Kun Feng

Dr. Kun Feng


Research Focus & Metrics

Academic Career Length

10Years (2017-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 October 2020 in IEEE Access
Verified
Journal Article
Open access
YHYa He
Ya He
MHMinghui Hu
Minghui Hu
KFKun Feng
Kun Feng
ZJZhinong Jiang
Zhinong Jiang
Machine learning methods have made great development in data-driven fault diagnosis of rolling bearings. But the intelligent fault diagnosis of intershaft bearing faces the following two dilemmas: 1) the fault vibration is extremely weak, and it is difficult to extract features that can distinguish different classes; 2) due to the complex and variable working condition, the intershaft bearing does not always fail in same working conditions. That is, the labeled training data is not sufficient for every source domain. These challenges lead to the failure of traditional machine learning based fault diagnosis for intershaft bearings. Therefore, a novel intelligent fault diagnosis scheme is investigated for intershaft bearings of dual-rotor equipment under variable working conditions. The paper focuses on two key issues: 1) developing a feature extraction approach with which the fault features with excellent clustering and separation are extracted from vibration signals. This approach addresses the dilemma of weak fault feature extraction of intershaft bearing and creates a feasible precondition for subsequent feature transfer; 2) proposing a feature transfer method transforming the labeled sample features in multiple source domains into the trainable sample features in the target domain. This new transfer achieves the sharing of labeled training samples under working conditions and enriches the trainable samples in target domain. Ultimately, the faults of intershaft bearings can be diagnosed with the help of the neural network classifier trained by the transferred samples with labels. Experimental results verify that this established scheme is effective and superior to other comparable methods for the transfer diagnosis task from multiple source domains to target domain.

12 September 2017 in Applied Sciences
Verified
Journal Article
Open access
PDF file
ZJZhinong Jiang
Zhinong Jiang
MHMinghui Hu
Minghui Hu
KFKun Feng
Kun Feng
YHYa He
Ya He
Because both the inner and outer rings rotate, the intershaft bearings used in gas turbines do not have fixed bearing housings. As a result, the vibration of intershaft bearings cannot be measured directly. Therefore, a vibration signal can only be collected through indirect measurement. First, it must be transferred to adjacent bearings through the shafting. Then, it should be transferred by the elastic supports and complex structure of the thin-walled strut. The vibration signal is severely weakened during transmission under the influences of the transfer path. In the meantime, in the vibration of other components, a huge amount of noise is produced by the air flow, and the variable speeds of the inner and outer rings of the intershaft bearings make it harder to analyze the signal. Hence, it is very difficult to extract the vibration fault features of intershaft bearings. To deal with the variable speed of dual rotors, as well as the weak signal, a fault feature extraction scheme for the weak fault signals of intershaft bearings is proposed in this paper. This scheme is based on linear prediction, spectral kurtosis, and order tracking in the rotation speed difference domain. First, a prewhitening process, based on linear prediction, is applied to the fault signal of the intershaft bearings to eliminate the stationary component. Thus, the remaining components, including the impulse signal of faulty bearings and nonstationary noise, can retain the features of the vibrational bearings, in addition to reducing the noise. Second, the optimal center frequency and bandwidth of the band-pass filter, applied to resonant demodulation, are selected by spectral kurtosis. Subsequently, the enveloped signal containing the features of the faults found in the intershaft bearings is obtained by resonance demodulation. The quasi-stationary signal in the angle domain is acquired by the even angle resampling of the nonstationary envelope signal, as a result of the variable speed. The final order spectrum is obtained through a Fourier transform. Fault diagnosis can be conducted for the intershaft bearings by comparing this spectrum with the feature order of the bearing fault. Experiments were conducted to verify the validity of the proposed scheme.

31 January 2024 in Sensors
Verified
Journal Article
Open access
PDF file
DMDongyan Miao
Dongyan Miao
KFKun Feng
Kun Feng
YXYuan Xiao
Yuan Xiao
PZPeng Zhang
Peng Zhang
JGJinji Gao
Jinji Gao
Gas turbine vibration data may exhibit considerable differences under time-varying conditions, which poses challenges for neural network anomaly detection. We first propose a framework for a gas turbine vibration frequency spectra process under time-varying operation conditions, assisting neural networks’ ability to capture weak information. The framework involves scaling spectra for aligning all frequency components related to rotational speed and normalizing frequency amplitude in a self-adaptive way. Degressive beta variational autoencoder is employed for learning spectra characteristics and anomaly detection, while a multi-category anomaly index is proposed to accommodate various operating conditions. Finally, a dataset of blade Foreign Object Damage (FOD) fault occurring under time-varying operating conditions was used to validate the framework and anomaly detection. The results demonstrate that the proposed method can effectively reduce the spectra differences under time-varying conditions, and also detect FOD fault during operation, which are challenging to identify using conventional methods.

4 January 2018 in Shock and Vibration
Verified
Journal Article
Open access
ZJZhinong Jiang
Zhinong Jiang
MHMinghui Hu
Minghui Hu
KFKun Feng
Kun Feng
HWHao Wang
Hao Wang
Under frequently time‐varying operating conditions, equipment with dual rotors like gas turbines is influenced by two rotors with different rotating speeds. Alarm methods of fixed threshold are unable to consider the influences of time‐varying operating conditions. Hence, those methods are not suitable for monitoring dual‐rotor equipment. An early warning method for dual‐rotor equipment under time‐varying operating conditions is proposed in this paper. The influences of time‐varying rotating speeds of dual rotors on alarm thresholds have been considered. Firstly, the operating conditions are divided into several limited intervals according to rotating speeds of dual rotors. Secondly, the train data within each interval is processed by SVDD and the allowable ranges (i.e., the alarm threshold) of the vibration are determined. The alarm threshold of each interval of operating conditions is obtained. The alarm threshold can be expressed as a sphere, whose controlling parameters are the coordinate of the center and the radius. Then, the cluster center of the test data, whose alarm state is to be judged, can be extracted through K‐means. Finally, the alarm state can be obtained by comparing the cluster center with the corresponding sphere. Experiments are conducted to validate the proposed method.

19 July 2021 in Energies
Verified
Journal Article
Open access
PDF file
BYBinbin Yan
Binbin Yan
MHMinghui Hu
Minghui Hu
KFKun Feng
Kun Feng
ZJZhinong Jiang
Zhinong Jiang
Accurate component analytical solution is very important to gas path prognostics and diagnostics of a gas turbine. However, due to the highly complex nonlinear behavior of component performance, the nonlinear relationships between various key parameters still should be further studied. For this purpose, a new component analytical solution is proposed to enhance the current adaptation and diagnostics scheme of gas turbines. First, the tuning factors are defined to construct the enhanced component analytical solution and identify the nonlinear behaviors more accurately. Second, a sensitivity analysis for tuning factors is performed to understand the effect of each factor on the shape of component maps. Then, a particle swarm optimization algorithm is used to capture the optimal tuning factors, and then the performance adaptation is implemented. Finally, the proposed method has been validated with normal field data and fouling fault field data of a PGT25+ gas turbine. Compared with two earlier off-design point adaptation methods, the proposed method shows some advantages in performance adaptation and diagnostics.

16 August 2022 in IEEE Access
Verified
Journal Article
Open access
ZLZhouzheng Li
Zhouzheng Li
DMDongyan Miao
Dongyan Miao
KFKun Feng
Kun Feng
Thresholds are commonly used in condition monitoring and fault diagnosis of gas turbine engines as they are the criteria for health status discrimination. We discover that most thresholds should be varying according to gas turbine working condition, but many traditional fixed threshold designs didn’t consider this and therefore can potentially be replaced with dynamic ones for higher monitoring precision. In this paper, we propose an objective function that can be applied in any neural network model to learn to output a set of parameters for a distribution based on the maximum likelihood theory, and a matching dynamic threshold determination method for health state discrimination in example of a residual-based monitoring approach. The proposed methods are examined progressively with simulated data and finally an industrial gas turbine failure case. Results show that 1) the proposed objective function can learn the parameters of the distribution well; 2) the dynamic thresholds can effectively reduce the false-positive and false-negative rates when there is a varying noise comparing to fixed thresholds.