Acoustic signal-based structural defects diagnosis of hydropower runners using enhanced CNN-GRU via GILA algorithm

Diagnosing structural defective faults of hydro-turbine runners is crucial for the safe and stable operation of hydropower units. This study analyzes acoustic signals from cracked and dropped runners, proposing a CNN-GRU (convolutional neural network-gated recurrent unit) fault diagnosis model optimized using the good point set incomprehensible but intelligible-in-time logic algorithm (GILA). Constructing a CNN-GRU model enhances time-series data processing. GILA improves the ILA’s population initialization with good points, accelerating convergence and enhancing global search capability. Hyperparameter optimization via GILA boosts the model’s stability and robustness. The results show that the acoustic spectrum of the runner will also change after it has cracked and dropped faults, and this property can be used as a potential basis for determining whether the runner has cracked and dropped faults. The training and testing accuracies of the GILA-CNN-GRU model reach 94.4% and 90.8%, respectively, which are better than those of the CNN, GRU, and CNN-GRU models. Compared with the ILA-CNN-GRU model, the accuracy is improved by 6.3% and 3.8%, respectively. The model exhibits higher stability, convergence speed, and generalization ability. This study can be used as a helpful supplement to the existing hydro-turbine condition monitoring and fault diagnosis system.