CruiseBench: A New Benchmark for Evaluating Neural Networks on Aircraft Engine Lifetime Prediction

In July 2026, a group of researchers published CruiseBench on arXiv, a benchmark for predicting the remaining useful life (RUL) of aircraft engines that extracts only the cruise phase of flight from the N-CMAPSS dataset. In baseline tests, the TSMixer model achieved the best accuracy, with a mean RMSE of 3.4.

What is CruiseBench

CruiseBench is a specialized benchmark for comparing neural networks that predict the remaining useful life (RUL) of an aircraft engine, built on top of the N-CMAPSS dataset. RUL prediction estimates how much longer an engine can safely operate, and it underpins maintenance planning. The core of the benchmark is an artifact called CPM-N-CMAPSS (Cruising-Period Mask for N-CMAPSS): a mask that stores per-cycle cruise-flight intervals, identified with the common-altitude method, for all nine available sub-datasets.

• The benchmark is derived from the N-CMAPSS dataset, an extension of the classic C-MAPSS
• CPM-N-CMAPSS is a mask of cruise intervals for 9 sub-datasets
• The protocol's inputs are scenario descriptors and readings from real sensors
• Virtual sensors, health parameters and metadata are excluded from the features
• Baseline models: LSTM, GRU, TCN and TSMixer

Why only the cruise phase?

The cruise phase removes the noise from changes in operating mode, which gets in the way of measuring engine degradation fairly. N-CMAPSS stores full time series within each flight rather than per-cycle snapshots, as the earlier C-MAPSS does. According to the authors, this realism both increases the volume of data and entangles wear signals with variations in operating mode: takeoff, climb, descent. As a result, the choice of preprocessing starts to affect the final metrics, and direct comparisons between models become unreliable. CruiseBench answers this with a fixed protocol: it works only with "masked" cruise rows, keeps windows at their original resolution, and applies separate RUL caps for each of the nine sub-datasets.

Which model was more accurate?

TSMixer delivered the best result. Under the CruiseBench-eta5-W256-S10 configuration, it reached the lowest mean RMSE of 3.4±1.71 and a Saxena score of (2.50±2.99)×10⁴, outperforming the recurrent LSTM and GRU as well as the convolutional TCN. The authors ran all four models as baselines. RMSE measures the average forecast error in cycles, while the Saxena score penalizes late failure warnings more heavily than premature ones. The authors stress that the figures are sensitive to three factors, as their ablation experiments showed: the choice of flight phase, the temporal downscaling method, and the RUL cap threshold.

"CruiseBench provides a reproducible sub-benchmark for controlled

comparison of RUL models, and CPM-N-CMAPSS provides a stage-specific data foundation for future research on transfer learning and domain adaptation," the paper's abstract on arXiv states.

What this means

Predictive maintenance for aircraft gets a fairer yardstick for comparing models. By fixing a single phase of flight, CruiseBench reduces the influence of preprocessing on the final numbers and improves reproducibility, a persistent problem for tasks built on raw industrial data. The CPM-N-CMAPSS mask is valuable in its own right: the authors position it as a foundation for future work on transfer learning and domain adaptation, where a model trained on one operating mode or engine is transferred to another.

Frequently asked questions

What is RUL in the context of aircraft engines?

RUL (Remaining Useful Life) is an estimate of how much longer an engine can safely operate. According to the authors, it is the central quantity for planning maintenance.

How does N-CMAPSS differ from C-MAPSS?

N-CMAPSS extends C-MAPSS: it simulates engine trajectories to failure based on recorded real flight profiles, and it keeps full time series within each flight rather than per-cycle snapshots. This level of detail makes the data more realistic, but also harder to process.