Huimin Song
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Huimin Song has a total of 47 co-authors. Here are the top co-authors by number of shared publications:
Min Jia
Air Force Engn Univ, Sci & Technol Plasma Dynam Lab, Xian 710038, Peoples R China
5 shared publicationsYun Wu
4 shared publications- Zhibo Zhang
Air Force Engineering University
4 shared publications Wei Cui
3 shared publications- Yun Wu
National Key Lab of Aerospace Power System and Plasma Technology, Xi’an Jiaotong University, Xi’an 710049, People’s Republic of China
3 shared publications Di Jin
Air Force Engn Univ, Sci & Technol Plasma Dynam Lab, Xian 710038, Shaanxi, Peoples R China
2 shared publications
Latest Publications
- 18 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-13
An Incremental Learning Method With Feature-Attention Distillation and Logit Adjustment for Rotating Machinery Fault Diagnosis
Deep learning-based diagnosis methods can accurately identify the fault mode, which have attracted widespread attention from researchers. For mechanical systems, in actual industrial environments, data from different fault modes will continue to emerge, which requires the model to be updated in a timely manner and maintain high diagnosis accuracy. Class incremental learning (CIL) is proposed for classification problems with a continuously increasing number of modes, which meets the requirement of industrial diagnosis. However, directly incorporating new class data into the training set to optimize the network will cause the model to forget old class knowledge, resulting in irreversible performance degradation, known as catastrophic forgetting. To solve this problem, this paper constructs an Incremental Learning method with Feature-Attention Distillation and Logit Adjustment (ILD-FADLA) for fault diagnosis. Specifically, a residual network composed of convolutional blocks is utilized as the feature extractor, and channel attention and spatial attention are superimposed in each residual block to enhance the feature extraction capability. To alleviate catastrophic forgetting, the proposed ILD-FADLA distills the attention weights at each layer and feature relation information before classifier. In addition, logit adjustment cross-entropy loss is employed to mitigate the bias of the classifier towards new classes. Experimental results on two private datasets show that the proposed ILD-FADLA improves the average accuracy of the incremental phase by 17.76% and 22.42% over the baseline method.
- 18 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-13
An Incremental Learning Method With Feature-Attention Distillation and Logit Adjustment for Rotating Machinery Fault Diagnosis
Deep learning-based diagnosis methods can accurately identify the fault mode, which have attracted widespread attention from researchers. For mechanical systems, in actual industrial environments, data from different fault modes will continue to emerge, which requires the model to be updated in a timely manner and maintain high diagnosis accuracy. Class incremental learning (CIL) is proposed for classification problems with a continuously increasing number of modes, which meets the requirement of industrial diagnosis. However, directly incorporating new class data into the training set to optimize the network will cause the model to forget old class knowledge, resulting in irreversible performance degradation, known as catastrophic forgetting. To solve this problem, this paper constructs an Incremental Learning method with Feature-Attention Distillation and Logit Adjustment (ILD-FADLA) for fault diagnosis. Specifically, a residual network composed of convolutional blocks is utilized as the feature extractor, and channel attention and spatial attention are superimposed in each residual block to enhance the feature extraction capability. To alleviate catastrophic forgetting, the proposed ILD-FADLA distills the attention weights at each layer and feature relation information before classifier. In addition, logit adjustment cross-entropy loss is employed to mitigate the bias of the classifier towards new classes. Experimental results on two private datasets show that the proposed ILD-FADLA improves the average accuracy of the incremental phase by 17.76% and 22.42% over the baseline method.
- 5 December 2024
- Published by Springer Nature
- p. 312-321
Induced Flow Characteristics and Applications of Plasma Actuation Based on High-Repetition-Rate Fast-Front NS-DBD
- 5 December 2024
- Published by Springer Nature
- p. 312-321
Induced Flow Characteristics and Applications of Plasma Actuation Based on High-Repetition-Rate Fast-Front NS-DBD
- Open Access
- 31 October 2024
- Published by Atlantis Press SARL
- p. 137-142
Discussion on the Teaching Improvement of Professional Basic Courses for Military Academy Students, Taking Non-electrical Measurement Techniques as Example
- 31 October 2024
- Published by Atlantis Press SARL
- p. 137-142
Open AccessDiscussion on the Teaching Improvement of Professional Basic Courses for Military Academy Students, Taking Non-electrical Measurement Techniques as Example
- 1 September 2024
- Published by AIP Publishing in Physics of Fluids
- Vol. 36 (9)
A field programmable gate array-based deep reinforcement learning framework for experimental active flow control and its application in airfoil flow separation elimination
Although deep reinforcement learning (DRL) has gained increasing popularity in numerical studies of active flow control, practical implementations of this technique in experiments remain challenging, particularly for fast air flows. In this study, we proposed a field programmable gate array (FPGA)-based DRL framework for high-speed experimental active flow control. By splitting the training and execution process of artificial neural network and implementing them effectively in central processing unit (CPU) and FPGA, experimental DRL control with an interaction frequency up to 10–100 kHz can be realized, which is two orders higher than that of the traditional CPU-based DRL framework. Such a new framework is tested in the classical problem of airfoil leading flow separation control at Rec = 2.13 × 105, with a plasma actuator and a hotwire serving as the flow disturber and the state sensor, respectively. As a result, experimental DRL control is rather attractive in terms of the time cost, able to find a converged closed-loop control strategy in only one run of 5 min, eliminating the tedious parameter tuning process in open-loop control (time cost: dozens of minutes to hours). Moreover, the magnitude of lift increment in the case of optimal DRL control is 3.2% higher than that of the best open-loop periodical control strategy.
- 1 September 2024
- Published by AIP Publishing in Physics of Fluids
- Vol. 36 (9)
A field programmable gate array-based deep reinforcement learning framework for experimental active flow control and its application in airfoil flow separation elimination
Although deep reinforcement learning (DRL) has gained increasing popularity in numerical studies of active flow control, practical implementations of this technique in experiments remain challenging, particularly for fast air flows. In this study, we proposed a field programmable gate array (FPGA)-based DRL framework for high-speed experimental active flow control. By splitting the training and execution process of artificial neural network and implementing them effectively in central processing unit (CPU) and FPGA, experimental DRL control with an interaction frequency up to 10–100 kHz can be realized, which is two orders higher than that of the traditional CPU-based DRL framework. Such a new framework is tested in the classical problem of airfoil leading flow separation control at Rec = 2.13 × 105, with a plasma actuator and a hotwire serving as the flow disturber and the state sensor, respectively. As a result, experimental DRL control is rather attractive in terms of the time cost, able to find a converged closed-loop control strategy in only one run of 5 min, eliminating the tedious parameter tuning process in open-loop control (time cost: dozens of minutes to hours). Moreover, the magnitude of lift increment in the case of optimal DRL control is 3.2% higher than that of the best open-loop periodical control strategy.
- Top Cited Papers
- 1 January 2024
- Published by AIP Publishing in Physics of Fluids
- Vol. 36 (1)
Closed-loop plasma flow control of a turbulent cylinder wake flow using machine learning at Reynolds number of 28 000
Machine learning is increasingly used for active flow control. In this experimental study, alternating-current dielectric barrier discharge plasma actuators are deployed for the closed-loop intelligent control of the flow around a cylinder at a Reynolds number of 28 000 based on the velocity feedback from two hot-wire sensors placed in the wake. Variations in the cylinder drag are monitored by a load cell, and the temporal response of the wake flow field is visualized by a high-speed particle image velocimetry system working at 1 kHz. The high-speed control law is operated using a field programmable gate array optimized by genetic programing (GP). The results show that the peak drag reduction achieved by machine learning is of similar magnitude to that of conventional steady actuation (∼15%), while the power saving ratio is 35% higher than with conventional techniques because of the reduced power consumption. Analysis of the best GP control laws shows that the intensity of plasma actuation should be kept at a medium level to maximize the power-saving ratio. When compared with the baseline uncontrolled flow, the best controlled cases constrain the meandering motion of the cylinder wake, resulting in a narrow stabilized velocity deficit zone in the time-averaged sense. According to the results of proper orthogonal decomposition and dynamic mode decomposition, Karman vortex shedding is promoted under the best GP control.
- 1 January 2024
- Published by AIP Publishing in Physics of Fluids
- Vol. 36 (1)
Top Cited PapersClosed-loop plasma flow control of a turbulent cylinder wake flow using machine learning at Reynolds number of 28 000
Machine learning is increasingly used for active flow control. In this experimental study, alternating-current dielectric barrier discharge plasma actuators are deployed for the closed-loop intelligent control of the flow around a cylinder at a Reynolds number of 28 000 based on the velocity feedback from two hot-wire sensors placed in the wake. Variations in the cylinder drag are monitored by a load cell, and the temporal response of the wake flow field is visualized by a high-speed particle image velocimetry system working at 1 kHz. The high-speed control law is operated using a field programmable gate array optimized by genetic programing (GP). The results show that the peak drag reduction achieved by machine learning is of similar magnitude to that of conventional steady actuation (∼15%), while the power saving ratio is 35% higher than with conventional techniques because of the reduced power consumption. Analysis of the best GP control laws shows that the intensity of plasma actuation should be kept at a medium level to maximize the power-saving ratio. When compared with the baseline uncontrolled flow, the best controlled cases constrain the meandering motion of the cylinder wake, resulting in a narrow stabilized velocity deficit zone in the time-averaged sense. According to the results of proper orthogonal decomposition and dynamic mode decomposition, Karman vortex shedding is promoted under the best GP control.
- Open Access
Experimental Investigation on Ignition of Hyperburner Based on Gliding Arc Plasma Igniter Driven by Pressure Difference
The hyperburner is a key component of a TBCC engine, and its reliable ignition and stable operation are critical. The gliding arc plasma igniter driven by differential pressure has the technical advantages of low energy consumption and high jet temperature. In this paper, the electrical and flow characteristics of the gliding arc plasma igniter are studied, and the basic ignition experiment in the hyperburner is carried out. Electrical characteristic experiments show that the discharge duration, the evolution of the gliding arc and the fracture frequency are affected by the pressure difference between the inlet and outlet of the igniter (Δp). With the increase in Δp, the frequency of the trapezoidal envelope in the voltage and current waveforms increases, and the frequency of the evolution and fracture of the gliding arc increases. The continuous discharge time of the gliding arc decreases when Δp = 550 Torr. The flow characteristic experiments show that the velocity of the swirl sheath is increased and the protective effect on the gliding arc is enhanced with the increase in Δp. In the range of 20–550 Torr, the jet length first increases and then decreases with the increase in Δp. The jet length reaches a maximum of 31 mm at Δp = 50 Torr. Basic ignition experiments show that proper Δp can widen the lean ignition limit and shorten the ignition delay time. In the working conditions of this paper, the ignition effect is the best when Δp = 350 Torr, which can widen the lean ignition limit by 37.5% and shorten the ignition delay time by 17%. After increasing the oil–gas ratios, the combustion is more complete and the ignition delay time can be shortened by 93.1% at most.
Open AccessExperimental Investigation on Ignition of Hyperburner Based on Gliding Arc Plasma Igniter Driven by Pressure Difference
The hyperburner is a key component of a TBCC engine, and its reliable ignition and stable operation are critical. The gliding arc plasma igniter driven by differential pressure has the technical advantages of low energy consumption and high jet temperature. In this paper, the electrical and flow characteristics of the gliding arc plasma igniter are studied, and the basic ignition experiment in the hyperburner is carried out. Electrical characteristic experiments show that the discharge duration, the evolution of the gliding arc and the fracture frequency are affected by the pressure difference between the inlet and outlet of the igniter (Δp). With the increase in Δp, the frequency of the trapezoidal envelope in the voltage and current waveforms increases, and the frequency of the evolution and fracture of the gliding arc increases. The continuous discharge time of the gliding arc decreases when Δp = 550 Torr. The flow characteristic experiments show that the velocity of the swirl sheath is increased and the protective effect on the gliding arc is enhanced with the increase in Δp. In the range of 20–550 Torr, the jet length first increases and then decreases with the increase in Δp. The jet length reaches a maximum of 31 mm at Δp = 50 Torr. Basic ignition experiments show that proper Δp can widen the lean ignition limit and shorten the ignition delay time. In the working conditions of this paper, the ignition effect is the best when Δp = 350 Torr, which can widen the lean ignition limit by 37.5% and shorten the ignition delay time by 17%. After increasing the oil–gas ratios, the combustion is more complete and the ignition delay time can be shortened by 93.1% at most.
- Open Access
Experimental Investigation on Flow Characteristics and Ignition Performance of Plasma-Actuated Flame Holder
Improving the performance of flame holders has been a key focus of research on ramjet combustors. The plasma actuator has the potential to improve the ignition performance by manipulating the flow field of the flame holder. In this study, a plasma-actuated flame holder was designed. The aim of this study is to improve the performance of ramjet combustor by applying plasma discharge to the flame holder. The aerodynamic effects and ignition performance of the flame holder were investigated. The results demonstrated that the induced jet direction of the surface arc discharge was perpendicular to the actuator. The induced jet dissipated faster at lower pressures. The aerodynamic actuation intensity and jet area increased with the number of channels of surface arc discharges. Increasing discharge frequencies can increase the discharge times and jet height. The aerodynamic effects under a microsecond pulse duration were better than those under a nanosecond pulse duration. Actuators installed on the inside surface showed better performance than those installed outside. Under different total flow temperature conditions, the plasma-actuated flame holder significantly extended the ignition pressure limit and increased the combustion efficiency by 9.12% and 4.3% on average, respectively.
Open AccessExperimental Investigation on Flow Characteristics and Ignition Performance of Plasma-Actuated Flame Holder
Improving the performance of flame holders has been a key focus of research on ramjet combustors. The plasma actuator has the potential to improve the ignition performance by manipulating the flow field of the flame holder. In this study, a plasma-actuated flame holder was designed. The aim of this study is to improve the performance of ramjet combustor by applying plasma discharge to the flame holder. The aerodynamic effects and ignition performance of the flame holder were investigated. The results demonstrated that the induced jet direction of the surface arc discharge was perpendicular to the actuator. The induced jet dissipated faster at lower pressures. The aerodynamic actuation intensity and jet area increased with the number of channels of surface arc discharges. Increasing discharge frequencies can increase the discharge times and jet height. The aerodynamic effects under a microsecond pulse duration were better than those under a nanosecond pulse duration. Actuators installed on the inside surface showed better performance than those installed outside. Under different total flow temperature conditions, the plasma-actuated flame holder significantly extended the ignition pressure limit and increased the combustion efficiency by 9.12% and 4.3% on average, respectively.
- 1 June 2020
- Published by IOP Publishing in Chinese Physics B
- Vol. 29 (6) , 065207
Experimental study on energy characteristics and ignition performance of recessed multichannel plasma igniter*
In the extreme conditions of high altitude, low temperature, low pressure, and high speed, the aircraft engine is prone to flameout and difficult to start secondary ignition, which makes reliable ignition of combustion chamber at high altitude become a worldwide problem. To solve this problem, a kind of multichannel plasma igniter with round cavity is proposed in this paper, the three-channel and five-channel igniters are compared with the traditional ones. The discharge energy of the three igniters was compared based on the electric energy test and the thermal energy test, and ignition experiments was conducted in the simulated high-altitude environment of the component combustion chamber. The results show that the recessed multichannel plasma igniter has higher discharge energy than the conventional spark igniter, which can increase the conversion efficiency of electric energy from 26% to 43%, and the conversion efficiency of thermal energy from 25% to 73%. The recessed multichannel plasma igniter can achieve greater spark penetration depth and excitation area, which both increase with the increase of height. At the same height, the inlet flow helps to increase the penetration depth of the spark. The recessed multichannel plasma igniter can widen the lean ignition boundary, and the maximum enrichment percentage of lean ignition boundary can reach 31%.
- 1 June 2020
- Published by IOP Publishing in Chinese Physics B
- Vol. 29 (6) , 065207
Experimental study on energy characteristics and ignition performance of recessed multichannel plasma igniter*
In the extreme conditions of high altitude, low temperature, low pressure, and high speed, the aircraft engine is prone to flameout and difficult to start secondary ignition, which makes reliable ignition of combustion chamber at high altitude become a worldwide problem. To solve this problem, a kind of multichannel plasma igniter with round cavity is proposed in this paper, the three-channel and five-channel igniters are compared with the traditional ones. The discharge energy of the three igniters was compared based on the electric energy test and the thermal energy test, and ignition experiments was conducted in the simulated high-altitude environment of the component combustion chamber. The results show that the recessed multichannel plasma igniter has higher discharge energy than the conventional spark igniter, which can increase the conversion efficiency of electric energy from 26% to 43%, and the conversion efficiency of thermal energy from 25% to 73%. The recessed multichannel plasma igniter can achieve greater spark penetration depth and excitation area, which both increase with the increase of height. At the same height, the inlet flow helps to increase the penetration depth of the spark. The recessed multichannel plasma igniter can widen the lean ignition boundary, and the maximum enrichment percentage of lean ignition boundary can reach 31%.
- 1 February 2019
- Published by IOP Publishing in Chinese Physics B
- Vol. 28 (2)
Aerodynamic actuation characteristics of radio-frequency discharge plasma and control of supersonic flow
- 1 February 2019
- Published by IOP Publishing in Chinese Physics B
- Vol. 28 (2)
Aerodynamic actuation characteristics of radio-frequency discharge plasma and control of supersonic flow
- Open Access
- 1 September 2018
- Published by MDPI AG in Applied Sciences
- Vol. 8 (9) , 1534
Analytic Model and the Influence of Actuator Number on the Performance of Plasma Synthetic Jet Actuator Array
Coupled with the multichannel discharge model and plasma synthetic jet actuator (PSJA) aerodynamic model, an analytical model to predict the performance of the PSJA array is put forward. The multichannel discharge model takes these factors into consideration, the delay time in the breakdown process, the electrical transformation of the discharge channel from a capacitor to a resistor induced by the air breakdown, and the varying plasma resistance in the discharge process. The PSJA aerodynamic model is developed based on the conservation equations of mass, momentum, energy, and the lumped capacitance method. The multichannel discharge model can simulate the multichannel discharge process and give the discharge energy in the plasma channel. With a constant heating efficiency, the time-independent heating energy deposition power in a discharge channel is obtained. Importing the heating energy, the PSJA aerodynamic model presents the evolution process of the jet. Simulation results show that the jet strength induced by a single actuator decreases with the number of actuators in the PSJA array. When the actuator number increases from 1 to 20, the weakening extent of mass ejected, peak jet velocity, and jet duration time is 62%, 54%, and 33%, respectively. The discharge efficiency increases with the actuator number, while the thermodynamic efficiency decreases with the actuator number. As a result, the total energy efficiency doesn’t always increase with an increase in the number of actuators. When the discharge efficiency of a conventional one channel discharge has been a relatively large value, the total energy efficiency actually decreases with the growth of actuator number.
- 1 September 2018
- Published by MDPI AG in Applied Sciences
- Vol. 8 (9) , 1534
Open AccessAnalytic Model and the Influence of Actuator Number on the Performance of Plasma Synthetic Jet Actuator Array
Coupled with the multichannel discharge model and plasma synthetic jet actuator (PSJA) aerodynamic model, an analytical model to predict the performance of the PSJA array is put forward. The multichannel discharge model takes these factors into consideration, the delay time in the breakdown process, the electrical transformation of the discharge channel from a capacitor to a resistor induced by the air breakdown, and the varying plasma resistance in the discharge process. The PSJA aerodynamic model is developed based on the conservation equations of mass, momentum, energy, and the lumped capacitance method. The multichannel discharge model can simulate the multichannel discharge process and give the discharge energy in the plasma channel. With a constant heating efficiency, the time-independent heating energy deposition power in a discharge channel is obtained. Importing the heating energy, the PSJA aerodynamic model presents the evolution process of the jet. Simulation results show that the jet strength induced by a single actuator decreases with the number of actuators in the PSJA array. When the actuator number increases from 1 to 20, the weakening extent of mass ejected, peak jet velocity, and jet duration time is 62%, 54%, and 33%, respectively. The discharge efficiency increases with the actuator number, while the thermodynamic efficiency decreases with the actuator number. As a result, the total energy efficiency doesn’t always increase with an increase in the number of actuators. When the discharge efficiency of a conventional one channel discharge has been a relatively large value, the total energy efficiency actually decreases with the growth of actuator number.