Research On Retrieval And Intelligent Design Techniques For 3D Engineering Models

Engineering CAD model is of great importance for engineering design and construction.With the expanding investment in engineering industry,the number of engineering CAD models is increasing rapidly,as well as the model scale.It poses great challenges to model management,consistency check and design.In order to solve the problems above,this thesis is dedicated to the model retrieval and intelligent design techniques for process plant CAD models.On one hand,model retrieval serves as a powerful tool for model management,consistency check and reuse.On the other hand,design efficiency can be greatly boosted by exploiting intelligent design.Accordingly,the representation and similarity measuring criteria for engineering CAD models are firstly proposed.Then,four mesuring algorithms are presented,as are a global similarity measurement,a partial retrieval algorithm,as well as two 2D engineering drawing&3D model matching algorithms.At last,a rule mining and intelligent prediction method is introduced.The major contributions are listed as follows:1.The representation and similarity measuring criteria for engineering CAD modelsIn accordance with the characteristics and graphical representations of engineering CAD models,a unified representation for heterogeneous engineering CAD models is firstly presented.As the core of engineering model lies in its topology structure,this representation seeks to transform CAD models into attribute graphs and evaluate model similarity through their graph similarity.Then,the similarity measuring criteria for engineering CAD models is proposed.Engineering CAD models could be considered as a match,if they share an identical topology structure.The model representation and similarity measuring criteria lay the groundwork for the following similarity measuring research.2.A topological relationship distribution based similarity algorithmThis thesis proposes a global similarity measurement based on Topological Relationship Distribution(TRD)feature.First,a Relation Tree(RT)model for extracting TRD is proposed.The RT model attains and stores a process plant model's relationship statistics by traversing all attributes and topological relationships of components.Second,as to achieve the comparable feature vectors,standardization is performed via mapping relationship statistics into vector space.Last,a hybrid similarity function combining both directional and numerical differences in feature vectors is proposed to evaluate model similarity.3.An edit distance based partial retrieval algorithmAs Min-Edit Distance is a common solution of graph similarity,this thesis introduces an edit distance based partial retrieval algorithm.First,transform all process plant models and the model to be retrieved into attribute graph structures.Second,compute the minimum edit distances between each model.and the model to be retrieved.Last,each distance is compared with a predefined threshold to determine whether the model exists in the current process plant model or not.4.Topology feature based 2D engineering drawing and 3D model matching algorithmsThis thesis presents two topology feature based algorithms to calculate the matching degree between 2D engineering drawing and 3D model,as are a feature indentification based similarity measurement and a SimHash based similarity measurement.First,both of these algorithms preprocess 2D engineering drawing and 3D model into graph structures.Then,they extract each model's relationship types from the graph and regard them as a model's topology feature.At last,the alogrithms calculate feature similarity by their feature vectors and hash values respectively to measure the matching degree between their corresponding models.5.A rule mining and intelligent prediction method in process plant designProcess plant design involves a variety of engineering backgrounds and specialized knowledge.With abundant of latent design rules not being explicitly extracted,the existing modeling methods present the disadvantages of low efficiency and model quality.In that sense,a Frequent-Type-Tree model for mining process plant design rules is proposed.In the offline stage,this model constructs a Frequent Type Tree by analyzing component attributes and topological correlations in plant models.Then in the online stage,it provides real-time intelligent predictions by querying the tree for future design.