Performance Sensing Data Prediction for an Aircraft Auxiliary Power Unit Using the Optimized Extreme Learning Machine

Xiaolei Liu, Liansheng Liu, Lulu Wang, Qing Guo, Xiyuan PengSensors 2019 · DOI

TL;DR

This paper tackles the challenge of predicting performance data for aircraft Auxiliary Power Units (APUs), which is crucial for condition-based maintenance. They propose an optimized Extreme Learning Machine (ELM) that uses a Restricted Boltzmann Machine (RBM) to overcome the ELM's inherent instability, leading to more accurate and reliable predictions.

What Problem Does This Paper Address?

Aircraft APUs are vital for cabin environmental control and engine starting, but their maintenance is costly and often reactive (after failure). To improve reliability and reduce costs, condition-based maintenance is preferred, which relies on accurate prediction of APU performance data. However, APU performance data exhibits complex, nonlinear characteristics, making accurate prediction difficult. Traditional neural networks can struggle with this nonlinearity and have slow training times. While Extreme Learning Machines (ELMs) offer faster training and better generalization, their randomly initialized weights and thresholds can lead to unstable and unreliable predictions, hindering their practical application for critical systems like APUs.

Methods

The core of this paper is the proposed method of optimizing the Extreme Learning Machine (ELM) using a Restricted Boltzmann Machine (RBM).

  • Extreme Learning Machine (ELM): The ELM is a single-hidden-layer feedforward neural network known for its rapid training speed and good generalization capabilities, unlike traditional backpropagation networks. It achieves this by randomly initializing input weights and biases and then analytically determining the output weights.
  • Restricted Boltzmann Machine (RBM): The RBM is a generative stochastic neural network with two layers (visible and hidden) that can learn a probability distribution over its inputs. Crucially, RBMs are known for their ability to escape local minima during training due to their probabilistic nature, which helps in finding more robust solutions.
  • RBM-Optimized ELM: The proposed approach uses a pre-trained RBM to initialize the weights and thresholds of the ELM's input and hidden layers. This pre-training step leverages the RBM's ability to learn meaningful feature representations and avoid the instability associated with random initialization in standard ELMs. The trained RBM's parameters are then transferred to the ELM, which is subsequently trained. This hybrid approach aims to combine the speed of ELM with the stability and robustness gained from RBM-based feature extraction.

Datasets

The study utilized real APU sensing data from the China Southern Airlines Company Limited Shenyang Maintenance Base. The paper describes three groups of experiments with varying numbers of training and testing data points, ranging from 240 to 320 training data points and 21 to 41 test data points. No new benchmark datasets were introduced.

Key Insights and Ideas

  • ELM Instability is a Real Problem: The paper clearly demonstrates that standard ELMs, due to random initialization, produce highly variable and unstable predictions, making them unreliable for critical applications.
  • RBMs as a Stabilizing Force: The RBM's ability to learn robust feature representations and escape local minima is effectively leveraged to overcome ELM's instability. By using RBM-trained weights and biases, the ELM becomes significantly more stable.
  • Predicting APU Health is Crucial: The research underscores the importance of accurate APU performance prediction for enabling condition-based maintenance, which can lead to significant cost savings and improved flight safety.
  • Exhaust Gas Temperature (EGT) as a Key Indicator: The study identifies EGT as a critical performance parameter for predicting APU degradation.

Key Results

The proposed RBM-optimized ELM method demonstrated superior performance compared to the original ELM:

  • Improved Accuracy: The proposed method achieved significantly lower Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) across all experimental groups. For instance, in Group 1, MAE, MAPE, and RMSE were reduced to 35.7%, 36.1%, and 45.7% of their original ELM values, respectively.
  • Enhanced Stability: Visualizations and evaluation metrics clearly showed that the RBM-optimized ELM produced much more stable and consistent predictions, unlike the erratic outputs of the standard ELM.
  • Practical Applicability: The method was validated using real-world APU data, demonstrating its potential for practical implementation in aircraft maintenance systems.

Limitations and Future Work

  • Environmental Factors Not Considered: The authors acknowledge that environmental factors such as ambient temperature and atmospheric pressure were not considered in their EGT sensing data analysis.
  • Future Work: They plan to incorporate a physical model of bleed air performance to further refine EGT data and enhance prediction accuracy.

Summary updated

Abstract

The aircraft auxiliary power unit (APU) is responsible for environmental control in the cabin and the main engines starting the aircraft. The prediction of its performance sensing data is significant for condition-based maintenance. As a complex system, its performance sensing data have a typically nonlinear feature. In order to monitor this process, a model with strong nonlinear fitting ability needs to be formulated. A neural network has advantages of solving a nonlinear problem. Compared with the traditional back propagation neural network algorithm, an extreme learning machine (ELM) has features of a faster learning speed and better generalization performance. To enhance the training of the neural network with a back propagation algorithm, an ELM is employed to predict the performance sensing data of the APU in this study. However, the randomly generated weights and thresholds of the ELM often may result in unstable prediction results. To address this problem, a restricted Boltzmann machine (RBM) is utilized to optimize the ELM. In this way, a stable performance parameter prediction model of the APU can be obtained and better performance parameter prediction results can be achieved. The proposed method is evaluated by the real APU sensing data of China Southern Airlines Company Limited Shenyang Maintenance Base. Experimental results show that the optimized ELM with an RBM is more stable and can obtain more accurate prediction results.
BibTeX
@inproceedings{liu2019performance,
  author = {Xiaolei Liu and Liansheng Liu and Lulu Wang and Qing Guo and Xiyuan Peng},
  title = {{Performance Sensing Data Prediction for an Aircraft Auxiliary Power Unit Using the Optimized Extreme Learning Machine}},
  booktitle = {Sensors},
  year = {2019},
  doi = {10.3390/s19183935},
}
Figure