具将应用新颖的优化和快速响应算法来系统地改进和量化模型中的不确定性。这将使CAT能够更有效地利用数据，并使用模型来帮助优化发动机性能，包括气体和颗粒（烟灰）排放。该项目分为四个主要合作伙伴。剑桥大学的计算建模（CoMo）小组将进行基础研究和开发，将MODSUITE应用于cat提出的应用。剑桥计算建模有限公司（CMCL）是一家工程软件和服务公司，将专注于开发用户界面和测试该软件的应用。CAT和博格华纳有限公司（见www.borgwarner.com）将提供实际应用的实验数据和模型，以支持软件测试。Federal Mogul Ltd.（见www.federalmogul.com）将作为CAT的一些模型开发的分包商。该软件将使用三个演示应用程序开发和测试，以增加系统复杂性的水平。该项目的初始阶段将在剑桥的CMCL工厂进行。在初步测试成功完成后，该软件将通过基于互联网的用户界面部署在CAT。这将使软件在CMCL上运行，同时它仍处于开发阶段，而应用程序模型将在CAT上运行，使用分布式计算技术，允许CAT利用其处置的大型计算资源并保持对模型和数据的控制。基于web的界面和分布式计算设计提供了一个简单但功能强大的解决方案。CoMo集团最初将通过研究MODSUITE目前尚未实施的优化方法来为该项目做出贡献。将识别并添加与CAT所提供的应用程序最相关的应用程序。该小组将扩展MODSUITE，使其能够自动读取和处理CAT提供的大量实验数据，促进数据驱动模型的创建。将研究和实施更广泛的不同响应面方法和自动响应生成和选择方法。这将增加工具的多功能性，以便为每个特定的测试应用生成合适的响应面。这些改进将促进生成合适的数据驱动模型以及快速代理模型的能力。将开发先进的优化方法，重点是自校准和鲁棒性，以提供一致和可靠的结果，而不需要特定算法的专家知识。这些先进的算法将与不确定性传播和分析工具相结合，以量化模型中的不确定性。modsuite代码将被调整，以允许它在分布式计算系统上运行。

The project will develop and deploy MODSUITE within the Caterpillar UK Engines Company Ltd. (CAT). MODSUITE is anovel data analysis and optimisation tool and will be applied to analyse engine and machine test data against predictivephysics-based models of the processes occurring within the engine. The tool will apply novel optimisation and fastresponsealgorithms to systematically refine and quantify the uncertainty within the models. This will enable CAT to use themake more effective use of the data and use the models to help optimise engine performance, including gaseous andparticulate (soot) emissions.The project is split between four main partners. The Computational Modelling (CoMo) Group at the University of Cambridgewill perform the fundamental research and development required to apply MODSUITE to the applications presented byCAT. Cambridge Computational Modelling Ltd. (CMCL), an engineering software and services company, will focus ondeveloping a user interface and testing the application of the software. CAT and BorgWarner Ltd. (seewww.borgwarner.com) will provide experimental data and models for real applications to support the software testing.Federal Mogul Ltd. (see www.federalmogul.com) will act as a subcontractor to CAT for some of the model development.The software will be developed and tested using three demonstration applications, at increasing levels of systemcomplexity. The initial phase of the project will be performed at CMCL facilities in Cambridge. Following successfulcompletion of the initial testing, the software will be deployed at CAT via an internet based user interface. This will enablethe software to be run at CMCL whilst it is still in development, whereas the application models will be run at CAT usingdistributed computing technology, allowing CAT to harness the large computing resource at their disposal and maintaincontrol over the models and data. The web-based interface and distributed computing design offer a simple, but powerfulsolution.The CoMo Group will initially contribute to the project by investigating optimisation methods that are not currentlyimplemented within MODSUITE. The ones that are most relevant to the applications presented by CAT will be identifiedand added. The group will extend MODSUITE so that it can automatically read and process the large quantities ofexperimental data made available by CAT, facilitating the creation of data driven models. A wider range of differentresponse surface methods and an automated response generation and selection method will be investigated andimplemented. This will increase the versatility of the tool such that suitable response surfaces can be generated for eachspecific test application. These improvements will facilitate the ability to generate suitable data driven models as well asfast surrogate models. Advanced optimisation methods will be developed with a focus on self-calibration and robustness toprovide consistent and reliable results without the need for expert knowledge of the specific algorithms. These advancedalgorithms will be combined with uncertainty propagation and analysis tools to quantify the uncertainties in the model. TheMODSUITE code will be adapted to allow it to run over a distributed computing system.