基于雕塑形体概率图文法的产品外观自动反求创新理论与方法

It's a great challenge for traditional reverse engineering method to deal with more and more complex objects in an efficient and effective way, which stir up the exploration for new theoretical basis. In this project, directing at the core of the problem-complex shape reverse design, a new automatic reverse workflow is put forth based on probabilistic graph grammar of sculptured objects.A unified framework to representation, learning, and parsing complex shapes is obtained, which is used to guide the process of reverse engineering. The extraction of design intent is equivalent to infering the parse graph of the object, and the mesh segment, feature recognition, constraint recognition, and so on, are seamlessly integrated together to get a most reasonable interpretation of the point clouds.In our project, the strategy is to divide and rule the problem, and object category is the base to resolve. In this way, we solve the problem of how to judge a reverse result the best one for the first time in Bayesian framework, which laid the foundation of reconstruction automation. The knowledge of industrial design is difficult to captured by other methods, which can be easily expressed as a graph grammar learned from small sample sets. Besides efficiency, the reconstruction results of our method are more qualified, and can be used more effectively for design problem. The deep design law may be found from the unfied structural reconstruction of exist products, and some new design pattern can be distinguished from the failure of the automatic process. The deep understanding of the product category is inevitable for our new method,which made it more suitable for design than any other reverse scheme. The method of our project can be broadly used in fields of product inspection on-line, part restoration, and reconstruction of archaeological fragments, in a unique style.

传统的反求设计，不仅实施过程繁杂、结果的一致性不能保证，更难以积累现有产品的设计知识与经验，几乎背负了"拷贝、抄袭"的恶名。本项目针对复杂产品外观设计，提出了一套新的、自动化反求创新设计理论与方法。以雕塑形体的概率图文法为基础，建立了复杂产品外观的设计表达、学习、理解及创新的统一框架；设计意图的理解，转化为产品图文法的推断；并在这种产品图文法的指导下，自动创新出合理设计。本项目提出的分层次、多视图文法模型，完整表达了设计人员对产品外观的理解，并采用分层、分解、分步策略设计可行的产品文法学习、解译算法，保证了反求设计严密、精确地自动实施。项目成果，不仅可用于产品的创新设计，在产品检测、修复、文物复原等领域，也具有独到的广泛应用前景。 传统反求流程中的区域分割、特征识别、约束识别等环节在图文法框架下融为一体、自动实现，以获取对测量数据的最合理解释与重建。

传统反求,过程繁杂、结果难以重用,很难恢复原有的设计意图。数据驱动、基于知识的反求是突破当前瓶颈的有效途径。本项目针对复杂产品外观设计，以经验数据积累为基础,借助概率图文法学习理解产品形状,以实现自动化反求建模与创新设计。通过1D/2D/3D形状的概率图文法,积累产品结构的多层次分解知识;对产品概率图文法中典型样例,添加可调模板,以此表达产品的建模特征、建模过程知识。在此基础上,首次实现了从测量数据到参数化特征模型的自动反求重建,解决了反求领域的一个难题。在产品质量控制方面，利用概率图文法表达变形特征分布,提出了一种柔性部件无夹具检测的全新方法。本项目成果，不仅可用于产品的创新设计，在产品检测、修复、文物复原等领域，也具有独到的广泛应用前景。