Bayesian Assessments of Aeroengine Performance

Pranay Seshadri, Andrew B. Duncan, George Thorne, Geoffrey T. Parks, Mark Girolami

TL;DR

This paper introduces a Bayesian framework for inferring aeroengine temperature and pressure profiles from sparse sensor data. By leveraging Gaussian Processes and a novel "planar covariance kernel," the method effectively transfers information across measurement planes, leading to more accurate area averages and a principled decomposition of uncertainty compared to traditional approaches.

What Problem Does This Paper Address?

Aeroengines generate complex internal flow fields characterized by temperature and pressure variations. Accurately measuring these profiles is crucial for engine design, performance monitoring, and predicting remaining useful life. However, practical limitations in sensor placement within an engine result in sparse data. Traditional methods for calculating performance metrics like area averages often rely on simplified assumptions that can lead to inaccuracies, especially with non-uniform flow. Furthermore, existing methods struggle to rigorously quantify and decompose the sources of uncertainty (e.g., sensor error vs. insufficient sampling), making it difficult to decide whether to add more sensors or improve existing ones. This paper aims to address these challenges by developing a statistically principled Bayesian approach to infer these spatial profiles and their associated uncertainties from limited data.

Methods

The core of the proposed method is a Bayesian Gaussian Process (GP) regression model. This model represents the spatial temperature or pressure profile at an axial station as a Gaussian random field. Specifically:

  • Circumferential variations are modeled using a Fourier basis.
  • Radial variations are modeled using a squared exponential kernel.

The key innovation is the extension of this model to handle multiple axial measurement planes using a novel planar covariance kernel. This kernel allows for transfer learning, enabling information to be shared between planes. This is particularly useful when some planes have more instrumentation than others.

To handle situations where the exact frequencies of the circumferential variations are unknown, the authors employ a sparsity-promoting prior on the Fourier coefficients. This encourages the model to identify and utilize only the most significant harmonics, leading to more parsimonious and robust representations.

The framework also provides a principled way to decompose the posterior uncertainty into two components:

  1. Impact of measurement imprecision: Uncertainty arising from sensor noise and calibration errors.
  2. Spatial sampling uncertainty: Uncertainty due to limited sensor coverage.

This decomposition helps in making informed decisions about instrumentation upgrades.

Datasets

The paper utilizes experimental data from a research turbine rig. Specific details about the dataset include:

  • Temperature measurements from an exit station in a high-pressure turbine.
  • Data collected at various traverse locations (hub, mid-span, tip) and circumferential positions.
  • The study simulates scenarios with varying numbers of rakes (sets of probes) and probes per rake to evaluate the model's performance under different instrumentation levels.
  • The paper also discusses using data from multiple research engines to demonstrate transfer learning across a fleet.

Key Insights and Ideas

  • Transfer Learning is Powerful for Sparse Data: By explicitly modeling the correlation between different measurement planes, the method can significantly improve inference, especially when some planes are sparsely instrumented. This is demonstrated by transferring knowledge from a more instrumented plane to a less instrumented one.
  • Sparsity is Key for Aeroengine Flows: Aeroengine flow fields often exhibit dominant harmonic components. Utilizing sparsity-promoting priors effectively captures this characteristic, leading to more accurate and robust models, particularly when the exact harmonic content is unknown.
  • Principled Uncertainty Decomposition: The ability to separate uncertainty into "measurement imprecision" and "spatial sampling" is a significant advantage. This provides actionable insights for engine manufacturers, guiding decisions on whether to improve sensor accuracy or increase sensor density.
  • Bayesian Area Average Outperforms Traditional Methods: The proposed Bayesian area average is shown to be more precise and less sensitive to rake placement compared to traditional sector area averaging.

Key Results

  • The Bayesian approach, particularly with transfer learning, significantly reduces uncertainty in inferring spatial temperature distributions, especially in sparsely instrumented planes.
  • The method successfully delineates complex spatial patterns (e.g., hot spots) even with limited sensor data.
  • The decomposition of uncertainty shows that spatial sampling uncertainty is often dominant over measurement imprecision, highlighting the importance of sensor placement.
  • In transfer learning scenarios, the model achieved improved predictions and reduced uncertainty by leveraging data from multiple planes. For instance, in one case, the uncertainty was significantly reduced due to improved mean prediction.
  • The paper demonstrates that the Bayesian area average is more robust and converges better than sector area averaging, especially when dealing with randomized rake positions.

Limitations and Future Work

  • Reliance on Similarity Parameter 's': The current transfer learning framework relies on a user-defined parameter 's' to specify the similarity between measurement planes. The authors acknowledge that in some cases, determining these relationships can be difficult.
  • Future Work:
    • Develop methods to learn the similarity parameter 's' directly from data using Dirichlet prior models.
  • Explore the application of the model to real-time engine health monitoring and prognostic efforts.
  • Investigate the model's performance with 1-2 rakes of instrumentation when paired with more instrumented planes.
  • Extend the framework to other aerothermal quantities beyond temperature and pressure.

Summary updated

Abstract

Aeroengine performance is determined by temperature and pressure profiles along various axial stations within an engine. Given limited sensor measurements both along and between axial stations, we require a statistically principled approach to inferring these profiles. In this paper we detail a Bayesian methodology for interpolating the spatial temperature or pressure profile at axial stations within an aeroengine. The profile at any given axial station is represented as a spatial Gaussian random field on an annulus, with circumferential variations modelled using a Fourier basis and radial variations modelled with a squared exponential kernel. This Gaussian random field is extended to ingest data from multiple axial measurement planes, with the aim of transferring information across the planes. To facilitate this type of transfer learning, a novel planar covariance kernel is proposed, with hyperparameters that characterise the correlation between any two measurement planes. In the scenario where precise frequencies comprising the temperature field are unknown, we utilise a sparsity-promoting prior on the frequencies to encourage sparse representations. This easily extends to cases with multiple engine planes whilst accommodating frequency variations between the planes. The main quantity of interest, the spatial area average is readily obtained in closed form. We term this the Bayesian area average and demonstrate how this metric offers far more precise averages than a sector area average -- a widely used area averaging approach. Furthermore, the Bayesian area average naturally decomposes the posterior uncertainty into terms characterising insufficient sampling and sensor measurement error respectively. This too provides a significant improvement over prior standard deviation based uncertainty breakdowns.
BibTeX
@inproceedings{seshadri2022bayesian,
  author = {Pranay Seshadri and Andrew B. Duncan and George Thorne and Geoffrey T. Parks and Mark Girolami},
  title = {{Bayesian Assessments of Aeroengine Performance}},
  booktitle = {Data-Centric},
  year = {2022},
  doi = {10.1017/dce.2022.29},
}
Figure