Applications of AI to Computer-Aided Engineering

My research includes applications of artificial intelligence to problems within the broad domain of Computer-Aided Engineering (CAE). These problems are mostly related to data management in large engineering digital libraries, such as automated engineering format classification, and search and retrieval of engineering artifacts, such as Computer-Aided Design (CAD) models based upon structural similarity.

Automated Engineering Format Classification

There are hundreds of distinct 3D, CAD, and engineering file formats. As engineering design and analysis has become increasingly digital, the proliferation of file formats has created many problems for data preservation, data exchange, and interoperability. In some situations, physical file objects exist on legacy media and must be identified and interpreted for reuse. In other cases, file objects may have varying representational expressiveness. In an article in the journal Computers & Graphics titled A Flexible and Extensible Approach to Automated CAD/CAM Format Classification, we introduced the problem of automated file recognition and classification in emerging digital engineering environments, where all design, manufacturing and production activities are digital from the start. The result is that massive quantities and varieties of data objects are created during the product life-cycle. Additionally, we presented an approach to automated identification of engineering file formats that operates independent of any modeling tools and can identify families of related file objects as well as variations in versions. This problem is challenging as it cannot assume any a priori knowledge about the nature of the physical file object. Applications for these methods include support for a number of emerging applications in areas such as forensic analysis, data translation, as well as digital curation and long-term data management.

Automated Management of Computer Aided Design Digital Libraries

My earliest research experience was focused on problems related to search and retrieval of engineering artifacts from computer-aided design (CAD) digital libraries based upon similarity, essentially a 3D search where the query is a 3D model of an engineering artifact, and the objective of the search is to retrieve objects that are similar to the query object. This work began with my MS Thesis, Intelligent Retrieval of Solid Models, which explored representations of the design features of an artifact along with algorithms for comparing objects to determine which are similar in design. This MS Thesis introduced a graph-theoretic approach involving a graph called a Model Dependency Graph (MDG) representing the design features of a solid model and the dependencies among those features. Additionally, algorithms for comparing these graphs were developed including a local hill climbing algorithm as well as an algorithm called A* Subgraph Isomorphism Checker (ASIC). Further details of the approach are presented in the paper Resolving Non-Uniqueness in Design Feature Histories, including proofs of various properties of the MDGs as well as defining an approximate algorithm for checking whether two MDGs are D-morphic or subgraph D-morphic. These and related activities are discussed further in Managing Digital Libraries for Computer-Aided Design.

The early research discussed above focused on comparing 3D solid models based on design feature histories, i.e., the similarity of artifacts in terms of how they were designed. Later, we explored how to find engineering artifacts that would be machined similarly. This is a harder problem. Our approach begins by extracting STEP AP 224 machining features from the CAD models of the artifacts, computing feature interactions, and forming an MDG from the features and their interactions. We developed a local search algorithm for the NP-Hard problem, Largest Common Subgraph, and defined a measure of similarity based on the size of the Largest Common Subgraph of the MDG of a query artifact, and a potential match from the digital library. This work was presented in the paper Machining Feature-Based Comparison of Mechanical Parts, presented at the International Conference on Shape Modeling and Applications. We further refined the matching approach, developing the Greedy Subgraph Isomorphism Checker (GSIC), among other improvements in a journal article published in Artificial Intelligence for Engineering Design, Analysis and Manufacturing titled An Approach to a Feature-based Comparison of Solid Models of Machined Parts. This work also resulted in a patent: Method for Comparing Solid Models.

Selected Publications