The present research lines at AAIL are:

1. Evolutionary computation and other metaheuristics applied to engineering;
2. Neural networks and fuzzy logic in the operation of industrial plants;
3. Neural networks and fuzzy logic in non-destructive essays;
4. Fuzzy logic and "data mining" support to decision making

1.1 Objective

This research line applies evolutionary computation techniques and other metaheuristics
to investigate the solution to several nuclear engineering problems, with emphasis, among
others, on the application of

• genetic algorithms (GA);
• genetic programming (GP);
• particle swarm optimization (PSO);

to:

• optimizing nuclear fuel reload patterns
• neutronic design;
• planning of maintenance policies and monitoring tests for nuclear systems;
• transient identification;
• thermal-hydraulic design;
• control;

1.2 Present research

1.2.1 Parallel Evolutionary Computation to Support the Operation of Nuclear
Reactors

Coordinator: Cláudio M. N. A. Pereira
Researchers: Cláudio M. N. A. Pereira, Antônio C. A. Mol, Marcel Waintraub, Rafael P.
Baptista, Roberto Schirru (COPPE)
Research type/external sponsorship: CNPq Project (Edital Universal
PROC.474889/2004-1)
Summary:
The overall objective of this project is to investigate and develop parallel evolutionary
computation models as a support to the operation of nuclear power plants. We point out,
however, two important distinct targets: I) investigation and development of parallel GA
models for reactor reload optimization; II) investigation and development of parallel
Genetic Programming (GP) models, a GA extension whose objective is knowledge
extraction (discovery) for application in transient identification in nuclear systems.

1.2.2 Parallel Genetic algorithms with Niche Techniques Applied to Nuclear Reactor
Design Optimization

Coordinator: Cláudio M. N. A. Pereira
Researchers: Cláudio M. N. A. Pereira, Marcel Waintraub e Rafael P. Baptista
Research type/external sponsorship: CNPq Project (Productivity in Research)
Summary:
The present project has the main objective of improving the optimization process of
complex problems through genetic algorithms. Here, the use of a hybrid methodology of
niche techniques (NGA) and parallel GA models (PGA) is proposed. Then, the
development of a Parallel Genetic Algorithm with Niche Techniques (PGAN) is proposed
regarding its application to an optimization problem in nuclear reactor design.

1.2.3 Optimization of Nuclear and Thermoelectric Power Plants Thermal Efficiency through Genetic Algorithms

Coordinator: Nelbia S. Lapa, Cláudio M. N. A. Pereira
Researchers: Nelbia S. Lapa, Cláudio M. N. A. Pereira, Celso M. F. Lapa, Wagner Sacco
Research type/external sponsorship: Doctoral thesis COPPE/UFRJ
Summary:
The present work proposes the utilization of a genetic algorithm for extraction optimization
in the turbines of a PWR-type reactor’s secondary system so that thermal efficiency can be
maximized. Such methodology can be applied, in the same way, to thermal plants.

1.2.4 Optimization of a PWR reactor design considering thermal-hydraulic safety
and feedback aspects

Coordinator: Celso M. F. Lapa
Researchers: Celso M. F. Lapa, Eugênio Martins, Antônio C. M. Alvim (COPPE)
Research type/external sponsorship: Doctoral thesis COPPE/UFRJ
Summary:
This work investigates the utilization of a genetic algorithm in the optimization of a PWR
reactor design, considering, besides neutronic, aspects of reactor safety and thermal-
hydraulic feedback, which bring important restrictions to the project.

1.2.5 GA and PSO parallel models applied to the optimization of the reload of PWR type reactors

Coordinator: Cláudio M. N. A. Pereira
Researchers: Cláudio M. N. A. Pereira, Marcel Waintraub
Research type/external sponsorship: Doctoral thesis COPPE/UFRJ
Summary:
In the reload problem, the objective is to find out the best nucleus load pattern that leads to
a longer burning period, thus increasing the extraction of energy from the combustible
elements. To do so, “heavy” reactor physics codes need to be run hundreds or thousands
of times, which makes this application not feasible for simple personal computers. Parallel
computation is, therefore, a way to overcome such difficulty. To avoid the utilization of
high-priced special computers, a low cost option (but which in this case does not imply
being less efficient) is the use of clusters of personal computers. The purpose of this
project is to study efficient models for paralleling genetic algorithms and PSO at IEN’s
cluster, for application to the problem of nuclear reactor reload.

1.2.6 Parallel genetic programming in scheduling and task assignment problems

Coordinator: Cláudio M. N. A. Pereira
Researchers: Cláudio M. N. A. Pereira, Rafael P. Baptista, Roberto Schirru (COPPE)
Research type/external sponsorship: Doctoral thesis COPPE/UFRJ
Summary:
This work studies efficient parallel models of restricted genetic programming for application
in problems of scheduling and task and resource assignment related
to the operation of nuclear power plants.

1.2.7 Application of genetic algorithms to the integrated optimization of the
availability of safety systems

Coordinator: Cláudio M. N. A. Pereira
Researchers: Cláudio M. N. A. Pereira, Vinícius Damaso e Paulo F. F. F. e Melo
(COPPE)
Research type/external sponsorship: Doctoral thesis COPPE/UFRJ
Summary:
This work presents a genetic algorithm based method of integrated optimization of system
performance. Its objective is to maximize the gain obtained with the operation of a
simplified system, where aspects related to structuring and design, reliability, redundancy
allocation, maintenance and tests planning and costs are considered simultaneously and
in an integrated way. The availability modeling deals with the original non-linear problem
without making any change. The set of solutions obtained supported decision taking in
situations subject to budget and safety restrictions. The results presented show the
benefits of an integrated optimization, which contemplates the interaction between design
and availability, when maximizing the performance.

1.2.8 Planning preventive maintenance and periodic tests policies via PSO

Coordinators: Cláudio M. N. A. Pereira and Celso M. F. Lapa
Researchers: Cláudio M. N. A. Pereira, Celso M. F. Lapa and Newton Norat
Research type/external sponsorship: Master's Thesis IEN/CNEN
Summary:
This research proposes the utilization of PSO as an alternative method of optimizing
policies of preventive maintenance and periodic tests in NPP, what has been done
successfully through genetic algorithms. The main idea is to find the best intervention
policy that leads to a greater integrated availability for a given period.

1.2.9 Design of reduced-scale thermal-hydraulic experiments using genetic
algorithms

Coordinators: Celso M. F. Lapa and David Botelho
Researchers: Celso M. F. Lapa, David Botelho, Cláudio M. N. A. Pereira, Paulo A. B.
Sampaio, Maria de Lourdes Moreira
Research type/external sponsorship: IRIS Project (AIEA)
Summary:
In the design of reduced-scale thermal-hydraulic experiments, geometry, flows, thermal
power, pressure, etc, need to be determined so as to reproduce the desired characteristics
of the real scale system. For that, it is necessary that certain adimensional numbers (which
characterize similarity between systems) be reproduced in the reduced scale experiment.
This research investigates the utilization of genetic algorithms in the determination of the
reduced scale design characteristics of the IRIS pressurizer.

1.2.10 Multi-objective optimization of availability and costs in the planning of
policies of preventive maintenance and monitoring tests via genetic algorithms.

Coordinators: Celso M. F. Lapa and Cláudio M. N. A. Pereira
Researchers: Celso M. F. Lapa, Cláudio M. N. A. Pereira and Newton Norat
Research type/external sponsorship: IEN Research
Summary:
This research has the objective of investigating the optimization of intervention policies for
maintenance and monitoring tests in nuclear systems, with two objectives: i) increase
availability and ii) minimize costs, considering practical questions such as seasonality,
availability of maintenance teams, etc.

2 Neural networks and fuzzy logic in industrial plant operation

2.1 Objective

• transient identification;
• signal validation;

2.2 Present research

2.2.1 Identifying nuclear transients with “don't know” response through neural
networks and fuzzy logic

Coordinator: Antônio C. A. Mol
Researchers: Antônio C. A. Mol, Cláudio M. N. A. Pereira, and Mauro Victor
Research type/external sponsorship:
Summary:
This project has the objective of developing efficient methodologies to identify transients in nuclear power plants. For that, approaches based on neural networks and fuzzy logic are used to increase identification efficiency and include the “don’t know” response capability in the case of unlabeled transients.

2.2.2 Neural redundancy in signal validation
Coordinator: Antônio C. A. Mol
Researchers: Antônio C. A. Mol, Cláudio M. N. A. Pereira,
Research type/external sponsorship:
Summary:
This research introduces the concept of neural redundancy and applies it to the parity
space method to overcome a deficiency inherent to this method: the determination of the
best measurement estimate in the case of completely inconsistent redundant
measurements. The neural redundancy concept consists of the calculus of a redundancy
utilizing artificial neural networks (NNs) trained with the recent history of its own state
variable. This way, NNs dynamically trained with the temporal series estimate a value for
the measurement, which in turn will be used as the “referee” for the redundant
measurements in the parity state. To be able to reproduce the temporal series trend even
in an accident condition, NN dynamic training privileges the series recent points. The tests carried out with simulated nuclear plant data showed that neural redundancy applied to the parity space method improves the signal validation process.

2.2.3 Methodology for signal validation using empirical models with artificial
intelligence techniques applied to a nuclear reactor

Coordinator: Mauro V. de Oliveira
Researchers: Mauro V. de Oliveira, Roberto Schirru
Research type/external sponsorship: Doctoral thesis COPPE/UFRJ
Summary:
This work develops a methodology for building empirical models for signal validation, with
artificial intelligence techniques, in order to perform the analytical redundancy of monitored
signals in an industrial plant.

2.2.4 Failure mode and effect analysis via fuzzy logic for the assessment of nuclear power plant risk and operational life extension.

Coordinator: Celso M. F. Lapa
Researchers: Antônio C. F. Guimarães, Celso M. F. Lapa
Research type/external sponsorship: FAPERJ Project
Summary:
This project aims to develop a fuzzy-logic-based methodology for failure mode and effects analysis to assess nuclear plant risk and operational life extension.

3 Neural networks and fuzzy logic in non-destructive essays

3.1 Objective

This research line investigates the utilization of neural networks and fuzzy logic to identify
the signal pattern originating in non-destructive essays, involving, among others:

• applied nuclear techniques;
• applied ultrasound techniques;

In problems such as:

• triphasic flow identification
• detection of welding flaws
• others

3.2 Present research lines

3.2.1 Real-time triphasic flow measurement with gamma radiation attenuation and artificial intelligence techniques for use in the petroleum industry

Coordinators: Luis E. B. Brandão and Cláudio M. N. A. Pereira
Researchers: Luis E. B. Brandão, Cláudio M. N. A. Pereira and César M. Salgado
Research type/external sponsorship: Doctoral thesis COPPE/UFRJ
Summary:
The accurate real-time assessment of the efficiency of the triphasic separators used in the
petroleum industry is fundamental to the optimization of petroleum production and
transportation costs. The application of the gamma radiation attenuation technique is
proposed in replacement of the present measurement systems, with the advantage of
being non-invasive. The utilization of artificial intelligence becomes an important
differential due to the little a priori knowledge about outflow characteristics.

3.2.2 Real-time triphasic flow measurement with neutron and artificial intelligence techniques in the petroleum industry

Coordinators: Luis E. B. Brandão and Cláudio M. N. A. Pereira
Researchers: Luis E. B. Brandão, Cláudio M. N. A. Pereira and Robson Ramos
Research type/external sponsorship: Doctoral thesis COPPE/UFRJ
Summary:
The accurate real-time evaluation of the efficiency of triphasic separators used in the
petroleum industry is fundamental to optimize petroleum production/transportation costs.
The application of a nuclear technique (neutron emission), proposed in replacement of the
present measurement systems, has the advantage of being non-invasive. The utilization of
artificial intelligence becomes an important differential due to the little a priori knowledge
about outflow characteristics.

4 Fuzzy Logic and data mining for decision taking support

This research line has the objective of investigating the utilization of fuzzy logic in the
decision taking process, comprehending, among others, the areas of:

• technological innovation;
• technology transfer;
• logistic;

4.2 Present research lines

1.2.1 Fuzzy logic in strategic technology transfer assessment

Coordinators: Ana Gabriella A. A. Pereira
Researchers: Ana Gabriella A. A. Pereira, Cláudio M. N. A. Pereira
Research type/external sponsorship: Doctoral thesis COPPE/UFRJ
Summary:
This work investigates the development of a decision support system using fuzzy logic to
help universities and research institutions to evaluate the effectiveness of a prospective
technology transfer.