Monitoring of Flow Rate Measurements Using LSTM Autoencoder

Integration of automation in operation and maintenance (O&M) tasks can enhance early detection of faults in advanced reactors (AR). Data-driven machine learning (ML) methods provide a viable solution for efficiently monitoring equipment in such systems. We use a long short-term memory (LSTM) autoencoder to monitor flow rate measurements in a water pump using multiple features, including flow and pressure measurements. Results show that the prediction error increases as less correlated features are introduced in the model. Moreover, prediction error slightly increases with lookback window length.