Application of machine learning algorithms in real-time fouling monitoring of plate heat exchangers

The utilization of plate heat exchangers is prevalent in aerospace, nuclear power, petrochemical, and other industries; however, operational challenges arise due to scaling issues. If not addressed promptly, it will diminish its heat transfer efficiency, resulting in energy wastage, shortened lifespan, equipment congestion, and even safety hazards. Long Short-Term Memory (LSTM) can effectively filter and store important information and can solve the problem of vanishing and exploding gradients. It is also capable of processing input data in real time, providing short- and long-term forecast results and monitoring heat transfer efficiency. The LSTM algorithm model is employed to monitor the health status of plate heat exchangers under various configurations of hidden layers, neurons, and discard rate in order to address this issue. The LSTM algorithm model with the highest predictive accuracy was combined with a Linear model to create a more sophisticated integrated model for monitoring the health status of plate heat exchangers. The LSTM 2 x 64 + Linear Model C was found to exhibit the highest prediction accuracy 0.9943. Since the fouling layer in the plate heat exchanger cannot be directly monitored, this paper firstly establishes a simulation programme for the plate heat exchanger through MATLAB. The outlet temperature of the cold measurement was changed by adding fouling to the cold side of the plate heat exchanger, which had a fouling thermal resistance of 0.0003 m2.K/W on the cold side when the efficiency of the plate heat exchanger was reduced to 50 %. Based on this result, in the LSTM algorithm, we use 0.0003 m2.K/W as the alarm threshold for the operation of the plate heat exchanger. This provides a feasible technical path for plate heat exchanger fouling assessment and long term performance diagnosis.