Abstract

Power electronics are widely serving as core components of propulsion systems in electric vertical takeoff and landing (eVTOL) aircraft. Nevertheless, affected by the Paschen effect, the breakdown voltage of these electronics during flight is significantly lower than that on the ground, which could deteriorate system stability, or even cause aircraft faults or crashes, increasing safety risks for the public. As such, condition monitoring of power electronics has become critical for the safe operation of eVTOL. To achieve nonintrusive monitoring, acoustic emission (AE) sensors have gained traction with their prominent resistance to electromagnetic interference and high temperatures. However, existing studies yield conflicting results regarding whether an increase in loading voltage impacts the internal mechanical stress of electronics. To address this issue, we present this work to probe the existence and the pattern of a relationship between load voltage and stress wave, since the change of mechanical stress could generate acoustic waves that can be acquired by AE sensors. In this study, an AE sensor was applied onto an oscillator placed upon an epoxy substrate to characterize the acoustic waves emitted from the device under various input loads. In the experiment, we observed a unique AE signal pattern characterized by two distinct components that consistently intersect at a specific frequency. The time required to establish such intersections progressively lengthens as the load voltage increases. Through time-domain and frequency-domain analyses of the unique AE signals under different load voltages, it was discovered that certain features of the unique AE signals have an approximately linear relationship with the load voltage.

1 Introduction

Nowadays, electric vertical takeoff and landing (eVTOL) aircraft are extensively utilized in various fields such as disaster relief [1], environmental monitoring [2], oil/gas pipeline inspections [3], and agricultural spraying [4]. These missions often require eVTOL aircraft to engage in long-distance flights. The success of such missions is made possible by high-energy-density batteries and high-power electronic devices. However, power electronics also have some vulnerabilities. For instance, during high-altitude flights, the semiconductor components of power electronics are more susceptible to breakdown, leading to aircraft shutdowns or crashes. This is attributed to the impact of the Paschen effect, where the breakdown voltage of these basic components, transistors, is significantly lower at high altitudes than on the ground [5]. According to a report, semiconductor components contribute to a substantial majority of all failures related to integrated circuits (ICs), accounting for 20% of all power electronic failures [6]. Nevertheless, maintaining and repairing these ICs has proven to be both challenging and expensive [7]. Consequently, there has been an ongoing pursuit of research and development for methodologies to monitor the condition of ICs within these power electronics.

Conventional condition monitoring techniques for ICs have primarily focused on utilizing electrical, magnetic, and thermal parameters. Electromagnetic monitoring employs magnetic field disruption detection to identify abnormalities such as cracks, dents [8], and corrosion present within electrical components [9]. Prominent testing methodologies involving electromagnetic measurements encompass eddy-current testing [10], magnetic flux leakage analysis [11], and magnetic memory evaluation. Thermal-based condition monitoring is typically associated with high-voltage devices like transformers [12] and relies on thermocouples integrated within the transistors to detect potential failures. Nonetheless, none of these methods are universally accepted or extensively adopted as an effective means of condition monitoring [1315]. This is because electromagnetic data are susceptible to equipment degradation, resulting in low monitoring reliability. On the other hand, relying on thermal measurements makes it difficult to identify potential faults in power electronic devices proactively.

Different from electromagnetic and thermal-based monitoring, acoustic sensing can capture real-time subtle changes and signs of faults within power electronics. This function enhances the comprehensiveness of monitoring, contributing to an improved capability for early fault detection. More importantly, acoustic sensing represents a noninvasive method for monitoring power electronic devices. Unlike invasive techniques based on electromagnetic and thermal methods, it does not cause interference to an electronic system. In recent years, researchers have endeavored to assess the reliability of electronic devices through acoustic sensing. Karkkainen et al. [16] found that acoustic events can originate from the switching state of an insulated-gate bipolar transistor (IGBT), also referred to as a switching stress wave. Furthermore, they [17] identified acoustic signals generated by IGBT failures and discovered that different types of faults could lead to distinct signal patterns. Müller et al. [18] implemented the acoustic sensing technique to assess the state of health of power modules, their relationship was verified in an experiment with a 200 A/650 V three-phase inverter for automotive applications. Davari et al. [19] investigated whether it is possible to detect the degradation of semiconductor components through acoustic emissions (AEs) during the operation of an inductive switch. Silicon IGBT with antiparallel diodes is used to prove its feasibility.

Through literature review, it has been established that acoustic sensors can discern when transistors alternate between operating modes by measuring the induced mechanical stress on semiconductor devices [20]. However, the precise mechanism underlying the generation of mechanical stress remains uncertain. Moreover, conflicting results still exist in studies regarding whether an increase in load voltage affects mechanical stress [21,22]. Therefore, this paper aims to investigate how the load voltage of ICs affects the AE signals. The contribution can be summarized as follows: An acoustic monitoring system is established for power electronic ICs. Utilizing this system, we identified a unique acoustic emission behavior within an IC. Furthermore, we have experimentally explored the effects of load voltage variations on the time-domain and frequency-domain features of this unique AE behavior.

The rest of this paper is organized as follows. Section 2 introduces the acoustic monitoring system. Section 3 presents the experimental results and discussion. Finally, Sec. 4 concludes this work.

2 The Acoustic Monitoring System

To investigate AE behaviors within ICs, an AE monitoring system is designed to acquire acoustic signals at various load voltages. The structure of the developed system is illustrated in Fig. 1(a). In this system, an acoustic emission sensor is positioned on an IC. This sensor is linked to a preamplifier and an A/D converter to amplify the signal and convert it from analog to digital format. Subsequently, the digitized acoustic signal is transmitted to a computer for analysis. A trigger threshold above ambient noise level is predefined on the computer to initiate data collection. Through signal and correlation analysis performed on the computer, a unique AE behavior can be identified as they occur.

Fig. 1
The proposed integrated circuit monitoring system: (a) Schematic of the AE–electrical monitoring system and (b) The experiment setup in this paper
Fig. 1
The proposed integrated circuit monitoring system: (a) Schematic of the AE–electrical monitoring system and (b) The experiment setup in this paper
Close modal

In our experiments, the system configuration is shown in Fig. 1(b). The IC is tested under different voltage levels within its rated voltage (4.5–16 V). Detailed materials and testing methodologies are provided below. The testing circuit devised for this study is depicted in Fig. 2. It constitutes a multiharmonic oscillation circuit capable of generating square waves. The printed circuit board (PCB) was fabricated utilizing a Voltera V-One PCB printer, incorporating additive manufacturing techniques to generate the circuit comprising conductive silver tracks on a 53 mm × 83 mm silicon FR4 epoxy fiberglass substrate. Jumper cables are subsequently connected to a power supply, providing power to the circuit terminals. An light-emitting diode is positioned on the circuit, alongside an oscilloscope employed to assess the output voltage state.

Fig. 2
The multiharmonic oscillation circuit designed in this paper for square wave generation
Fig. 2
The multiharmonic oscillation circuit designed in this paper for square wave generation
Close modal
The IC under test is an NE555 oscillator, assembled to produce a 1.24 Hz square wave with a 60% duty cycle. The calculation formulas for frequency (F) and duty cycle (D) are provided as follows:
(1)
(2)

The frequency and duty cycle of a square wave are determined by the external resistors and capacitors. The external capacitor (C1) starts charging when the DC power is supplied to the chip. Pin 5 of the oscillator is unconnected as shown in Fig. 2. Therefore, when the voltage of C1 reaches 2/3 of the input voltage (Vcc), pin 6 is enabled and pin 3, i.e., the output pin, remains at a low level. Meanwhile, the voltage of pin 7 is pulled to a low level. C1 starts to discharge. When the voltage of C1 falls to 1/3 of Vcc, pin 2 is enabled and the output pin changes to a high level. Meanwhile, pin 7 is pulled to a high level, and a new charging cycle begins.

In the developed system, a Mistras R3a acoustic sensor is connected to a USB-AE Node that provides internal pre-amplification for wave sensing. The acoustic data are collected using mistrasaewin software. The trigger threshold for transient sensing is determined if noise was present and if it would play a role in characterization when the acoustic sensor is also placed upon the epoxy substrate, the resistors, and the capacitors.

3 Experiment and Discussion

The flowchart of AE generation and its capture is given in Fig. 3. AE can be generated when the material property changes in an IC, in which the changes result from the on/off of transistors. AE sensor converts the vibration velocity of acoustic into a voltage signal through its sensitive components. When the amplified voltage signal strength exceeds the set threshold, the acquisition system is triggered, initiating data recording until the stop condition is met. Such a recorded waveform is referred to as one hit. Before testing, a consistent 5 V voltage was applied to the positive terminal within the circuit to determine the status of the device. During the test, different load voltages are set to assess the impact of load voltage on AE signals. For each load voltage, hits with different time–frequency features have been detected. Within these hits, a unique signal pattern can be detected. As illustrated in Fig. 4, it presents an example of this unique signal pattern and its spectrogram. In this spectrogram, two prominent signal components originate at a higher frequency (around 450 kHz) and a lower frequency (approximately 300 kHz). Initially, the high-frequency initialized component holds a relatively high power. As time progresses, the power of this component gradually decreases. Conversely, the power of the low-frequency initialized component increases with a decreasing slope, eventually reaching an upper bound. Therefore, at the beginning of the time-domain signal, the high-power, high-frequency component predominates, resulting in greater voltage. Over time, as the power of the high-frequency initialized component wanes, these fluctuations diminish. The point at which the high and low-frequency initialized components intersect marks the period of least fluctuation in the time-domain signal. Subsequently, as the low-frequency initialized component's frequency and power increase, the magnitude of fluctuations in the time domain correspondingly rises. However, these signal components consistently intersect around a specific frequency. And this intersection frequency also remains constant with voltage. This AE components' intersection may relate to the on/off switching at the chip output, i.e., at the rising or falling edge of the output square wave. Although further research is needed to validate this explanation, our paper still represents the first observation and reporting of such acoustic emission behavior within an oscillator.

Fig. 3
The schematic of acoustic emission generation and capture
Fig. 3
The schematic of acoustic emission generation and capture
Close modal
Fig. 4
A raw acoustic signal and its spectrum. (a) The captured waveform in the time domain. (b) Spectrogram of the waveform. Two prominent signal components originate at a higher frequency and a lower frequency can be found intersecting at a specific frequency.
Fig. 4
A raw acoustic signal and its spectrum. (a) The captured waveform in the time domain. (b) Spectrogram of the waveform. Two prominent signal components originate at a higher frequency and a lower frequency can be found intersecting at a specific frequency.
Close modal

More interestingly, we observed that as the input voltage increases, the time required to establish this distinctive intersection also escalates, as shown in Fig. 5. In this figure, the unique spectrograms at four voltages including 5 V, 7.5 V, 10 V, and 12 V are provided. The spectrograms are obtained using the short-time Fourier transform with a window length of 128 and an overlap of 120. The intersection-establish-time at the first three voltages can be easily picked out from the spectrogram and are approximately 0.9025 ms, 2.207 ms, and 4.3631 ms, respectively. For the 12 V load voltage, however, its intersection-establish-time is estimated based on its spectrogram. This is because the longest waveform our AE system can capture consists of 7168 data points (approximately 7 ms at a sampling frequency of 1 MHz). However, intersection-establish-time at 12 V exceeds the maximum duration we can achieve. Although the precision of intersection-establish-time at 12 V is compromised, our observed inference still holds true: as the input voltage increases, the time required to establish this unique intersection extends.

Fig. 5
The left figure includes the spectrograms of AE signals at voltage 5 V, 7.5 V, 10 V, and 12 V. These spectrograms visualize the power levels in decibels (dB), as indicated by the adjacent color scale. The dots pinpoint the establishment timestamp of the intersection for the identified unique AE signal pattern. The bar and line graph on the right shows the relationship between load voltage and establishment time of the intersections. The increasing trend indicates that the intersection establishment time is positively related to the load voltage.
Fig. 5
The left figure includes the spectrograms of AE signals at voltage 5 V, 7.5 V, 10 V, and 12 V. These spectrograms visualize the power levels in decibels (dB), as indicated by the adjacent color scale. The dots pinpoint the establishment timestamp of the intersection for the identified unique AE signal pattern. The bar and line graph on the right shows the relationship between load voltage and establishment time of the intersections. The increasing trend indicates that the intersection establishment time is positively related to the load voltage.
Close modal

Except for the intersection, the time and frequency analysis are also conducted in this work to inquire about the impact of load voltage on the unique AE behavior. In this section, ten time-domain features and six frequency-domain features are extracted from the four raw signals [23]. Their calculation formulas are provided in Tables 1 and 2. In these tables, xi represent the raw time-domain signal, and s(k) represent its fast Fourier transform spectrum. For frequency-domain features, the frequency centroid

Table 1

Time-domain features

No.FormulaNameNo.FormulaName
T11ni=1n(xix¯)2Standard deviation (Std)T6max(|xi|)1ni=1n|xi|Pulse metric
T21ni=1n(xi)2Root‐mean‐square (RMS)T7max(|xi|)min(|xi|)Peak-to-peak
T3max(|xi|)MaxT8max(|xi|)1ni=1n(xi)2Peak metric
T4i=1n(xix¯)4n×(1ni=1n(xix¯)2)2KurtosisT91ni=1n(xi)21ni=1n|xi|Waveform metric
T5max(|xi|)(1ni=1n|xi|)2Margin metricT101ni=1n(xix¯)3(1ni=1n(xix¯)2)3Skewness
No.FormulaNameNo.FormulaName
T11ni=1n(xix¯)2Standard deviation (Std)T6max(|xi|)1ni=1n|xi|Pulse metric
T21ni=1n(xi)2Root‐mean‐square (RMS)T7max(|xi|)min(|xi|)Peak-to-peak
T3max(|xi|)MaxT8max(|xi|)1ni=1n(xi)2Peak metric
T4i=1n(xix¯)4n×(1ni=1n(xix¯)2)2KurtosisT91ni=1n(xi)21ni=1n|xi|Waveform metric
T5max(|xi|)(1ni=1n|xi|)2Margin metricT101ni=1n(xix¯)3(1ni=1n(xix¯)2)3Skewness
Table 2

Frequency-domain features

No.FormulaNameNo.FormulaName
F11Kk=1Ks(k)Spectrum meanF4k=1K(fkF7)3·s(k)(K1)·(F3)3None
F21K1k=1K(s(k)F1)2Spectrum RMSF5k=1K(fkF7)·s(k)(K1)·F3Std frequency
F31K1k=1K(fkF7)2·s(k)NoneF6k=1K(fkF7)4·s(k)(K1)·(F3)4None
No.FormulaNameNo.FormulaName
F11Kk=1Ks(k)Spectrum meanF4k=1K(fkF7)3·s(k)(K1)·(F3)3None
F21K1k=1K(s(k)F1)2Spectrum RMSF5k=1K(fkF7)·s(k)(K1)·F3Std frequency
F31K1k=1K(fkF7)2·s(k)NoneF6k=1K(fkF7)4·s(k)(K1)·(F3)4None

Note: s(k) is the spectrum of signal x(i), where k = 1, 2, …, K are the spectral lines, and fk is the frequency of the kth spectral line.

(3)
is used to obtain F3,F4,F5, and F6. Given the disparate ranges inherent in these features, a normalization process is applied to each feature type according to formula (4), enabling them to be displayed on one figure. These normalized time-domain and frequency-domain features are shown in Figs. 6 and 7, respectively.
(4)
Fig. 6
Normalized time-domain features of the identified unique acoustic emission signal at different load voltages. Features T1–T9 (including std, RMS, etc.) exhibit an upward trend as the voltage increases, while T10 shows a downward trend which indicates an increase in the skewness of the data distribution with rising voltage.
Fig. 6
Normalized time-domain features of the identified unique acoustic emission signal at different load voltages. Features T1–T9 (including std, RMS, etc.) exhibit an upward trend as the voltage increases, while T10 shows a downward trend which indicates an increase in the skewness of the data distribution with rising voltage.
Close modal
Fig. 7
Normalized frequency-domain features of the distinct acoustic emission at different load voltages. Most of the features show an overall positive trend, suggesting potential applications in state monitoring for ICs. Frequency centroid-associated feature F6 inversely correlates with load voltage, which might signify a more even dispersion of vibratory or noise sources within the IC at higher voltages.
Fig. 7
Normalized frequency-domain features of the distinct acoustic emission at different load voltages. Most of the features show an overall positive trend, suggesting potential applications in state monitoring for ICs. Frequency centroid-associated feature F6 inversely correlates with load voltage, which might signify a more even dispersion of vibratory or noise sources within the IC at higher voltages.
Close modal

The analysis of normalized time-domain features indicates a consistent pattern across most of the features. With the increase of load voltage, T1 to T9 all show a consistent pattern of variation, i.e., an approximately linear correlation with load voltage. It is noteworthy that the feature T10, which represents skewness, exhibits an opposite trend compared to all other time-domain features, decreasing as the voltage increases. For the four increasing voltage levels, the skewness data before normalization are −0.057, −0.123, −0.192, and −0.282, respectively. The skewness is consistently negative for all four AE signals, indicating a leftward skew of their data distribution as compared to a normal distribution. Additionally, there is also an increase in the skewness of the data distribution with the rise in voltage.

Turning to the frequency-domain features, similarly, the majority of frequency-domain features exhibit an increase with rising voltage. Among them, the normalized mean of the frequency spectrum, F1, displays an almost perfect linear relationship (R2=0.9951), which could suggest that the source of the acoustic emissions intensifies or becomes more active with an increase in load voltage. This distinct linear relationship could be utilized for IC state-of-health monitoring, diagnostics, or prediction. Features F2 to F5, despite minor variations at 5 V and 7.5 V, still show an overall positive trend. Such variations are reasonable, considering that each data point represents only a sample. Conducting large-scale AE sampling across multiple voltage values would enhance the regression accuracy of these features. Additionally, the feature F6, which is related to the frequency centroid, demonstrates a negative correlation with the load voltage. This indicates a transition of the signal spectrum's morphology from a sharp, heavy-tailed distribution to a flatter, lighter-tailed one. Physically, this could imply that as the voltage increases, the source of vibrations or noise within the IC becomes more uniformly dispersed, without a pronounced amplification at any specific frequency. This suggests that the system's dynamic response is growing increasingly stable.

4 Conclusion

To explore the mechanism underlying the generation of mechanical stress within ICs and understand how the load voltage of ICs affects the mechanical stress. An acoustic monitoring system is presented in this work for power electronics ICs. Using this system, sufficient experiments were conducted. Based on the experimental results, the following conclusions can be summarized:

  1. A distinct acoustic emission signal was observed during the generation of square waves from an oscillator. This unique AE signal contains two prominent components that intersect at some specific frequency. The time required to establish such an intersection extends as the input voltage increases. To the best of our knowledge, this paper represents the first report on the phenomenon.

  2. There is a clear correlation between load voltages applied to the IC and features of captured AE signals in both the time and frequency domain. This result provides a definitive answer to the question of whether the load voltage affects the mechanical stress of ICs. It also corroborates our hypothesis that different types of features may exhibit varying patterns of change with the increase of load voltage.

It should be noted that, currently, the correlations described in this paper have been validated only on oscillators. Researchers can extend this discovery to similar IC families that use transistors as primary components. This finding can also offer a suggestion to engineers who use AE sensors for IC monitoring, suggesting that an adjusted threshold may be necessary for failure warning under different voltage loads. Moreover, this work is carried out under resistive load conditions, which means that our circuit is just a simplified model. Consequently, it may not fully replicate the inductive, complex, or dynamic loads encountered by propulsion systems under real operational conditions. Therefore, when researchers working on similar topics should consider whether the load used allows for a linear response. Future work could focus on identifying the source of AE generation, which involves testing various resistive loads and capacitors to gauge their impact on acoustic emissions.

Funding Data

  • University of Arkansas (Chancellor's GAP Fund and Chancellor's Fund for Commercialization; Funder ID: 10.13039/100007756).

Data Availability Statement

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.

Footnotes

1

International Technical Conference and Exhibition on Packaging and Integration of Electronic and Photonic Microsystems, Doubletree by Hilton San Diego Mission. InterPACK2023.

References

1.
Doo
,
J.
,
2022
, “
Unsettled Issues Regarding the Use of eVTOL Aircraft During Natural Disasters
,”
SAE
Paper No. EPR2022001.10.4271/EPR2022001
2.
Vieira
,
D. R.
,
Silva
,
D.
, and
Bravo
,
A.
,
2019
, “
Electric VTOL Aircraft: The Future of Urban Air Mobility (Background, Advantages and Challenges)
,”
Int. J. Sustainable Aviat.
,
5
(
2
), pp.
101
118
.10.1504/IJSA.2019.101746
3.
Datta
,
A.
,
2021
, “
PEM Fuel Cell Model for Conceptual Design of Hydrogen eVTOL Aircraft
,” NASA Ames Research Center, Moffett Field, CA, Report No.
20210000284
.https://ntrs.nasa.gov/api/citations/20210000284/downloads/1502_Datta__CR%2020210000284_081821.pdf
4.
Okulski
,
M.
, and
Ławryńczuk
,
M.
,
2022
, “
A Small UAV Optimized for Efficient Long-Range and VTOL Missions: An Experimental Tandem-Wing Quadplane Drone
,”
Appl. Sci.
,
12
(
14
), p.
7059
.10.3390/app12147059
5.
Hu
,
J.
,
Zhao
,
X.
,
Ravi
,
L.
,
Burgos
,
R.
, and
Dong
,
D.
,
2021
, “
Enhanced Gate Driver Design for SiC-Based Generator Rectifier Unit for Airborne Applications
,” Conference Proceedings of the IEEE Applied Power Electronics Conference and Exposition (
APEC
), Phoenix, AZ, June 14–17, pp.
2526
2531
.10.1109/APEC42165.2021.9487425
6.
Shrivastava
,
V.
,
2022
, “
Condition Monitoring of Power Electronic Devices—A Review
,”
2022 2nd International Conference on Advance Computing and Innovative Technologies in Engineering
(
ICACITE
), Greater Noida, India, Apr. 28–29,
pp.
749
753
.10.1109/ICACITE53722.2022.9823555
7.
Manohar
,
S. S.
,
Sahoo
,
A.
,
Subramaniam
,
A.
, and
Panda
,
S. K.
,
2017
, “
Condition Monitoring of Power Electronic Converters in Power Plants—A Review
,” 2017 20th International Conference on Electrical Machines and Systems (
ICEMS
), Sydney, NSW, Australia, Aug. 11–14,
pp.
1
5
.10.1109/ICEMS.2017.8056371
8.
Zheng
,
W.
, and
Li
,
J.
,
2024
, “
Method for Detecting and Locating Cracks in Aluminum Plates Based on Balanced Field Electromagnetic Technique
,”
Measurement
,
229
, p.
114505
.10.1016/j.measurement.2024.114505
9.
Kang
,
X.
,
Xiong
,
Z.
,
Zhang
,
L.
,
He
,
R.
,
Meng
,
X.
,
Chen
,
J.
, and
Zhang
,
X.
,
2021
, “
Simulation and Experimental Research on Magnetic Flux Leakage Detection Method of Long-Distance Pipeline Local Dents Stress
,”
J. Phys.: Conf. Ser.
,
1894
, p.
012088
.10.1088/1742-6596/1894/1/012088
10.
Lu
,
C.
,
Li
,
L.
,
Liu
,
Z.
,
Xu
,
C.
,
Xin
,
M.
,
Fu
,
G.
,
Wang
,
T.
, and
Wang
,
X.
,
2022
, “
Location and Corrosion Detection of Tower Grounding Conductors Based on Electromagnetic Measurement
,”
Measurement
,
199
, p.
111469
.10.1016/j.measurement.2022.111469
11.
Theodoulidis
,
T. P.
,
Tsiboukis
,
T. D.
, and
Kriezis
,
E. E.
,
1995
, “
Analytical Solutions in Eddy Current Testing of Layered Metals With Continuous Conductivity Profiles
,”
IEEE Trans. Magn.
,
31
(
3
), pp.
2254
2260
.10.1109/20.376236
12.
Sulaiman
,
E.
, and
Zakaria
,
S.
,
2015
, “
Magnetic Flux Analysis of a New e-Core HEFSM With Various Slot-Pole Combinations for HEV
,” 2015 IEEE International Magnetics Conference (
INTERMAG
), Beijing, China, May 11–15,
p.
1
.10.1109/INTMAG.2015.7156703
13.
Gutten
,
M.
,
Kucera
,
M.
,
Cefer
,
V.
,
Brncal
,
P.
, and
Jarina
,
R.
,
2020
, “
Analysis of Transformers by Acoustic Emission
,” 2020 IEEE 61th International Scientific Conference on Power and Electrical Engineering of Riga Technical University (
RTUCON
), Riga, Latvia, Nov. 5–7,
pp.
1
4
.10.1109/RTUCON51174.2020.9316564
14.
Li
,
C.
,
Guo
,
L.
,
Gao
,
H.
, and
Li
,
Y.
,
2021
, “
Similarity-Measured Isolation Forest: Anomaly Detection Method for Machine Monitoring Data
,”
IEEE Trans. Instrum. Meas.
,
70
, pp.
1
12
.10.1109/TIM.2021.3062684
15.
Yang
,
S.
,
Xiang
,
D.
,
Bryant
,
A.
,
Mawby
,
P.
,
Ran
,
L.
, and
Tavner
,
P.
,
2010
, “
Condition Monitoring for Device Reliability in Power Electronic Converters: A Review
,”
IEEE Trans. Power Electron.
,
25
(
11
), pp.
2734
2752
.10.1109/TPEL.2010.2049377
16.
Li
,
C.
,
Guo
,
L.
,
Lei
,
Y.
,
Gao
,
H.
, and
Zio
,
E.
,
2023
, “
A Signal Segmentation Method for CFRP/CFRP Stacks Drilling-Countersinking Monitoring
,”
Mech. Syst. Signal Process.
,
196
, p.
110332
.10.1016/j.ymssp.2023.110332
17.
Karkkainen
,
T. J.
,
Talvitie
,
J. P.
,
Kuisma
,
M.
,
Hannonen
,
J.
,
Strom
,
J.-P.
,
Mengotti
,
E.
, and
Silventoinen
,
P.
,
2014
, “
Acoustic Emission in Power Semiconductor Modules—First Observations
,”
IEEE Trans. Power Electron.
,
29
(
11
), pp.
6081
6086
.10.1109/TPEL.2013.2295460
18.
Karkkainen
,
T. J.
,
Talvitie
,
J. P.
,
Kuisma
,
M.
,
Silventoinen
,
P.
, and
Mengotti
,
E.
,
2015
, “
Acoustic Emission Caused by the Failure of a Power Transistor
,” Conference Proceedings of the IEEE Applied Power Electronics Conference and Exposition (
APEC
), Charlotte, NC, Mar. 15–19,
pp.
2481
2484
.10.1109/APEC.2015.7104697
19.
Müller
,
S.
,
Drechsler
,
C.
,
Heinkel
,
U.
, and
Herold
,
C.
,
2016
, “
Acoustic Emission for State-of-Health Determination in Power Modules
,” 13th International Multi-Conference on Systems, Signals and Devices (
SSD 2016
), Leipzig, Germany, Mar. 21–24, pp.
468
471
.10.1109/SSD.2016.7473704
20.
Davari
,
P.
,
Kristensen
,
O.
, and
Iannuzzo
,
F.
,
2018
, “
Investigation of Acoustic Emission as a Non-Invasive Method for Detection of Power Semiconductor Aging
,”
Microelectron. Reliab.
,
88–90
, pp.
545
549
.10.1016/j.microrel.2018.06.074
21.
Brauhn
,
T. J.
,
Sheng
,
M.
,
Dow
,
B. A.
,
Nogawa
,
H.
, and
Lorenz
,
R. D.
,
2016
, “
Module-Integrated GMR-Based Current Sensing for Closed-Loop Control of a Motor Drive
,”
IEEE Trans. Ind. Appl.
,
53
(
1
), pp.
222
231
.10.1109/TIA.2016.2614771
22.
Zou
,
X.
,
He
,
Y.
,
Zhang
,
Z.
,
Li
,
M.
,
She
,
S.
,
Geng
,
X.
,
Bai
,
Y.
,
Dang
,
X.
,
Ren
,
D.
, and
Chen
,
Z.
,
2021
, “
Experimental Study and Signal Analysis of Acoustic Emission From Power MOSFET
,”
Microelectron. Reliab.
,
127
, p.
114411
.10.1016/j.microrel.2021.114411
23.
Liu
,
F.
,
Zeng
,
C.
,
Cheng
,
L.
,
He
,
Y.
,
Bai
,
Y.
,
Geng
,
X.
,
Liu
,
S.
,
Ren
,
D.
, and
Wang
,
L.
,
2022
, “
Condition Monitoring of Inverter Power Devices Based on Electromagnetic Acoustic Emissions
,”
J. Power Electron.
,
22
(
12
), pp.
2122
2135
.10.1007/s43236-022-00502-1
24.
Lei
,
Y.
,
2016
,
Intelligent Fault Diagnosis and Remaining Useful Life Prediction of Rotating Machinery
, Butterworth-Heinemann, Oxford, UK.