Hybrid Quantum Recurrent Neural Network For Remaining Useful Life Prediction
Olga Tsurkan, Aleksandra Konstantinova, Aleksandr Sedykh, Dmitrii Zhiganov, Arsenii Senokosov, Daniil Tarpanov, Matvei Anoshin, Leonid Fedichkin · 2025
Read Hybrid Quantum Recurrent Neural Network For Remaining Useful Life Prediction by Olga Tsurkan, Aleksandra Konstantinova, Aleksandr Sedykh, Dmitrii Zhiganov, Arsenii Senokosov, Daniil Tarpanov, Matvei Anoshin, Leonid Fedichkin (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-20823.txt.
Introduction
Olga Tsurkan and co-authors build a hybrid quantum recurrent network to forecast how much life a jet engine has left, in a paper from two thousand twenty-five. Such forecasts drive safe, cost-wise maintenance. Their design swaps the linear maps inside memory gates for quantum circuits, then adds classical layers on top.
On a NASA turbofan benchmark, the hybrid beat matched classical memory networks by about five percent on error scores across ten seeds. It also topped forest, convolutional, and plain network baselines. Circuit analyses backed its compactness and trainability. Top joint deep models still won, so the authors pitch quantum parts as components, not replacements.
Catalog record
| CREATOR | Olga Tsurkan, Aleksandra Konstantinova, Aleksandr Sedykh, Dmitrii Zhiganov, Arsenii Senokosov, Daniil Tarpanov, Matvei Anoshin, Leonid Fedichkin |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-20823
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Hybrid Quantum Recurrent Neural Network For Remaining Useful Life Prediction Preprint: APS/123-QED Olga Tsurkan, Aleksandra Konstantinova, Aleksandr Sedykh, Dmitrii Zhiganov, Arsenii Senokosov, Daniil Tarpanov, Matvei Anoshin, Leonid Fedichkin Abstract Predictive maintenance in aerospace heavily relies on accurate estimation of the remaining useful life of jet engines. In this paper, we introduce a Hybrid Quantum Recurrent Neural Network framework, combining Quantum Long Short-Term Memory layers with classical dense layers for Remaining Useful Life forecasting on NASA’s Commercial Modular Aero-Propulsion System Simulation dataset. Each Quantum Long Short-Term Memory gate replaces conventional linear transformations with Quantum Depth-Infused circuits, allowing the network to learn high-frequency components more effectively. Experimental results demonstrate that, despite having fewer trainable parameters, the Hybrid Quantum Recurrent Neural Network achieves up to a 5% improvement over a Recurrent Neural Network based on stacked Long Short-Term Memory layers in terms of mean root mean squared error and mean absolute error. Moreover, a thorough comparison of our method with established techniques, including Random Forest, Convolutional Neural Network, and Multilayer Perceptron, demonstrates that our approach, which achieves a Root Mean Squared Error of 15.46, surpasses these baselines by approximately 13.68%, 16.21%, and 7.87%, respectively. Nevertheless, it remains outperformed by certain advanced joint architectures. Our findings highlight the potential of hybrid quantum-classical approaches for robust time-series forecasting under limited data conditions, offering new avenues for enhancing reliability in predictive maintenance tasks. I Introduction Accurate estimation of the remaining useful life (RUL) of critical machinery is a cornerstone of modern reliability and risk-management strategies si2011remaining ; lee2014prognostics ; berghout2022systematic . Nowhere is this more evident than in the aviation industry: the timely prediction of gas turbine jet engine failures not only bolsters safety by preventing catastrophic breakdowns zhang2016multiobjective , but also serves to streamline maintenance and resource allocation kang2021remaining . Indeed, robust forecasting models enable operators to defer costly repairs until necessary while avoiding the risks of overdue overhauls, ultimately leading to reduced downtime and more efficient fleet utilization huang2019bidirectional ; ferreira2022remaining . Time-series forecasting methods for RUL estimation can be broadly categorized into statistical approaches and machine learning techniques. Traditional statistical methods, including autoregressive, moving average, and autoregressive integrated moving average models, typically excel under assumptions of stationarity and linearity box2015time ; shumway2000time . These models are prized for their interpretability and rigorous theoretical foundation but can struggle to capture the nonlinear and interdependent patterns that characterize real-world industrial and aerospace data zhang2003time . Machine learning (ML) has exhibited remarkable success in tackling complex predictive tasks across numerous fields, including economics stock2006forecasting , healthcare ML2023Healthcare , energy ML2025Energy , meteorology zhou2019forecasting , and industry Ml2025Industry . Within this landscape, recurrent neural networks (RNNs) stand out due to their ability to capture temporal dependencies in data. However, they are often hindered by vanishing and exploding gradients bengio1994learning , which complicate model training. Long short-term memory (LSTM) networks hochreiter1997long ; gers2000learning ; graves2013speech ; sak2014long alleviate these issues by effectively modeling nonlinear interactions, long-term trends, and seasonal or cyclical patterns, thus finding broad applicability in time series tasks, including RUL estimation LSTM2024Summary . Despite these achievements, classical ML strategies can be severely weakened by limited or noisy data, high-dimensional feature spaces, and intricate fault dynamics Bishop2007Pattern ; Goodfellow2016Deep . In recent years, quantum computing has progressed from theoretical constructs to initial practical demonstrations, leveraging quantum properties such as entanglement and superposition to enable computational processes beyond the reach of traditional hardware Nielsen2011Quantum ; biamonte2017quantum . This development has led to quantum algorithms designed for optimization, simulation, and cryptography, thereby paving the way for more specialized implementations montanaro2016quantum ; Preskill2018quantumcomputingin . Finally, the new tool that allows the use of the advantages of ML algorithms and quantum computations emerged — quantum machine learning (QML), which proved its usability by solving problems typified by constrained datasets or high-dimensional feature spaces rebentrost2014quantum ; ciliberto2018quantum ; schuld2018supervised ; cao2017quantum ; Alharbi2024QCNN . QML utilises high-dimensional Hilbert spaces to encode input features, allowing it to detect more complex patterns with fewer parameters than its classical counterpart havlivcek2019supervised ; schuld2019quantum . This proved to be especially useful when applied to problems with nonstationary or noisy signals, which are frequently encountered in industrial maintenance and prognostic applications sagingalieva2023photovoltaic ; emmanoulopoulos2022quantummachinelearningfinance ; protasevich2025multi ; lee2025predictive . A promising approach for leveraging these quantum advantages involves hybrid quantum-classical neural networks (HQNNs), which combine classical and quantum ML in one single framework arthur2022hybridquantumclassicalneuralnetwork ; sagingalieva2023photovoltaic ; sagingalieva2023hybrid ; Bischof2025Hybrid ; SUN2025130226 . Inside this structure, the quantum component plays a specific role in data processing or feature encoding while leaving computationally hard tasks like backpropagation and parameter optimization for the classical routines. By selectively introducing quantum effects, HQNNs retain the scalability and reliability of conventional deep learning pipelines broughton2021tensorflowquantumsoftwareframework . Preliminary studies indicate that HQNNs can match or exceed the performance of their classical counterparts, often exhibiting greater resilience against overfitting abbas2021power ; Berberich2024training . The inherent parameter efficiency of HQNNs further enhances their suitability for scenarios involving complex, high-dimensional inputs or limited data availability — situations frequently observed in predictive maintenance sagingalieva2023hybrid ; anoshin2024hybrid . In this paper, we propose a novel Hybrid Quantum Recurrent Neural Network (HQRNN) tailored for jet engine RUL forecasting. Our approach replaces the linear transformations in each gate of a classical LSTM with a Quantum Depth-Infused (QDI) circuit sagingalieva2023hybrid ; anoshin2024hybrid ; lusnig2024hybrid , creating a quantum-enhanced LSTM (QLSTM) chen2022quantum ; sagingalieva2023photovoltaic . By encoding sensor measurements in a quantum Hilbert space, these QDI layers effectively capture complex temporal patterns while requiring fewer trainable parameters than a purely classical baseline chen2022quantum . We evaluate our method on the NASA Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset saxena2008turbofan , a widely used benchmark in turbofan engine degradation studies, and demonstrate improved predictive performance in terms of both root mean squared error (RMSE) and mean absolute error (MAE). Moreover, experimental results suggest that our hybrid approach can help mitigate the challenges associated with limited data, underscoring its potential benefits in industrial time-series applications sagingalieva2023hybrid . The C-MAPSS dataset is a widely used benchmark for predicting RUL of aircraft engines. Recent studies addressing this prediction problem include approaches based on Temporal Convolutional Neural Networks (TCNN) combined with Transformers wang2021remaining and joint CNN-LSTM models enhanced with attention mechanisms deng2024prediction ; both articles also provide comprehensive summaries of alternative methods. In Section VI , we present a detailed comparative analysis of our proposed HQRNN against classical machine learning techniques and simple neural network architectures (Table 2 ), as well as recent state-of-the-art joint deep learning approaches (Table 3 ). Our HQRNN approach demonstrates superior performance compared to traditional ML and simple neural network models, though it currently underperforms relative to several advanced joint models. These results highlight promising opportunities for future research focused on integrating quantum-enhanced recurrent structures with complementary deep learning methods. The rest of the paper is structured as follows: Section II describes the C-MAPSS dataset used in this study. Section III outlines our proposed HQRNN architecture and the underlying QDI circuits. We detail the experimental setup and report empirical findings in Section IV , followed by a deeper analysis of the quantum circuit properties in Section V . Lastly, Section VI concludes with a discussion of the implications and future directions for