Hybrid quantum recurrent neural network for remaining useful life prediction of turbofan engines

Abstract

Accurate remaining useful life (RUL) estimation underpins safe operation and cost-effective maintenance of aerospace propulsion systems. We propose a Hybrid Quantum Recurrent Neural Network (HQRNN) for jet-engine RUL forecasting on the NASA C-MAPSS FD001 benchmark. The HQRNN stacks Quantum Long Short-Term Memory (QLSTM) layers, replacing each LSTM gate's linear transformation with a Quantum Depth-Infused (QDI) circuit; this is followed by classical dense layers. Quantum and hybrid quantum-classical methods for turbofan RUL prediction are still at an early stage. Our study is therefore among the first to evaluate a gate-based QLSTM hybrid at matched parameter counts, comparing it against classical and joint state-of-the-art models on this benchmark and complementing that comparison with a circuit-level analysis of the quantum layer. Encoding the gate signals in a quantum feature space is intended to help the network represent high-frequency degradation patterns with fewer trainable parameters than a matched classical counterpart. The HQRNN improves mean RMSE and mean MAE by about 5% over matched-parameter stacked-LSTM RNNs across 10 random seeds, and attains a test RMSE of 15.46, outperforming Random Forest, CNN, and MLP baselines. ZX calculus, Fisher information, and Fourier analyses indicate that the QDI circuit is compact, trainable, and expressive. Advanced joint deep-learning models still outperform the stand-alone HQRNN, indicating that quantum-enhanced recurrent modules are best deployed as components within composite prognostics pipelines rather than stand-alone predictors.

17 pages, 7 figures, 5 tables

References in corpus (16)

• Quantum error correction below the surface code threshold
• Quantum Error Mitigation
• Generalization in quantum machine learning from few training data
• Hybrid quantum neural network for drug response prediction
• Generalization despite overfitting in quantum machine learning models
• Equivalence Checking of Quantum Circuits with the ZX-Calculus
• Hybrid quantum image classification and federated learning for hepatic steatosis diagnosis
• Classical versus Quantum: comparing Tensor Network-based Quantum Circuits on LHC data
• Hybrid quantum cycle generative adversarial network for small molecule generation
• Training robust and generalizable quantum models
• An exponentially-growing family of universal quantum circuits
• Forecasting steam mass flow in power plants using the parallel hybrid network
• Differentiating and Integrating ZX Diagrams with Applications to Quantum Machine Learning
• Method for noise-induced regularization in quantum neural networks
• Multi-objective optimization and quantum hybridization of equivariant deep learning interatomic potentials
• Shot-based quantum encoding: a data-loading paradigm for quantum neural networks