Predicting ASTI automated weather station (AWS) failure based on data-forwarding behavior
The exposure of the DOST-ASTI automated weather stations (AWS) to extreme weather events makes conventional condition-based maintenance methods insufficient in terms of preventive maintenance for proactive or corrective actions. The integration of machine learning to early failure detection, meanwhile, has become part of continuous developments on predictive maintenance (PdM). This study shows that by employing machine learning algorithms, failure of an AWS in one to eight weeks of time based on its data-forwarding behavior is predictable. There is some relation between the 7-day mean and variance of an AWS data-forwarding frequency and the condition of an AWS weeks in advance. Results also show that applying synthetic minority oversampling technique (SMOTE) in preprocessing is an effective method to balancing the dataset.