Machine learning prediction of downstream oil carryover in air-oil separators for vacuum application

Air or oil filters are commonly used in vacuum conditions in pumps within various industries such as healthcare, pharmaceutical, and many more. Some of these filters are exposed to an upstream challenge of 1000-20000 mg/m³ of oil particles mixed with air, which needs to be reduced to 3-5 mg/m³ downstream. Accurate determination of the carryover rate of oil by the filters is crucial for meeting environmental compliance and enhancing operational efficiency. Traditional methods for measuring the carryover rate rely on time-consuming and costly experiments, making them impractical for large-scale production and real-time quality assessment. Therefore, the main objective of this study is to develop digital models using mathematical and machine learning approaches to accurately predict the carryover rate in filters, thereby reducing reliance on physical testing. To this end,224 sample experimental datasets were utilized, preprocessed, and cleaned to develop several multilayered Artificial Neural Networks (ANNs) and a multilinear regression model. Three optimisation strategies, including Levenberg-Marquardt, Bayesian Regularisation, and Particle Swarm Optimisation, have been used to tune the developed ANNs, which consist of various hidden layers and different neurons. For the purpose of comparison, the velocity-based physics model was applied to predict the oil carryover rate. The results of the study revealed that the single hidden layer with 20 neurons ANN optimized with the BR algorithm performed the best among all the models, with a Mean Square Error of 0.0648 and a correlation coefficient value of 0.942 for predicting the oil carryover rate. The developed model is validated and can be used for the fast computation of the carryover rate,informing the optimisation strategy of the filters