You are on page 1of 7

Proceedings of the 29th Annual Hawaii International Conference on System Sciences - 1996

Applications of Neural Network in Manufacturing


Ramesh Rajagopalan,Ph.D. * and Purnima Rajagopalan+
* Industrial and SystemsEngineering, Mercer University, Macon, GA 3 1207
rajr@egr.mercer.peachnet.edu
+ M IS Manager, Galaxy Services Group, Macon, GA 3 1210

Abstract called neurons. Neurons get data from the surrounding


neurons, perform some computations, pass the results to
Neural network is a model of brains’s cognitive process. other neurons. Connections between the neurons have
Neural network originated as a model of how the brain weight associated with them. In neural network, the
works. Neural network research has its beginnings in knowledge is stored in the network’s interconnection
psychology. Today neural network methods are being weights in an implicit manner, learning takes place within
used to solve numerous problems associated with the system and plays the most important role in the
manufacturing operations. A review of neural network construction of an neural network system. The neural
applications to problems in production and operations network system learns by determining the interconnection
management is presented. Applications reviewed in this weights from a set of given data [40]. Learning in neural
paper include character, image andpattern recognition, network can be supervised, unsupervised or based on a
managerial decision making, manufacturing cell design, combined unsupervised-supervised training [20].
tool condition monitoring, real-time robot scheduling and
statistical process control. Methods and structures of In supervised learning, a set of data, called a training
neural network are explained. data set, is used to help the network in arriving at the
appropriate weights [40]. A teacher teaches the network
and gives results of the output corresponding to the input.
I. Introduction to neural network
The inputs as well as side information indicating the
correct outputs are presented to the network[20]. The
Over the past several years, various methods based on
network is also ‘programmed’ to know the procedure to be
such areas as operations research, statistics, computer applied to adjust the weights and thus the network has the
simulation, control theory have been developed and means to determine whether or not its output was correct
applied to solve a wide spectrum of problems in and the means to apply the learning law to adjust its
manufacturing. Today’s manufacturing environment is weights in response to the resulting errors[20].
characterized by complexity, inter-disciplinary
manufacturing functions and ever growing demand for Weights are generally modified on the basis of the
new tools and techniques to solve difficult problems. errors between desired and actual outputs in an iterative
Neural network offers a new and intelligent alternative to fashion and one of the widely used training algorithms is
investigate and analyze challenging issues related to the “Delta Rule” [20]. The neural network learns the
manufacturing. In this section, an introduction to neural desired outputs by adjusting its internal connection
network and two commonly used neural network methods weights to minimize the discrepancy between the actual
will be provided. outputs of the system and the desired outputs [32].

Neural network is used to capture the general In unsupervised learning a neural network operates in
relationship between variables of a system that are self-organization mode. In this self-organization mode, a
difficult to analytically relate. Neural network has been competition mechanism is used to select processing
described as “brain metaphor of information processing” elements due to which their weights are modified. The
or as “a biologically inspired statistical too1”[35]. It has first part of the competition involves competition between
the capability to learn or to be trained about a particular the simple-elements that he within same sample. Winners
task, its computational capabilities and the ability to from these competitions compete with other processing
formulate abstractions and generalizations. Neural elements in their layer that have won similar competitions
network has an organization similar to that of a human in order to be single processing element on their layer
brain and it is a network made up of processing elements

447
1060-3425/96 $5.00 0 1996 IEEE

Proceedings of the 1996 Hawaii International Conference on System Sciences (HICSS-29)


1060-3425/96 $10.00 © 1996 IEEE
Proceedings of the 29th Annual Hawaii International Conference on System Sciences - 1996

which has the weight changed during training. In self- III. Neural network applications
organized training, the network is given many input
characters and no information is given to the network as to Neural network offers a method for incorporating and
what each example corresponds to. There are many layers processing qualitative knowledge and have the additional
and each layer corresponds to one character more than the advantage of formalizing machine learning in an explicit
other layers. The importance of unsupervised learning is manner, it aspires to imitate human intelligence in its
that the system does not need to know the correct answer totality [40]. Neural network has been applied to a wide
in order to solve a problem. The system learns a pattern variety of problems, ranging from traveling salesman
from repeated exposure to it and is able to recall the optimization to vision problems. Neural network is best at
learned pattern when it solves a categorization or pattern solving classification problems and it also has the added
matching problem [40]. advantage of performing successfully where other
methods often fail - recognizing and matching
Neural network can also employ a hybrid approach in complicated, vague, or incomplete patterns [40].
which learning is based on combined unsupervised-
supervised learning. The hybrid approach first uses Neural network is used to convert text to speech, for
unsupervised learning to form clusters and the labels are natural language processing, for example, for deriving
then assigned to the clusters identified and a supervised language rules, recognition of characters and handwriting,
training follows[20]. image processing and pattern recognition. This line of
research has immediate use in banking, credit card
II. Neural network methods processing, and other financial services where reading and
recognizing handwriting on documents is crucial [40].
Two methods are commonly used in neural network
applications - neocognitron and back propagation. Due to rapid development of neural network methods
Neocognitron is a hierarchical network made up of many and tools, neural network has generated tremendous
layers and it’s organization is like that of the visual cortex. amount of interest solving manufacturing related
Neocognitron is a method of pattern recognition. It has the problems. Neural network is used to address issues
capability to recognize shapes and sizes of characters even relating to manufacturing process planning and process
if it involves noise and distortion. Back propagation control, manufacturing system design, operational decision
allows the training ofmulti-layer networks, it is a powerful making, and resource scheduling. The Table 1 provides
and practical tool for solving problems that would be quite a summary of various neural network applications in the
difficult using conventional computer science techniques general and manufacturing categories. The remainder of
and these problems range horn image processing to speech this section is organized into two subsections. Section III. 1
recognition to character recognition forecasting to describes general applications of neural network in pattern
optimization[5]. and character recognition. Applications of neural network
to manufacturing problems are presented in section III.2.
Back propagation is the best-known supervised
learning method for Neural Network with three or more III.1 General applications
layers[40]. In the back-propagation procedure, the
network fast uses the input data set to produce its own A character recognition problem using neocognitron to
output, and then compares this with the desired output, if translate alphabets in one language to alphabets in another
there is no difference, no learning takes place else the language is the described by Sankamarayanan [ 3 11. The
weights are changed to reduce the error term [lo]. One of author used the method of neocognitron to translate
the most important steps in the development of a neural English letters to Greek letters. The method is quite
network is the development of the training data set and the versatile and may find widespread application as a general
training data set must be constructed carefully to include pattern-recognition paradigm. A software system using
examples of many different operational scenarios. PASCAL was developed [31]. The software system
Approximately equal numbers of examples of each implements a neocognitron based self-learning neural
scenario should be included in the data set. In the network. The input is a character with low-resolution.
following section applications of neural network to a wide Neocognitron is not affected by scale change, shift in
variety of manufacturing planning, design and control position, noise and distortion. It shares the weight during
problems are presented. training. This weight sharing method is used so that
whenever the weight of an element in a layer is modified,

448

Proceedings of the 1996 Hawaii International Conference on System Sciences (HICSS-29)


1060-3425/96 $10.00 © 1996 IEEE
Proceedings of the 29th Annual Hawaii International Conference on System Sciences - 1996

through a visual system and recognized correctly by the


Table 1. Summary of neural network applications. multilayered hierarchical network using self-organization.

To recognize handwritten digits using a machine and


using pattern recognition is difficult. This is of some
interest to recognize ZIP codes in addresses written by
hand in mail. Neural network and its computational
properties have attracted the interest of researchers in the
area of machine perception by presenting an exciting,
complementary alternative to symbolic processing
paradigms[29]. The neural network gives results quickly
Fukushirna et a1.[13j using unsupervised training. There are many different
neural network models.

The method of neocognitron is used to recognize


handwritten alphanumeric characters. This neural network
can acquire the ability to recognize patterns by learning,
and can be trained to recognize any set of patterns, it also
has a large power of generalization, presentation of only a
few typical examples of deformed patterns is enough for
the learning [13].

The neural network not only makes use of classic


notions of dynamic learning systems, but also has at its
core, the idea that a system may use relatively static sets of
previously learned information analogous to what
cognitive psychologists might refer to as schemas or
strategies[25]. In particular applications, such as optical
character recognition, it is possible to train a network by
presenting it with the finite universe of prototypical sets of
pixels that may lead to a correct solution most of the
time[25]. When the training is completed the neural
network is found to be stationary. Rarely when the neural
network is static, it fails to recognize the input figure and
in that case the dynamic neural network system is used.
all other processing elements in the layer immediately These two networks are combined in an image recognition
adopt new weights. The input to the network is in the unit that has the capability to exceed the expected
form of zeroes and ones (binary). Tests are conducted for performance of human beings at this task[25]. Neural
many iterations. The goal of the network is to identify network is used in the image recognition and pattern
accurately character present in the input image. At each recognition methods. The neural network model should be
layer of network input image is analyzed within the synthesized in order to endow it with pattern recognition
surrounding area that becomes bigger and bigger till the capability like that of a human being. A cell in a deeper
entire input character is classified. Several examples of layer generally has a tendency to respond selectively to a
translation of English letters such as “a” and “b” to more complicated feature of the stimulus patterns and, at
corresponding Greek letters such as “~1”and “p” are given the same time, has a larger receptive field and is less
in [31]. sensitive to shifts in position of the stimulus patterns, SO
each cell of the deepest layer of the network responds
Neural network using neocognitron can recognize selectively to a specific stimulus pattern and is not affected
handwritten numerals of various styles of penmanship by the distortion in shape or the shift in position of the
correctly, even if they are considerably distorted in pattem[l4].
shape[31]. Although the author shows results for the
recognition of alphabet, the neocognitron can be trained to III.2 Manufacturing applications
recognize other set of patterns such as Arabic numerals,
geometrical shapes, or others. A pattern can be seen Moon and Chi [27] have utilized the generalization

449

Proceedings of the 1996 Hawaii International Conference on System Sciences (HICSS-29)


1060-3425/96 $10.00 © 1996 IEEE
Proceedings of the 29th Annual Hawaii International Conference on System Sciences - 1996

capability of neural network models to solve the part network is started by providing an output signal to the
family formation problem. The approach presented by the chosen neuron. A threshold value, which determines the
authors, combines the useful capacities of the neural number of clusters as well as the degree of similarity
network technique with the flexibility of the similarity within a cluster is then decided. Neurons with activation
coefficient method [27]. Manufacturing information, such values greater than the threshold value are grouped
as the sequence of operations, lot size, multiple process together. Thus part families and machine groups are
plans were given special consideration in their approach to identified. The method then chooses a neuron arbitrarily
solve generalized part family formation problem. The which has not been assigned to any family or group.
authors also point out that the method is flexible and can These steps are iteratively carried out until all the neurons
be efficiently integrated with other manufacturing are clustered.
functions. The authors conclude that neural network can
address the Group Technology (GT) family formation Wu describes two applications of neural networks to
problem efficiently. solve manufacturing problems - to form group technology
based manufacturing cells and to monitor cutting tool
Kaparthi and Suresh [20] present a neural network condition.
clustering method for the part-machine grouping problem.
This method is based on a neural network algorithm to To form a manufacturing cell, a neural network was
support procedures like production flow analysis. A made to operate on the classical machine/components
neural network clustering algorithm using similarity matrix Patterns representing respectively the machine
coefficients, is used to solve the part-machine grouping characteristics and the component characteristics are used
problem. The neural network method is based on as input to the network. This approach was tested on two
unsupervised learning. In the part-machine matrix, each such matrices, one consisting of 10 jobs and 15 machines,
row (part) is considered as a vector in a higher and the other 14 machines and 24 parts. The author
dimensional space and every dimension corresponds to a concluded that the results produced by the neural network
machine type, and the number of dimensions is given by approach compare well with the classical clustering
the total number of machine types required for all parts techniques [ 3 71.
[20]. The authors have shown that the neural network is
capable of handling large data sets. The monitoring of tool condition during metal cutting
operations is a classical production topic for which a
Lee et al., [24] present a method for part family substantial amount of research work has been carried out,
formation, machine cell identification, bottleneck machine the problem is to monitor accurately the condition of the
detection and the natural cluster generation is done using cutting tool on-line so that a worn-out tool can be detected
a self-organizing neural network. The authors argue that quickly and subsequently replaced [37]. The methods to
the generalization ability of the neural network makes it detect tool failure are based on the level of cutting force,
possible to assign the new parts to the existing machine vibration or acoustic emission, and these are measured by
cells without repeating the entire computational process. sensors that are attached to the tool. But the main
The authors show that neural networks can learn from a difficulty is how to detect the signal pattern produced by
given set of patterns and are able to generalize this a worn tool.
knowledge to other similar problems. This property
makes them useful in small and medium-size batch- Some of the signal components are selected from a
manufacturing systems where training data are limited and number of raw input data which are more sensitive to tool
new parts are continuously encountered. The authors ware, but less sensitive to process noise. A set of training
point out that their method based on neural networks is not inputs are chosen which consists of signal patterns
significantly influenced by the size of the machine-part representative of both fresh tool cutting, and worn tool
matrix and hence it is appropriate for solving large-scale cutting and these are used to train a single hidden layer
industrial problems. back-propagation neural network with a 8-3-l structure,
i.e. 8 input processing units, 3 hidden processing units and
Moon [26] presents a neuro-computing model to 1 output processing unit, processing units are the building
identify part family/machine groups. Similarity bricks of neural network [37]. Target outputs of 0.01 and
coefficients based on, for example McAuley’s method, is 0.99 aTe set for fresh and worn cutting patterns.
determined. The similarity coefficients are used as
connection weight values for neurons within the machine After training, each layer of the neural network acts as
and part layer. A neuron from either the part layer of the a signal filter so that the neural network suppresses noise
machine layer is then chosen arbitrarily. The neural and increases the signal/noise ratio step by step as the input

450

Proceedings of the 1996 Hawaii International Conference on System Sciences (HICSS-29)


1060-3425/96 $10.00 © 1996 IEEE
Proceedings of the 29th Annual Hawaii International Conference on System Sciences - 1996

patterns propagate through the network, the trained neural The relative importance of various manufacturing
network produced a 94% success rate for worn pattern decision making criteria and the overall performance of a
detection [37]. Neural network monitors successfully the manufacturing system is impossible to analytically
tool condition in metal-cutting operations. estabhsh[ll]. Chryssolouris et al., [l l] explore the use of
neural networks for identifying the relative importance of
Yih et al.,[39] uses a hybrid method that combines decision criteria. Simulation and a neural network are
human intelligence, optimization and neural network to used to establish adequate weights of the criteria for the
solve a real-time robot scheduling problem in a circuit decision-making process at the work center level. A
board production line. The hybrid method is divided into procedure for determining operational policies for
three phases. ln phase I, a simulator collects data from manufacturing systems has been presented. The authors
human schedulers. Phase II constructs Semi-Markov use simulation results to train a neural network which then
decision models which are then used to fmd solutions to prescribes an operational policy suitable for achieving a set
the scheduling problem. Solutions derived from phase II of goal performance measures [ 111. The proposed neural
are then fed as training data for the neural network. In the network procedure determines suitable criteria weights for
neural network model, there exists three layers: an input an entire sequence of multiple-criteria decisions, The
layer, hidden layer, and output layer. The nodes between authors conclude that the method is better suited to
the input layer and hidden layer are fully connected, as are complex applications involving chains of decisions, such
those between the hidden layer and the output layer. The as job shop scheduling, while the conventional methods
input layer contains nine nodes, each correspond to one are likely better suited to isolated, single decisions [ 111.
attribute in the state definition and the output layer has six
nodes that represent six possible decisions in the system. Yao et a1.,[38] uses a distributed neural network of
For the hidden layer twenty nodes are selected after coupled oscillators to solve an industrial pattern
systematically experimenting with various hidden nodes recognition problem. The problem addressed is machine
based on the performance of the resulting neural network. recognition of industrial screws, bolts, etc. in simulated
The neural network was trained for twenty hours on a 486 real time in accordance with tolerated deviations from
IBM-compatible personal computer. The duration of manufacturing specifications. Inputs are preprocessed and
training was determined by observing the level of errors of represented as 1 x 64 binary vectors. The supervised
the neural network and the decrement in error rate. neural network uses the backpropagation method to
Training the neural network was concluded when the error accomplish the pattern recognition task. The authors state
levels were low and decrement in error rate was that the neural network method performs better than a
approaching zero. The resulting network model was then standard Bayesian statistical method.
evaluated in the real-time robot scheduling problem [39].
IV. Summary
Cook and Shannon [lo] present a methodology to
predict the occurrence of out-of-control process conditions The methods, structures and applications of neural
in a composite board manufacturing facility. This method networks are presented in this paper. Ever since the
is developed using neural network theory. The neural beginnings of research in artificial intelligence, neural
network, using back-propagation method, was network has shown tremendous potential in solving
successfully trained to represent the process parameters. complex problems. Neural network methods based on
The trained neural network was able to successfully supervised and unsupervised learning are used in
predict the state of control of the specific manufacturing numerous manufacturing applications. Solution methods
process parameters with 70% accuracy. The learning rule based on neural network are elegant and efficient in
used in this research was the generalized delta rule which solving a wide range of problems in manufacturing that
is an error-correcting rule that has been used in various are critical for companies’ survival in today’s competitive
applications including converting printed text to speech, environment. Applications of neural network presented in
controlling robot arms, and selecting good loan this paper are only a small sample of a large number of
applications [lo]. The back-propagation learning potential applications.
algorithm uses a gradient- or steepest-descent heuristic that
helps a network to organize itself in ways that improve its BIBLIOGRAPHY
performance over time. Continuous manufacturing
processes with measurable process parameters are VI. Abutaleb, A.S., “A Neural Network for estimation of
promising application areas for neural network modeling forces acting on Radar Targets”, Neural Networks,
[lOI. vo1.4, 1991, pp.667-678.

451

Proceedings of the 1996 Hawaii International Conference on System Sciences (HICSS-29)


1060-3425/96 $10.00 © 1996 IEEE
Proceedings of the 29th Annual Hawaii International Conference on System Sciences - 1996

PI. Atiya, F. A., “An Unsupervised Learning Technique D51. Fukushima, K., and Miyake, S., “Neocognitron: A
for Artificial Neural Networks,” Neural Networks, New Algorithm for Pattern Recognition Tolerant of
vo1.3, 1990, pp. 707-711. Deformations and Shifts in Position,” Pattern
Recognition 15(6), 1982, pp. 455-469.
[33. Baran, N., “The Outlook for Pen Computing,” BYTE,
September 1992, The Worldwide Computing V61. Fukushima, K., “Cognitron: A Self-organizing
Authority. Multilayered Neural Network,” Biological Cybernetics
20, 1975, pp. 121-136.
M. Bebis, G.N., and Papadourakis, G.M., “Object
Recognition using invariant object boundary P71. Jacobs, R.A., “Increased Rates of Convergence
representations and Neural Network Models,” Pattern through Learning Rate Adaption,” Neural Networks,
Recognition, Vo1.25, No.1, 1992, pp.25-44. ~01.1, 1988, pp.295-307.

r51. Blum, A., Neural Networks in C++ - An Obiect- U81. Johnson, K., Daniell, C., and Burman, J., “Feature
Oriented Framework for Building Connectionist Extraction in the Neocognitron,” IEEE International
Systems, John Wiley and Sons, New York, NY, 1992, Conference on Neural Networkq Vol. 2, 1988, pp.
pp.l-10. 117-126.

161. Carpenter, G.A., and Grossberg, S., “The ART of D91. Kaparthi, S., and Sure&, N.C., “A neural network
adaptive pattern recognition by a self-organizing system for shape-based classification and coding of
neural network,” Computer, 21 (3), 1988, pp.77-88. rotational parts,” International Journal ofProduction
Research, 29 (9), 1991,pp.1771-1784.
[71. Caudill, M., “Back Propagation Networks,” Naturallv
Intelligent Svstems, Chapter 14, 1990. PO]. Kaparthi, S., and Suresh, N.C., “Machine-component
Cell Formation in Group Technology: A Neural
PI. Caudill, M., “Neural Networks, Primer Part II,” AI Network Approach,” International Journal of
Expert, February, 1988, pp. 55-61. Production Research 1992, Vol. 30, No. 6, pp. 1353-
1367.
PI. Caudill, M., and Butler, C., “A Neural Network Model
for Selective Attention,” Proceedings of the IEEE Pll. Khotanzad, A. and Lu, J.H., “Distortion Invariant
First International Conference on Neural Networks. Character Recognition by a Multi-layer Perceptron and
Vol. 2, 1987, pp.ll-18. Back-Propagation Learning,” IEEE International
Conference on Neural Networks 1988 - Volume I,
uw Cook D.F., and Shannon, R.E., “A predictive neural pp.625632.
network modelling system for manufacturing process
parameters,” International Journal of Production P21. Lippmann, R.P., “Pattern classification using neural
Research, 1992, Vol. 30, No.7, pp.1537-1550. networks,” IEEE Communications Magan’ne,
November, 1989, pp.47-64.
[ill. Chryssolouris, G., Lee, M., Domroese, M., “The Use
of Neural Networks in determining Operational ~231. Lippmann, R.P., “An introduction to computing with
Policies for Manufacturing Systems,” Journal of nemal nets,”IEEE A S S P Magazine, April, 1987, pp.4-
ManufactutingSystems, Volume lOiNo.2, pp.166175. 22.

WI. Eberts, R., “Using Neural Nets for Assistance in ~241. Lee, H., Malave, CO., and Ramachandran, S., “A self-
Interacting with Computerized Systems,” Proceedings organizing neural network approach for the design of
ofthe I992 NSF Design and Manufacturing Systems, cellular manufacturing systems,” Journal ofIntelligent
pp. 687-690. Manufactuting (1992)3,325-332.

P31. Fukushima, K., and Wake, N., “Handwritten ~251. Mehr, D., and Richfield, S., “Neural Net Application
Alphanumeric Character Recognition by the to Optical Character Recognition,” IEEE Zntemafional
Neocognitron,” IEEE Transactions on Neural Conference on Neural Networks, Vol. 4, 1987, pp,
Networks. Vol. 2,1991, pp. 355-365. 771-776.

[141. Fukushima, K., Miyake, S., and Ito, T., [261. Moon, Y.B., “Establishment of a neurocomputing
“Neocognitron: A Neural Network Model for a model for part family/machine group identification,”
Mechanism of Visual Pattern Recognition,” IEEE. JoumalofIntelligentManufacturing, (1992)3, pp.173-
Trans. Systems,Man, and Cybernetics, Vol. SMC-13, 182.
No. 5, 1983, pp. 826-834.

452

Proceedings of the 1996 Hawaii International Conference on System Sciences (HICSS-29)


1060-3425/96 $10.00 © 1996 IEEE
Proceedings of the 29th Annual Hawaii International Conference on System Sciences - 1996

v71. Moon, Y.B. and Chi, SC.,” Generalized Part Family [401. Zahedi, F., “An Introduction to Neural Networks and
Formation Using Neural Network Techniques,” a Comparison with Artificial Intelligence and Expert
Journal ofManufacturing Systems, Volume 1 l/No. 3, Systems,“Znterfnces 21: 2 March -April 1991, pp.25-
pp.149-159. 38.

P81. Pao, Y., Adautive Pattern Recoenition and Neural


Networks, Addison-Wesley, New York 1989.

P91. Pawlicki, T. F., Lee, D-F., Hull, J.J., and Srihari, S.N.,
“Neural Network Models and their Application to
Handwritten Digit Recognition,” IEEE International
Conference on Neural Networks, Vol. 2., 1988, pp. 63-
10.

[301. Rumelhart, D.E., Hinton, G.E., Williams, R.J.,


“Learning Internal Representations by Error
Propagation,” Parallel Distributed Processing, Chapter
8, vol. 1, 1986.

[311. Sankamarayanan, P., “Application of the


Neocognitron to the Character Recognition Problem,”
unpublished Master’s Thesis, Central Michigan
University, December 1992.

~321. Srinivasa, N., Kalle, P., Ziegert, J.C., and Smith, S.,
“Neural Network Prediction of Machine Tool Error
Maps,” Proceedings of the I992 National Science
Foundation (NSF) Design and Manufactun’ng
Systems, pp. 1099-1103.

[331. Vogl, T.P., et al., “Accelerating the Convergence of the


Back-Propagation Method,” Biological Cybernetics,
vol. 59,198X, pp.257-263.

[341. Wasserman, &Ural Computing, pp. 167-188.

[351. Wilson, R., and Sharda, R., “Neural Networks,”


OR/MS TODAY, A Joint Publication of ORSA and
TIMS, August 1992, pp. 36-42.

[361. Winograd, T., and Flores, F., m


Comnuters and Coenition.

[371. Wu, B., ” An introduction to neural networks and their


applications in manufacturing,” Journal of Intelligent
Manufacturing (1992)3, pp.391-403.

1381. Yao, Y., Freeman, W.J., Burke, B., and Yang, Q.,
“Pattern Recognition by a Distributed Neural Network:
An Industrial Application,” Neural Networks, Vol. 4,
1991, pp.103-121.

L391. Yih, Y., Liang, T-P, and Moskowitz, H., “Robot


Scheduling in a Circuit Board Production Line: A
Hybrid OR/ANN Approach,” ZZE Transactions,
Volume 25, Number 2, March 1993, pp.26-33.

453

Proceedings of the 1996 Hawaii International Conference on System Sciences (HICSS-29)


1060-3425/96 $10.00 © 1996 IEEE

You might also like