Predictive fouling detection in dairy industry heat exchangers using sequence-to-sequence deep learning models

Fouling in heat exchangers challenges heat transfer efficiency, making its predictive detection crucial for the dairy industry.Despite numerous sensors integrated into production facilities, valuable data relevant to fouling prediction often remains underutilized.This study evaluates three Sequence-to-Sequence deep learning models-Long Short-Term Memory (LSTM), Encoder-Decoder LSTM, and Convolutional Neural Network LSTM (CNN-LSTM) to predict sensor profiles several minutes into the future and to estimate fouling based on a dataset from a full-scale dairy heat exchanger.Sensor data such as inlet hot water temperature, outlet product pressure, and pump power were treated as indicator for fouling and analyzed to estimate it.Evaluation metrics include mean squared error and mean absolute scaled error.Of the models tested, CNN-LSTM showed the lowest prediction error 12.3 for estimating the outlet product pressure, offering potential for early fouling detection in industrial applications.These findings demonstrate the feasibility of deep learning for improving heat exchanger maintenance strategies.