Filter Airflow prediction model development

The implementation of air filters on commercial swine farms has effectively reduced the frequency of airborne disease transmission. However, efficiently managing filter lifespan remains a challenge and an unknown operational cost for filtered swine facilities. Individual filter testing protocols are time consuming and expensive for producers. The objective of this study was to develop a predictive model for estimating airflow for an individual filter in situ by comparing multiple machine learning models to eliminate the need for manual, individual filter testing for filter resistance. The data set was generated from a custom Air Filter Environmental Testing Chamber that mimics on farm operational conditions with a low static pressure drop per filter and ground level installation. Model parameters were developed from a six-month long data set. The models were developed when the chamber was running with a new set of pre-filters and a set of five-month old v-bank filters. The developed models include a single input linear regression, multiple linear regression, and random forest models. A single input linear regression was not an effective method for predicting the chamber airflow, R2=0.08. The multiple linear regression moderately explained the variation in the data, R2=0.77. The random forest models performed the best for predicting the test chamber airflow with both models featuring R2= 0.98. The results and models from this study will be used to determine the feasibility of an on-farm application.