This article develops new deep learning methods, namely, residual shrinkage networks, to improve the feature ability from highly noised vibration signals and achieve a high fault diagnosing accuracy. Soft thresholding is inserted as nonlinear transformation layers into architectures eliminate unimportant features. Moreover, considering that it generally challenging set proper values for thresholds, developed networks integrate few specialized neural trainable modules automatically determine...

Vibration signals under the same health state often have large differences due to changes in operating conditions. Likewise, among vibration different states can be small some Traditional deep learning methods apply fixed nonlinear transformations all input signals, which a negative impact on discriminative feature ability, i.e., projecting intraclass into region and interclass distant regions. Aiming at this issue, article develops new activation function, adaptively parametric rectifier...

One of the significant tasks in remaining useful life (RUL) prediction is to find a good health indicator (HI) that can effectively represent degradation process system. However, it difficult for traditional data-driven methods construct accurate HIs due their incomprehensive consideration temporal dependencies within monitoring data, especially aeroengines working under nonstationary operating conditions (OCs). Aiming at this problem, article develops novel unsupervised deep neural network,...

Engineering data are often highly nonlinear and contain high-frequency noise, so the Levenberg–Marquardt (LM) algorithm may not converge when a neural network optimized by is trained with engineering data. In this work, we analyzed reasons for LM network’s poor convergence commonly associated algorithm. Specifically, effects of different activation functions such as Sigmoid, Tanh, Rectified Linear Unit (RELU) Parametric (PRLU) were evaluated on general performance networks, special values...

Many existing aircraft engine fault detection methods are highly dependent on performance deviation data that provided by the original equipment manufacturer. To improve independent ability, Aircraft Communications Addressing and Reporting System (ACARS) can be used. However, owing to characteristics of high dimension, complex correlations between parameters, large noise content, it is difficult for detect faults effectively using ACARS data. solve this problem, a novel method based...

Effective fault detection and identification methods are crucial in gas turbine maintenance. To express the performance of symptom state precisely to reduce individual differences different turbines, a novel deviation model based on real-life operation data turbines is proposed this paper. A backpropagation neural network adopted establish model. Performance values calculated by regarded as signatures turbines. enhance accuracy diagnosis, multikernel support vector machine employed...

Gas turbine engine anomaly detection is a critical means to ensure the safety and economic efficiency of flight. As gas path faults make up sizeable proportion all faults, an method was proposed in present article. Inspired by recent progress deep learning, we explored that combined learning with traditional improve accuracy detection. Firstly stacked denoising autoencoders model built learn robust features from datasets without labels. Then, used learned as input algorithm based on Gaussian...

The variations in gas path parameter deviations can fully reflect the healthy state of aero-engine components and units; therefore, airlines usually take them as key parameters for monitoring performance conducting fault diagnosis. In past, could not obtain autonomously. At present, a data-driven method based on an dataset with large sample size be utilized to deviations. However, it is still difficult utilize datasets small sizes establish regression models deep neural networks. To autonomy...

In general, deep learning-based fault diagnosis methods need a large number of training samples, which are often not available in real applications. Aiming at this problem, article develops new data augmentation method, i.e., randomized wavelet expansion (RWE), to generate set synthesis samples that share similar characteristics with the original sample. The first key point is amplitudes coefficients randomly selected frequency band enlarged through random expansion. Another processed have...

Considering that large mechanical equipment often has various excitation sources, the signals generated by these sources are not simply added or multiplied together, but nonlinearly mixed, which exhibit complex non-stationary characteristics, making classical algorithms difficult to extract fault features. Especially when faults just occur, symptom is weak and submerged noise, resulting in low diagnosing accuracy. Accordingly, this article develops a new deep attention method, namely...

Monitoring gas turbines' health, in particular, detecting abnormal behaviors time, is critical ensuring turbine operating safety and preventing costly unplanned maintenance. One most popular anomaly detection method to obtain a classification-prediction model by training classifier using the real-life data of turbine. The excellent ability this attributed enough annotated samples, especially samples. Nevertheless, monitoring data, normal far more than even no data. Advanced technologies that...