Predict the remaining useful life in HVAC filters using a hybrid strategy

Abstract

This article discusses an engineering prediction-oriented method to monitor and predict the healthy conditions of air filters in heating, ventilation and air conditioning (HVAC) installations in the construction industry. In the literature, many researchers have studied hybrid prognostic methods for monitoring and predicting filter clogging, and experimental studies have been conducted to develop degradation models to demonstrate the mechanism of filter clogging. The common methods usually predict residual useful life based on a physics-based degradation model along with a prognostic model based on measured data. However, if there is not run-to-fail data or it is costly to prepare, another method is needed. The method used in the present work is useful when the data of entire operating period is not available, instead part of the operational range is obtained during the operation of the air handling unit (AHU). The method described in this article includes a combination of physically-based models and acquired operational data. An appropriate health indicator (HI) is calculated based on measurements. Learning algorithms are used to calibrate a carefully designed filter degradation model. The remaining useful life (RUL) of the filter is estimated using the dimensional reduction method, in particular principal component analysis (PCA) technique. The proposed method has been tested on a real air conditioning unit installed in a building located in Tallinn. The results show that the selected degradation model provides the best fit based on the data observed from the field. In addition, using dimensional reduction methods to estimate remaining useful life is feasible for HVAC filter clogging prediction. This is based on a comparison between an acceptable remaining useful life estimate and experimental data. The performance analysis results show that predictive maintenance methods can provide accurate prognostic indications. The application of a hybrid prediction model allows accurate estimation of the characteristics of the remaining useful life of the target component. It should be noted that the use of predictive maintenance strategies in this situation has increased the life of filters in buildings by a significant amount compared to replacement time schedule.