Metodologia Aplicada ao Desenvolvimento de Gêmeo Digital de Sistemas de Bombeamento de Água

Abstract

Water supply companies are always being modernized with the implementation of new technologies, such as supervisory control systems and data acquisition. With the arrival of Industry 4.0 concept, the next step is the development of Digital Twins (DT) in order to find solutions in the lowest possible time and support operators training. The studies related to DT are recent and have few applications in water supply network systems, where most part of the research are related to concepts and how it can be improved with its implementation. This article describes the steps to develop a DT in order to support decision-making process, detect anomalies and assist in the training of operators of a hydraulic plant. The study was divided into two parts, the first consists of the develop of a Digital Model, and the second of a Digital Shadow. Artificial Neural Networks were used for the Digital Model implementation, in which two different activation functions were considered, the sigmoid and ReLU, thus looking for the best result of the mean square error between the two activation functions. The best model was used in parallel with the real hydraulic plant, thus observing its performance in relation to the real data. Among the developed models, the ReLU activation function presented the best performance , being in the sequence used in Digital Shadow implementation, which even having a result with a greater error than that observed in its training and validation step, but presenting a good behavior in relation to the actual plant data. The results obtained in the Digital Model and Digital Shadow stages are relevant for the development of DT.