Research On Question Answering System Of Product Part Design Based On Sketch Recognition

Industrial design plays a crucial role in promoting the transformation and upgrading of the manufacturing industry,but the current design capabilities in the manufacturing sector still needs to be improved.In the process of product research and development,it is common for industrial designers to fail to retrieve the expected parts due to lack of relevant knowledge in the conceptual design stage,resulting in modeling design that does not meet the requirements of structural design and manufacturing and processing,which affects the product development process.However,the existing CAD system is still difficult to provide effective support for designers in the retrieval and matching of design knowledge in the early stage of design conception,conceptual design and so on.This paper presents a question answering system for product part design based on sketch recognition.The system can use the sketch to retrieve the expected part when the user knows the function and shape of the part but does not know the name of the part and cannot accurately retrieve the part.Besides,it will also provide the basic information,three-dimensional model of the part and other related parts information,which can provide relevant knowledge for designers in the conceptual design stage,accelerate the product design process,and promote design reuse.The main work of this paper is as follows:Firstly,by analyzing the current research status in the field of part recognition,researching the existing image recognition algorithms,and combining the characteristics of the part sketches in this article,the part sketch model（VGG-Caps Net）which combines pre-training network（VGG）and capsule network（Caps Net）is proposed.The between-group experiment and within-group experiment were designed,and two comparison models,the r VGG-13 model and the r CNN-13 model,were proposed.Secondly,taking the chemical fiber winding machine as the research object,five designers were recruited to sketch parts,and build 23 kinds of freehand sketch of parts including standard parts and non-standard parts.The part sketch recognition model of VGG-Caps Net was constructed,and the recognition results were compared with the other two models.The experimental results show that the average accuracy of the VGG-Caps Net model in the between-group and within-group experiments is higher than the other two models.Then,the construction process of the part knowledge graph was proposed,the data source and data storage method of the part knowledge graph were determined,and the definitions of the part entities,attributes and relationships among entities were defined.The parts related information and triad data were collected,and the data were stored in the graph database to complete the construction of the part knowledge graph in this paper.Finally,the Python language and related framework are used to develop the prototype of the question answering system,and the function of retrieving parts based on sketches and providing parts basic information,three-dimensional model and relevant parts information is realized.Finally,the application of the question answering system is verified to prove its practicability and effectiveness.