Assistente virtual baseada em processamento de linguagem natural para automação de consulta em banco de dados industrial
Carregando...
Data
2021-11-22
Autores
Fernandez, Douglas de Pádua
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
O progresso nas indústrias vem se desenvolvendo constantemente e, atualmente, a automação tradicional tem sido substituída por aplicações de tecnologias como a Internet das Coisas, Computação em Nuvem, Inteligência Artificial, Big Data, protocolos de comunicação via redes sem fio e acessos remotos de dados, decorrentes da Indústria 4.0. Visando uma automação mais colaborativa para com as interações humanas e que favoreça a eficiência nas tomadas de decisão dentro de uma fábrica, da integração entre tais tecnologias emergentes surge uma tendência que vêm sendo posta em evidência no âmbito industrial: a utilização de assistentes virtuais, também denominadas de sistemas de conversação orientados a tarefas. A partir desta tendência, no presente trabalho foi desenvolvida uma assistente virtual industrial composta por três pilares da Indústria 4.0: Internet das Coisas, Inteligência Artificial e Acesso Remoto a Dados. Tal sistema desenvolvido é responsável por permitir e facilitar o acesso a dados históricos de sensores de nível, pressão, vazão e temperatura de uma planta-piloto industrial para o usuário final, a partir de qualquer dispositivo conectado à rede de internet da planta. Na implementação da aplicação da assistente virtual, foi desenvolvida uma dashboard gráfica Node-RED, como interface entre a aplicação e o usuário. Além disso, foram desenvolvidos métodos de predição em Processamento de Linguagem Natural para realizar a conversão de linguagem natural em queries do banco de dados InfluxDB. Tais métodos foram explorados para os algoritmos Logistic Regression, MultinomialNB, AdaBoost, Decision Tree e Dummy nas estratégias estratificado e mais frequente, e, após a investigação e otimização de parâmetros, obteve-se uma acurácia de predição de 93.26% para o algoritmo Decision Tree, com um intervalo de confiança entre 88.19% e 98.33%.
The industrial advancement had been constantly developed and, nowadays, the traditional automation has been replaced by technological applications such as Internet of Things, Cloud Computing, Artificial Intelligence, Big Data, wireless communication protocols and remote access to data, due to Industry 4.0. Aiming at a more collaborative automation in human interactions and at interactions that help the decision taking efficiency inside a factory, from the integration between those emerging technologies, arise a trend that has been put in focus on the industrial environment: the use of virtual assistants, also known as task oriented conversational systems. From this trend, in the present paper, an industrial virtual assistant has been developed based on three pillars from Industry 4.0: Internet of Things, Artificial Intelligence and Remote Access to Data. This developed system is responsible to allow and favor the access to level, pressure, flow and temperature sensor data from the industrial plant to the final user, from any device connected to the industrial facility’s internet network. In the implementation of the virtual assistant application, a graphic Node-RED dashboard has been developed, as the interface between the user and the application. Furthermore, Natural Language Processing prediction methods have been developed to convert natural language into InfluxDB database queries. These methods were explored for Logistic Regression, MultinomialNB, AdaBoost, Decision Tree and Dummy (with stratified and most frequent strategies) algorithms, and after some parameters and algorithms investigation and optimization, it resulted in a prediction accuracy of 93.26% for the Decision Tree algorithm and an error range between 88.19% and 98.33%.
The industrial advancement had been constantly developed and, nowadays, the traditional automation has been replaced by technological applications such as Internet of Things, Cloud Computing, Artificial Intelligence, Big Data, wireless communication protocols and remote access to data, due to Industry 4.0. Aiming at a more collaborative automation in human interactions and at interactions that help the decision taking efficiency inside a factory, from the integration between those emerging technologies, arise a trend that has been put in focus on the industrial environment: the use of virtual assistants, also known as task oriented conversational systems. From this trend, in the present paper, an industrial virtual assistant has been developed based on three pillars from Industry 4.0: Internet of Things, Artificial Intelligence and Remote Access to Data. This developed system is responsible to allow and favor the access to level, pressure, flow and temperature sensor data from the industrial plant to the final user, from any device connected to the industrial facility’s internet network. In the implementation of the virtual assistant application, a graphic Node-RED dashboard has been developed, as the interface between the user and the application. Furthermore, Natural Language Processing prediction methods have been developed to convert natural language into InfluxDB database queries. These methods were explored for Logistic Regression, MultinomialNB, AdaBoost, Decision Tree and Dummy (with stratified and most frequent strategies) algorithms, and after some parameters and algorithms investigation and optimization, it resulted in a prediction accuracy of 93.26% for the Decision Tree algorithm and an error range between 88.19% and 98.33%.
Descrição
Palavras-chave
Aplicações industriais, Internet das coisas, Processamento de linguagem natural, Banco de dados, Industrial applications, Internet of Things, Natural language processing, Database