Mark R. Henderson

Automatic form-feature recognition using neural-network-based techniques on boundary representations of solid models

Abstract A new technique for performing form-feature recognition using the principles of neural networks is discussed. Neural nets require parallel input of data, which, in this case, are B-rep solid models of parts. An input format has been developed which includes face descriptions and face-face relationships. An algorithm for recognition using neural-net-based techniques has been developed, and a suitable net architecture, which is similar to the multilayer perceptron in function and which implements the algorithm, has been designed. The net architecture is described, and a few examples are presented which highlight the strengths and weaknesses of the recognition algorithm.

Graph-based extraction of protrusions and depressions from boundary representations

Abstract A feature extraction technique to isolate protrusions and depressions from boundary models is discussed. The technique involves identification of faces with multiple edge loops as candidates for entrance faces of such features. It is observed that protrusions and depressions constitute biconnected components in the edge-face graphs of the boundary models. Heuristics are used to enhance an algorithm to decompose the edge-face graph of a boundary representation into its biconnected components. Advantages and disadvantages of this method are compared against the classic graph theoretic algorithm to obtain biconnected components from a graph.

Three-dimensional shape pattern recognition using vertex-edge graphs

Abstract A method for computer recognition of shape patterns from a three-dimensional (3D) boundary representation of a solid object is described. The vertices of the object in the solid modelling database are classified by analysing the topology and geometric properties surrounding the vertices. Using vertex types to label the nodes of a vertex-edge (V-E) graph of a designed object, a labelled graph with sufficient shape information for recognition is established. The graphs for regional shape patterns can be defined in the forms of these labelled graphs. The recognition of each single regional pattern requires matching the pattern graph to subgraphs embedded in the labelled graph for the object. A hierarchy can be established for shape patterns identified by this method. The role of pattern recognition for CAD/CAM integration has been discussed in this research.

A meta-model for mechanical products based upon the mechanical design process

This paper describes preliminary work toward the develoment of a framework and a system for modeling the “meta-physical” information of mechanical products. Meta-physical information is that information which describes the nature or reason for existence of objects in the physical product model. Such information includes product and feature functionality, design intent, relations, constraints and viewpoint-dependent definitions. This effort has resulted in an initial model structure and a prototype system. The product model consists of a meta-physical product model with attached physical product models containing, among other information, geometry, dimensions, tolerances, and features. The content and structure of the product model correspond directly to the information used and produced during the mechanical design process. The prototype system integrates a solid modeler, a feature modeler, a dimension and tolerance modeler, and a meta-physical modeler. This paper provides an overview of the meta-model structure, usage and potential.

Boundary Representation-based Feature Identification

Abstract This chapter describes the automated recognition of form features from boundary representations of solid models. The methods covered include rule-based, graph-based and neural net based techniques. Each method is briefly described with specific instances of each referenced and pros and cons listed.

Compound feature recognition by web grammar parsing

A method is presented to automatically recognize compound features using web grammar parsing on a solid model graph. A compound feature is a class of regional shapes with a variable topology. Examples of compound features are connected features, which contain a variable number of simple features connected together, and intersecting features, which intersect one another creating different shapes from the original simple features conceived. The simple features are regional shapes related to preestablished application processes. A solid object, stored as a boundary representation, is transformed to a web representation (a node-labeled graph) which is the input of a web parsing system for feature recognition. And a compound feature is described as a web grammar. From the web representation of an object shape, the compound feature recognition can be accomplished by parsing the web with the web grammar. The application of a web grammar enables a user to define a compound feature and makes feature recognition a formalized process for subsequent recognition.

Generating resource based flexible form manufacturing features through objective driven clustering

The development of a new feature based technique for automated manufacturability evaluation (ME) of machined parts is reported in this article. Key to this approach is a new type of feature called a resource based flexible form manufacturing feature. This type of manufacturing feature incorporates available factory resources and permits unlimited variations in the geometric form as dictated by tool accessibility. A ME system based on this new feature definition is overviewed. Through a process of automatic feature recognition, a manufacturing feature based description of a part is generated which is then used as a form of high level operation plan on which accurate estimates of production cost and time can be made. This paper focuses on the feature recognition algorithm, which is termed Objective Driven Clustering. The recognition algorithm consists of generating feature primitives, which are operational subplans for subregions of a part. Subsequently, primitives are intelligently selected and grouped in a clustering process that uses heuristics, constraints and a user defined evaluation objective to form manufacturing features. The methodology accommodates parts with complex surfaces and interacting form features. It is also sensitive to a variety of part, factory and evaluation related parameters including the evaluation objective, accessibility, part material, D&T, available machines and tools, tool cost, tool change time and setup change time. A prototype system Arizona State University Manufacturability Evaluator (ASUME) used in validating the methodology is discussed.

Feature based object decomposition for finite element meshing

Performing a finite element analysis requires overlaying an object with a mesh of varying density based on the expected stress levels within the part. Attempts have been made in the past to automat the finite element meshing procedure. The method presented here is “intelligent” in the sense that it examines the complete part for potential stress gradients and decomposes the part into hexahedral regions according to the geometry gradients in the part. High geometry gradients are regions of high curvature, especially edges. The algorithm segregates high gradient features into isolation volumes. It then continues to decompose each isolation volume dependent on the particular geometry contained in the feature. The result is a set of hexahedral bricks suitable for passing to an automatic meshing routine.

Using a feature-based model for automatic determination of assembly handling codes

Abstract Automated concurrent engineering requires, among other things, that the assemblability of parts be evaluated during the design process. This evaluation can be made by determining an assembly part code from a solid model of the components. This research attempts to automatically derive an assembly handling part code from a solid model boundary representation. A rule based recognition system called AHPC (Automatic Handling Part Coder) was developed to extract the pertinent assembly features and then to convert these feature attributes to code digits. The UMass coding scheme is used in which the digits represent the overall shape, symmetry, and other assembly characteristics. Several previously uninvestigated features such as axial and transverse symmetries and other patterns of high-level features have been investigated and recognized, in addition to features such as holes, protrusions, slots, and steps. A technique called Feature Point Attribute Symmetry (FPAS) has been developed which converts a part to a feature-point model. Symmetry evaluation can be performed then by using this concise model and merely testing the symmetry of the feature points and the overall body shape. The system is implemented in Prolog on the Vax/Unix operating system. The test parts were created using the Romulus solid modeler. This system has been successfully implemented to generate assembly handling part code digits for a limited domain of parts.

Manufacturing feature identification

Integration of CAD/CAM involves the use of a design description, design database model and applications while ensuring data consistency. The data types needed in a variety of applications may be different from the data type stored in a design database. In order to automate the CAD/CAM process, computer understanding of the meta-knowledge in the database or description and automatic data conversion to the next analysis stage are necessary (Woodwark, 1988).