This paper describes the design and implementation of an OntologyBased System for Features Recognition and Design Rules Checking in the domain of sheet-metal engineering using Semantic Web technologies. The system was implemented by means of the Protégé Application Programming Interface (API), a rule engine and a reasoner. Using ontology in the core of the system enabled the representation of manufacturing rules and Automatic Features recognition. Rules and Queries were not hard coded in the program, giving the system a high level of maintainability and reusability. Features were classified as general and specific, easing the work of classifying newer features as they appear given their previous classification. Most common sheet metal features referred to in the literature were recognized by the system.
Sheet metal parts are elements commonly used in several products manufactured in modern industry; aerospace, automotive, appliances, machine tools, etc. are only some of these. Although various software tools are available for making digital designs for such metal parts, trying to use those designs as input for process planning and manufacturing is not straightforward. The majority of standards for Computer Aided Design (CAD) do not represent certain features that are needed when manufacturing. For instance, declaring when a circle corresponds to an inner edge or an outer edge has significant consequences during manufacturing although not necessary when displaying designs. Facilitating the interoperability across the CADCAM (Computer Aided Manufacturing) process should make it possible for designers to select optimal manufacturing conditions. For instance, to choose a sheet thickness that would prevent crashes when punching holes.
In order to achieve such interoperability, manufacturing features must be extracted from CAD files. Extracting these features entails identifying certain patterns from the CAD files that are to receive additional manufacturing-relevant enrichment. This information about features is used a posteriori to determine the machining tools and manufacturing processes required to manufacture a given design [1]. Feature information can also then be used to pre-check designs in order to detect production rules violations. If these violations are not detected in the early stage of design, the life cycle of its development increases; raising both production costs and time to market [2]. Manual recognition of the targeted features is not a viable alternative; Automated Features Recognition (AFR) is thus required. However, even nowadays AFR is not fully integrated with CAD software tools. It is applicable only to parts with relatively simple geometry and still requires human intervention to obtain the features identified. Moreover, such techniques are generally supported only on expensive CAD software tools that are beyond the reach of small-scale industry [3].
To address these shortcomings of AFR from sheet metal designs and design checking, in this paper we propose an ontology-based framework that facilitates interoperability across the entire CAD-CAM process. Our system integrates both a CAD and a feature ontology; these ontologies were written in OWL (Web Ontology Language) and represent 2D primitives as lines, arcs and circles in the CAD ontology and as edges, slots, tab and holes in the features ontology. In order to provide rules and query support, OWL was complemented with SWRL [4] and SQWRL [5]. The result is an ontology-based system that allows us to automatically extract the most common manufacturing features referred to in the literature.
The paper is organized as follows. We begin with related work in Section 2. Section 3 then describes the components in the architecture of the implemented system. Section 4 shows the implementations and some results of the evaluation on a sample design and Section 5 concludes with an outlook for future work. 2
Henderson & Anderson [6] reported on early experiences with the use of production rules that were not hard-coded into the system using Prolog, while Meeran & Pratt [7] proposed to extract features from 2D drawings to generate process planning and machining instructions from CAD. The latter approach was also based on logic programming, adopting Prolog as the implementation language. These authors mentioned that Prolog presented certain limitations when dealing with trigonometric functions.
Other authors, such as Vasilakis [8] and Krima et al [9], did not use logic programming. Rules were represented instead as a set of nested IF <list of conditions> THEN <hypothesis><list of actions>. In the case of Vasilakis [8], when a design rule was violated the system showed the rule violated but did not suggest any corrective action. Radhakrishnan [2] integrated a rule checker within a design adviser; this approach identifies distance violations amongst features. The general tendency in these contributions was the development of feature recognition and design advisory systems with hard-coded rules.
In summary, two main classes of approach can be identified from the literature. On the one hand there are approaches based on logic programming. Here, rules are separated from code, facilitating interoperability, shareability and maintainability; these approaches have generally used Prolog as implementation language. On the other hand, there are also approaches relying on procedural programming, commonly using C and C++ as implementation languages. As these are not rule-based, maintainability and share-ability are limited. Babic et al [10] consider that emerging approaches should be hybrids.
The limitations of CAD software tools have motivated researchers to develop frameworks that overcome those limitations and some of these have already adopted ontological approaches. Vasilakis & Andersen [11] and Krima et al [12], for example, developed ontologies for the Standard for the Exchange of Product model data (STEP). Here, Krima et al [12] acknowledged the limitation of STEP and accordingly developed OntoSTEP. OntoSTEP is based on OWL precisely because of its support of logic reasoning and inference mechanisms.
Approaches addressing the problem of interoperability across CAD standards have been explored by Ghafour et al [13], Sun & Ding [14] and Ramos [15]. Ghafour et al. proposed a common Design Features Ontology written in OWL and one domain ontology from each specific standard, an approach similar to that of Sun & Ding. Ramos focused on an ontology for CAD primitives populating a CAD ontology. In addition, Grüninger & Delaval [16] developed an ontology for the cutting process of sheet metal parts; this approach was proposed as a theoretical step from which the planning process could be supported. Franke et al [17] developed an ontology-based tool for quality control in CAD design. They aimed to verify cases in which overlapping occurred in order to determine if this constituted a design mistake. Their architecture included their own software tool (OntoDMU) and the system HETS [18].
Defining features in the manufacturing domain is not straightforward [1]; a feature is understood here as a part of a mechanical design that has a specific functionality and that is semantically significant within the manufacturing process. Farsi & Arezoo [19] classified sheet metal features according to whether those features were internal or external; internal features are holes and external features are different types of notches. Radhakrishnan et al [2] classified rules into two types: intrafeature rules and interfeature rules. Interfeature rules are related only to the feature itself and its constraints, indicating minimum values of its dimensional parameters. Interfeature rules describe restrictions across features and contours. In Fig. 1 some of the most common features and rules found in the literature are described. The figure includes features names and some of their most important parameters. The indicated rules are related directly with the production process. It is necessary to take such rules into account in order to provide the necessary structural strength to the design´s features. Improperly formed features will contribute to waste of material in the finished products due to buckle, cracks or breaks.
Circular Hole
Notch or slot d H
& d W d T & d 2.5mm T= Thickness d = diameter H = High W = Wide W 1.5T & W 0.5mm H W
5 dbh T dbh 1.5 mm dhb T dhb 2 T dhb 0.8 mm
Protrusion or Tab
Oblong holes
Corner
W 1.5T & W 0.5m H W
5 d H
& d W d T & d 2.5mm r 0.7 T Distance to holes dbh: distance between holes dhb: distance hole border
The approach developed here builds on the state of the art we have seen and takes this further, relying on a modular architecture that makes use of the Protégé API. In the following subsection we describe the approach’s components and the way in which they interact with each other. Although the organization and classes of the two ontologies are often comparable, it is important to see that they reflect fundamentally different kinds of ontological information. To the extent that similarities are found, then this renders the subsequent mapping process more straightforward, but there is no requirement that the ontologies correspond precisely. Since the identification criteria for the CAD and the features ontologies are quite distinct—the former relating to descriptions of design, the latter to manufacturability —it is both theoretically and architecturally cleaner to maintain them separately. We return to this important consideration and the issues of perspectives below.
The first component to be discussed is the CAD ontology. During the development of
this ontology several CAD formats, such as DXF [20] and IGES [
As is illustrated in the figure, we divided CAD_Features into the two subclasses: Qualitative_Features and Quantitative_Features. Among the former, Primitives can be either Open or Closed. Under these classes, Line, Arc, Circle and Ellipse are understood as Primitive elements that are present in any CAD standard that provides at least a 2D representation. Derived elements are then Qualitative Features that may vary and are standard dependent; they appear as a combination of certain primitive elements. For example, a Vertex appears when two primitive lines are connected at one end. Definition of concepts, such as Vertex, is dimensional-space dependent.
For the development of the ontology of features, available at http://bit.ly/pyQGEE, we considered information such as that shown in Fig. 1. An extract of the resulting ontology is presented in Fig 3. There are two main concepts: these are Shape_Features and NonGeometry. Shape_Features was divided further into Qualitative_Features and Quantitative_Features, analogously to the top-level distinction in the CAD ontology. Named_Feature involves features found in the literature that can be precisely defined and identified using their properties or attributes. Atomic_Feature is then a Named_Feature consisting of exactly one identifiable element which has meaning in a manufacturing domain. These elements can have Open or Closed edges.
Composed_Features include features that appear as a combination of
Atomic_Features. Partial_Composed includes identifiable features that need to be
part of a whole to be consistent. Total_Composed covers a feature that exists as a
whole. Partial_Composed features have to be part of a Total_Composed feature.
hasPart and isPartof were included following a standard mereological [
One of the main issues of AFR is the level of abstraction involved in the
manufacturing domain [
In Fig. 4, we introduce the framework implemented for dealing with this
situation. Knowledge transfer between domain ontologies has been facilitated by
means of a third, mapping ontology, which keeps track of the related concepts of both
domains. This mapping ontology was developed using the Prompt plug-in [
Features were recognized by querying the features ontology with SQWRL. These
queries were performed at two levels of abstraction. A first set of queries was
implemented to facilitate a general classification of features and a second set was
executed to facilitate specialized classification. These two sets of queries reduced the
impact of not extracted or unidentified features referred to in the literature [
For example, a protrusion pattern is formed by three connected edges having a specific slope. The query for extracting this pattern was expressed as follows: Open_Primitive(?l1) ∧ Slope(?l1, "-0.0") ∧ Open_Primitive(?l2) ∧ Slope(?l2, "Infinity") ∧ Open_Primitive(?l3) ∧ Slope(?l3, "+0.0") ∧ hasendpoint(?l1, ?p1) ∧ hasstartpoint(?l2, ?p1) ∧ hasendpoint(?l2, ?p2) ∧ hasstartpoint(?l3, ?p2) ˚ sqwrl:makeSet(?s, ?l3) → sqwrl:select(?s, ?l1, ?l2, ?l3) The number of sets obtained and their composition allows the generation of composed features. Once the query has been executed, composed features are related to atomic features by means of the has-part property.
Although AFR research is receiving considerable attention, a recognized feature still
does not indicate anything about its quality. This aspect is fundamental in order to
reduce the cost of manufacturing a given design [
PartialComposed(?pc) ∧ hasEdge(?pc, ?e) ∧ slope(?e, "0.0") ∧ hasEndPoint(?e, ?p2) ∧ hasYcoordinate(?p2, ?y2) ∧ hasStartPoint(?e, ?p1) ∧ hasYcoordinate(?p1, ?y1) ∧ swrlb:subtract(?v1, ?y2, ?y1) ∧ swrlb:abs(?v1) → width(?pc, ?v1)
The second set of rules was implemented for checking the design. For instance, given the thickness of a given material, a minimum width is mandatory to make its manufacturability feasible. For some materials the width must be 1.5 times greater than the given thickness. This rule was expressed in SWRL as follows: PartialComposed(?pc) ∧ width(?pc, ?w) ∧ heigh(?pc, ?h) ∧ thickness(?pc, ?t) ∧ swrlb:multiply(?v1, ?t, 1.5) ∧ swrlb:greaterThan(?w, ?t) → proper_width(?pc, true)
The reasoning capabilities of the Semantic Web framework gives us the possibility of
classifying instances of classes against the ontologies of our application. Our
reasoning was done in two steps. Firstly, using SWRLTab [
Secondly, using the SWRL API [
The implementation was tested with several designs of shapes. In Fig. 5 one of these
is displayed by means of a CADViewer plug-in for Protégé [15]. Our first evaluation
of the design consisted of diagnosing the topological correctness of the model and
verifying the connectivity of edges and vertex [
The features described in Fig. 1 were recognized and extracted into the features ontology. Manufacturing rules were run on the OWL model adding newer knowledge. This new knowledge was then used by the reasoner to classify features into sets of proper and improper features.
AFR is a research and application area devoted to bridging the gap between design and manufacturing. Most of the approaches used in AFR are divided into procedural and declarative approaches. We have developed an Ontology-Based AFR and Design Rules Checker System that combines the advantages of both approaches. An ontology of CAD and an ontology of features are the fundamental components of our system. The necessary interoperation between both ontologies was achieved by means of a mapping ontology generated in a semi-automatic manner using the Prompt Protégé plug-in.
SWRL was used to perform engineering calculations in order to add appropriate semantics to the features ontology. SQWRL was used for modeling feature patterns commonly referred to in the literature. Sheet metal features were extracted as the result of queries applied to the features ontology. SQWRL queries and SWRL rules were written at several levels of abstraction in order to make a progressive identification of features possible. By using such combination, SQWRL and SWRL, we have identified a significant number of features.
In this paper we have demonstrated that complex rule of mechanical and manufacturing engineering processes can be expressed using OWL and SWRL. Similarly, we have demonstrated that complex engineering information can be retrieved from the ontology using the SQWRL language. These rules and queries were also used for inferring knowledge and driving appropriate decision-making information. New features, rules and patterns can be integrated into our system by means of standard ontology editors. In the near future, we plan to integrate a recommendation ontology. Thus, as soon as a design violation is found, a recommendation will be presented to the user. We will also investigate how to deal with intersecting mechanical features.
The resulting set of recognized features could be used as an input in Computer Aided Process Planning (CAPP) systems, from which process planning will be obtained. We will continue this line of development as a Semantic Web CAPP system, focusing on the accuracy of the kind of planning generated with the Semantic Web Technologies taking into consideration other aspects of the engineering environment and investigating scalability. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19.