| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.86 MB | Adobe PDF |
Advisor(s)
Abstract(s)
A Indústria 4.0 traz associado o conceito de fábricas inteligentes, que surgem para atender
a uma necessidade crescente de elevada flexibilidade e eficiência na fabricação dos
produtos. Os motores de indução trifásicos já são utilizados em larga escala pela indústria
e terão um papel primordial nas Smart Factories. A fim de mantê-los em funcionamento,
com elevado grau de fiabilidade, e de reduzir os custos associados a paragens e intervenções,
torna-se necessária a identificação de avarias, ainda em estágio precoce, de forma a
programar a manutenção antes da paragem total daqueles equipamentos.
Os métodos de diagnóstico de avarias online vêm sendo alvo de estudo há muitos anos. A
avaliação dos resultados da aplicação desses métodos de diagnóstico depende de um
especialista para uma interpretação e um diagnóstico precisos. Este trabalho aborda o uso
de Inteligência Artificial para a deteção de avarias em motores elétricos, de forma
automatizada, ainda em estágio precoce. Foram utilizados algoritmos de Machine
Learning, nomeadamente Support Vector Machines e Decision Trees para a deteção de
curtos-circuitos entre espiras, de forma preditiva, baseando-se em dados reais adquiridos
no Laboratório de Sistemas Electromecatrónicos do CISE - Centro de Investigação em
Sistemas Electromecatrónicos.
Como principal diferença em relação a outros trabalhos, será apresentada uma abordagem
baseada no Extended Park’s Vector Approach, que já é atualmente utilizada para esta
finalidade. Trata-se, portanto, de uma abordagem híbrida e tem como principal objetivo
obter o máximo de eficiência com o mínimo de features, a partir de uma ferramenta já
utilizada para este fim.
Como resultado, é apresentado um sistema de diagnóstico automatizado para a deteção de
curtos-circuitos estatóricos em motores de indução trifásicos que utiliza Support Vector
Machine e Decision Tree, contemplando cenários de funcionamento distintos,
nomeadamente condição normal e condição de desequilíbrio na rede de alimentação,
relativo ao deslocamento de fase ou à amplitude das tensões. Após a otimização dos
modelos, será realizada a validação em dados novos e os seus resultados serão discutidos.
Industry 4.0 brings the concept of intelligent factories, that arise to meet a growing need for high flexibility and efficiency in the manufacture of products. Three-phase induction motors are already used on a large scale by the industry and will play a key role in Smart Factories. In order to keep them in operation with a high degree of reliability, and to reduce the costs associated with stoppages and interventions, it is necessary to identify faults at an early stage, in order to schedule maintenance before the electric machine stops completely. Online fault diagnostics methods have been studied for many years. However, the assessment of their results depends on an expert for accurate interpretation and diagnosis. This study addresses the use of Artificial Intelligence to detect malfunctions in electric motors, in an automated way, at an early stage. Machine Learning algorithms, namely Support Vector Machines and Decision Trees, were used for the detection of stator interturn short circuits, in a predictive way, based on real data acquired at the Laboratory of Electromechatronic Systems of CISE - Electromechatronic Systems Research Centre. As a main difference in relation to other works, an approach based on the Extended Park's Vector Approach, which is currently used for this purpose, will be presented. Therefore, it is a hybrid approach, and its main objective is to obtain maximum efficiency with minimum features, from a tool already used for this purpose. As a result, an automated diagnostic system, that uses Support Vector Machine and Decision Tree, is presented for the detection of stator short circuits in three-phase induction motors, which includes scenarios of normal and unbalance in the phase or amplitude of the supply voltages. After optimizing the models, validation will be performed on raw data and its results will be discussed.
Industry 4.0 brings the concept of intelligent factories, that arise to meet a growing need for high flexibility and efficiency in the manufacture of products. Three-phase induction motors are already used on a large scale by the industry and will play a key role in Smart Factories. In order to keep them in operation with a high degree of reliability, and to reduce the costs associated with stoppages and interventions, it is necessary to identify faults at an early stage, in order to schedule maintenance before the electric machine stops completely. Online fault diagnostics methods have been studied for many years. However, the assessment of their results depends on an expert for accurate interpretation and diagnosis. This study addresses the use of Artificial Intelligence to detect malfunctions in electric motors, in an automated way, at an early stage. Machine Learning algorithms, namely Support Vector Machines and Decision Trees, were used for the detection of stator interturn short circuits, in a predictive way, based on real data acquired at the Laboratory of Electromechatronic Systems of CISE - Electromechatronic Systems Research Centre. As a main difference in relation to other works, an approach based on the Extended Park's Vector Approach, which is currently used for this purpose, will be presented. Therefore, it is a hybrid approach, and its main objective is to obtain maximum efficiency with minimum features, from a tool already used for this purpose. As a result, an automated diagnostic system, that uses Support Vector Machine and Decision Tree, is presented for the detection of stator short circuits in three-phase induction motors, which includes scenarios of normal and unbalance in the phase or amplitude of the supply voltages. After optimizing the models, validation will be performed on raw data and its results will be discussed.
Description
Keywords
Curto-Circuito Diagnóstico de Avarias Estatóricas Extended Park’S Machine Learning Motor Current Signature Analysis Park’S Vector Approach Power Current Signature Vector Approach.