The nearest perspectives of applications of artificial neural networks for diagnostics, modeling, and control in science and industry (methods, results, demonstrations

1
The nearest perspectives of applications of
artificial neural networks for diagnostics,
modeling, and control in science and industry
(methods, results, demonstrations hands-on)
Chuvash State University, Russia Department of
Thermo Physics
Victor S. Abrukov (2008) e-mail
abrukov_at_yandex.ru
2
Outline

• 1. Introduction
• Ex-1 Deflagration to detonation transition
• Ex-2 Automatic control system of steam
  generating unit
• Ex-3 Wave propagation on a free surface of a
  fluid
• Ex-4 Burning Rate vs Temperature profile
• 2. Perspectives of others (laser scattering
  technique for remote sensing, , smart sensors for
  monitoring of industrial equipment (smart watch,
  soft sensors), etc.
• 3. Conclusions

3
Introduction - 1

• Commercial applications of Data Mining (ANN, etc)
  in areas such as e-commerce, market-basket
  analysis, text-mining, and web-mining have taken
  on a central focus in last years.
• There is a significant amount of innovative Data
  Mining work taking place in the context of
  scientific and engineering applications but that
  is not well-represented in the mainstream of Data
  Mining.

4
Introduction - 2

• Scientific Data Mining techniques are being
  applied to diverse fields such as physics,
  chemistry, astronomy, engineering, etc.
• Data Mining frequently enhances existing analysis
  methods based on statistics and exploratory data
  analysis.

5
The goal

• The goal of the seminar is
• Share experiences
• Learn how ANN can be applied to their data
• Practice in it (You will the first user of Data
  Fusor in all of the world!)

6
Introduction (main!)

• No Science without Data Base
• and
• Data Analysis

7
Data Base

• There are three ways for creation Data Base
• Real experiment
• Numeral experiment
• Analytical Solution
• We have obtained DB. What do we have to do
  further?

8
Data Analysis

• We have to create a model of DB that help us to
  use DB (for understanding, diagnostics, testing,
  control, pleasures).
• What is the best way?
• I think Artificial Intelligence.
• Data Fusor is a part of AI tools
• ANN is a part of Data Fusor

9
EX-1 Modeling of Deflagration-to-Detonation
Transition for Pulse Detonation Engine.
10
Ex-1

• A deflagration-to-detonation transition under
  various experiment conditions was studied by
  Santora R.J. at al (USA).
• The data that were presented by it had many
  blanks (about 60). A task of filling them by
  means of ANN was set.
• The results obtained are presented in Table.

11
Ex-1
12
Ex-2 ANN Model of Automatic Control System of
Boiler Unit

• The work has been aimed on a development of
  principles of creation of a new model of
  automatic control system of a boiler unit in
  transient (non-linear) regimes (change of load,
  launch and stop of the boiler aggregate, etc) on
  a basis of ANN.

13
The following problems were set and solved -1

• - A complete set of non-linear differential
  equations for the two-phase flow in cylindrical
  coordinate system circumscribing processes in a
  superheater was formulated. It described also a
  process of injection of water into a superheater
  during control procedure of vapor temperature in
  a superheater.
• - The obtained set of equations was converted
  into a system of finite-difference equations.

14
Equations
15
Equations
16
The following problems were set and solved - 2

• Adjustable and controlling parameters were
  selected by means of experimental data and
  experts experience.
• The rate of a change of vapor temperature on an
  exit of the boiler aggregate was selected as a
  controlled parameter. The water mass flow via
  injector was selected as controlling parameter.
• The analytical connection between them was
  determined by means of the mathematical apparatus
  of Lee derivatives.

17
The analytical connections
18
The following problems were set and solved - 3

• The database for training of ANN was created by
  means of the obtained analytical connection and
  an optimal architecture of ANN was determined.
• The training of ANN was executed and a model of a
  control system as well as a skeleton diagram of a
  position of a gate valve governing a water
  discharge on injection were created.

19
Data Base for training
20
The following problems were set and solved - 4

• The prototype of an automatic control system of
  temperature vapor in a superheater of a boiler
  unit TGME-464 was designed.
• It consists of the program emulator of ANN
  established on the personal computer, industrial
  microcontroller Philips P89LPC935, cable lines
  and switch gears.  
• The obtained prototype can work both in a mode of
  advice, and in a mode of automatic control. It
  can be simply enough integrated in present
  control systems.
• The features of ANN technologies of control
  (property of adaptability to new conditions and
  ability to self-training) provide simplicity of
  modernization and escalating of capabilities.

21
Scheme of adaptive control
22
Ex 3 Wave propagation on a free surface of a
fluid

• The tasks of hard shock about a layer of fluid
  and of wave propagation on a free surface of a
  fluid were the problems (at an impulse
  formulation, the deformation of the bottom of an
  earthquake occurs for a rather short time, i.e.,
  there is a bottom shock about a fluid)
• With the help of an available analytical solution
  the data base consisting of values of
  dimensionless coordinates, times and the velocity
  on a free surface of a fluid were obtained.

23
Ex 3 A part of Data Base

• The values of coordinates and times were
  considered the input characteristics, and the
  values of velocities the output.

24
Ex 3 Scheme of ANN

25
Ex 3 results

• The results of ANN calculations of the velocity
  values as related to different coordinates and
  times together with analytical results are
  presented in Figures.

26
Ex 3 results (t2 sec)(coordinate y is U
velocity)

27
Ex 3 results (t3,5 sec)

28
Ex 3 results (t4 sec)

29
Ex 4 Burning Rate vs Temperature Profile

• The black box computational model of a flame
  temperature profile prediction is created.
• It allows to predict the temperature profiles by
  means of data about heat of combustion, burning
  rate, and pressure
• WITHOUT NEW MEASUREMENT EVERY TIME

30
EX - 4

31
Ex - 4

• The scheme of construction of an ANN PCW model
  was as follows.
• We have taken experimental data (from scientific
  and applied literature) about
• pressure of burning, P,
• burning rate, U,
• heat of propellant burning, Q,
• as well as temperature profile (values of
  temperature, T and coordinates, x) for some
  propellants

32
Ex - 4

33
Ex 4

• The different sets of values of pressure,
  burning rate, heat of combustion and coordinates
  were installed on the input layer of ANN. The
  corresponding values of temperature were
  installed on the output layer of ANN.

34
Ex - 4

• By means of a training tool named the method of
  back propagation of errors1, we have created
  the ANN PCW model.
• This model is a black box type which consists
  of latent connections between variables.
• The black box obtained we have used in an
  invented experimental investigation for the
  prediction of a temperature profile as follows.
  One of the experimental data that have not been
  used for training of the ANN we have used for
  checking of the black box obtained. The values
  of pressure, burning rate, heat of burning and
  coordinates were installed on the input layer of
  the black box. Then, the correspondence values
  of temperature T, were obtained on the exit of
  the output layer of the black box.
• The results we have obtained are presented in
  Table 2.

35
Ex - 4

36
Ex - 4

• An analysis of Table depicts that the black box
  enough correctly determines the temperature
  profile by means of data about burning rate,
  pressure and heat of burning (or maximum
  temperature of flame).
• In our opinion, this way of determination the
  temperature profiles could be considered as
  perspective unique tool for cases when we have no
  any experimental results for concrete propellants
  and would like to have them.
• In order to make it all we need to do is to
  collect a lot of experimental data published in
  scientific and technical literature concerning
  temperature profiles for various propellants upon
  various condition.
• Then we could create the common black box
  model Burning Rate vs Temperature Profile.

37
Perspectives

38
Conclusions

• Practice is the best of all instructors