The researchers, affiliated with Hefei University of Technology, call their model a physics-informed uncertainty-aware patch-level graph temporal network, abbreviated UP2-GTN. Its central insight is that aero-engines are not clean laboratory objects. They are sprawling, tightly coupled systems whose sensors are buffeted by measurement noise, sudden electrical interference, sensor aging, thermal drift, and constantly shifting operating conditions. Most existing deep learning approaches to RUL prediction, whether built on convolutional neural networks, recurrent architectures, temporal convolutional networks, or Transformers, are optimized for point prediction accuracy under relatively ideal data. When deployed in the messy real world, their reliability can degrade sharply, because the noise and disturbances that pervade actual sensor streams are rarely modeled explicitly.

The new framework attacks uncertainty on three fronts simultaneously: at the level of the raw data, at the level of time, and at the level of the engine’s physical structure. During training, the model deliberately contaminates its inputs with a cocktail of engineered noise types. Gaussian noise mimics the continuous, random measurement errors that accumulate as sensors age. Impulse noise injects high-amplitude outliers resembling electrical interference or sudden external shocks. Scaling noise perturbs signal amplitudes to simulate gain errors, while drift noise adds a slowly growing offset that mirrors the systematic biases sensors develop over years of service. The team even blends these perturbations in weighted combinations, weighting Gaussian and scaling noise most heavily, to produce training data that better reflects the multi-source uncertainty of genuine flight operations.

The temporal innovation is perhaps the most elegant. Degradation in an aero-engine unfolds on multiple time scales at once: slow, global trends of performance decline coexist with sharp, short-lived fluctuations caused by local disturbances. Traditional point-wise models process sensor readings one instant at a time, which allows a single noisy measurement to ripple through an entire long sequence, accumulating error as it goes. UP2-GTN instead slices the sensor history into short, non-overlapping segments called patches. Within each patch, a one-dimensional convolution extracts local temporal features, while auxiliary average and max pooling branches capture the segment’s overall trend and its extreme excursions. A gated fusion mechanism then adaptively balances these dynamic and statistical views, allowing transient anomalies to be dampened before they can contaminate the broader degradation narrative.

The structural layer is where physics enters the picture. Rather than letting a graph neural network freely infer which sensors are related based on statistical correlation alone, a strategy that can manufacture spurious connections when data are noisy, the model builds its sensor graph under explicit physical constraints derived from the engine’s architecture. Sensors are treated as nodes, and edges are permitted only where genuine physical coupling exists: sequential coupling along the airflow path, energy coupling through heat and power transfer between components, and control-feedback coupling imposed by the engine control system. When an intermediate physical variable cannot be directly observed, the physical path is compressed to connect the observable sensor nodes, preserving the dominant dependency pathway. A multi-head attention mechanism then learns the dynamic strength of these physically admissible connections, so the graph can evolve as operating conditions change while never inventing relationships the hardware does not actually support.

Once the patch-level graphs are constructed, a temporal convolutional network models their evolution over time. Causal convolutions ensure that predictions depend only on past information, a requirement for real-time online prognostics, while dilated convolutions expand the receptive field to capture long-range degradation trends without exploding computational cost. The final features pass through convolutional and fully connected layers to produce a single RUL estimate, trained end-to-end with a mean squared error objective. Notably, the framework achieves its robustness without probabilistic output layers or Bayesian machinery; uncertainty is suppressed at the representation and structure level rather than merely quantified at the output.

The experimental evidence is drawn from the two most widely used aero-engine benchmarks, C-MAPSS and its high-fidelity successor N-CMAPSS, which simulate run-to-failure trajectories under single and multiple operating conditions. Across the four C-MAPSS subsets, the model achieved root mean square errors of 11.12, 13.61, 11.09, and 17.30 cycles, outperforming a long list of recent graph-based and Transformer-based competitors on most subsets, with particularly strong gains on the challenging multi-condition FD002 and FD004 scenarios. On N-CMAPSS, the model posted an overall RMSE of 5.57 and, strikingly, a score function of 6669.42 on the aggregated test set, a metric that penalizes late, safety-critical predictions asymmetrically and where the model’s advantage over prior methods was most dramatic.

Ablation studies confirmed that each of the three mechanisms earns its place. Removing uncertainty augmentation, patch-level modeling, or the physics-informed graph constraints each degraded performance, with the patch mechanism proving especially important for stabilizing features under noise. Sensitivity analyses showed that patch length involves a genuine trade-off: short patches preserve fine-grained dynamics but are more vulnerable to impulse noise, while longer patches smooth out transient disturbances at the cost of temporal resolution. Under noise intensities up to ten percent, the simpler single-condition subsets remained remarkably stable, with RMSE fluctuations generally within one cycle, while the multi-condition subsets revealed drift noise as the most damaging disturbance, precisely because its slow accumulation corrupts the long-term degradation trends the model is trying to learn.

Perhaps most importantly for real-world deployment, the model is computationally modest. It contains roughly 305,000 trainable parameters, requires under five million floating-point operations per input window, and delivers inference in about 2.3 milliseconds per sample on a consumer-grade GPU, with peak inference memory of just 173.5 megabytes. That combination of accuracy, robustness, and efficiency makes the approach a plausible candidate for onboard or near-line prognostic systems. The authors acknowledge that challenges remain, including adaptively selecting degradation scales and modeling richer uncertainty and component coupling, but the study marks a meaningful step toward predictive maintenance systems that do not just forecast failure, but know how much they can trust their own forecast when the data grow noisy.

Subject of Research: Physics-informed deep learning for uncertainty-aware remaining useful life prediction of aero-engines

Article Title: Physics-informed uncertainty-aware patch-level graph temporal network for aero-engine remaining useful life prediction

Article References: Peng, P., Fan, X., Lin, B., Sun, X., Yin, Z., & Guo, Z. (2026). Physics-informed uncertainty-aware patch-level graph temporal network for aero-engine remaining useful life prediction. Results in Engineering, 32, Article 113421. https://doi.org/10.1016/j.rineng.2026.113421

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113421

Keywords: aero-engine, remaining useful life, predictive maintenance, graph neural network, uncertainty quantification, physics-informed machine learning, temporal convolutional network, sensor noise, prognostics and health management, C-MAPSS, N-CMAPSS, deep learning

The research, carried out by Chenchen Wang, Keqiang Miao, Tong Xin, Weiqun Fan, Chunyan Hu, and Wei Li of the Institute of Engineering Thermophysics at the Chinese Academy of Sciences, addresses a long-standing headache in altitude ground testing. These facilities simulate the temperature, pressure, and flow conditions of flight by feeding an engine through a sophisticated intake system that mixes hot and cold gas streams. In steady operation, standard proportional–integral controllers can hold the inlet pressure steady. But the moment the engine accelerates, its rotational speed climbs and the intake mass flow demand changes rapidly, dragging the inlet pressure along an unpredictable, nonlinear trajectory. Add unmeasured disturbances — fluctuations in the mixing of gas streams, downstream cabin pressure shifts, and the coupled thermal dynamics of a massive piping network — and the problem becomes one that classical control theory has struggled to solve cleanly.

Previous attempts to tame the transient have followed two broad paths. On the control side, engineers have deployed gain-scheduled PI schemes, fuzzy controllers, and increasingly sophisticated active disturbance rejection control (ADRC) and its linear variant, LADRC, including cascade and fixed-time formulations designed for high-flow intake systems. These approaches estimate and cancel disturbances in real time, but they react to deviations after they begin to develop. On the modeling side, researchers have built data-driven simulations of engine transients, recognizing that physics-based models of the full facility are too slow and too uncertain for real-time prediction. What has been missing, the authors argue, is a prediction tool that both understands the character of an acceleration or deceleration event as it unfolds and produces forecasts accurate enough to serve as feedforward compensation — telling the control system what is about to happen before it happens.

The new method attacks the problem with an architecture that mirrors the physics of the transient itself. At its core sits a CNN–GRU sequence encoder: a convolutional neural network layer stack that scans short windows of sensor history to pick out local, sharp features — the telltale signature of a throttle slam or a rapid deceleration — followed by a gated recurrent unit network that integrates those features across time, capturing how the event’s dynamics evolve. This pairing is well suited to the task because acceleration and deceleration are fundamentally event-like: they have sharp onsets, distinctive local patterns, and temporal dependencies that a pure convolutional model or a pure recurrent model tends to blur. The convolutional front end acts like a pattern detector for the early moments of a transient, while the recurrent backend maintains the running memory needed to track the pressure trajectory seconds into the future.

The second innovation is how the network outputs its answer. Rather than a single regression head, the model uses a dual-head, multi-task structure: one head predicts the pressure increment — the step-to-step change in inlet pressure — while the other predicts the absolute pressure trajectory directly. Predicting increments emphasizes the shape and timing of the transient, which is precisely where errors hurt most, while predicting absolute values anchors the forecast to physically meaningful levels. The two tasks reinforce each other, but the authors go a step further with what they call a dynamic-consistency constraint: a loss term that forces the increment head’s predictions to agree with the temporal differences of the absolute head’s trajectory. If the two views of the same physics diverge, the training process penalizes the inconsistency, effectively making the network internally check its own work. The consistency is weighted dynamically by an event-aware scheme that applies extra weight during periods when engine rotational speed is changing rapidly — exactly the moments when accuracy matters most and when unmeasured disturbances are strongest.

Training and validation relied on real measured data from an aero-engine altitude test facility, covering genuine acceleration and deceleration events with their full complement of nonlinear, time-varying behavior. The team benchmarked their model against a demanding lineup of alternatives, including a standalone one-dimensional CNN, a deep GRU network, a temporal convolutional network with dilated causal convolutions, a Transformer encoder, and hybrid TCN–GRU and CNN–GRU variants with single or dual output heads. Every model saw the same event-level training, validation, and test splits, the same normalization, and the same optimizer configuration — Adam with a learning rate of 0.001, a batch size of 128, up to 500 epochs, and early stopping with a patience of 200 epochs. To guard against lucky random seeds, each network was trained five times with different initializations, and results were reported as means and standard deviations.

The verdict was consistent: the full dynamic-consistency-guided multi-task CNN–GRU model beat every baseline on prediction accuracy, transient trend tracking, and robustness. The ablation structure of the comparison tells an instructive story. Moving from a single-head CNN–GRU to the dual-head multi-task version improves performance by letting the network learn complementary representations of increments and levels. Adding the dynamic-consistency constraint improves it further by eliminating the drift between the two heads that otherwise accumulates during long, fast transients. And the event-aware weighting concentrates the model’s capacity where it counts, sharpening predictions precisely during the steep portions of the pressure curve when engine speed is sweeping through its fastest changes. Metrics based on mean absolute error and root mean square error confirmed that the improvements were systematic rather than artifacts of a single test case.

The practical payoff lies in what predictive accuracy enables downstream. Because the model can characterize transient inlet pressure fluctuations before they fully develop, its output can feed a feedforward compensation loop: the inlet pressure control system receives an anticipatory signal and adjusts the mixing of hot and cold gas flows proactively, rather than scrambling to correct a deviation that has already corrupted a measurement. The authors also position the framework as a design tool for inlet pressure control systems, allowing engineers to evaluate candidate control laws against realistic predicted transients without burning test-cell time. In an era when aero-engine development cycles are compressed and altitude test facilities are booked solid, shaving the uncertainty out of transient test data has direct implications for cost, schedule, and ultimately the safety certification of engines that carry passengers.

Broader significance of the work extends beyond test cells. The core ideas — event-aware temporal weighting, dual-head multi-task prediction, and self-consistency between increment and level forecasts — are architecture-agnostic and could transfer to other industrial transient prediction problems: wind tunnel pressure control, power plant load following, or any process where a fast actuation event disturbs a slow thermal system. The study also joins a growing literature showing that hybrid physics-informed and data-driven models are overtaking both purely physical simulations and generic deep networks for aerospace applications, from engine health management to supersonic isolator flow prediction. The data underpinning the study remain confidential and are available from the corresponding author on request, a reminder that in the world of full-scale engine testing, the hardest data to get are often the most valuable. Supported by funding under China’s Engine Thermophysical Test Apparatus program and built on test data from the Institute of Engineering Thermophysics’ test-cell team, the research signals a shift in how ground test facilities may be operated: not as passive measurement chambers that react to the engine, but as predictive systems that see the transient coming and meet it with computed foresight. As neural networks learn the rhythms of acceleration and deceleration, the boundary between flight and simulated flight grows correspondingly thinner.

Subject of Research: Machine learning prediction of transient inlet pressure in aero-engine altitude simulation testing

Article Title: Prediction of Inlet Pressure During Aero-Engine Acceleration and Deceleration Using a Hybrid Neural Network

Article References: Wang, C., Miao, K., Xin, T., Fan, W., Hu, C., & Li, W. (2026). Prediction of Inlet Pressure During Aero-Engine Acceleration and Deceleration Using a Hybrid Neural Network. International Journal of Aeronautical and Space Sciences. https://doi.org/10.1007/s42405-026-01265-w

Image Credits: AI Generated

DOI: 10.1007/s42405-026-01265-w

Keywords: aero-engine, altitude simulation tests, inlet pressure prediction, hybrid neural network, CNN–GRU, multi-task learning, transient pressure, active disturbance rejection, feedforward compensation, deep learning, altitude test facility, engine acceleration