Intelligent composites forming - simulations for faster, higher quality manufacture

智能复合材料成型 - 通过模拟实现更快、更高质量的制造

• 批准号: 2443421
• 金额: --
• 依托单位: University of Bristol
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2020
• 资助国家: 英国
• 起止时间: 2020 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2443421
• 关键词: Intelligent; composites; forming; simulations; faster

This project aims to explore an intelligent way to optimize and accelerate our computer assistant modelling tool which simulates the textile forming process. A machine learning (ML) based surrogate model is being developed, which aims to provide live prediction to fabric forming industry. This surrogate model is trained by a set of data generated by finite element (FE) simulation tool. Considering the high calculation cost when obtaining an accurate FE simulation data point, the method of sampling and supplementary point selecting should be well designed, in order to make the training cost as low as possible meanwhile control the predicting deviations.First of all, in the aspect of FE simulation, a way to the destination of reducing defect level in dry textile forming process is explored. A shell-membrane hybrid FE modelling tool is adopted to simulate the behaviour of textile during forming on an industry-inspired tool. A series of springs or other controlling method will be adopted to adjust the forming controls in the FE model, in order to simulate the different wrinkling and bridging level under different forming parameters. In current research, the positions and stiffnesses, together with the pressure applied on the top, are regarded as input parameters and can be modified to control the deformation of textile during forming. By providing a set of combinations of input parameters, hundreds of simulations will be conducted to obtain a data set, which will be used as the training set for surrogate model.The long-term research will explore the feasibility of the framework of using ML-based time-efficient surrogate model to assist forming process optimization. In the current work, the Gaussian Process Regression (GPR) method is used to develop the surrogate model, for its applicability on small data set problem. On the other hand, GPR method naturally features uncertainty quantification ability, which can be used to predict and quantify the potential variances in the forming process. This method will be tested and developed as a tool for our industry partners, which is expected to greatly reduce forming defects as well as shorten the parameter test period. With the maturation of this surrogate model, the model together with the entire method can be compiled into software and integrated in forming process equipment in the future. With the use of sensors, real-time parameters such as the local temperature, tensile force, and shear angle of fabrics and moulds can be detected and collected during the process. By importing these real-time data into the surrogate model, the software can calculate the point with the highest probability of producing the optimal result, so that the forming rig can fine-tune the control parameters to optimize the quality of the forming.This technology can not only be applied in the forming process. In the various steps of composites part production, such as compaction, curing or AFP layup, this technology can be used for real-time optimization of processing quality. It can be said that this technology is an important way for the intelligent upgrading of the manufacturing industry.The project will explore the possibility of applying artificial intelligence and machine learning methods to optimization of composites fabric forming process, and furthermore, other complex manufacturing process. In the context of the rapid development of machine learning theories and computing capabilities, the use of artificial intelligence for process optimization will be a major trend in the development of the manufacturing industry in the future. This will help the country's manufacturing level to continue to maintain its leading position in the world and bring a lot of benefits.

本项目旨在探索一种智能的方法来优化和加速我们的计算机辅助建模工具，该工具模拟纺织成型过程。开发了一种基于机器学习(ML)的代理模型，旨在为织物成型行业提供实时预测。该代理模型由有限元(FE)仿真工具生成的一组数据进行训练。考虑到获得准确的有限元模拟数据点的计算成本较高，应合理设计采样和补充点选择的方法，以使训练成本尽可能低，同时控制预测偏差。首先，在有限元模拟方面，探索了一条降低干法纺织品成形过程缺陷水平的途径。采用壳-膜混合有限元模拟工具，在一种受工业启发的工具上模拟纺织品成形过程中的行为。采用一系列弹簧或其他控制方法来调整有限元模型中的成形控制，以模拟不同成形参数下不同的起皱和桥联程度。在目前的研究中，将位置和刚度以及施加在顶部的压力作为输入参数，并可以对其进行修改以控制纺织品在成形过程中的变形。通过提供一组输入参数的组合，进行数百次模拟得到一个数据集，作为代理模型的训练集，长期研究将探索使用基于ML的时效性代理模型框架来辅助成形工艺优化的可行性。由于高斯过程回归(GPR)方法适用于小数据集问题，因此本文采用GPR方法来建立代理模型。另一方面，探地雷达方法具有天然的不确定性量化能力，可以用来预测和量化成形过程中的潜在差异。该方法将作为行业合作伙伴的工具进行测试和开发，预计将大大减少成形缺陷，并缩短参数测试周期。随着该代理模型的成熟，该模型和整个方法可以被编译成软件，并集成到未来的成形工艺设备中。通过传感器的使用，可以实时检测和采集织物和模具的局部温度、拉力和剪切角等参数。通过将这些实时数据输入到代理模型中，软件可以计算出产生最优结果的概率最大的点，从而使成形设备能够微调控制参数，以优化成形质量。在复合材料零件生产的各个步骤中，如压实、固化或AFP铺层，该技术可用于加工质量的实时优化。可以说，这项技术是制造业智能化升级的重要途径。该项目将探索将人工智能和机器学习方法应用于复合材料织物成型过程优化，进而应用于其他复杂的制造过程。在机器学习理论和计算能力快速发展的背景下，利用人工智能进行流程优化将是未来制造业发展的一大趋势。这将有助于该国制造业水平继续保持世界领先地位，并带来很多好处。