META

The word feature is a very general term, which often indicates certain characteristics of shape and/or function pertaining to an artefact. In the context of manufacturing, features such as holes, slots and pockets are shapes, which the desired component should possess, arising as a consequence of applying some manufacturing process. Thus, a feature can be described as "a physical constituent of a part, mappable to a generic shape and having engineering significance".

The past decade has seen an increased use of solid modelling and an awareness of the need to structure the geometric information in a manner useful to manufacturing, assembly and process planning systems. The underlying thesis is that, properly utilised, solid models together with features technology (in particular feature recognition) form the central element in the iterative process of design conceptualisation and prototype development.

While CAD provides an excellent means of representing the geometry and topology of artefacts, much of the information crucial to manufacturing processes is not geometric. In integrating CAD with manufacturing applications one has to overcome a major problem - how to interpret low-level CAD information automatically in a manner that is useful for automated manufacture. Beyond the geometrical description of the component's shape the exchange of semantic (manufacture-oriented) information for use in all phases of product development is essential for its intelligent use. Such requirements may be met using features as semantically endowed objects that accompany the product development process from customer request through to production release.

Clearly, a solid model together with its feature information provides the opportunity to exploit its data in downstream CAM systems. Therefore, features are generally regarded as a key component in CADCAM integration. By employing feature recognition and classification as a communication medium between CAD and manufacturing applications a tool is created for design analysis and feedback This may include systems that automatically generate process plans and drive manufacturing processes.

Feature recognition methods

• Interactive feature definition: where the user defines features by picking entities from a geometric model.
• Design-by-features or Feature-base modelling: where the component geometry is directly designed using shape features from a given fixed library (i.e. geometric models are created in terms of features). 
• Automatic Feature Recognition (AFR): where a computer program interrogates the data structure of the geometric model to discover and extract features automatically.

In the first method, a feature description of a predefined CAD model is defined manually by the user who interactively selects the individual topological entities (such as edges, faces) associated with each individual feature. In design-by-features, features are incorporated in the component model during its creation. Generic feature definitions are placed in libraries from which features are instanced by specifying geometric parameters and various attributes. These attributes typically include semantic manufacturing information used for process planning and assembly.

The third method, Automatic Feature Recognition (AFR), relies on algorithms that extract geometric features from the CAD system database. It is difficult to classify recognition methods into a clean taxonomy (due to considerable overlap between various techniques) nevertheless the two important elements namely, the definition of features and the feature-recognition mechanism, remain common themes. The various mechanisms developed for automatic feature recognition can be divided informally into the following six categories:

• Syntactic pattern recognition
• Rule-based methods
• Graph-based methods
• Volume decomposition 
• Neural net approach 
• Hybrid

Project: Feature Recognition for NC-machining

Description:

The FeatureFinder software described in this paper has been developed to support a solids machining CAM package known as GNCSolid (CADCentre 1995) therefore it is appropriate to give an outline of the way such systems work in general.

Early software for generating CNC toolpaths was based on 2D profiles representing the periphery of material, which required clearing. Typically these profiles were either entered manually or imported from engineering drawings. In contrast, solids machining packages use 3D models to represent both the shape of the stock and the final component. The volume of material to be removed is generated by subtracting the component from the stock to form what is termed the deltavolume (i.e. the material to be removed). A toolpath to clear (i.e. machine) a given delta-volume can then be generated automatically after the human operator has specified the size of cutting tool. If the chosen cutter is able to penetrate all of the delta-volume (without being constricted by the shape of the finished component) then manufacture could potentially be carried out in a single machining operation.

At first glance such rapid prototyping procedures appear to render feature recognition redundant. Fortunately this is not the case! Because such default clearance of an entire deltavolume must be done using a single tool the resulting NC-code is almost always extremely inefficient. Regions of the deltavolume which could in theory be rapidly removed by a large diameter cutter in a couple of passes, are in practice swept many times by a small tool. To generate efficient part programs the delta-volume must be sub-divided into a number of discrete volumes (i.e. feature volumes), each of which can be associated with a different size and type of cutting tool. The role of the recognition system described here is to support this subdivision and provide an efficient alternative to standard Boolean operations. FeatureFinder acts as an intelligent aid to delta-volume subdivision by exploiting three important observations:

• In any given machining set-up the engineer's main objective is to identify feature volumes accessible from the tool approach direction.
• The feature volumes generated automatically should be modifiable manually.
• Unrecognised parts of the delta-volume can be left for subsequent manual interpretation.

Using haptic feedback and 3D stereo-rendering this project will create novel application interfaces to a range of macro/micro manufacturing and assembly tasks. This is an area of considerable current interest to both users and developers of CAD/CAM systems who know that better interfaces will lead to lower engineering costs and lead-times. This project will also deliver quantitative assessments of applications that use haptic feedback and 3D stereo-rendering for interactive design and manufacturing.

Project: Laminae-based feature recognition and applications in manufacturing

Description:

The limiting factor for the majority of reported feature recognition (AFR) algorithms lie in their inability to handle anything more complex than the restricted geometric domain of 2.5D machined components. A novel approach to recognising shape features on models comprising both simple and complex ruled surfaces uses the concept of 3D-laminae (i.e. faces), which enables feature volumes, bounded by complex ruled surfaces to be constructed. This generic feature recognition algorithm requires no predefined feature libraries and advocates the notion of neutral features, which separates the generic features identified by the extraction algorithm from those (features) classified subsequently to suit a discrete domain. The work concentrates on identifying machinable volumes (for manufacture by CNC machines) and the classifications presented apply specifically to this context. However, because the algorithm is capable of handling complex ruled surfaces, it is envisaged that the methodology will be applicable to industries involved with the manufacture of dies and moulds.

While many techniques have been developed for identifying features from solid models, the vast majority of these still lack the ability to handle such complex transitions (i.e. blends). This may be attributed to difficulties in extending fundamental concepts such as edge vexity or surface orientation to non-2.5D geometry. For example, a radius blend occurring along the base of a pocket removes the vital loop of concave edges that hint at the presence of the depression.

It has been widely accepted that features technology has a key role to play in CAD/CAM integration. However, effective integration remains elusive as existing process planning systems make little use of solid modelers and rarely perform geometric computations. AFR has long been seen as one way of interfacing the CAPP system with a modeller. The feature recogniser is often envisaged as simply outputting a set of features for a process planner with no feedback from the planner. For AFR to relate the CAD model and the computer-aided process planning (CAPP), it must be able to organise and decompose/group features into appropriate and 'manageable' machining volumes. Both heuristics and/or human interaction can accomplish this.

The objective here therefore is to design a feature recogniser capable of handling a wide variety of geometry. Once accomplished, a set of neutral features can be extracted, organised and decomposed. This set of neutral features can then be classified to suit a particular domain or field of interest. Because this work was motivated by the manufacture of mechanical components, the classifications will use this context.

Tool selection lies at the heart of manufacturing processes and is one of the key steps in process planning, allowing increased productivity and decreased machining cost. Many feature recognition systems have been described in the literature, which identify machine removal volumes to assist manufacturing. However, all of these (including the few commercially available CAM packages that have feature recognition incorporated) require human interaction for tool assignment/selection. Consequently, the process of selecting optimal tooling is a tedious, time-consuming, error prone task, especially if it involves numerous features. The point of view of the work is that by automating the process of associating tools with features the efficacy of current CAD/CAM systems can be increased considerably.

A methodology is designed for sizing and selecting optimal sets of tools for manufacturing collections of features. It exploits a novel algorithm for the calculation of exact tool access volumes (TAVs), which are computed by offsetting feature profiles. This tool access algorithm is used to create a Tool Access Distribution (TAD) and the Relative Delta-Volume Clearance (RDVC) data from which an automated optimum choice of tools can be made. The implementation of this method, referred to as QuickTest, provides a bridge between the machining features obtained from CAD models and the machining processes actually used to produce the component.

The approaches employed in tool selection cover a wide spectrum, the majority being concerned with tools that provide minimum cost and maximum production rates. It appears also that no current commercial Computer Aided Manufacturing (CAM) packages provide any analytic tools for geometrical analysis to support tool selection. In contrast, tool sizing, as defined here, concerns tool-to-feature association; such that when assigned to a feature, the effectiveness of its volumetric clearance can be observed. In essence, this process is a form of manufacturability analysis, which can also assist process planning. The objective is to study the geometric constraints imposed on tool selection, enabling both a qualitative (i.e. visual) and quantitative indication of their effectiveness under actual machining operation. Such tests also enable a planner to modify, or discard, features that present unrealistic machining requirements.

Scope:

The fundamental computational issues governing feature recognition are manifest in algorithms that enable machines to interpret the real world. For example, the identification of a hole and its orientation is trivial to a human who is blessed with a complex in-built vision system. On the other hand, for a machine to do likewise, it must be specifically programmed to decipher intelligently the geometrical and topological data structure of a suitable model of the artefact. In order to achieve this one has to define the problem, ask appropriate questions and then develop suitable algorithms for solving it. For completeness we also require a means of measuring the computational difficulty and robustness of such algorithms.

The objective is to further the development of systematic methodologies for the recognition and classification of features on a wide range of solid CAD models of mechanical components for downstream manufacturing applications. This requires a method for the recognition of machining features from generalised solid models of mechanical artefacts and to address the computational issues involved in building efficient solutions for automated feature recognition. It also includes functions for the classification of general features recognised (such as pockets, holes, protrusions, etc.) as well as features with overhang (such as undercuts and recesses). Additionally, an automatic tool-feature sizing aid provides a systematic method for associating optimal tools with sets of features.

The emphasis of this research is on understanding the computational issues involved in feature recognition and the application of features in manufacturing process planning, specifically in tool selection.