Turbofan Engine Sensor Fault Detection and Isolation Benchmark Dataset

Aircraft engine monitoring relies on sensor measurements to assess gas path condition and to distinguish gradual degradation from abrupt performance changes. In practice, however, sensor signals are influenced simultaneously by the engine state, changing operating conditions, and sensor side effects such as step, drift, random outliers, and measurement noise. This makes it difficult to determine whether an observed deviation originates from the engine, the environment, or the sensing system. For the development and fair comparison of sensor fault detection and isolation methods, a benchmark is required that represents these effects in a controlled and labelled manner. Most publicly available turbofan datasets, however, are primarily intended for remaining useful life prediction and do not provide standardised sensor fault cases for reproducible FDI evaluation. To address this gap, the Turbofan Sensor-FDI-Bench is introduced as a synthetic steady state benchmark dataset generated with a physics based turbofan performance model. The benchmark consists of cruise operating point snapshots and provides, for each flight, environmental conditions, an extended sensor package, and gradual multi component performance degradation. Structured sensor faults with controlled onset and severity are superimposed, including step and drift faults as well as stochastic measurement disturbances. The benchmark is organised as a progressive suite of subsets with increasing complexity, covering fixed and variable operating conditions as well as single fault and multi fault diagnosis settings. For each engine unit, clean reference sensor values are released alongside noisy or faulty measurements, enabling supervised denoising and controlled evaluation of sensor fault diagnosis methods. The resulting benchmark provides a reproducible basis for comparing sensor fault detection and isolation methods under degradation and operating variability.

航空发动机监测依赖传感器测量结果，以评估燃气路径状态，并区分性能的渐进式退化与突变性变化。然而在实际应用中，传感器信号同时受到发动机状态、变化的运行工况，以及阶跃故障、漂移故障、随机异常值与测量噪声等传感器干扰效应的影响。这使得难以判断观测到的信号偏差源自发动机、运行环境还是传感系统本身。为了开发传感器故障检测与隔离（Fault Detection and Isolation, FDI）方法并开展公平对比，需要一个能够以受控且带标注的方式复现上述各类影响的基准数据集。然而，当前多数公开可用的涡扇发动机数据集主要用于剩余使用寿命预测，并未提供可复现FDI评估所需的标准化传感器故障案例。为填补这一空白，本研究提出了Turbofan Sensor-FDI-Bench——一款基于物理建模的涡扇发动机性能模型生成的合成稳态基准数据集。该基准数据集包含巡航工况点的瞬时快照数据，为每个飞行工况提供了运行环境参数、扩展传感器组测量数据，以及多部件渐进式性能退化信息。数据集还叠加了受控起始时刻与严重程度的结构化传感器故障，包括阶跃故障、漂移故障以及随机测量扰动。该基准数据集按照复杂度递增的原则组织为渐进式子集套件，涵盖固定工况与变工况场景，以及单故障与多故障诊断设置。针对每个发动机单元，数据集同步提供了干净的参考传感器真值与带噪声或故障的测量数据，可用于监督去噪以及传感器故障诊断方法的受控评估。该基准数据集可为退化工况与运行工况变化下的传感器故障检测与隔离方法的对比提供可复现的研究基础。