Data-driven Friction Models for Simulation and Material Fabrication

用于模拟和材料制造的数据驱动摩擦模型

• 批准号: 571411-2021
• 负责人: Andrews, SheldonP
• 金额: $ 3.28万
• 依托单位: École de technologie supérieure
• 依托单位国家: 加拿大
• 项目类别: Alliance Grants
• 财政年份: 2022
• 资助国家: 加拿大
• 起止时间: 2022-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=754215
• 关键词: Data; driven; Friction; Models; Simulation

Friction modeling is an important consideration when designing machines, controlling robots, and for many healthcare applications. Examples include constructing more comfortable prostheses, predicting the physical behaviour of objects during robotic manipulation, or understanding human-machine interactions between a surgeon and a medical robot. However, standard friction models fail to capture the rich behaviour arising from interactions between real-world surfaces. Furthermore, common models, such as Coulomb, are empirical and hence they are limited in their capability for computational design applications based on surface microgeometry and related material properties.This research project is about developing a better understanding of the physical world and simulating friction using a data-driven approach. Furthermore, an improved understanding will lead to improved models for frictional simulation that can faithfully reproduce real-world behaviour. We then plan to leverage these simulations for computational inverse design applications, where frictional surfaces may be fabricated according to specifications or tasks defined in a numerical optimization pipeline.Our project combines interdisciplinary themes and a team of experts with complementary skills in contact simulation, data-driven modeling for virtual environments, and computational fabrication. The main objectives will be achieved by three sub-projects, which are summarized below.1. Measurement and Analysis of Friction: Humans have the amazing capability to identify the frictional properties of objects simply from the sense of touch. Inspired by this, we will use scans of surface microgeometry to infer frictional behaviour between surfaces by developing data-driven friction models. We will also use robots to collect large quantities of sliding data of different objects and artificial surfaces to help with the ultimate challenge of solving the inverse modelling problem-- designing friction from micro-surface structures. Additionally, this dataset will provide an invaluable resource for validating and benchmarking physics simulation involving friction, which is of interest for many research communities.2. Geometry driven and Data-driven Models of Frictional Behaviour: Formulating novel friction models and new computational approaches will be a key component of our work. We are motivated by recent machine learning techniques for their ability to predict and regress complex, non-linear functions. Aggregate friction effects arising from micro-surface interactions represent such a function, and we therefore intend to leverage modern techniques to learn mappings from surface micro-geometry to aggregate friction behaviour. The result will be a new class of models for realistic friction simulation.3. Computational Fabrication of Materials by Inverse Friction Modeling: Although friction is a functional aspect of many physical systems, methods for designing and fabricating rough surfaces that meet user-specified functional requirements are nearly non-existent. This presents us with an interesting research opportunity to develop inverse modeling and optimization techniques that are based on our data-driven friction models that will facilitate new technologies for automatic design and fabrication of friction. This has a wide range of applications, such as creating artificial skin for gloves to better grasp, designing task-specific shoe soles (e.g., for curling or other sports), or fabricating mechanical macrostructure 'coatings' of objects without using chemicals.

在设计机器、控制机器人和许多医疗保健应用时，摩擦建模是一个重要的考虑因素。例如，构建更舒适的假肢，预测机器人操作过程中物体的物理行为，或者理解外科医生和医疗机器人之间的人机交互。然而，标准的摩擦模型无法捕捉到现实世界表面之间相互作用产生的丰富行为。此外，常见的模型，如库仑，是经验的，因此它们在基于表面微观几何和相关材料特性的计算设计应用中的能力受到限制。这个研究项目是为了更好地理解物理世界，并使用数据驱动的方法模拟摩擦。此外，改进的理解将导致改进的摩擦模拟模型，可以忠实地再现现实世界的行为。然后，我们计划将这些模拟用于计算逆设计应用，其中摩擦表面可以根据数值优化管道中定义的规格或任务制造。我们的项目结合了跨学科的主题和专家团队，他们在接触模拟、虚拟环境的数据驱动建模和计算制造方面具有互补的技能。主要目标将通过三个次级项目实现，概述如下1。摩擦的测量和分析：人类有一种惊人的能力，仅仅通过触觉就能识别物体的摩擦特性。受此启发，我们将通过开发数据驱动的摩擦模型，使用表面微观几何扫描来推断表面之间的摩擦行为。我们还将使用机器人收集不同物体和人造表面的大量滑动数据，以帮助解决反建模问题的终极挑战-从微表面结构设计摩擦。此外，该数据集将为验证和基准化涉及摩擦的物理模拟提供宝贵的资源，这是许多研究团体感兴趣的。几何驱动和数据驱动的摩擦行为模型：制定新的摩擦模型和新的计算方法将是我们工作的关键组成部分。我们被最近的机器学习技术所激励，因为它们能够预测和回归复杂的非线性函数。微表面相互作用产生的聚集摩擦效应代表了这样一个函数，因此我们打算利用现代技术来学习从表面微观几何到聚集摩擦行为的映射。结果将为现实摩擦模拟提供一类新的模型。通过逆向摩擦建模的材料计算制造：尽管摩擦是许多物理系统的一个功能方面，但设计和制造满足用户指定功能要求的粗糙表面的方法几乎不存在。这为我们提供了一个有趣的研究机会，开发基于数据驱动的摩擦模型的逆建模和优化技术，这将促进自动设计和制造摩擦的新技术。这有广泛的应用，比如为手套创造更好的人造皮肤，设计特定任务的鞋底（例如，用于冰壶或其他运动），或者在不使用化学物质的情况下制造物体的机械宏观结构“涂层”。