Neural network model for turning operation.
Vaupotic, Bostjan ; Balic, Joze ; Cus, Franci 等
1. INTRODUCTION
Today, many different computer-aided CNC machine programming
systems are available on the market offering different solutions for
selection of cutting tools, how-ever, still always the programmer himself must select the optimum set of tools and their correct sequence
on the basis of his knowledge and experience. To that end, he applies
most frequently the recommendations of the tool makers in their catalogs
and the general instructions of CAM systems (Balic, 2004). Such
selection method requires the user to coordinate the recommendations
with his knowledge, experience and intuition. In practice that means
that still always the human's decision is necessary for selection
of the cutting tool. The use of CAM systems requires well qualified
users and a lot of time, since they are exacting for programming which
is very time-consuming, too. The CAM systems do not allow capturing of
the feed-back information, reached in the process of selection, and
repeated preparation of those data in similar cases, therefore they can
be considered to be inflexible.
The proposed intelligent system summarizes all the advantages and
does away with the disadvantages of such systems and can work
independently or it can be integrated into the CAM system.
2. AI AND NN IN MANUFACTURING SYSTEMS
The term "artificial intelligence" appeared in the
eighteenth century in science fiction. In the fifties it was made a
scientific discipline with the birth of "electronic brain"
(Kokol et al., 2001). Today, the artificial intelligence deals,
particularly, with the development of the techniques, methods and
arhitectures for solving logically complicated problems which would be
difficult or even impossible to solve by conventional methods.
The hitherto researches and findings in the area of the use of the
artificial intelligence techniques in the modern manufacturing system
show that the latter can considerably contribute to the improvement of
the necessary flexibility and efficiency required by modern
manufacturing systems. Since its appearance the artificial intelligence
has broken through into many areas in the manufacturing systems. While
during the early years the knowledge-based systems attracted the highest
attention, recently the soft systems have been in the foreground. Among
them the genetic algorithms, the case-based decisions, the fuzzy logic and the neural network are most widely used. The neural networks were
formed in the early forties of the previous century as one of the
branches of artificial intelligence. Haykin (Haykin, 1994) defines the
neural network as a mass parallel distributed processor that stores the
experimental knowledge and ensures its use . Thus, the neural network is
very similar to the brain, since the knowledge is gathered by the
network through the learning process and the interneural connections
known as synaptic weights are used for storing of that knowledge.
3. INTELLIGENT SELECTION OF CUTTING TOOL IN TURNING BY THE USE OF
ANN
The proposed intelligent system for automatic selection of the
cutting tool for the turning process is so conceived that it can select
independently, without the human's interference into the system,
the optimum set of cutting tools on the basis of the 3D CAD model and
important parameters. The system must be capable to learn in order to
work intelligently. However, learning only is not sufficient. To be
capable to learn at all, the system must have certain capabilities such
as sufficient memorizing capacities (memory), concluding capacity
(processor capacities), perceptive capacities (input and output) etc.
In addition, learning requires certain initial previous knowledge
which is inherited in the living systems. By learning the capability of
the system increases, consequently, its intelligence increases
(Hopfield, 1982). The artificial neural networks, used by our system,
have such capabilities.
3.1 Creation of data base for learning
For the needs of learning of the neural network the data base was
created. This is the model of the environment serving as a basis for the
control, decision making and execution of action. The data bases as such
are a kind of the driving medium for operation of intelligent systems
and/or a kind of "reservoirs" ensuring the existence of
intelligent systems (taking of data) at all.
The data base consists of four units. In each unit the relations,
on the one hand, between the geometrical features, workpiece material,
quality of the require surface and other parameters and, on the other
hand, between the cutting tools are defined.
In the unit one there are eight cutting tools for roughing, i.e.
four left and four right tools. The unit two contains twenty cutting
tools for finishing, out of which eight left, eight right and four
neutral cutting tools. The third unit contains four grooving tools for
different size of depth cutting and cut width. The fourth unit contains
four threading tools with two different thread pitches. The cutting
tools differ in geometry, material type to be machined, surface quality
after cutting and other tool parameters (Sandvik Coromant, 2008).
Figure 1 shows the maximum and/or minimum entering and/or outlet
angle on one workpiece for a specific tool.
[FIGURE 1 OMITTED]
Identically, also other cutting tools are limited. In the data base
each geometrical feature, with required surface quality and by
considering also other selection parameters, has the relevant cutting
tools which can machine it. To this end, the tool entry into the
workpiece and iths exit are taken into account.
3.2 3D CAD model
The starting point for operation of our system is the 3D CAD model,
which is an advantage nowadays. When designing the products, the users
more and more frequently use three-dimensional CAD systems because of
higher flexibility, accuracy and intuitiveness of three-dimensional
modeling.
The selection of appropriate cutting tools for the manufacture of
products requires the familiarization with all geometrical features of
the product. As, usually, the turning process takes place in two stages,
i.e.rough and fine turning, the geometrical features must be known for
the final as well as semi-finished product.
From the basic 3D CAD model, representing in the same time the
final product and/or containing the geometrical features for fine
turning the semi-finished product, after rough turning, as well as the
blank necessary for the manufacture of such product are automatically
created.
3.3 Operation of the system
The system works in several stages. The first stage covers the
preparation of the 3D CAD model. The blank, the workpiece for rough
turning and the workpiece for fine machining representing the final
product are automatically formed. The next step is the recognition of
geometrical features and/or higher geometrical elements, first on the
workpiece for rough and then on the workpiece for fine machining. The
features thus recognized are then coded and, together with the material
and the workpiece surface quality and other selection parameters, they
represent the input data in the neural network in the second stage of
the system. The second stage comprises the selection of the most
suitable sets of tools for the concerned workpiece, which totally
machine the product. In the meantime, two neural networks have learned
one to turn roughly and the other to turn finely on the basis of known
cutting tools, required surface quality, workpiece material and
features[degrees]Ccuring on the workpieces. For easier work and more
clear presentation of operation our research was limited to workpieces
with specific features, which, however, can be randomly superstructed
and/or extended. It must be pointed out that the two neural networks can
learn again new rules and laws by adding new cutting tools and features
into the base and, thus, the capacity of networks increases with the
extension of the base. After completion of the second stage the output
data from neural networks serve as input data into the other two neural
network (third stage).
Like the human the two neural networks have learnt to propose
solutions on the basis of solutions of similar cases. The two networks
have learnt one to turn roughly and the other to turn finely with the
examples of workpieces, where all the limiting features and tool sets
for those workpieces occur.
[FIGURE 2 OMITTED]
On the basis of their knowledge and experience the experts have
selected the best tool set. Those decisions were taken into account in
learning of those networks. Each example newly solved is stored in the
base, so that the two networks permanently learn and gain knowledge and
experience similarly as the human expert. Still other selection
parameters can be included into learning of the networks and, thus,
still more accurate results can be reached. The last stage (the fourth
stage) covers the selection of the best set of cutting tools, decoding and graphic representation of selection (Figure 2).
4. CONCLUSION
The model developed is simple, robust, user simple and cheap at the
price. The organization and the determination of parameters of the
neural network take place in the NeuroSolutions environment with
simplified transition of data from and into that environment through
data tables and graphs in the Excel programme which can be used by the
operator without deep knowledge about the neural network. The system
works independetly of the human, completely autonomoustly. The system
does away with the inconveniences which are a consequence of the human
factor (monotony, repeating operations, inaccuracy, slowness). If such a
system is included into other intelligent solutions which have been or
will be developed ( such as intelligent automatic systems for generation
of the tool path, for the determination of optimum cutting conditions,
for selection of optimum fixing device, for generation of the NC code
etc.), the "intelligent machine tool" will be able to make the
desired product on the basis of the 3D CAD model autonomously,
independently of the human, quickly, accurately, safely and reliably.
5. REFERENCES
Balic J. (2004). Inteligentni obdelovalni sistemi, Faculty of
mechanical engineering, Maribor
Kokol, P.; Hleb Babic, S.; Podgorelec, V.; Zorman, M. (2001)
Inteligentni sistemi, Faculty of electrical engineering and Computer
Science, Maribor
Haykin, S. (1994) Neural Networks. A Comprehensive Foundation,
Macmillan college publishing company, New York
Hopfield, J. J. (1982). Neural networks and physical systems with
emergent collective computational abilities, National Academy of
Sciences, vol. 79, pp. 2554-2558.
Sandvik Coromant (2008). Available from: http://www.
coromant.sandvik.com/ [20.2.2008].
