Alarmes inteligentes em processos de desmineralização da água para a produção de vidros baseados em descoberta do conhecimento

Abstract

The anticipation of deterioration in the desired behavior of a process variable is crucial for systems to operate efficiently and safely. In this context, alarms are essential for capturing operators' attention to the need for immediate corrective actions in the process. However, traditional alarms based on statistical process control follow methodologies that have long remained unchanged, failing to keep pace with modern industrial demands, such as predictive anomaly detection that relies solely on statistical process control methods, limited to observing the upper and lower bounds of the process variable. Therefore, as a way to improve the relationship between alarm signaling and variations in industrial process dynamics, this work proposes a methodology for creating intelligent alarms based on the combination of trend detection algorithms (linear regression) with rule extraction algorithms (based on data mining). Through the application of this method, the goal is to create an alarm system that indicates deterioration trends in a process variable before it reaches critical limits, providing sufficient time for corrective actions. Additionally, using a decision tree classification model, it is intended to extract knowledge from the data to offer suggestions regarding which process should be adjusted to halt the deterioration trend. The combination of these two techniques resulted in an alarm capable of detecting progressive deviations and providing valuable information for operations.