Automatic CAD model retrieval based on design documents using semantic processing and rule processing

Highlights

• Design documents that contain initial design information are used to retrieve relevant CAD models.

• Semantic gap between text in CAD models and text in design documents is resolved.

• Design information is converted to semantically structured information.

• Hidden design information in CAD models and design documents is extracted.

• The experimental results show that the proposed approach is better than keyword-based approach.

Abstract

Reuse of CAD model is important for manufacturing companies where similar products are produced. If existing CAD models have design information related to the design of a new product, their reuse saves costs and reduces time of product design. Since design documents generated in the early design stage contain product specifications and design directions for new product design, they can be used as input queries to retrieve existing CAD models. In this paper, we propose an approach to retrieving CAD models with design documents. A major challenge of the proposed approach was reduction of semantic gap between design documents and CAD models because design documents generated in the early design stage contain abstract design descriptions, while CAD models contain detailed design descriptions. In order to reduce the semantic gap between CAD models and design documents, we adopt semantic processing and rule processing. During the semantic processing, semantic representations are generated from the text in CAD models and text in design documents using a domain ontology and shallow natural language processing (NLP). Hidden design information is extracted from CAD models and design documents based on rule processing. The experimental results show that the proposed approach can outperform the keyword-based approach which does not overcome semantic gap.

Introduction

The product design process is roughly divided into two stages: early design and detail design. In the early design stage, engineers create initial design information described in sketches, rough 3D models, or design documents. Design documents at the early design stage generally include important information which can help engineers to create detailed design information. Design documents contain product specifications and design directions for a specific design using textual descriptions [1]. In the detail design stage, product design information is generally represented by CAD models, and it contains geometry, tolerance, and material information for a specific product design.

Reuse of CAD model is very important for manufacturing companies where similar products are produced. Engineers can efficiently make decisions for design details of new product from existing detailed design information of similar products. If existing CAD models have design information related to the design of a new product, their reuse saves costs and reduces time of product design. According to the statistics in industries [2], only about 20% of OEM's investment is on new design while about 80% is on the reuse of existing designs, with or without modification. Aberdeen Group [3] shows that design reuse helped top performing organizations to achieve 76% of their engineering targets and a 30% reduction in design time.

Since design documents generated in the early design stage contain product specifications and design directions for new product design, they can be used as input queries to retrieve existing CAD models. Most existing CAD model retrieval methods [4], [5], [6], [7] were limited to use design documents because they supported searching by 2D sketches or by 3D geometries. Searching by 2D sketches or by 3D geometry models requires 2D sketches or a detailed 3D model as an input query, respectively.

Product lifecycle management (PLM) systems provide engineers a function to manually create links between design documents and CAD models, but they do not have a function to automatically generate links based on the design information in design documents and CAD models. Engineers have to identify design information in design documents and CAD models to create links between them manually. It is a complex and time-consuming task if there were too many design documents and CAD models to be linked. It becomes worse if some design documents and CAD models were generated by different engineers. Engineers use the check-in/check-out function to manage their design outputs, but this function does not assist engineers to make meaningful links between the design outputs. CAD model retrieval approach based on design documents can help engineers to manage links between design documents and CAD models by using semantic processing and rule processing.

CAD models can contain sufficient text in the following circumstances: Commercial CAD systems provide predefined CAD objects (i.e., point, circle, surface) and those CAD objects have several attributes (i.e., name, type). Many manufacturing companies develop CAD templates or domain-specific user-defined features (UDF) and utilize them in the product design stage to reduce design cost and lead-time [8]. When engineers create CAD templates and UDFs, they use domain specific terms in order to share design intentions of the CAD templates and UDFs with other engineers. In addition, engineers give special names to specific CAD objects in a CAD model so that they can find the specific CAD objects quickly during the CAD modeling process. CAD models, therefore, include enough text to be searched by matching text in design documents.

There exists a semantic gap between text in CAD models and text in design documents. This semantic gap can occur for the following two reasons. First, domain-specific terms may have semantic ambiguity due to abbreviations, acronyms, synonyms, and polysemy. For example, the term “SW” in the tire design domain is used for different concepts (Sidewall or SectionWidth). The semantic ambiguity of domain-specific terms should be removed to improve retrieval performance. Second, text in design documents in the early design stage includes abstract design descriptions while text in CAD models has detailed design descriptions. For instance, it is difficult to directly match “3 channel patterns”, written in a tire design document and “circumferential groove”, referring to CAD objects in a tire CAD model. In order to match these two terms, it is necessary to know that “3 channel patterns” means a tread pattern that has 3 circumferential grooves. Thus, design information hidden behind text in CAD models and text in design documents should be recognized.

The semantic ambiguity of terms in engineering text can be resolved by using a domain ontology. Several studies [1], [9], [10], [11] have used a domain ontology and a concept distance measurement method to handle the semantic ambiguity issues. They extracted concepts from terms in engineering texts using domain ontology. If a word in a text matched multiple concepts, they removed the semantic ambiguity using concept distance between different concepts in the domain ontology. Min et al. [12] used WordNet to disambiguate meaning of extracted text that was associated with a particular 3D model. Yu and Hsu [13] used a domain corpus to moderate the semantic ambiguity of terms in CAD drawings.

Hidden design information in CAD models and design documents can be explicitly represented by using a domain ontology and rules together. Some studies have used a domain ontology and rules to represent design knowledge and to infer new knowledge. Wang et al. [14] used a 3D model ontology and relationship rules to search 3D models. Kim et al. [15] used an assembly ontology and rules to infer new assembly information.

In this paper, we propose an automatic CAD model retrieval approach based on design documents that contain initial design information. The proposed approach reduces the semantic gap between CAD models and design documents by using semantic processing and rule processing. Semantic processing is used to generate semantic representations from the text in CAD models and text in design documents using a domain ontology and shallow natural language processing (NLP). Since a semantic representation consists of concepts and their relationships, design information in CAD models and design documents can be converted to semantically structured information. The semantic ambiguity of terms is resolved by heuristic rules and a concept distance measurement method. Hidden design information in CAD models and design documents are extracted by a given rule base. Then, similarity calculation is conducted between semantic representations to search CAD models relevant to a design document. Tire design examples are used in the paper to demonstrate the feasibility of the proposed approach.

The remainder of this paper is organized as follows. Section 2 presents related works. Section 3 describes the overall architecture of the proposed approach. Section 4 provides an analysis of CAD models and design documents. Section 4 also shows technical details of how to generate semantic representations, how to reduce semantic gaps, and how to search CAD models with similarity calculation. Section 5 describes the implementation of the proposed approach. Section 6 provides an evaluation of our approach with real industry data. Section 7 discusses limitations of the proposed approach. Section 8 provides the conclusion and suggestions for future work.