Enhancing cooling tower performance with condition monitoring and machine learning based drift detection

Process cooling is crucial to many manufacturing processes. To monitor the performance of a cooling tower, it was equipped with extensive sensors for internal and environmental data acquisition. The aim is to improve reactive and predictive maintenance by estimating the actual condition as well as predicting defective behavior of the cooling tower. We designed a method, which derives the degree of defect from data of the non-defective cooling tower. A concept drift detection approach was implemented, which monitors the model estimation error of a multilayer perceptron model. Increasing model estimation error indicates changing system behavior and increasing risk of failure.