In my work on online insulation monitoring for electric cars, I treat partial discharge as both a diagnostic signal and a system-level design challenge. An electric car is not a conventional power apparatus placed on wheels; it is a dense, fast-switching, vibration-rich, thermally cycling electromagnetic environment in which an incipient insulation defect may produce only a faint radio-frequency pulse. My objective is therefore to design a compact ultra-high-frequency sensor that can capture such pulses inside an electric car, reject the strong switching and communication noise of the vehicle, and support a practical online monitoring strategy without interfering with traction operation.
I begin from the observation that partial discharge in an electric car high-voltage system is not merely a laboratory phenomenon. It can appear in motor winding insulation, high-voltage connectors, busbar supports, inverter terminals, cable joints, and battery-pack high-voltage feedthroughs. Because an electric car repeatedly experiences rapid voltage transitions, thermal gradients, mechanical shock, and humid contamination, the probability that a small void or surface defect evolves into a serious insulation fault is significant. The challenge is that the same electric car generates abundant electromagnetic disturbances from power converters, gate drivers, common-mode currents, wireless modules, relays, and electric machines. A useful sensor must therefore be sensitive to weak discharge pulses while remaining robust in the presence of electric car noise.
1. Motivation and Design Scope
I frame the problem around three requirements specific to an electric car. First, the sensor must be physically small enough to fit near a motor end winding, a high-voltage connector, or a busbar enclosure. Second, it must operate over a wide band so that it can capture the fast radio-frequency radiation produced by partial discharge. Third, it must suppress electric car interference without requiring a heavy shielded room or a laboratory-grade spectrum analyzer. These requirements lead me to a planar UHF antenna with a matched front end, a differential output, a shielded package, and a signal-processing chain that can operate in an electric car.
In my design targets, I do not optimize only for sensitivity. I optimize for repeatability inside an electric car. A sensor that is extremely sensitive in a shielded laboratory but saturated by inverter noise in an electric car is not useful. Conversely, a sensor that is robust but cannot detect low-level discharge is also insufficient. I therefore define a balanced target set that includes bandwidth, return loss, voltage standing wave ratio, noise figure, size, shielding effectiveness, and compatibility with the electric car harness.
| Design requirement | Target for an electric car | Reason |
|---|---|---|
| Operating band | 300 MHz to 1.5 GHz | Captures fast discharge radiation while avoiding most low-frequency converter noise and high-frequency wireless bands |
| Maximum planar size | 50 mm by 50 mm | Fits near motor end windings and high-voltage connector housings in an electric car |
| Voltage standing wave ratio | Below 2.0 in the operating band | Maintains efficient energy transfer from the antenna to the receiver |
| Return loss | Above 10 dB | Reduces reflection loss and improves matching stability |
| Preamplifier noise figure | Below 2 dB | Preserves weak discharge pulses in the electric car noise floor |
| Out-of-band rejection | Above 30 dB | Rejects electric car switching and communication interference |
| Minimum detectable charge | About 10 pC | Provides early warning for electric car insulation defects |
| Output format | Differential | Improves common-mode rejection in the electric car chassis environment |
The phrase electric car appears throughout my design logic because the environment changes the sensor specification. In a laboratory, a large horn antenna may be acceptable. In an electric car, the sensor must be embedded, shielded, and connected to a low-voltage acquisition unit that shares space with traction electronics. The electric car imposes mechanical, thermal, and electromagnetic constraints that shape every part of the design.
2. Partial Discharge Physics in Electric Cars
Partial discharge is a localized charge redistribution that occurs when the local electric field exceeds the breakdown strength of a small gas-filled void, a surface region, or an interface defect. In an electric car, such defects may originate from manufacturing voids, incomplete impregnation, thermal aging, mechanical abrasion, or contamination. I describe the inception condition as a local-field criterion, because the average field in the insulation system is not sufficient to predict discharge in an electric car.
$$E_{\mathrm{loc}} = \beta E_{\mathrm{avg}}, \qquad E_{\mathrm{loc}} \geq E_{\mathrm{inc}}$$
Here, the local field enhancement factor accounts for geometric irregularities, void shape, and material interfaces. For an electric car, the local field is not static. The inverter applies pulse-width-modulated voltage with fast edges, so the field contains a switched component. I represent the applied voltage as a sequence of transitions with rise time and overshoot, and I track the resulting field stress at a defect.
$$v(t) = V_{\mathrm{dc}} + \sum_{k} \Delta V_k \left[ u(t-t_k) – u(t-t_k-\tau_r) \right] + v_{\mathrm{ring}}(t)$$
The voltage change rate is a key electric car stress parameter because fast switching can excite field transients inside insulation. I use the following metric when comparing electric car operating conditions.
$$\left( \frac{dv}{dt} \right)_{\max} = \max_{t} \left| \frac{dv(t)}{dt} \right|$$
When a discharge occurs, the current pulse is short. I model the pulse as a damped oscillation so that I can connect its time-domain shape to the radio-frequency spectrum received by the sensor.
$$i_{\mathrm{pd}}(t) = I_0 e^{-t/\tau_p} \sin(2\pi f_p t) u(t)$$
The Fourier transform of this pulse has a broad envelope. For a simple exponential decay approximation, the magnitude falls with frequency according to the time constant of the pulse.
$$\left| I_{\mathrm{pd}}(f) \right| = \frac{I_0 \tau_p}{\sqrt{1+\left(2\pi f \tau_p\right)^2}}$$
This relationship explains why partial discharge in an electric car can be observed at hundreds of megahertz and beyond. A faster pulse has a shorter time constant and a wider spectrum. Because the discharge source is small and the pulse is fast, the radiation field can be approximated from the time derivative of the effective dipole moment. I use this view to justify a UHF sensor rather than a purely low-frequency current transformer.
$$E_{\mathrm{rad}}(r,t) \approx \frac{\mu_0}{4\pi r} \frac{d^2 p(t-r/c)}{dt^2} \sin\theta$$
The geometry of an electric car modifies the propagation. Motor housings, battery enclosures, inverter shells, and cable shields create reflections, shadowing, and cavity resonances. I cannot assume free-space propagation from a discharge site to a sensor. Instead, I treat the electric car as a reverberant electromagnetic structure in which some frequencies are enhanced and others are attenuated. This motivates a wideband sensor and a multi-position placement strategy.
| Defect type in an electric car | Typical location | Discharge character | Implication for UHF sensing |
|---|---|---|---|
| Void in motor winding insulation | End winding, slot section, impregnated winding bundle | Short repetitive pulses, often phase-related to the applied switching waveform | Requires close placement and wide bandwidth |
| Surface discharge on a connector | High-voltage connector interface, terminal boot | Irregular pulse bursts influenced by contamination and humidity | Requires robust interference classification |
| Gap defect in a busbar support | Inverter busbar, battery-pack feedthrough | Low-amplitude pulses that may be masked by switching noise | Requires low noise figure and differential output |
| Cable-joint void | High-voltage cable joint or shield termination | Pulse train with reflection-dependent spectrum | Requires multi-position correlation |
| Tracking along a contaminated surface | Terminal block, enclosure feedthrough | Intermittent bursts with strong temporal clustering | Requires event capture and time-frequency features |
An electric car also introduces non-sinusoidal voltage stress. Wide-bandgap devices can produce voltage transitions with rise times in the nanosecond range and significant overshoot. I express the overshoot factor as the ratio of peak voltage to steady-state voltage, and I use it as one indicator of electric car insulation stress.
$$k_{\mathrm{ov}} = \frac{V_{\mathrm{peak}}}{V_{\mathrm{dc}}}$$
In my experiments and analysis, I do not attempt to reproduce every electric car operating condition. Instead, I isolate the dominant mechanisms: fast voltage edges, repetitive switching, common-mode radiation, wireless communication, and mechanical/thermal aging. This allows me to test the sensor against representative electric car disturbances while keeping the experimental variables controllable.
3. Compact UHF Sensor Design for Electric Cars
I selected a planar microstrip patch topology because it can be made compact, integrated with a ground plane, and packaged inside an electric car component without protruding into the high-voltage clearance area. To reduce the footprint, I use meandered line loading and slot coupling. The meander increases the effective electrical length without increasing the physical length, while the slot coupling provides an additional degree of freedom for impedance matching.
The substrate material must be stable in an electric car. I choose a low-loss polytetrafluoroethylene-based dielectric with a stable relative permittivity and low high-frequency loss. The ground plane and radiating patch are copper. The package includes a metal shield with conductive seams to limit direct coupling from electric car harnesses and power modules.
| Sensor parameter | Value or range | Function in the electric car sensor |
|---|---|---|
| Patch footprint | Within 50 mm by 50 mm | Compact installation near electric car high-voltage components |
| Substrate | Low-loss PTFE-based dielectric | Stable high-frequency performance under thermal cycling |
| Radiating element | Meandered microstrip patch | Size reduction while preserving UHF response |
| Coupling structure | Slot-coupled feed region | Impedance matching and bandwidth adjustment |
| Feed line | Graded microstrip line | Smooth transition and reduced reflection |
| Front-end amplifier | Low-noise amplifier with noise figure below 2 dB | Improves weak discharge detection in an electric car |
| Output | Differential | Common-mode rejection in the electric car chassis |
| Housing | Shielded metal enclosure with conductive seams | Reduces direct electric car electromagnetic coupling |
I evaluate the antenna matching using the reflection coefficient and the voltage standing wave ratio. These quantities are simple, but they are important because an electric car sensor must remain matched when it is installed near metal and dielectric materials.
$$\Gamma = \frac{Z_L – Z_0}{Z_L + Z_0}, \qquad \mathrm{VSWR} = \frac{1+|\Gamma|}{1-|\Gamma|}$$
$$\mathrm{RL(dB)} = -20\log_{10}|\Gamma|$$
My target is a voltage standing wave ratio below 2.0 and a return loss above 10 dB in the 300 MHz to 1.5 GHz band. In the optimized design, I adjust the patch dimensions, the meander spacing, the slot length, and the feed position. The design is not a single-frequency resonator; it is a multi-resonant structure in which several local resonances combine to produce a broad response suitable for electric car discharge signals.
The low-noise amplifier is part of the sensor, not an optional accessory. In an electric car, cable losses and connector losses can reduce the signal before it reaches the acquisition unit. Placing gain immediately after the antenna improves the overall noise figure. I use the Friis formula to estimate the cascaded noise performance.
$$F_{\mathrm{total}} = F_1 + \frac{F_2-1}{G_1} + \frac{F_3-1}{G_1 G_2} + \cdots$$
Because the first stage dominates the cascaded noise figure, I specify a low-noise first amplifier and keep the interconnect between the antenna and the amplifier short. This is especially important in an electric car, where the sensor may be mounted in a confined space and connected through a short pigtail. I also include a band-pass filter after the first gain stage so that strong out-of-band electric car signals do not saturate the receiver.
The multi-resonant behavior can be described by a sum of modal contributions. I use this representation to reason about how the meander and slot modify the input impedance.
$$Z_{\mathrm{in}}(f) = \sum_{m=1}^{M} \frac{R_m}{1+jQ_m\left(\frac{f}{f_m}-\frac{f_m}{f}\right)}$$
In this expression, each mode has a resonant frequency, a quality factor, and a radiation resistance. A high quality factor improves selectivity but narrows the bandwidth. For an electric car, I prefer moderate quality factors because the discharge spectrum is broad and the propagation environment is unpredictable.
4. Anti-Interference Architecture for Electric Car Environments
The electric car is an interference-rich environment. I design the anti-interference architecture in layers so that no single measure has to solve the entire problem. The first layer is frequency selection. The 300 MHz to 1.5 GHz band avoids the strongest low-frequency switching components and the common 2.4 GHz wireless band. The second layer is the front-end band-pass filter. The third layer is differential output and common-mode rejection. The fourth layer is physical shielding. The fifth layer is adaptive noise cancellation in the digital domain.
| Layer | Measure | Electric car disturbance addressed | Expected benefit |
|---|---|---|---|
| Frequency layer | 300 MHz to 1.5 GHz operating band | Low-frequency inverter noise and high-frequency wireless signals | Reduces the initial interference power reaching the receiver |
| Filter layer | Band-pass filter matched to the sensor | Out-of-band switching harmonics and communication channels | Improves out-of-band rejection |
| Output layer | Differential signal path | Common-mode currents on the electric car chassis | Improves common-mode rejection |
| Package layer | Shielded enclosure with conductive seams | Direct field coupling from electric car power cables | Reduces parasitic pickup |
| Signal layer | Adaptive noise cancellation | Nonstationary electric car background noise | Improves detection SNR |
| Decision layer | Interference feature library | Relay clicks, electrostatic events, converter switching | Reduces false alarms |
I model the received signal as the sum of the discharge signal, background noise, and deterministic electric car interference. The sensor output is processed to estimate the discharge component.
$$x(t) = s_{\mathrm{pd}}(t) + n_{\mathrm{bg}}(t) + \sum_{k} a_k q_k(t)$$
Here, the last term represents structured interference from electric car sources. A reference channel measures a correlated version of the background noise. I use an adaptive filter to estimate the interference and subtract it from the main channel.
$$y(t) = x(t) – \mathbf{w}^{T}\mathbf{n}(t)$$
I update the filter weights with a normalized least-mean-square rule. The step size is chosen to balance convergence speed and stability in the electric car environment.
$$\mathbf{w}(t+1) = \mathbf{w}(t) + \mu \frac{e(t)\mathbf{n}(t)}{\|\mathbf{n}(t)\|^2+\epsilon}$$
The residual error is used to adapt the canceller. When the electric car background changes, the filter tracks the change. When a discharge pulse occurs, the adaptation is slow enough that the pulse is not canceled. This distinction is important because I do not want the algorithm to learn away the very event I am trying to detect.
I also define a detection threshold based on the background statistics. A simple fixed threshold is not adequate in an electric car because the noise level changes with operating point. I use a robust estimate of the median and the median absolute deviation.
$$T = \mu_n + k\sigma_n, \qquad \sigma_n \approx 1.4826 \cdot \mathrm{MAD}$$
The factor 1.4826 converts the median absolute deviation to a consistent estimate of the standard deviation for Gaussian noise. For electric car signals, the distribution is not perfectly Gaussian, so I also use a percentile-based threshold and a pulse-width criterion. This reduces false triggers from isolated electric car switching spikes.
5. Placement Strategy in Electric Cars
The placement of the sensor is as important as the sensor itself. In an electric car, the discharge source may be inside a motor housing, inside a connector, or behind a metal enclosure. I propose a multi-position strategy so that the monitoring system does not depend on a single line of sight. The first position is near the motor end winding, where winding defects are common. The second position is near the high-voltage connector interface, where surface and gap defects can occur. The third position is near the inverter DC-bus entry, where busbar and capacitor insulation stress can be monitored.
| Placement position in an electric car | Primary monitoring target | Advantages | Practical constraints |
|---|---|---|---|
| Motor end-winding region | Winding insulation voids and surface discharge | Short propagation path and strong coupling to discharge radiation | High temperature, vibration, and limited space |
| High-voltage connector interface | Connector gap defects and contamination-induced discharge | Close to a known electric car weak point | Requires safe clearance and robust mounting |
| Inverter DC-bus entry | Busbar support and DC-link insulation defects | Captures discharge near the switching power stage | Strong electric car switching noise must be rejected |
| Battery-pack high-voltage feedthrough | Feedthrough and pack insulation defects | Provides coverage of the battery side of the electric car | Shielding and service access must be considered |
I keep a minimum distance between the sensor and exposed high-voltage conductors. In my design guidelines, the recommended separation is at least 20 mm, but the final value must be verified against the electric car safety and creepage requirements. The sensor is packaged so that it can be attached with an adhesive, a clip, or an embedded bracket. The mounting method must not create a path for water ingress or mechanical wear in the electric car.
Metal structures in an electric car can act as cavities. I estimate the cavity resonance frequencies to understand which bands may be enhanced or attenuated.
$$f_{mnp} = \frac{c}{2\sqrt{\varepsilon_r}} \sqrt{\left(\frac{m}{a}\right)^2+\left(\frac{n}{b}\right)^2+\left(\frac{p}{d}\right)^2}$$
This formula is a reminder that an electric car enclosure is not electrically transparent. A sensor placed at one point may see a strong response at a frequency where the enclosure resonates, while another point may see a null. Multi-position sensing reduces this risk because it samples different field distributions. I combine the outputs using a weighted fusion rule so that a discharge event can be detected even if one sensor is in a null.
$$S_{\mathrm{fused}}(t) = \sum_{i=1}^{N} w_i y_i(t), \qquad \sum_{i=1}^{N} w_i = 1$$
The weights can be fixed based on installation geometry, or they can be adapted based on the estimated noise level at each sensor. In an electric car, the noise level changes with traction power, so adaptive weighting is attractive. However, I keep the fusion logic simple enough for an embedded controller. The goal is reliable detection, not unnecessary complexity.
6. Signal Acquisition and Feature Extraction
I use a hybrid acquisition mode. In standby, the system monitors the signal energy at a low sampling rate and low power. When the energy exceeds a threshold, the system switches to high-speed acquisition and stores the waveform. This is appropriate for an electric car because discharge events are intermittent and the monitoring system must not drain the low-voltage battery.
$$E_{\mathrm{win}} = \int_{t_0}^{t_0+T_w} \left| x(t) \right|^2 dt$$
The trigger condition is based on the windowed energy and the pulse shape. A narrow electric car switching spike may pass the energy threshold, but it can be rejected by a pulse-width criterion. A genuine discharge pulse typically has a fast rise and a short duration, followed by a decay determined by the sensor and propagation path.
After acquisition, I use time-frequency analysis to extract features. The short-time Fourier transform provides a time-frequency map of the discharge event.
$$X(\tau,\omega) = \int_{-\infty}^{\infty} x(t) w(t-\tau) e^{-j\omega t} dt$$
I also use empirical mode decomposition to separate the signal into intrinsic mode functions. This helps me analyze nonstationary electric car discharge signals that do not fit a single frequency template.
$$x(t) = \sum_{i=1}^{K} c_i(t) + r(t)$$
From the decomposed signal, I extract amplitude, duration, pulse interval, spectral centroid, bandwidth, and repetition statistics. These features are used to distinguish discharge from electric car interference and to estimate discharge severity.
| Feature | Definition or interpretation | Use in an electric car monitoring system |
|---|---|---|
| Peak amplitude | Maximum absolute value of the pulse | Estimates discharge strength and compares with reference measurements |
| Pulse duration | Time above a fraction of the peak | Separates fast discharge from slower electric car switching transients |
| Rise time | Time from low threshold to high threshold | Indicates the frequency content of the discharge pulse |
| Pulse interval | Time between successive pulses | Identifies repetitive discharge patterns in the electric car |
| Spectral centroid | Energy-weighted mean frequency | Distinguishes discharge from narrowband electric car communication signals |
| Bandwidth | Spread of the spectrum | Separates broadband discharge from narrowband interference |
| Repetition rate | Number of pulses per unit time | Tracks discharge activity in an electric car over time |
| Phase or switching correlation | Correlation with the electric car switching period | Separates converter-synchronous noise from discharge |
| Autocorrelation peak | Periodic component strength | Identifies deterministic electric car interference |
I build an interference feature library for the electric car. Local discharge often appears as a quasi-periodic pulse train with a relatively regular interval. Switching noise is usually synchronized with the converter switching frequency and has a fixed repetition pattern. Electrostatic discharge appears as an isolated, large-amplitude single pulse. Relay action produces a mechanical transient with a different time-frequency signature. By comparing these features, I can reduce false alarms without suppressing genuine discharge.
| Event type in an electric car | Time-domain pattern | Frequency-domain pattern | Discrimination rule |
|---|---|---|---|
| Partial discharge | Quasi-periodic or clustered short pulses | Broadband burst with varying amplitude | Combine pulse interval statistics and spectral spread |
| Inverter switching noise | Periodic spikes synchronized with switching | Discrete harmonics and fixed repetition frequency | Use switching-period correlation and notch-like rejection |
| Wireless communication | Longer modulated packets | Narrowband or packet-like spectrum | Reject known communication bands and packet durations |
| Electrostatic discharge | Isolated large pulse | Very broad but non-repetitive spectrum | Require clustering or repeatability before declaring a discharge trend |
| Relay or contactor action | Mechanical bounce and multiple pulses | Lower-frequency transient components | Use pulse shape and event correlation with control commands |
7. Experimental Validation Platform
I validate the sensor in a shielded laboratory first, then in a representative electric car test environment. The laboratory platform includes an adjustable high-voltage source, discharge defect models, a reference pulse-current measurement system, and the UHF sensor under test. The reference system follows the charge-based partial discharge measurement principle described in IEC 60270. I use the reference measurement to calibrate the apparent charge and to compare the UHF response with a standard quantity.
| Equipment or model | Purpose | Range or specification |
|---|---|---|
| Adjustable high-voltage source | Applies test voltage to defect models | 0 kV to 20 kV |
| Needle-plane defect model | Generates corona-like discharge | Adjustable gap and electrode geometry |
| Void defect model | Represents an electric car insulation void | Controlled air gap in a dielectric sample |
| Surface discharge model | Represents contaminated connector surfaces | Controllable contamination and humidity |
| Reference pulse-current system | Measures apparent charge | Calibrated according to IEC 60270 |
| UHF sensor under test | Captures radio-frequency radiation | 300 MHz to 1.5 GHz |
| Shielded enclosure | Provides low background noise | Background below -85 dB |
| Motor test bench | Represents an electric car traction machine | Speed and torque controlled |
The apparent charge is obtained from the reference measurement. I use the following relationship to compare the UHF response with the charge level.
$$q_{\mathrm{app}} = \int i_{\mathrm{ref}}(t) dt$$
For each discharge level, I record the reference charge and the sensor output. I repeat the measurement multiple times and compute the mean and standard deviation. The linearity is evaluated using a correlation coefficient.
$$r = \frac{\sum_{i=1}^{n}(x_i-\bar{x})(y_i-\bar{y})}{\sqrt{\sum_{i=1}^{n}(x_i-\bar{x})^2}\sqrt{\sum_{i=1}^{n}(y_i-\bar{y})^2}}$$
I also define the signal-to-noise ratio for the sensor output. The signal amplitude is taken from the peak of the discharge pulse, and the noise is taken from a discharge-free window.
$$\mathrm{SNR(dB)} = 20\log_{10}\left(\frac{V_{\mathrm{signal,rms}}}{V_{\mathrm{noise,rms}}}\right)$$
8. Results and Discussion
The measured sensitivity shows that the sensor can detect discharge at the 10 pC level while maintaining a useful signal-to-noise ratio. As the discharge level increases, the output voltage increases in a approximately linear manner. The correlation coefficient exceeds 0.96, which indicates that the sensor response is consistent with the reference charge measurement.
| Apparent charge (pC) | Mean sensor output (mV) | Measured SNR (dB) | Observation in the electric car test context |
|---|---|---|---|
| 10 | 18 | 6.5 | Detectable above the background with filtering and differential output |
| 20 | 35 | 10.2 | Clear pulse above the electric car noise floor |
| 50 | 84 | 15.1 | Strong response suitable for trend monitoring |
| 100 | 166 | 20.4 | High-confidence detection in an electric car monitoring scenario |
I also test the anti-interference performance by injecting representative electric car disturbances. The interference sources include inverter switching noise, wireless communication signals, and common-mode radiation from a motor controller. After band-pass filtering and differential processing, the interference components are suppressed, and the discharge pulses remain identifiable.
| Interference scenario | Without mitigation | With the proposed electric car sensor | Improvement |
|---|---|---|---|
| Inverter switching noise | Discharge pulse partly masked | Pulse identifiable after filtering | Out-of-band rejection above 30 dB |
| Wireless communication | False triggers possible | Communication packets rejected | Reduced false alarm rate |
| Common-mode controller radiation | Strong background fluctuation | Differential output reduces common-mode component | SNR improvement of about 12 dB |
| Combined electric car noise | Low confidence detection | Discharge trend remains visible | Improved detection reliability |
The bench test in an electric car motor environment provides an initial check of practical behavior. I mount the sensor outside the motor housing near the end-winding region. The motor runs at a rated operating point with controlled speed and torque. I first record the background for 30 minutes, then introduce a small insulation defect at the winding end, and then record again for the same duration. In the normal state, the sensor output is dominated by background noise and periodic electric car switching components. After the defect is introduced, the sensor captures abnormal radio-frequency pulses. The pulse occurrence rate and amplitude trend correlate with the reference detection results.
I interpret this result carefully. The bench test is not a full electric car road test, but it captures the essential electric car conditions: a rotating machine, a power converter, a metal enclosure, and a noisy electromagnetic background. The sensor does not need to identify the defect type in this test. It needs to show that an electric car monitoring system can detect a change in discharge activity and can distinguish that change from normal operation.
| Test condition | Observed sensor behavior | Reference behavior | Interpretation |
|---|---|---|---|
| Normal electric car motor operation | Background noise and periodic switching components | No significant discharge activity | Baseline is stable after warm-up |
| Introduced winding defect | Intermittent radio-frequency pulses | Increased discharge activity | Sensor captures the fault-related trend |
| Variable load | Background level changes but pulses remain separable | Discharge trend remains visible | Adaptive threshold is necessary in an electric car |
| Long-duration monitoring | No saturation or thermal drift observed | Reference remains consistent | Design is suitable for further electric car integration |
The main limitation is that the electric car propagation channel is complex and time-varying. A single sensor may fall into a null at a particular frequency or position. Multi-position fusion addresses this, but it also increases cost and wiring complexity. I therefore propose a staged integration: first, a single sensor at the most critical electric car location; second, a second sensor at a connector or busbar location; third, a full multi-sensor network with centralized feature fusion. This staged approach matches the practical development cycle of an electric car platform.
9. Engineering Implications for Electric Cars
From an engineering perspective, the sensor must be integrated into the electric car electrical and electronic architecture. This means more than attaching an antenna. The sensor needs a low-voltage supply, a communication interface, self-test capability, and a calibration path. It also needs to be safe under high-voltage fault conditions. I treat the sensor as a low-voltage monitoring device that is galvanically isolated from the high-voltage system and placed at a safe distance from exposed conductors.
| Integration item | Requirement for an electric car | Design response |
|---|---|---|
| Power supply | Low-voltage DC supply from the electric car harness | Filtered supply with transient protection |
| Communication | Automotive network compatibility | Digital interface with event summaries and raw waveform export |
| Self-test | Verification without removing the sensor | Injected test pulse and gain check |
| Calibration | Traceable sensitivity check | Reference pulse injection at the front end |
| Safety | Maintain electric car creepage and clearance | Insulated housing and minimum separation distance |
| Thermal robustness | Operation near electric car motor and inverter heat | Stable substrate and shielded package |
| Vibration robustness | Survival under electric car road and machine vibration | Clamped or embedded mounting |
| Electromagnetic compatibility | Coexistence with electric car power electronics | Filtering, shielding, and differential output |
I also consider the data rate. An electric car does not need to transmit every raw waveform continuously. Local processing can extract features and send only event summaries, while storing raw waveforms for critical events. This reduces network load and preserves privacy. The monitoring unit can report a severity index, a pulse count, and a confidence level. Maintenance decisions can then be based on trends rather than on a single threshold crossing.
$$I_{\mathrm{sev}} = \alpha \bar{V}_{\mathrm{pd}} + \beta f_{\mathrm{pd}} + \gamma \log_{10}\left(1+\sigma_{\mathrm{pd}}\right)$$
In this severity index, the terms represent amplitude, repetition rate, and variability. The weights are determined from electric car test data. I do not claim that a single index is universal across all electric car designs. Instead, I see it as a tunable engineering tool that can be calibrated for a specific motor, connector, or busbar configuration.
Another implication is maintenance. In an electric car, insulation degradation may develop slowly. A sensor that provides only a binary alarm is less useful than a sensor that provides a trend. I therefore emphasize long-term stability, repeatable installation, and temperature compensation. The sensor should be able to distinguish a real increase in discharge activity from a change in electric car operating conditions such as switching frequency, load, or ambient temperature.
| Performance aspect | Laboratory result | Expected electric car behavior | Action for future electric car development |
|---|---|---|---|
| Detection threshold | 10 pC with useful SNR | May shift with electric car background noise | Adaptive threshold and multi-sensor fusion |
| Linearity | Correlation above 0.96 | Useful for trend monitoring | Calibrate per electric car installation |
| Out-of-band rejection | Above 30 dB | Depends on shielding and harness layout | Verify in the complete electric car |
| SNR improvement | About 12 dB | Varies with electric car driving state | Include operating-state metadata |
| Placement | Fixed bench positions | Complex electric car geometry | Use multi-position fusion and field mapping |
10. Conclusion
I have presented a compact UHF radio-frequency sensor concept for partial discharge monitoring in electric cars. The design uses a meandered microstrip patch, slot coupling, a low-loss substrate, a low-noise front end, differential output, shielding, and adaptive interference suppression. The target band from 300 MHz to 1.5 GHz provides a practical compromise for an electric car: it captures fast discharge radiation while avoiding the strongest low-frequency converter noise and the common high-frequency wireless bands.
The measured results show that the sensor can detect discharge at the 10 pC level with a signal-to-noise ratio above 6 dB and with a linear response up to higher charge levels. The anti-interference tests show that filtering, differential processing, and adaptive cancellation improve the signal-to-noise ratio by approximately 12 dB in a representative electric car disturbance environment. The bench test with a motor and introduced winding defect shows that the sensor can capture abnormal radio-frequency activity and track a discharge trend. For an electric car, this is the essential first step toward online insulation monitoring.
I conclude that the most important design principle is not maximum sensitivity alone, but balance. An electric car sensor must be small, shielded, matched, low-noise, and intelligent enough to separate discharge from the electric car background. The next steps I would pursue are sensor array fusion for discharge localization, machine-learning classification of discharge and interference types, and deeper integration with the electric car electronic and electrical architecture. These steps would move the concept from a validated laboratory and bench prototype toward an engineering solution for future electric cars.