APPLICATION%20OF%20THE%20METHOD%20AND%20COMBINED%20ALGORITHM%20ON%20THE%20BASIS%20OF%20IMMUNE%20NETWORK%20AND%20NEGATIVE%20SELECTION%20FOR%20IDENTIFICATION%20OF%20TURBINE%20ENGINE%20SURGING

1
APPLICATION OF THE METHOD AND COMBINED ALGORITHM
ON THE BASIS OF IMMUNE NETWORK AND NEGATIVE
SELECTION FOR IDENTIFICATION OF TURBINE ENGINE
SURGING

• Lytvynenko Volodymyr
• KHERSON NATIONAL TECHNICAL UNIVERSITY
• Ukraine

2
Contents

• I. Problem statement
• 1.1 Turbine engine surging
• 1.2 How it is possible to minimized
  consequences Surging gas turbine engine (GTE)?
• 1.3 What are used now methods of the
  decision of the given problem?
• II. Solving of the problem
• 2.1. Use of artificial immune systems
• - Algorithm of negative selection
• - Problems of use of algorithm of negative
  selection
• 2.2. The decision of problems of algorithm of
  negative selection
• - Artificial immune network
• - Adaptation of detectors
• - The developed combined algorithm
• III. Experiments
• 3.1. The first experiment
• 3.2. The second experiment
• 3.3. The third experiment
• IV. Current researches
• V. The future researches
• VI. Conclusion.

3
I. Problem statement
4
1.1 Turbine engine surging
In the given report the algorithm of definition
turbine engine surging is offered
What is Surging ?
Surging (fr. pompage) is stalled operating
mode of aviation gas turbine engine (GTE),
infringement of its gas-dynamic stability of
functioning accompanied by claps, sharp decrease
of thrust and powerful vibrations which are
capable to destroy the engine

• 1.2 How it is possible to minimized consequences
  Surging gas turbine engine (GTE)?

Prevention of the coming surging demands a
possibility of forecasting of approaching to
these modes and their instant registration.
1.3 What are used now methods of the decision of
the given problem?

• Method Fourier transform
• Wavelet-analysis
• Neural networks
• Robust statistics

5
II. Solving of the problem
6
Our decision of a problem

• To use for the decision of the given problem
  artificial immune systems
• To examine the decision of the given problem as a
  task of detection of anomalies

• We examine anomaly as a status of system which is
  not compatible to normal behavior of this system.
• According to this, an anomaly detection system
  will perform a continuous monitoring of the
  system and an explicit classification of each
  state as normal or abnormal.

7
2.1. Use of artificial immune systems
8
What methods the given problem by means of
artificial immune systems dares?

• For the decision of the given problem methods
  based on algorithm of negative selection are
  used.

9
Algorithm of negative selection
10
Algorithm of negative selection

• Formally it is possible to present algorithm of
  negative selection in the form of expression

11
In what an essence of algorithm of negative
selection?

1. Initialization randomly generate strings and
   place them in a set P of immature T-cells, Assume
   all molecules (receptors and self-peptides)
   represented as binary strings of same length L
2. Affinity evaluation determine the affinity of
   all T-cells in V with all elements of the self
   set S
3. Generation of the available repertoire if the
   affinity of an immature T cell (element of P)
   with at least one self-peptide is greater than or
   equal to a give cross reactive threshold, then
   the T-cell recognizes this self-peptide and has
   to be eliminated (negative selection) else the
   T-cell is introduced into the available
   repertoire A.

The process of generating the available
repertoire in the negative selection algorithm
was termed censoring phase by the authors. The
algorithm is also composed of a monitoring phase.
In the monitoring phase, a set S of protected
strings is matched against the elements of the
available repertoire A. The set S might be the
own set S, a completely new set, or composed of
elements of S. If recognition occurs, then a
non-self pattern (string) is detected.
Even the random generation of the repertoire P
results in algorithms with some drawbacks. First,
this approach results in an exponential cost to
generate the available repertoire A in relation
to the number of self strings in S. Second,
randomly generating P does not account for any
adaptability in the algorithm and neither any
information contained in the set S.
The negative selection algorithm suggests the
random generation of strings, until an available
repertoire A of appropriate size is generated.
This approach could be adopted in both algorithms.
12
Graphic representation of objects of algorithm

• U universum and set S of vectors which are
  classified as Self, and S? U

13
Problems of use of algorithm of negative selection
14
Limitation of algorithm of negative selection

• Casual generation of detectors does not give
  possibility to define their is minimum necessary
  quantity, sufficient for a covering of all set of
  "Non-Self
• High probability of education of "cavities" that
  worsens quality of recognition since "cavities"
  are areas in space of "Non-Self" which are not
  recognized by any of detectors
• Generation too a considerable quantity of
  detectors essentially slows down a recognition
  phase since any entering image is necessary for
  comparing to each of the created detectors

15
2.2. The decision of problems of algorithm of
negative selection
16
What it is necessary to make to eliminate
limitations of this algorithm?

• We have set for ourselves a problem to improve a
  method of generation of detectors which is
  applied at training of algorithm of negative
  selection which is capable is adaptive to select
  their options, quantity and an arrangement in
  phase space of an investigated signal

17
How we suggest to solve the given problem?

• We offer at generation of detectors for their
  adaptive and options, and also definitions of
  their optimum quantity and an arrangement in
  phase space of an investigated signal to use an
  artificial immune network.

18
Artificial immune network
19
What is the artificial immune network?
20
Artificial immune network
Network compression
Initial data (antigenes)
Network generation
Memory formation
Network activation
The trained network
21
Adaptation of detectors
22
Adaptation of detectors of an immune network for
a problem of negative selection
Ab
1. Representation of an individual (antibody)
Ag
2. Population of antigenes
Set of vectors of the training image representing
a phase portrait of a normal signal in
k-dimensional space
23
Adaptation of detectors of an immune network for
a problem of negative selection
? min
3. Calculation of affinity "antibody-antigene"
- Euclidean distance
- The parameter defining the importance
cross-reactivity a threshold r
24
Adaptation of detectors of an immune network for
a problem of negative selection
4. Calculation of affinity "antibody-antibody"
Depending on value fAb-Ab following situations
are possible
25
The developed combined algorithm
26
THE GENERALIZED SCHEME OF THE COMBINED NEGATIVE
SELECTION ALGORITHM AND AN IMMUNE NETWORK
27
III. Experiments
28
Experimental researches 1
Signal without anomalies (a training signal)
Phase portrait of a training signal (yt, yt1)
Class Self
Training sample of 200 points. The size of a
window 2
Results of learning AIS
kr 0.01
kr 0.1
Less steady decision
Steadier decision
29
Experimental researches 1
Signal with anomaly (a test signal)
Phase portrait of a test signal
Anomaly deviations on a phase portrait are
observed
Results of testing
It is recognized by 5th detectors
The histogram of the found out anomaly
(activation of detectors)
It is recognized by 3th detectors
30
Experimental researches 2 (Anomaly of parametre)
Investigated signal
Training data 1-100
Test data 100-200
Normal signal ? 4.0
Anomaly of parametre (? 3.6), data 112-121
Structure trained AIS
Activation of detectors in a place of occurrence
of anomaly
31
Experimental researches 3IDENTIFICATION OF
TURBINE ENGINE SURGINGFor the third experiment
the data have been used received on the test bed
for the aviation gas turbine engine. The data
represent four time series (Vk_3, Vk_P, Vv_3,
Vv_P) the signals received from gauges of
vibration of support on which the engine has been
fixed.
The graphs of time series, representing the
vibration
32
Structure of the trained immune network for
various values
33
THE HISTOGRAMS OF DETECTORS ACTIVATION
34
IV. Current researches

• A Hardware-Based realizations of the developed
  algorithm

35
(No Transcript)
36
(No Transcript)
37
V. The future researches and development

• In the further researches we plan
• To carry out comparative researches at the
  decision of the given problem with such methods
  as Method Fourier transform, the Analysis of a
  small wave, the Neural networks, the Steady
  statistics.
• To investigate identification possibility turbine
  engine surging on other parameters.
• To investigate possibility of the forecast on
  approximating wavelets-coefficients.
• To unite the given algorithm with the Bayes
  network

38
VI. Conclusion
1. The algorithm using mechanisms of artificial
immune networks for the decision of a problem of
detection of anomalies by a method of negative
selection is developed
2. Distinctive feature of algorithm is updating
of process of training thanks to which
possibility of adaptive selection of options is
realized, quantities and arrangements of detectors
3.The experimental study has shown a high
efficiency of the offered algorithm which is
linked to its computing stability thanks to
adaptive selection of the cross-reactive
threshold. Also optimality is achieved owing to
adaptive adjustment of the size of an immune
network, i.e. quantity of necessary detectors
high accuracy of detecting is shown, owing to
reduction of quantity and the sizes of "cavities"
created.
4.To compare the results of the algorithm an
exact benchmark diagnostics was used, supported
by experts. Results of diagnostics testify to
affinity of the estimates produced by the
experts, and the estimates generated by means of
the method and algorithm developed.
39
Thanks for attention!