1. Network Planning
A telephone network is given with specified flows between each pair of nodes. The problem is to efficiently cover the network by a set of rings with specific properties, so that all flows are transferred through the rings without exceeding their capacities. Criteria for efficiency are: equipment cost and flexibility, namely the maximal ratio of the flow in a ring and its capacity. A planning tool was constructed which allows the user to define and edit a network with its pertinent data and obtain efficient sets of rings and optimal sets of paths and flows.
The problem consists in detecting a segment of a border of a
part. In some cases it is a difficult
problem due to light effects. The
suggested algorithm first searches for (within a pre-specified region of
interests) an initial point belonging to the considered edge and then proceeds
along the margin. The algorithm finds
the next point maximizing the suggested threshold criterion within the square
with the center at the previous found point.
To retain the found points near the border the algorithm adaptively
determines upper bounds of gray level in the considered rectangles. Another
algorithm based on a special gradient procedure was used for exact detect of
the edges. The algorithm uses very
small sliding window and provides for sub-pixellian
accuracy of edge detection in the case of a clear edge. The sub-pixellian detector is based on very specific technique
originating from the control theory and real-time robust differentiators.
3.
Railroad Transportation
The Israeli railroads company needs a planning tool to schedule the transportation of empty freight cars so as to enable the daily realization of freight transportation plans. Three problems have been investigated: (1) an algorithm to select which transportation orders to satisfy from among the incoming stream considering the expected revenue and some administrative constraints; (2) application of the transportation problem with restrictions to determine the number of cars to be transported from points where they are found in the evening to points where they are needed next morning; (3) a combination of optimization and heuristic methods to plan the transportation by existing trains. An appropriate scheme for data collection and instruction distribution has been designed.
4. Image Understanding
An
image of an area is given, and it is required to define the locations of the
area and of the photographing camera within some pre-defined scene, as well as
the camera parameters. The method of solution is as
follows: details of the scene are extracted from a sequence of its images and
fed into a database. The recognition system tries to identify the details in
the image and to obtain equations comprising parameters of the camera and of
the found details. The solution of the equations provides the required data.
The problem is generalized for sequences of images.
5. Defect recognition on cutting tool surface
The problem is to
recognize surface defects of a cutting tool in order to perform an automatic
selection of damaged tools. The known
methods of pattern recognition do not solve the problem because they find
numerous objects not being defects. A
new algorithm was suggested taking into account the specific essential features
of the considered defects. The suggested
algorithm is applicable for various types of cutting tools (including carbide
inserts) and many examples confirm it.
It is necessary to emphasize that all gradient-based ideas do
not work here, for the gradients at the points of the defect boundary are much
smaller than at the points of most other “natural” edges. Contour tracing algorithms cannot be applied
as well, for they find domains that are more color-saturated than the
considered defects.
6. Fault
Management System
The Fault Management System receives signals about faults
found in the managed network. These alarm signals are analyzed in order to
pinpoint the fault that has caused the alarm and the eliminate it. Since the
flow of signals can contain hundreds and even thousands of practically simultaneous events, advanced
methods of analysis are required to
decode the signals and in particular to reveal correlation among them. The purpose
of establishing the correlation is to concentrate on the leading, determinant
events, but not on their corollaries, and thereby identify the most probable
root causes of the faults even in caseswhen no signal
is received directly from the root of the problem. The decision on the Alarm Correlation Problem is based on
the search for such propagation of faults and alarms which will most probably
result in a pattern of alarms identical
or close to that which was actually received at the given moment of time.
7. optimAL tank farms for a pipeline
The Eilat-Ashqelon pipeline serves to transport crude oil on
land from the
In the
course of the work on this subject a measure of performance for the system had
to be selected, and after analyzing the operational features it was taken to be
the average length of the queue of tankers that might form at both terminals.
The length of the queue directly affects the total expenses on the operation of
the entire system. The queue length is affected by a diversity of factors some
of which are uncontrollable, such as conditions at sea, and some that result
from human decisions, such as the order in which different types of oil are
pumped through the pipeline. Planning and designing the infrastructure of the
system depends significantly on its highly random operational behavior.
The tool
constructed to attack this problem not only reveals undesirable developments
resulting from inadequacies of the tank farms, but also produces
recommendations as to where and when additional tanks are expedient for
different annual flow rates. The tool also permits to ascertain the measure in
which the entire quality of operations is affected by various decisions, therefore it can be used for improving the
operational policy. The tool is realized in friendly and convenient software.
8. Planning device for Wafer Fabrication
The problem may be described as follows. N (~4000)
lots are given and each lot kÎN has its own sequence of technological stages Sk={sk1, …, skm}, where m (~300) is the number of
stages for the individual lot k. Stage skl
can be processed on the determined set of machines P(skl),
and on machine jÎ P(skl)
the processing time
is tj(skl).
Each lot k is to be completed before the due date dk.
We need to obtain a work plan, that is, for each lot to
determine the machine for stage skl
and the time for the beginning of this stage. The plan is more detailed for the
next shift and less detailed for several subsequent shifts. The quality of the
plan is estimated by several optimization criteria.