International Journal of Computer Theory and Engineering, Vol. 9, No.

1, February 2017

Development of an Intelligent Database System to

Automate the Recognition of Machining Features from a
Solid Model Using Graph Theory
Rachna Verma and A. K. Verma

 not possible to create a predefined library of it, hence shapes

Abstract—Automatic recognition of machining features is are stored in the solid model database in terms of low level
essential for the integration of CAD and CAM. Graph-based geometrical entities such as vertex, edges and faces in the
recognition is the most researched feature recognition method as form of a graph or in some other suitable structure. Hence
the B-Rep CAD modelers’ database uses graph to store the
model data. A graph-based feature recognition system uses
there is a requirement of a system that can extract patterns in
attributed graphs to store CAD models as well as machining this low level database that can be used to automate the
feature templates. The graph isomorphism is used to extract manufacturing of the given shape. There are a number of
features in the model graph and template graphs. There are two techniques used for this purpose but graph-based recognition
main research issues in this system- (1) Efficiently recognize the is the most researched feature recognition method due to
features as the graph isomorphism is computationally very sound mathematical background of graph theory and a
expensive and (2) incrementally expanding the feature template
database to include new features, without any structural change graph’s structural similarity with B-Rep CAD modelers’
in the recognizer. In this paper, the application of feature database. A graph-based feature recognition system
vectors (a heuristic developed by the authors that converts a represents geometry and topology information of a part using
feature graph into a unique vector of integers, irrespective of the a graph structure. The graph carries all the useful information
node-labeling scheme used by B-Rep modelers), to automatically of part features as certain attributes assigned to its arcs and/or
expand the recognizer’s feature template database, is presented.
nodes. This explains the inherent advantage of B-Rep-based
It facilitates automatic inclusion of new features in a feature
database, without requiring any additional programming effort solid models over others in graph-based feature recognition.
from the user or any changes in the structure of the recognizer. Pre-defined manufacturing features, known as manufacturing
The proposed system has been implemented in Visual C++ and primitives are also represented by similar graphs. Graph
ACIS solid modeling toolkit. Further, the proposed system is isomorphism is then used to extract features.
intelligent as it has the capabilities to learn from the examples to A feature recognition system uses a pre-defined database of
incrementally build the feature database.
feature templates. Thus, a simple mechanism is needed to
define new features and expand the template database to
Index Terms—Machining feature, feature recognition, graph
matching, solid model. include user-defined features without any structural change in
the recognizer. This led to development of various methods to
represent features. Pratt [1] proposed a non-manifold feature
I. INTRODUCTION representation scheme using B-Rep for volumetric features.
An intelligent database is a full-text database that employs He introduced the concept of implicit and explicit feature
some artificial intelligence to return the most relevant representations. In explicit representation, a full geometric
information possible requested by the user depending on the shape is defined, while in implicit representation, minimal
context. This is in contrast to a traditional database, which is information is used to define the feature but other details have
searchable only by keywords and verbatim phrases with some to be computed when needed. Shah and Rogers [2] and Wang
relational operators. In this paper, definition of intelligent and Ozsoy [3] used hybrid CSG and B-Rep schemes for
database is extended to include geometrical data along with feature representation. None of these schemes, however, is
the textual data. Further a system is presented that interact suitable for feature recognition as the B-Rep or CSG
with this raw geometrical database to extract useful representations of a shape are not unique.
machining features for automated manufacturing of the The ASU feature test bed [4], a system developed by
various parts. The manufacturing process and operation Laakko and Mantyla [5], and STEP [International Standard
selection is highly dependent of the shape of the part to be for Exchange of Product data] all use feature definition
produced. Since there are infinite possibilities of shapes it is languages to define new features [6].
In this paper a heuristic has been reported that converts a
feature graph into a vector of integers, called feature vector
[7], [8]. A feature vector uniquely represents a feature graph,
Manuscript received May 16, 2015; revised November 22, 2015. irrespective of the node-labelling scheme used by B-Rep
Rachna Verma is with the Department of Computer Science and modelers. It can be generated automatically from the modeler
Engineering, J.N.V. University, Jodhpur, India (e-mail: database. Its usage is intended to reduce the graph matching
[email protected]).
A. K. Verma is with the Department of Production and Industrial time complexity to a polynomial of order three. It also
Engineering, J.N.V. University, Jodhpur, India (e-mail: facilitates automatic inclusion of new features in a feature
[email protected]).

DOI: 10.7763/IJCTE.2017.V9.1112 58
 International Journal of Computer Theory and Engineering, Vol. 9, No. 1, February 2017

library, without requiring any additional programming effort

from the user or any changes in the structure of the recognizer.

II. SHAPE AND ITS GRAPH REPRESENTATION

Fig. 1(a) shows a part and Fig. 1(b) represents the part
using an attributed adjacency graph (AAG). The solid edges
show concave edges and dashed lines show the convex edges.
Features in the form of subgraphs of this part useful for
manufacturing is shown in Fig. 2 along with their computer (a) A part
representation in the form of attributed adjacency matrix
(AAM). The subgraphs are generated by removing all the
convex edges and searching for connected graphs in the
remaining graph. This part is machinable on a vertical CNC
milling machine. Now the task is to extract various features
such as holes, slots and packets from this graph and generate
relevant instruction for the CNC machine. The CNC program
for a feature can be manually generated and stored in the form
of a template database. However, a generic part may have
different types of features that may not be present in the
feature database. For a new pattern that does not exist in the
library, the proposed system atomically generate a unique (b) The AAG of the part (a)
code and store that into the library for future use. Fig. 1. A part and its attributed adjacency graph.

Subgraph AAM FV Recognised as

14 26 2 Step
a 26 14 14 0 1 1 
26 1 0
16 19 2 Step
b 19 16 16 0 1 1 
19 1 0
2
22 23 24 25
3
22 0 1 1 1
11
c …
2 2 23 1 0 1 0
7  Blind
4.1 2 12 Slot

4
24 1 1 0 1
12
2 25 1 0 1 0
5
29 27 28 29 30 31 32
31 
27
28
0
1
1
0
0
1
1
0
0
0
1
1
56 
d 28 32 29 0 1 0 0 0 1 38  Edge

30 1 0 0 0 1 1 37  Pocket

31 0 0 0 1 0 1  48 
27 30 31 32 1 1 1 1 1 0  40 
1 2 3 4 5 6 7 8 9 10 11 12
1 0 1 0 0 0 0 0 0 0 0 1 1
2 1 0 1 0 0 0 0 0 0 0 0 1
1 2 3 4 5 3 0 1 0 1 0 0 0 0 0 0 0 1
4 0 0 1 0 0 0 0 0 0 0 0 1
5 0 0 0 0 0 1 0 0 0 0 0 1
6 0 0 0 0 1 0 1 0 0 0 0 1
11 12 6 7 0 0 0 0 0 1 0 1 0 0 0 1
e
8 0 0 0 0 0 0 1 0 1 0 0 1
9 0 0 0 0 0 0 0 1 0 1 0 1
10 0 0 0 0 0 0 0 0 1 0 1 1
10 9 8 7 11 1 0 0 0 0 0 0 0 0 1 0 1
12 1 1 1 1 1 1 1 1 1 1 1 0

2047, 3584, 3088, 2562, 2432, 2308, 2144, 2116, 2072, 2057, 2176, 2080
Generic pocket
Fig. 2. Sub graphs, AAMs and feature vectors.

subgraph is described in the following steps.

III. FEATURE VECTOR GENERATION HEURISTIC 1) Rows and columns of AAM (i.e. vertices of subgraph)
A heuristic to generate feature vector from an AAM of a are arranged in descending order of their degrees. Degree

59
 International Journal of Computer Theory and Engineering, Vol. 9, No. 1, February 2017

of a vertex is the number of proper edges incident with it subgraphs in an arbitrary part graph. Architecture of the
plus twice the number of self-loops at it. In the present proposed feature recognition system based on feature vector
context, there are no self-loops. generation heuristic is shown in Fig. 3. It is implemented in
2) The arranged AAM is partitioned into overlapping zones. Microsoft® Visual C++ along with built-in application
Each zone contains those rows and columns that have the programming interfaces (APIs) of ACIS® 3D solid modeling
same degree. Intersection of rows and columns of a zone toolkit. The APIs are used for loading, saving, topological and
forms its pivot block. geometric reasoning.
3) For each zone: The architecture is built on the following modules.
Rows are arranged from top to bottom in descending Module 1. AAG of a part is generated from its B-Rep
order of number of 1s present in the pivot block of that zone. model and stored as an AAM. Edges of the part are obtained
Columns are arranged from left to right in descending by calling api_get_edges (). Adjacent faces of each edge are
order of the column value, treating 1 and 0 as binary numbers, found by following the links of B-Rep data structure of ACIS
and top row as the most significant bit. and calling api_get_faces () with edge as its parameter.
Similarly, rows are re-arranged from top to bottom in a Convexity of each edge is calculated at its mid point by
descending order of the row value, taking first column as the calling api_edge_convexity_param (). It is assumed that the
most significant bit. convexity remains the same throughout the length of the edge,
4) Finally, considering each row of the AAM as a binary which is true for all milled parts with only analytical surfaces.
number, feature vector is computed as the ordered set of Module 2. Subgraphs from the part AAG are generated
decimal integer values of the rows. using standard vertex fusion method.
This arrangement of an AAM uniquely represents an AAG, Module 3. Feature vector corresponding to each subgraph
irrespective of node labelling scheme used in the AAG. To having more than one face is generated. Single node features
match any two AAGs, their AMMs are converted to such as through cylindrical hole can be handled separately.
respective feature vectors as discussed above. Isomorphism Module 4. The generated feature vector is matched with
checking of the two AAGs then becomes a matter of templates in the feature library. In the proposed architecture,
comparing the two feature vectors. feature library is stored as a text file and loaded in computer
memory at the start of recognition process. It is updated
whenever a user defined feature is appended to it. The
IV. ARCHITECTURE OF FEATURE RECOGNIZER AND ITS modified library is reloaded in the memory. Most of the
IMPLEMENTATION features have unique AAGs; however, there exist some
situations where an AAG may represent more than one
feature. Specific features in this situation are differentiated by
using additional geometric reasoning which can be provided
by maintaining link with ACIS database. Presently, such
geometric rules are hard coded and require modification of
the source code in order to be included in the feature library.
A recognized feature is reported as a group of pointers
pointing to the faces constituting the feature. This facilitates
future analysis of the feature such as manufacturability
analysis.
Module 5. If module 4 fails to recognize a cluster of faces,
the corresponding feature vector is analyzed to extract generic
pocket.
Module 6. Unrecognized cluster of faces not recognized by
the previous modules is handled in this module. The user is
informed about no-match situation with template features in
the library. The system hands over control to the user for
breaking the compound feature into primitives. If he feels
appropriate, this cluster of faces can be added to the library. In
future this feature will be available as a template feature.

Fig. 3. Architecture of the proposed system.

V. CONCLUSION
Features in a part are clusters of faces having some
engineering significance. Manufacturing features are Feature vector is a unique representation scheme for a
subclasses of features that have some manufacturing feature AAG. It converts an arbitrarily labeled AAG into a
significance. They are used for planning manufacturing unique vector of integers. It can be automatically generated
activities. In graph theory terminology, if a part is represented from a B-Rep model. This helps in incrementally building a
by a graph then its features are subgraphs of the part graph feature library as per requirements of specific domain. The
representing pre-defined shapes. The graph-based feature proposed system is intelligent enough to extract useful
recognition is the process of searching these pre-defined information from the raw database of solid modeler and has

60
 International Journal of Computer Theory and Engineering, Vol. 9, No. 1, February 2017

the capabilities to include new patterns found in the data of its [8] A. K. Verma and S. Rajotia, “Review of machining feature recognition
methodologies,” Int. J. of CIM, vol. 23, no. 4, pp. 353-368, 2010.
own.

REFERENCES Rachna Verma received the bachelor of science from

University of Delhi in 1996, the master of computer
[1] M. J. Pratt, “A hybrid feature based modelling system,” in Advanced
applications from MBM Engineering College, Jodhpur,
Geometric Modelling for Engineering Applications, F. L. Krause and
India in 1999 and the PhD form J.N.V University,
H. Jansen, Eds. Elsevier Science Publishers, 1990, ch. 14, pp. 189-201.
Jodhpur, India in 2013.
[2] J. J. Shah and M. T. Rogers, “Expert form feature modelling shell,”
She is currently working as an assistant professor in
Computer Aided Design, vol. 20, no. 9, pp. 515-524, 1988.
the Department of Computer Science and Engineering,
[3] N. Wang and T. M. Ozsoy, “A scheme to represent features,
J.N.V. University, Jodhpur, India.
dimensions and tolerances in geometric modeling,” J. of
Manufacturing Systems, vol. 10, no. 3, pp. 233-240, 1991.
[4] J. J. Shah, D. Anderson, Y. S. Kim, and S. Joshi, “A Discourse on
A. K. Verma received the bachelor of engineering
geometric feature recognition from CAD models,” J. of Computing
degree in industrial engineering from IIT Roorkee, India
and Information Science in Engineering, vol. 1, no. 1, pp. 41-51, 2001.
in 1992 and the master of engineering in production and
[5] T. Laakko and M. Mantyla, “Feature modelling by incremental feature
industrial systems engineering in 1994 from the same
recognition,” Computer Aided Design, vol. 25, no. 8, pp. 479-492,
institute. He received his PhD degree from MBM
1993.
Engineering College, Jodhpur, India in 2005.
[6] E. S. A. Nasr, A. A. Khan, A. M. Alahmari, and H. M. A. Hussein, “A
He is currently working as an associate professor in
feature recognition system using geometrical reasoning,” in Proc.
the Department of Production and Industrial
International Conference on Manufactuirng of Lightweight
Engineering, J.N.V. University, Jodhpur, India.
Components – Manulight, 2014, pp. 238-243.
[7] A. K. Verma and S. Rajotia, “Feature vector: A graph based feature
recognition methodology,” Int. J. of Production Research, vol. 42, no.
16, pp. 3219-3234, 2004.

61