Remaining useful life prediction is one of the essential processes for machine system prognostics and health management. Although there are many new approaches based on deep learning for remaining useful life prediction emerging in recent years, these methods still have the following weaknesses: (1) The correlation between the information collected by each sensor and the remaining useful life of the machinery is not sufficiently considered. (2) The accuracy of deep learning algorithms for remaining useful life prediction is low due to the high noise, over-dimensionality, and non-linear signals generated during the operation of complex systems. To overcome the above weaknesses, a general deep long short memory network-based approach for mechanical remaining useful life prediction is proposed in this paper. Firstly, a two-step maximum information coefficient method was built to calculate the correlation between the sensor data and the remaining useful life. Secondly, the kernel principal component analysis with a simple moving average method was designed to eliminate noise, reduce dimensionality, and extract nonlinear features. Finally, a deep long short memory network-based deep learning method is presented to predict remaining useful life. The efficiency of the proposed method for remaining useful life prediction of a nonlinear degradation process is demonstrated by a test case of NASA’s commercial modular aero-propulsion system simulation data. The experimental results also show that the proposed method has better prediction accuracy than other state-of-the-art methods.

In view of the high false alarm rate in the oil debris monitoring results of the triple-coil inductive sensor in the transmission lubrication system of the aero-engine, a new debris signal processing method based on inductive monitoring is proposed. A time domain analysis is carried out first, and the signal energy is the most effective index to distinguish the debris signature from the noise signature. On this basis, signal energy values within a fixed-length sliding window is processed through the histogram. Finally, a threshold is set for the detection of the debris signature, which is based on the distribution of data within the histogram. This method is applied to the experimental data from a test run of an aero-engine, and the results show that all the debris is detected even if part of it appears during a change in the working condition of the aero-engine. Therefore, this method shows satisfactory results in debris detection accuracy and especially the inhibition of false alarms. It is also applicable for real-time monitoring due to the similarity between the movement of the sliding window and real-time data acquisition. In addition, it is applicable for various sensing principles, including but not limited to the inductive sensor signal, since the detection performance is only related to the signal itself.

Array-type electrostatic monitoring is gradually becoming an effective tool for aero-engine fault diagnosis. In this paper, we innovatively apply the surrogate optimization method to the optimization of the sensor array structure in order to meet the need of improving the particle information recognition capability of the electrostatic sensor array (ESA). A structure optimization method of ESA based on the Kriging surrogate model and improved NSGA-II algorithm is proposed. In this paper, a finite element simulation model of ESA is established, and the array optimization problem is abstracted as the solution of a mixed-integer optimization problem. This paper reduces the large-scale numerical simulations in the full-variable space with the help of the Kriging surrogate model. In addition, an improved NSGA-II algorithm for mixed-integer optimization is proposed. The simulation experiment verified that the average absolute error of the sensor before and after optimization for the identification of particle position and charge quantity was reduced by 69.88% and 49.68%, respectively. The array structure optimization method proposed in this paper facilitates the acceleration of the design process of electrostatic sensors and provides a scientific design method for their specific design for airborne application.

Aero-engine electrostatic monitoring technology (EMT) is a novel and effective condition monitoring technology. With the help of EMT, effective monitoring of early failures can be achieved. Since the electrostatic monitoring of the running engine will be strongly interfered, the sampled electrostatic signal has various noise components and low signal-to-noise ratio (SNR). After analyzing the source of the noise components carried by the electrostatic signal, this paper proposes a method for electrostatic signal denoising in a strong interference environment, which is based on the nonlinear total variation theory. In the experiments, the simulated electrostatic measurement signal and the actual test-run electrostatic measurement signal were used as the analysis objects, and the denoising test was carried out by using the proposed method. Meanwhile, the denoising effect was compared and analyzed with other classical methods. The experimental results show that the proposed denoising method can effectively remove random noise, electromagnetic pulse and periodic noise in electrostatic signal, and is more applicable to the measured electrostatic signal with low SNR than the classical electrostatic signal denoising methods such as wavelet threshold denoising method and empirical mode decomposition method.

Auxiliary Power Unit (APU) is an indispensable component utilized in modern aircraft, which provides electrical and pneumatic power to the aircraft independently. What’s more, APU can help the main engines restart in case of main engine failure during flight. Thus there exists the need of APU monitoring. However, APU has not received sufficient attention in maintenance due to its relatively low cost and safety requirements compared to main engines. Line maintenance shows that APU is likely to fail in the start stage. Based on the Quick Access Recorder data supported by the airline, a fault detection method for APU is proposed in this paper. In order to improve the accuracy of fault detection, feature selection is carried out firstly to determine the optimal subset of monitoring parameters by Recursive Feature Elimination. Then fault detection is carried out by Support Vector Machine. Results show that Exhaust Gas Temperature is the most effective index among all monitoring parameters, and feature selection has a significant improvement on the accuracy of fault detection, which means higher accuracy with less monitoring cost. The method proposed in this paper also disposes that some monitoring parameters should be paid more attention in terms of the condition monitoring of APU.