Improved Probabilistic Neural Network Based Fault Diagnosis of Control Valve

Control valves are important actuators in process industries. The blockage and leakage of valves are the most common faults in harsh environments. This paper proposes a fault diagnosis model based on an enhanced probability neural network (PNN), where a modified dung beetle optimization (DBO) algorithm is developed to optimize the smooth factors of PNN so as to improve classification accuracy and robustness. Meanwhile, k-medoids clustering algorithm is applied to optimize the PNN structure. The proposed model is validated by using the dataset from the experiments of fault classification for an electrical control valve. The experiment results indicate that the diagnostic accuracy reaches 99.3% by the developed IDBO-PNN model, which demonstrated the effectiveness of the proposed method.