Algoritmos de IA aplicados em IIOT

Abstract

Predictive maintenance in industrial refrigeration systems is essential for preventing unplanned downtime and reducing energy consumption. This study pre- sents a comparative analysis of Artificial Intelligence models tailored for time-series data from industrial sensors. This evaluated architectures range from Recurrent Neural Network (LSTM) and linear decomposition models (DLinear, NLinear) to state-of-art models based on attention mechanisms and hierarchy, such as NHITS, Autoformer, and PatchTST. The primary objective is to determine the effectiveness of these networks in identifying machine behavior and enabling the early detec- tion of operational anomalies. The results demonstrate that models like PatchTST offer significant advantages in handling long-term dependencies, while linear mo- dels provide a robust, low-computational cost baseline. The research concludes that integrating these AI models facilitates an efficient transition from preventive to predictive maintenance, optimizing the lifespan of industril assets