Hybrid Quantum Convolutional Neural Network for Defect Detection in a Wind Turbine Gearbox

Convolutional neural networks (CNNs) have been acknowledged for their effectiveness in vibration-based fault detection. However, when used to model high-dimensional vibration signals, the training cost increases exponentially. In this study, we present a hybrid quantum-classical approach that leverages the computational efficiency of quantum states to improve the training of a CNN fault diagnostic model. The proposed framework is validated with vibration signals from the intermediate speed shaft bearing of a wind turbine gearbox. We assess the performance of the hybrid quantum classical CNN (HQC-CNN) model across various optimizers, including adaptive moment estimation (Adam), stochastic gradient descent (SGD), and adaptive gradient algorithm (Adagrad). The Adam and SGD- based HQC-CNN models both showed superior resilience to overfitting at higher training epochs; however, while the Adam- based model required 93 seconds of run time to reach optimal classification accuracy, the SGD-based model required 243 seconds. All models achieved above 99.2% gearbox health state prediction accuracy. The Adam optimizer is recommended for integration into the HQC-CNN model to minimize computational resources while ensuring precise diagnosis of wind turbine gearbox health conditions.