Análise de tradução homem-máquina utilizando mecanismos de atenção para sistemas baseados em SQL para indústria 4.0

Abstract

The use of relational databases is increasingly present in the industry. Applications in medical, IoT and Industry 4.0 are examples of this. Despite the large capacity and efficiency in storing and retrieving data, this type of database requires technical knowledge in specific query languages to access this information, which distances these types of application from the non-expert public. In this work, we propose an application of recent Deep Learning models in natural language processing that uses attention mechanisms for translation from natural language in English to SQL applied to a database which stores data from sensors, focused on the concept of Industry 4.0. Paired examples of natural language phrases were generated with their corresponding SQL query to be used for training and validation. The model was database scheme agnostic, in a way that it only handles the input and output sequences regardless of the database structure. Data come from a typical process historians used in industrial scenarios. By training the deep neural network, it was obtained a language model with an accuracy of approximately 92% in the validation set.