O uso do aprendizado de máquina no monitoramento de incrustações em trocadores de calor: uma revisão sistemática da literatura

Abstract

The research aimed to discover the main machine learning methods used to inspect fouling in heat exchangers, through a systematic review. Furthermore, we sought to analyze the challenges related to the use of these algorithms in this inspection, and future perspectives for this area of research were also examined. The study was based on the use of StArt software to assist in the preparation of the review protocol. As a protocol proposal, four databases were used - SCOPUS, Science Direct, IEEE and Web of Science - to search for articles. In this way, seven relevant articles were obtained, following the established protocol. Among the most common methods for monitoring fouling, SVM and k-NN stand out, followed by the decision tree. The LSTM and NARX algorithms, along with the long and short-term predictors, stood out as the most successful, followed by the GPR, SVM, k-NN and IGWO-SVR algorithms. However, the studies evaluated are still in the testing phase and have not been applied to real heat exchangers. Therefore, in the near future, it is expected that there will be investments in this area, enabling the effective practical implementation of these algorithms.