Bayesian optimized LSTM-based sensor fault diagnosis of organic Rankine cycle system

Abstract

As the energy crisis intensifies, the organic Rankine cycle (ORC) is increasingly employed for efficient recovery of low-temperature waste heat. The operation of the ORC system necessitates the use of numerous sensors to monitor its status. Over time, these sensors may become faulty, rendering accurate and timely diagnosis is critical for proper and safe functioning of the ORC system. Currently, there is a lack of rapid diagnostic methods for sensor faults in ORC systems. This study establishes an ORC test bench utilizing cyclopentane as the working fluid. Experimental data incorporating induced faults from the ORC test bench is employed to train machine learning-based models for sensor fault diagnosis. The test results indicate that the diagnostic model developed in this study can accurately diagnose various sensor faults in the ORC system, thereby ensuring its safe operation. Notably, the method based on Bayesian-optimized long short-term memory network (BO-LSTM) achieved the highest diagnostic accuracy, reaching up to 95.92 %.