Insulation monitoring and fault diagnosis method and device for PCB stator axial flux motor

Abstract

This invention discloses a method and apparatus for insulation monitoring and fault diagnosis of a PCB stator axial flux motor, comprising the following steps: acquiring leakage current and partial discharge pulse signals collected by a Rogowski coil, as well as temperature, angular velocity, and voltage signals acquired by auxiliary sensors; preprocessing the signal data and extracting features to construct a multi-dimensional feature vector; processing the multi-dimensional feature vector using a multi-physics coupled health assessment model, identifying insulation resistance and interlayer capacitance using an extended Kalman filter algorithm, and calculating the insulation health index after correcting the insulation resistance by introducing an electric field distortion coefficient; comparing the insulation health index with a preset threshold range to implement a graded defense strategy, and using a fault type classifier to obtain the insulation fault type and confidence level. This invention improves the sensitivity of capturing leakage current and high-frequency partial discharge pulses, solves the problem of false alarms under complex operating conditions, effectively extends the service life of the motor, and provides a basis for maintenance decisions.

Description

Technical Field

[0001] This invention relates to the field of PCB stator axial flux motor technology, specifically to a method and apparatus for insulation monitoring and fault diagnosis of PCB stator axial flux motors. Background Technology

[0002] PCB stators use printed circuit board substrates such as FR-4 or polyimide as winding carriers. Compared to traditional enameled wire windings, their dielectric strength is significantly reduced, and the multilayer winding structure of PCBs has a large interlayer parasitic capacitance. Under the high-frequency switching conditions of silicon carbide inverters, PCB stators face serious problems of partial discharge and accelerated insulation aging.

[0003] However, existing technologies for monitoring and diagnosing motor insulation conditions have several shortcomings. Traditional insulation monitoring solutions typically use general-purpose sensors such as high-frequency current transformers or capacitively coupled sensors to collect insulation signals, making it difficult to distinguish between high-frequency common-mode noise generated by silicon carbide inverter switching and actual insulation degradation signals. Especially in 800V high-voltage systems, the switching speed of silicon carbide devices is much faster than that of traditional silicon-based devices, and the switching noise spectrum generated by them overlaps with the partial discharge pulse spectrum, further increasing the difficulty of signal separation. Existing insulation life prediction models are mostly empirical formulas based on statistical learning, failing to consider the impact of radial force waves unique to axial flux motors on the electric field distribution caused by micro-deformation between PCB layers. In addition, most existing insulation monitoring systems only trigger alarms after insulation breakdown occurs, lacking proactive defense mechanisms in the early stages of insulation degradation. When insulation health declines but has not yet reached the breakdown threshold, no measures can be taken to slow down the insulation degradation process, ultimately leading to the complete scrapping of the PCB stator due to sudden breakdown, resulting in economic losses.

[0004] Therefore, how to address the deficiencies and shortcomings of existing technologies through effective insulation monitoring methods has become an important issue that researchers in this field urgently need to solve. Summary of the Invention

[0005] The purpose of this invention is to address the above-mentioned problems by providing a method and apparatus for monitoring insulation and diagnosing faults in PCB stator axial flux motors.

[0006] The technical solution of this invention is: a method for insulation monitoring and fault diagnosis of a PCB stator axial flux motor, comprising the following steps:

[0007] S1: Acquire signal data, which includes: leakage current and partial discharge pulse signals collected by the Rogowski coil attached to the back of the stator winding of the PCB, and temperature, angular velocity and voltage signals of the motor obtained by auxiliary sensors; S2: Preprocess the signal data and extract features, and construct a multi-dimensional feature vector containing time-domain features, frequency-domain features, impedance features, temperature features and mechanical features based on the extracted features; S3: The multi-physics field coupled health assessment model is used to process the multi-dimensional feature vector. In the multi-physics field coupled health assessment model, based on the multi-dimensional feature vector, the extended Kalman filter algorithm is used to identify the insulation resistance and interlayer capacitance of the PCB stator winding. The electric field distortion coefficient calculated based on the mechanical features in the multi-dimensional feature vector is introduced to correct the insulation resistance. Then, the insulation health index is calculated. S4: The insulation health index is compared with multiple preset threshold ranges. Based on the comparison results, the motor is controlled to execute the corresponding hierarchical defense strategy by adjusting the PWM output parameters of the inverter. A preset fault type classifier is used to analyze and match the multidimensional feature vector to obtain the insulation fault type of the motor and the corresponding confidence level.

[0008] As an improvement to this embodiment of the invention, step S2, the process of preprocessing the signal data includes: performing wavelet denoising on the signal data using a dynamic soft thresholding method based on noise ratio; wherein, the dynamic soft thresholding method includes calculating the signal-to-noise ratio of the current signal segment, adjusting the wavelet denoising threshold according to the signal-to-noise ratio, and the wavelet denoising threshold is inversely proportional to the signal-to-noise ratio.

[0009] As an improvement to this embodiment of the invention, step S2, the preprocessing of the signal data further includes: using a pulse width discrimination algorithm to identify and suppress switching noise; the pulse width discrimination algorithm includes measuring the half-width at half maximum (WHM) of each pulse in the denoised signal data, determining that the WHM is a partial discharge pulse and retaining it when the WHM is less than a preset pulse width threshold, and determining that the WHM is switching noise and suppressing it when the WHM is greater than or equal to the preset pulse width threshold.

[0010] As an improvement of this embodiment of the invention, in step S2, the process of extracting features includes: analyzing the denoised signal data using short-time Fourier transform to obtain a time-spectrum map within the frequency band and calculating the energy proportion of each frequency band; using empirical mode decomposition to obtain several intrinsic mode function components for the frequency bands whose energy proportion exceeds a preset energy threshold; performing Hilbert transform on each intrinsic mode function component to extract the instantaneous frequency and instantaneous amplitude as the frequency domain features.

[0011] As an improvement to this embodiment of the invention, in step S3, the electric field distortion coefficient ,in, These are dimensionless empirical coefficients, calibrated experimentally, used to quantify the proportional relationship between radial force wave intensity and the degree of electric field distortion. This represents the torque ripple amplitude. The rated torque of the motor. This represents the number of pole pairs of the motor. The angular velocity, the It is calculated from the fluctuation of the angular velocity.

[0012] As an improvement to this embodiment of the invention, in step S3, the multiphysics coupled health assessment model further converts the temperature into a correction coefficient for the aging rate of the insulating material using the Arrhenius equation. The correction coefficient is used to calculate the predicted remaining lifetime. If the predicted remaining lifetime is less than the lifetime threshold, an early warning is triggered.

[0013] As an improvement to this embodiment of the invention, in step S3, the calculation process of the insulation health index includes: establishing a health benchmark model for each dimension feature in the multidimensional feature vector, calculating the real-time feature deviation based on the health benchmark model; weighting and fusing the real-time feature deviation of each dimension feature to obtain a comprehensive deviation, and mapping the comprehensive deviation to an insulation health index of 0~100%.

[0014] As an improvement to this embodiment of the invention, in step S4, the graded defense strategy includes a first-level active defense operation, a second-level compensation and isolation operation, and a third-level emergency shutdown operation; when the insulation health index is less than 90% and greater than or equal to 75%, the first-level active defense operation is executed; when the insulation health index is less than 75% and greater than or equal to 50%, the second-level compensation and isolation operation is executed; when the insulation health index is less than 50%, the third-level emergency shutdown operation is executed.

[0015] As an improvement of this embodiment of the invention, the secondary compensation and isolation operation includes: controlling the motor driver to reduce the duty cycle of the insulation-deteriorated phase and increase the duty cycle of the healthy phase; and simultaneously superimposing a reverse DC bias voltage on the AC voltage waveform of the insulation-deteriorated phase.

[0016] To achieve one of the above-mentioned objectives, one embodiment of the present invention provides a PCB stator axial flux motor insulation monitoring and fault diagnosis device, comprising the following modules: The data acquisition module is used to acquire signal data, which includes: leakage current and partial discharge pulse signals collected by a Rogowski coil attached to the back of the PCB stator winding, and temperature, angular velocity and voltage signals of the motor acquired by auxiliary sensors. The feature extraction module is used to preprocess the signal data and extract features, and construct a multi-dimensional feature vector containing time-domain features, frequency-domain features, impedance features, temperature features and mechanical features based on the extracted features; The health index calculation module is used to process the multi-dimensional feature vector using a multi-physics field coupled health assessment model. In the multi-physics field coupled health assessment model, based on the multi-dimensional feature vector, the extended Kalman filter algorithm is used to identify the insulation resistance and interlayer capacitance of the PCB stator winding, and the electric field distortion coefficient calculated based on the mechanical features in the multi-dimensional feature vector is introduced to correct the insulation resistance. Then, the insulation health index is calculated. The graded defense and fault diagnosis module is used to compare the insulation health index with multiple preset threshold ranges. Based on the comparison results, the motor is controlled to execute the corresponding graded defense strategy by adjusting the PWM output parameters of the inverter. A preset fault type classifier is used to analyze and match the multi-dimensional feature vector to obtain the insulation fault type of the motor and the corresponding confidence level.

[0017] The PCB stator axial flux motor insulation monitoring and fault diagnosis method and device provided in this invention have the following advantages: 1. By directly attaching the Rogowski coil to the back of the PCB stator winding, the sensitivity of capturing leakage current and high-frequency partial discharge pulses is improved.

[0018] 2. By using the extended Kalman filter algorithm to identify insulation resistance and interlayer capacitance, and introducing an electric field distortion coefficient to correct the insulation resistance, the system accurately reflects the periodic compression effect of the radial force wave unique to axial flux motors on the interlayer insulation of PCBs, thus solving the problem of false alarms under complex working conditions.

[0019] 3. A three-level defense strategy was implemented based on the insulation health index, realizing progressive control from early warning to protection, which effectively extended the service life of the motor.

[0020] 4. By using a pre-defined fault type classifier to analyze and match multi-dimensional feature vectors, it can automatically identify various insulation fault types and output the corresponding probability distribution and confidence level, providing a basis for maintenance decisions. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the method for insulation monitoring and fault diagnosis of PCB stator axial flux motors as described in this invention. Figure 2 This is a schematic diagram of the overall architecture of the insulation monitoring and fault diagnosis method described in this invention; Figure 3 This is a schematic diagram of the signal acquisition link described in this invention; Figure 4 This is a schematic diagram of the structure of the PCB stator axial flux motor insulation monitoring and fault diagnosis device described in this invention.

[0022] The circuit consists of: 1-Main power circuit, 2-Signal acquisition link, 21-Input protection and voltage divider circuit, 22-Instrumentation amplification and filtering circuit, 23-Output drive and protection circuit, R1-First resistor, R2-Second resistor, R3-Third resistor, R4-Fourth resistor, R5-Fifth resistor, R6-Sixth resistor, R8-Eighth resistor, R10-Tenth resistor, R11-Eleventh resistor, R12-Twelfth resistor, C1-First capacitor, C2-Second capacitor, C3-Third capacitor, C4-Fourth capacitor, D1-First diode, D2-Second diode, D3-Third diode, D4-Fourth diode, U1-First operational amplifier, U2-Magnetic isolation amplifier, U3-Second operational amplifier, and U4-Analog-to-digital converter. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Embodiment 1 of the present invention provides a method for insulation monitoring and fault diagnosis of PCB stator axial flux motors, such as... Figure 1 As shown, it includes the following steps: Step S1: Acquire signal data, which includes: leakage current and partial discharge pulse signals collected by the Rogowski coil attached to the back of the PCB stator winding, and temperature, angular velocity and voltage signals of the motor obtained by auxiliary sensors; like Figure 2 As shown, the Rogowski coil collects leakage current and partial discharge pulse signals from the back of the PCB stator winding and sends them to the digital processing and control circuit 3 via signal acquisition link 2. The auxiliary sensor simultaneously collects temperature, angular velocity, and voltage signals. The PWM parameters output by the digital processing and control circuit 3 are fed back to the inverter. Specifically, the 800V HVDC power supply provides high-voltage DC power to the main power circuit 1. After receiving the PWM parameters output by the digital processing and control circuit 3, the SiC MOSFET two-level inverter inverts the DC bus voltage into three-phase AC voltages (U-phase, V-phase, and W-phase) and drives the PCB stator winding to generate a rotating magnetic field. Due to the extremely high switching speed of the SiC MOSFET, a rapidly changing voltage rate of change occurs at the inverter output. This voltage rate of change, as shown, couples with the parasitic capacitance between the PCB stator winding and the chassis to form a high-frequency leakage current, simultaneously inducing partial discharge pulses at weak insulation points.

[0025] To monitor the aforementioned insulation degradation signals, this invention employs a Rogowski coil and an auxiliary sensor to construct signal acquisition link 2. For example... Figure 3 As shown, the signal acquisition link 2 includes an input protection and voltage divider circuit 21, an instrumentation amplifier and filter circuit 22, and an output drive and protection circuit 23. The leakage current or the partial discharge pulse signal is differentially amplified by the instrumentation amplifier after input protection, then suppressed by an RC filter to suppress high-frequency noise, and then electrically isolated between the high-voltage side and the low-voltage side by a magnetic isolation amplifier U2. Finally, after output stage amplification and clamping protection, it is sent to the analog-to-digital converter U4.

[0026] Specifically, the input protection and voltage divider circuit 21 is located at the very beginning of the signal conditioning link 2, and includes a first resistor R1, a second resistor R2, a third resistor R3, and a fourth resistor R4. One end of the first resistor R1 serves as the positive signal input terminal VIN+, and the other end of the first resistor R1 is connected in series with one end of the second resistor R2, the other end of which is grounded. One end of the third resistor R3 serves as the inverting signal input terminal VIN-, and the other end of the third resistor R3 is connected in series with one end of the fourth resistor R4, the other end of which is grounded. The common connection point of the first resistor R1 and the second resistor R2 is connected to the inverting input terminal of the first operational amplifier U1, and the common connection point of the third resistor R3 and the fourth resistor R4 is connected to the non-inverting input terminal of the first operational amplifier U1. The input protection and voltage divider circuit 21 attenuates the differential voltage signal output from the Rogowski coil to an acceptable input range for the instrumentation amplifier, preventing high-voltage spikes from damaging downstream devices.

[0027] The instrumentation amplification and filtering circuit 22 is located after the input protection and voltage divider circuit 21, and includes a first operational amplifier U1, a fifth resistor R5, a sixth resistor R6, a first capacitor C1, a second capacitor C2, and a magnetically isolated amplifier U2. The output terminal of the first operational amplifier U1 is connected to one end of the fifth resistor R5. The other end of the fifth resistor R5 is connected to one end of the first capacitor C1 and the positive signal input terminal of the magnetically isolated amplifier U2. The other end of the first capacitor C1 is connected to one end of the sixth resistor R6, and the other end of the sixth resistor R6 is grounded. One end of the second capacitor C2 is connected to the line between the fifth resistor R5 and the first capacitor C1, and the other end of the second capacitor C2 is grounded. The inverting signal input terminal of the magnetically isolated amplifier U2 is grounded. The first operational amplifier U1, the fifth resistor R5, and the sixth resistor R6 constitute a differential amplifier stage, which is used to effectively suppress common-mode interference in the high-frequency band and convert the differential signal into a single-ended signal; the first capacitor C1, the second capacitor C2, and the sixth resistor R6 together constitute an RC filter, which is used to filter out noise outside the frequency band of interest; the magnetic isolation amplifier U2 has an isolation withstand voltage level greater than or equal to 3kVrms, which is used to achieve electrical safety isolation between the high-voltage domain and the low-voltage control domain.

[0028] The output drive and protection circuit 23 is located at the rear end of the signal conditioning link 2, and includes a second operational amplifier U3, an eighth resistor R8, a tenth resistor R10, an eleventh resistor R11, a twelfth resistor R12, a third capacitor C3, a fourth capacitor C4, a first diode D1, a second diode D2, a third diode D3, a fourth diode D4, and an analog-to-digital converter U4. The positive voltage output terminal VOUT+ of the magnetically isolated amplifier U2 is connected to the non-inverting input terminal of the second operational amplifier U3, and the inverting voltage output terminal VOUT- of the magnetically isolated amplifier U2 is connected to one end of the tenth resistor R10. The other end of the tenth resistor R10 is connected to the inverting input terminal of the second operational amplifier U3. The eighth resistor R8 is connected between the output terminal and the inverting input terminal of the second operational amplifier U3. The output terminal of the second operational amplifier U3 is connected to one end of the eleventh resistor R11, and the other end of the eleventh resistor R11 is connected to the signal input terminal of the analog-to-digital converter U4. One end of the twelfth resistor R12 is connected to the line between the eleventh resistor R11 and the analog-to-digital converter U4. The other end of the twelfth resistor R12 is connected to one end of the third capacitor C3 and one end of the fourth capacitor C4. The other end of the third capacitor C3 and the other end of the fourth capacitor C4 are grounded. The first diode D1 and the second diode D2 are connected in series between the signal input terminal of the analog-to-digital converter U4 and the positive power supply. The third diode D3 and the fourth diode D4 are connected in series between the signal input terminal of the analog-to-digital converter U4 and ground, and the diodes are in opposite directions to form a bidirectional clamp. The second operational amplifier U3, the eighth resistor R8, and the tenth resistor R10 constitute the output amplification stage, converting the differential signal output from the magnetically isolated amplifier U2 into a single-ended signal and driving the analog-to-digital converter U4. The eleventh resistor R12, the third capacitor C3, and the fourth capacitor C4 constitute an anti-aliasing filter network to suppress high-frequency aliasing components during the sampling process. The first diode D1 to the fourth diode D4 constitute a bidirectional clamping protection circuit, limiting the output signal within the allowable input voltage range of the analog-to-digital converter U4 to prevent abnormal high-voltage pulses from damaging the ADC. The signal conditioned by the output drive and protection circuit 23 is finally sent to the analog-to-digital converter U4, where it is digitized by the 24-bit ADC at a sampling rate of not less than 2MHz to obtain the signal data.

[0029] It should be noted that, Figure 3 The resistors, capacitors, amplifiers, and isolators in this invention can be selected based on the actual signal bandwidth, isolation level, and cost requirements. All these specific selection variations fall within the protection scope of this invention.

[0030] In practice, the Rogowski coil can be an embedded flexible Rogowski coil. It is positioned between the U-phase, V-phase, and W-phase output terminals of the PCB stator and ground, and is attached to the back of the PCB stator windings. The output signal of the Rogowski coil is proportional to the rate of change of the current passing through the coil, and the waveform of the measured current can be reconstructed through integration. This invention directly attaches the Rogowski coil to the back of the PCB stator windings, using the PCB dielectric as insulation support, eliminating the need for additional air gaps between the Rogowski coil and the measured object, thus greatly improving the sensitivity for capturing weak leakage currents and high-frequency partial discharge pulses. The signal acquisition bandwidth of the Rogowski coil covers 0~5MHz, effectively capturing weak insulation degradation signals generated by the PCB stator under high-voltage, high-frequency operating conditions. The auxiliary sensors can include temperature sensors, angular velocity sensors, and voltage sensors, specifically, such as... Figure 2 As shown, the temperature sensor can be an NTC thermistor embedded inside the PCB stator for real-time monitoring of hot spot temperatures in the PCB stator windings; the angular velocity sensor can be a rotary encoder mounted on the rotor end for acquiring real-time angular velocity and rotor electrical angle; the voltage sensor can be a high-frequency differential probe mounted on the DC bus side with an isolation level of 5kV for monitoring bus voltage. It is understood that the type and deployment location of the auxiliary sensors can be adjusted according to the actual motor structure and do not constitute a limitation on the scope of protection of this invention.

[0031] Step S2: Preprocess the signal data and extract features, and construct a multi-dimensional feature vector containing time-domain features, frequency-domain features, impedance features, temperature features and mechanical features based on the extracted features; Here, the time-domain features are derived from the leakage current waveform acquired by the Rogowski coil. Statistical moment analysis is performed on the amplitude distribution of the preprocessed leakage current to extract skewness and kurtosis as time-domain features. Skewness reflects the degree of asymmetry in the signal amplitude distribution, and kurtosis reflects the sharpness of the signal amplitude distribution. When insulation degradation causes distortion of the leakage current waveform, skewness and kurtosis will change significantly. The frequency-domain features are derived from the partial discharge pulse signal acquired by the Rogowski coil, with the instantaneous frequency and instantaneous amplitude extracted through joint time-frequency analysis. The impedance features are derived from the voltage signal measured by the differential probe and the current signal measured by the Rogowski coil. The complex impedance is calculated by dividing the voltage signal by the current signal, and the imaginary part of the complex impedance is taken. The rate of change of this imaginary part over time is calculated, reflecting the dynamic change of the interlayer capacitance of the PCB winding. The temperature features are derived from the winding hot spot temperature measured by the NTC thermistor, with the temperature value of the NTC thermistor directly read as the temperature feature. The mechanical feature originates from the rotor angular velocity measured by the rotary encoder. The angular velocity signal is differentiated to obtain the angular acceleration, and then the torque ripple amplitude is calculated. The torque ripple amplitude and the angular velocity signal together constitute the mechanical feature.

[0032] In this embodiment, the preprocessing of the signal data includes: performing wavelet denoising on the signal data using a dynamic soft thresholding method based on noise ratio; wherein, the dynamic soft thresholding method includes calculating the signal-to-noise ratio of the current signal segment, adjusting the wavelet denoising threshold according to the signal-to-noise ratio, and the wavelet denoising threshold is inversely proportional to the signal-to-noise ratio.

[0033] Here, the wavelet denoising threshold can be expressed by the formula: , , The calculation yielded that, The wavelet denoising threshold is... The baseline threshold is calculated by minimizing the mean squared error of the estimate, based on Stein's unbiased risk estimation principle. For signal-to-noise ratio, This represents the power of the partial discharge pulse signal. For noise power, This is a dynamic adjustment factor that is inversely proportional to the signal-to-noise ratio. The adjustment coefficient, calibrated experimentally, has a value range of 1.0 to 2.0. For reference signal-to-noise ratio (SNR), a value of 10 dB can be used. As shown in the formula above, the wavelet denoising threshold is inversely proportional to the SNR. Therefore, when the SNR is low, the denoising threshold increases to enhance denoising performance; when the SNR is high, the denoising threshold decreases to retain more pulse details.

[0034] In this embodiment, the preprocessing of the signal data further includes: using a pulse width discrimination algorithm to identify and suppress switching noise; the pulse width discrimination algorithm includes measuring the half-width at half maximum (WHM) of each pulse in the denoised signal data, and when the WHM is less than a preset pulse width threshold, it is determined to be a partial discharge pulse and retained, and when the WHM is greater than or equal to the preset pulse width threshold, it is determined to be switching noise and suppressed.

[0035] Here, the pulse width discrimination algorithm mainly utilizes the essential physical difference in pulse width between partial discharge pulses and SiC switching noise to distinguish them. Specifically, partial discharge pulses are generated by the breakdown of air gaps within the insulation, with typical pulse widths ranging from 1 to 50 nanoseconds and rise times less than 10 nanoseconds; while SiC MOSFET switching noise is generated by high-frequency switching oscillations, with pulse widths typically ranging from 100 to 500 nanoseconds or even wider. The pulse width discrimination algorithm first performs pulse detection on the denoised signal data, setting a pulse trigger threshold of 20% of the signal peak value. When the signal amplitude exceeds the pulse trigger threshold, a pulse is detected. Then, the half-width at half-maximum (WHM) of each detected pulse is measured. Here, the WHM refers to the full width of the pulse waveform at half the height of the peak value. When the WHM of the pulse is less than the preset pulse width threshold, it is determined to be a partial discharge pulse and retained; when the WHM of the pulse is greater than or equal to the preset pulse width threshold, it is determined to be switching noise and suppressed. The preset pulse width threshold can be 80 nanoseconds. Optionally, the pulse rise time can be used as an auxiliary criterion. The rise time of a partial discharge pulse is typically less than 10 nanoseconds, thereby further improving the discrimination accuracy. It is understood that the pulse width discrimination algorithm described above can effectively distinguish between partial discharge pulses and switching noise, preserving the true characteristics of insulation degradation signals.

[0036] In this embodiment, the feature extraction process includes: analyzing the denoised signal data using short-time Fourier transform to obtain the time-spectrum map within the frequency band and calculating the energy proportion of each frequency band; using empirical mode decomposition to obtain several intrinsic mode function components for the frequency bands whose energy proportion exceeds a preset energy threshold; performing Hilbert transform on each intrinsic mode function component to extract the instantaneous frequency and instantaneous amplitude as the frequency domain features.

[0037] In practice, a Hamming window can be used as the window function, with a window length of 256 points and an overlap rate of 50%. A short-time Fourier transform is performed on the signal to obtain the time-frequency spectrum within the 100kHz to 10MHz frequency band. Then, the energy proportion of each frequency band and its changing trend over time are calculated. When partial discharge activity expands from high-frequency bands to low-frequency bands, it indicates a worsening of insulation degradation. The energy proportion of each frequency band is compared with a preset energy threshold to identify abnormal frequency bands whose energy proportion exceeds the preset threshold. Subsequently, empirical mode decomposition (EMD) is applied to the frequency bands with energy proportions exceeding the preset threshold, decomposing the complex signal into several intrinsic mode function (IMF) components. Each IMF component represents an oscillation mode at different time scales in the signal. A Hilbert transform is performed on each IMF component to extract the instantaneous frequency and instantaneous amplitude as frequency domain features, used to characterize the frequency evolution of a single partial discharge pulse.

[0038] Step S3: The multi-dimensional feature vector is processed using a multi-physics field coupled health assessment model. In the multi-physics field coupled health assessment model, based on the multi-dimensional feature vector, the extended Kalman filter algorithm is used to identify the insulation resistance and interlayer capacitance of the PCB stator winding. The electric field distortion coefficient calculated based on the mechanical features in the multi-dimensional feature vector is introduced to correct the insulation resistance. Then, the insulation health index is calculated. Here, the extended Kalman filter algorithm is used to identify the insulation resistance and interlayer capacitance of the multidimensional feature vector. This includes establishing a parallel RC equivalent circuit model of the PCB stator winding as the state equation of the extended Kalman filter algorithm. In this equivalent circuit, the insulation resistance Rb and the interlayer capacitance Clayer are connected in parallel, reflecting the insulation state of the PCB winding. The state vector... Observation vector , The imaginary part of the complex impedance is the derivative with respect to time, expressed in Ω / s. Based on the physical relationships of the parallel RC equivalent circuit, leakage current equals the applied voltage divided by the vector sum of the parallel impedance of the insulation resistance, interlayer capacitance, and capacitive reactance; the phase angle depends on the ratio of insulation resistance to capacitive reactance; the rate of change of the imaginary part of the complex impedance reflects the rate of change of the interlayer capacitance. These physical relationships are used as the observation functions of the Extended Kalman Filter (EKF) algorithm. The EKF algorithm achieves online identification of insulation parameters through iterative prediction and update phases. In the prediction phase, the current state value and covariance matrix are predicted based on the state estimate and state equation from the previous time step. In the update phase, the predicted value is corrected using the observables in the current multidimensional eigenvectors through Kalman gain to obtain the optimal state estimate for the current time step, which is the updated insulation resistance and interlayer capacitance. It can be understood that through continuous iteration, the dynamic changes of insulation parameters can be tracked in real time, achieving online monitoring of the insulation state.

[0039] During operation, the radial force waves generated by the rotor rotation of the axial flux motor periodically compress the interlayer insulation medium of the PCB, causing distortion of the electric field distribution and affecting the accurate assessment of insulation resistance. Therefore, in this embodiment, the electric field distortion coefficient... , ,in, These are dimensionless empirical coefficients, calibrated experimentally, used to quantify the proportional relationship between radial force wave intensity and the degree of electric field distortion. This represents the torque ripple amplitude. The rated torque of the motor. This represents the number of pole pairs of the motor. For rotor electrical angle, The rotor mechanical angle is obtained by a rotary encoder. It is calculated from the fluctuation of the angular velocity.

[0040] Here, the dimensionless empirical coefficients The calibration method can be as follows: Under known insulation conditions, measure the actual insulation resistance change under different torque pulsations, and obtain the result through fitting. value. The torque pulsation amplitude is calculated from the angular velocity fluctuation measured by the rotary encoder. Specifically, the angular velocity signal is analyzed in the frequency domain to extract the amplitude of the speed fluctuation component, and then converted into the torque pulsation amplitude based on the motor torque constant. Here, is the rated torque of the motor, and p is the number of pole pairs; both are motor design parameters. From the above formula, we can see that the electric field distortion coefficient... It consists of two parts: a constant 1 and a correction term. The square of the ratio of the torque pulsation amplitude to the rated torque in the correction term reflects the relative intensity of the radial force wave, while the angle-dependent sine function reflects the periodic variation of the radial force wave with rotor rotation. After calculating the electric field distortion coefficient, it is used to correct the insulation resistance value identified by the extended Kalman filter algorithm. The correction formula is: ,in, The corrected insulation resistance. To extend the insulation resistance identified by the Kalman filter algorithm, This is the electric field distortion coefficient. Because... The value of is always greater than or equal to 1. Less than or equal to That is, when mechanical stress is present, the evaluation value of insulation resistance is reduced to more conservatively reflect the true insulation state of the PCB stator under dynamic load.

[0041] In this embodiment, the multiphysics coupled health assessment model also converts the temperature into a correction coefficient for the aging rate of the insulation material using the Arrhenius equation. The correction coefficient is used to calculate the predicted remaining lifetime; if the predicted remaining lifetime is less than the lifetime threshold, an early warning is triggered.

[0042] Here, the temperature is the winding hot spot temperature measured by the temperature sensor, and the thermal aging acceleration coefficient is calculated using the Arrhenius equation. ,in, The activation energy of insulating materials is approximately 0.8 to 1.2 eV; is the Boltzmann constant, with a value of 8.617 × 10⁻⁶. -5 eV / K; For reference temperature, the value is taken as 25 degrees Celsius; The temperature is the hot spot temperature of the winding. As shown in the formula above, the thermal aging acceleration coefficient has an exponential relationship with temperature. When the winding temperature rises from 25℃ to 60℃, the thermal aging acceleration coefficient is approximately 20 to 50, meaning the aging rate of the insulation material increases by 20 to 50 times. Based on the thermal aging acceleration coefficient, and combined with electrical stress and mechanical stress factors, the remaining life is calculated and predicted. A modified Arrhenius-electrical stress joint life model is used to predict the remaining life. , Where L is the predicted remaining lifetime. The baseline lifespan is given by B, a material constant determined by thermal aging experiments on polyimide, with a value ranging from approximately 0.05 to 0.1 / K; ΔT is the temperature rise, and V is the measured bus voltage, obtained through differential probe measurement. is the rated voltage, and n is the voltage durability factor, which is determined by the electrical aging test of the insulating material and takes a value of approximately 8 to 12. This represents the electric field distortion coefficient. The physical meaning of the predicted remaining life formula is as follows: increased temperature accelerates thermal aging, increased voltage stress accelerates electrical aging, and increased mechanical stress accelerates insulation degradation. These three factors work together to shorten the predicted remaining life. An alarm is triggered when the predicted remaining life is less than a preset life threshold. Specifically, the life threshold can be set according to the motor application scenario, such as 500 hours. Even if the insulation health index has not yet dropped to the graded defense threshold, if the predicted remaining life is lower than the life threshold, an early warning will be triggered, prompting maintenance to be arranged, thus achieving preventative protection from a time perspective.

[0043] In this embodiment, the calculation process of the insulation health index includes: establishing a health benchmark model for each dimension feature in the multidimensional feature vector, calculating the real-time feature deviation based on the health benchmark model; weighting and fusing the real-time feature deviation of each dimension feature to obtain a comprehensive deviation, and mapping the comprehensive deviation to an insulation health index of 0~100%.

[0044] Here, the motor operates under its initial healthy condition for a period of time, collecting data on various features. Assuming each feature follows a Gaussian distribution, the mean and variance of each feature are calculated, serving as the health baseline model. Then, the real-time feature deviation is calculated based on the health baseline model. For each feature dimension, the formula for calculating the real-time feature deviation is: ,in, denoted as the real-time feature deviation of the i-th dimension; This represents the real-time value of the i-th dimension feature; Let be the mean of the i-th dimension feature in the health benchmark model; Let be the standard deviation of the i-th dimension feature in the health baseline model. The real-time feature deviation reflects the degree to which the current feature value deviates from the health baseline; a larger deviation indicates more severe insulation degradation. Next, the real-time feature deviations of each dimension feature are weighted and fused to obtain the comprehensive deviation, calculated using the following formula: Where D represents the overall deviation; The weights of the i-th dimension features; Let be the real-time feature deviation of the i-th dimension. The weights can be set according to the sensitivity of each dimension to insulation degradation, with the sum of the weights for each dimension being 1. Finally, the overall deviation is mapped to an insulation health index ranging from 0 to 100%, using the following formula: The insulation health index (SOH) ranges from 0 to 100%. It can be understood that when all dimensions are at a healthy baseline state, the overall deviation (D) is 0 and the SOH is 100%, indicating healthy insulation. When insulation is severely deteriorated, the overall deviation (D) approaches 1 and the SOH approaches 0%, indicating extremely dangerous insulation.

[0045] Step S4: Compare the insulation health index with multiple preset threshold ranges. Based on the comparison results, control the motor to execute the corresponding hierarchical defense strategy by adjusting the PWM output parameters of the inverter. Use a preset fault type classifier to analyze and match the multidimensional feature vector to obtain the insulation fault type of the motor and the corresponding confidence level.

[0046] Here, the fault type classifier can be constructed based on a fault feature pattern database. This database contains various insulation fault modes unique to PCB stators and their corresponding feature combinations. These insulation fault modes can include moisture, air gap discharge, carbonization, and mechanical fatigue. Specifically, the feature combination pattern for moisture is a significant decrease in insulation resistance and an increase in the dielectric loss tangent, but with normal temperature. The feature combination pattern for air gap discharge is a dramatic increase in the number of partial discharge pulses, with the phase-resolved partial discharge pattern exhibiting a typical rabbit-ear-shaped distribution. The feature combination pattern for carbonization is a continuous decrease in insulation resistance, an increase in the amplitude of partial discharge pulses, and an increase in temperature. The feature combination pattern for mechanical fatigue is a periodic fluctuation in the electric field distortion coefficient and an abnormal rate of change of the imaginary part of the complex impedance. The analysis process of the fault type classifier involves calculating the cosine similarity between the current multidimensional feature vector and each fault pattern vector in the fault feature pattern database. Then, the cosine similarity is normalized to a probability distribution using the Softmax function to obtain the probability that the motor belongs to each insulation fault type. The confidence level is calculated based on the significance level of feature deviation and the consistency of cross-validation of multiple features. When the deviation of a feature in a certain dimension exceeds 3 times the standard deviation, the diagnostic confidence of that dimension is high. When features in multiple dimensions point to the same fault type, the consistency of cross-validation is high, and the overall confidence level is also improved.

[0047] In this embodiment, the graded defense strategy includes a first-level active defense operation, a second-level compensation and isolation operation, and a third-level emergency shutdown operation. When the insulation health index is less than 90% but greater than or equal to 75%, the first-level active defense operation is executed; when the insulation health index is less than 75% but greater than or equal to 50%, the second-level compensation and isolation operation is executed; and when the insulation health index is less than 50%, the third-level emergency shutdown operation is executed.

[0048] In practice, the first-level active defense operation includes: reducing the PWM carrier frequency, for example, reducing the carrier frequency from 40kHz to 10kHz to 30kHz, to slow down the voltage change rate and reduce the electrical stress on the insulation material; limiting the maximum phase voltage amplitude to 85% of the rated value to reduce the voltage stress on the insulation; and increasing the cooling system speed to force cooling, reduce the winding hot spot temperature, and slow down the thermal aging rate. After executing the first-level active defense operation, the effectiveness of the strategy is verified: if the insulation health index recovers to above 90%, the system returns to normal monitoring; if the insulation health index continues to deteriorate, a second-level compensation and isolation operation is performed. The second-level compensation and isolation operation includes: controlling the motor driver to reduce the duty cycle of the insulation-deteriorated phase and increase the duty cycle of the healthy phase; and simultaneously superimposing a reverse DC bias voltage on the AC voltage waveform of the insulation-deteriorated phase. Specifically, the insulation-deteriorated phase and the healthy phase are determined by identifying the impedance characteristic changes of each phase in the multi-dimensional feature vector. Reducing the duty cycle of the insulation-deteriorated phase can decrease the voltage stress and voltage change rate of that phase winding, and reduce the intensity of partial discharge activity in that phase. Simultaneously, while maintaining the total power, increasing the duty cycle of the healthy phase compensates for the reduced power output of the insulation-deteriorated phase, maintaining motor torque balance. This is achieved by adjusting the reference voltage vector of each phase using carrier phase shifting or space vector modulation to keep the three-phase current symmetrical and avoid exacerbating torque ripple, although the actual voltage amplitude of the insulation-deteriorated phase has been reduced. Simultaneously, the motor is derated to 60% power and fault recording is initiated. Additionally, a reverse DC bias voltage is superimposed on the AC voltage waveform of the insulation-deteriorated phase. Specifically, by adjusting the PWM modulation strategy, a small reverse DC bias voltage is superimposed on the AC voltage waveform of the insulation-deteriorated phase at the inverter output. The amplitude of this DC bias voltage is typically 0.5% to 2% of the rated voltage. Its working principle is that during the AC voltage cycle, space charge accumulates inside the insulation material, which exacerbates local electric field distortion and promotes partial discharge. By superimposing a reverse DC bias voltage, a reverse electric field is generated near the zero-crossing of the AC voltage, which helps neutralize the accumulated space charge and suppress partial discharge activity. To prevent the DC bias from causing saturation of the motor core, current limiting protection needs to be superimposed on the phase current, and the detected DC component must not exceed 5% of the rated current. The three-level emergency shutdown operation includes: blocking the SiC drive signal, disconnecting the 800V bus, stopping the motor operation; and activating the braking resistor to discharge the back electromotive force. When the SiC drive is blocked, the motor rotor continues to rotate due to inertia. At this time, the permanent magnet rotor induces a back electromotive force in the stator winding. If the 800V bus has been disconnected, the back electromotive force cannot be fed back to the bus, which will form a dangerous overvoltage on the DC bus. The braking resistor is connected in parallel to the DC bus through a switch such as an IGBT or a contactor to dissipate the energy generated by the back electromotive force as heat, achieving safe and rapid discharge.The resistance value of the braking resistor is determined based on the maximum allowable voltage of the busbar and the braking power, which is typically 10% to 20% of the rated power. Simultaneously, fault waveform data is saved, and the system enters a physical isolation alarm state, awaiting manual intervention for reset. After reset, the insulation health index must be retested and graded for confirmation.

[0049] Embodiment 2 of the present invention provides an insulation monitoring and fault diagnosis device for a PCB stator axial flux motor, comprising the following modules: The data acquisition module 401 is used to acquire signal data, which includes: leakage current and partial discharge pulse signals collected by the Rogowski coil attached to the back of the stator winding of the PCB, and temperature, angular velocity and voltage signals of the motor acquired by auxiliary sensors. The feature extraction module 402 is used to preprocess the signal data and extract features, and construct a multi-dimensional feature vector containing time-domain features, frequency-domain features, impedance features, temperature features and mechanical features based on the extracted features; The health index calculation module 403 is used to process the multi-dimensional feature vector using a multi-physics field coupled health assessment model. In the multi-physics field coupled health assessment model, based on the multi-dimensional feature vector, the extended Kalman filter algorithm is used to identify the insulation resistance and interlayer capacitance of the PCB stator winding, and the electric field distortion coefficient calculated based on the mechanical features in the multi-dimensional feature vector is introduced to correct the insulation resistance. Then, the insulation health index is calculated. The graded defense and fault diagnosis module 404 is used to compare the insulation health index with multiple preset threshold ranges. Based on the comparison results, the motor is controlled to execute the corresponding graded defense strategy by adjusting the PWM output parameters of the inverter. The multi-dimensional feature vector is analyzed and matched using a preset fault type classifier to obtain the insulation fault type of the motor and the corresponding confidence level.

[0050] This invention can be an apparatus, method, and / or computer program product. A computer program product may include a readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0051] Storage media can be tangible devices that hold and store instructions for use by instruction execution devices. Storage media can include, but are not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof.

[0052] It should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0053] The detailed descriptions listed above are merely specific descriptions of feasible embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for insulation monitoring and fault diagnosis of a PCB stator axial flux motor, characterized by, Includes the following steps: S1: Acquire signal data, which includes: leakage current and partial discharge pulse signals collected by the Rogowski coil attached to the back of the stator winding of the PCB, and temperature, angular velocity and voltage signals of the motor obtained by auxiliary sensors; S2: Preprocess the signal data and extract features, and construct a multi-dimensional feature vector containing time-domain features, frequency-domain features, impedance features, temperature features and mechanical features based on the extracted features; S3: The multi-physics field coupled health assessment model is used to process the multi-dimensional feature vector. In the multi-physics field coupled health assessment model, based on the multi-dimensional feature vector, the extended Kalman filter algorithm is used to identify the insulation resistance and interlayer capacitance of the PCB stator winding. The electric field distortion coefficient calculated based on the mechanical features in the multi-dimensional feature vector is introduced to correct the insulation resistance. Then, the insulation health index is calculated. S4: The insulation health index is compared with multiple preset threshold ranges. Based on the comparison results, the motor is controlled to execute the corresponding hierarchical defense strategy by adjusting the PWM output parameters of the inverter. A preset fault type classifier is used to analyze and match the multidimensional feature vector to obtain the insulation fault type of the motor and the corresponding confidence level.

2. The method for monitoring and diagnosing insulation of a PCB stator axial flux motor according to claim 1, characterized in that, In step S2, the preprocessing of the signal data includes: performing wavelet denoising on the signal data using a dynamic soft thresholding method based on noise ratio; wherein, the dynamic soft thresholding method includes calculating the signal-to-noise ratio of the current signal segment, adjusting the wavelet denoising threshold according to the signal-to-noise ratio, and the wavelet denoising threshold is inversely proportional to the signal-to-noise ratio.

3. The method for monitoring and diagnosing insulation of a PCB stator axial flux motor according to claim 2, characterized in that, In step S2, the preprocessing of the signal data further includes: using a pulse width discrimination algorithm to identify and suppress switching noise; the pulse width discrimination algorithm includes measuring the half-width at half maximum (WHM) of each pulse in the denoised signal data, and when the WHM is less than a preset pulse width threshold, it is determined to be a partial discharge pulse and retained, and when the WHM is greater than or equal to the preset pulse width threshold, it is determined to be switching noise and suppressed.

4. The method for insulation monitoring and fault diagnosis of PCB stator axial flux motor according to claim 2, characterized in that, In step S2, the feature extraction process includes: analyzing the denoised signal data using short-time Fourier transform to obtain the time spectrum within the frequency band and calculating the energy proportion of each frequency band; using empirical mode decomposition to obtain several intrinsic mode function components for the frequency bands whose energy proportion exceeds a preset energy threshold; performing Hilbert transform on each intrinsic mode function component to extract the instantaneous frequency and instantaneous amplitude as the frequency domain features.

5. The method for insulation monitoring and fault diagnosis of PCB stator axial flux motor according to claim 1, characterized in that, In step S3, the electric field distortion coefficient ,in, These are dimensionless empirical coefficients, calibrated experimentally, used to quantify the proportional relationship between radial force wave intensity and the degree of electric field distortion. This represents the torque ripple amplitude. The rated torque of the motor. This represents the number of pole pairs of the motor. The angular velocity, the It is calculated from the fluctuation of the angular velocity.

6. The method for monitoring and diagnosing insulation of a PCB stator axial flux motor according to claim 1, characterized in that, In step S3, the multiphysics coupled health assessment model also converts the temperature into a correction coefficient for the aging rate of the insulation material using the Arrhenius equation. The correction coefficient is used to calculate the predicted remaining lifetime. If the predicted remaining lifetime is less than the lifetime threshold, an early warning is triggered.

7. The method for insulation monitoring and fault diagnosis of a PCB stator axial flux motor according to claim 1, characterized in that, In step S3, the calculation process of the insulation health index includes: establishing a health benchmark model for each dimension feature in the multidimensional feature vector, calculating the real-time feature deviation based on the health benchmark model; weighting and fusing the real-time feature deviation of each dimension feature to obtain a comprehensive deviation, and mapping the comprehensive deviation to an insulation health index of 0~100%.

8. The method for insulation monitoring and fault diagnosis of a PCB stator axial flux motor according to claim 1, characterized in that, In step S4, the graded defense strategy includes a first-level active defense operation, a second-level compensation and isolation operation, and a third-level emergency shutdown operation; when the insulation health index is less than 90% but greater than or equal to 75%, the first-level active defense operation is executed; when the insulation health index is less than 75% but greater than or equal to 50%, the second-level compensation and isolation operation is executed. When the insulation health index is less than 50%, the level three emergency shutdown operation is performed.

9. The method for insulation monitoring and fault diagnosis of a PCB stator axial flux motor according to claim 8, characterized in that, The secondary compensation and isolation operation includes: controlling the motor driver to reduce the duty cycle of the insulation-deteriorated phase and increase the duty cycle of the healthy phase; and simultaneously superimposing a reverse DC bias voltage on the AC voltage waveform of the insulation-deteriorated phase.

10. A device for monitoring and diagnosing insulation of a PCB stator axial flux motor, characterized in that, Includes the following modules: The data acquisition module is used to acquire signal data, which includes: leakage current and partial discharge pulse signals collected by a Rogowski coil attached to the back of the PCB stator winding, and temperature, angular velocity and voltage signals of the motor acquired by auxiliary sensors. The feature extraction module is used to preprocess the signal data and extract features, and construct a multi-dimensional feature vector containing time-domain features, frequency-domain features, impedance features, temperature features and mechanical features based on the extracted features; The health index calculation module is used to process the multi-dimensional feature vector using a multi-physics field coupled health assessment model. In the multi-physics field coupled health assessment model, based on the multi-dimensional feature vector, the extended Kalman filter algorithm is used to identify the insulation resistance and interlayer capacitance of the PCB stator winding, and the electric field distortion coefficient calculated based on the mechanical features in the multi-dimensional feature vector is introduced to correct the insulation resistance. Then, the insulation health index is calculated. The graded defense and fault diagnosis module is used to compare the insulation health index with multiple preset threshold ranges. Based on the comparison results, the motor is controlled to execute the corresponding graded defense strategy by adjusting the PWM output parameters of the inverter. A preset fault type classifier is used to analyze and match the multi-dimensional feature vector to obtain the insulation fault type of the motor and the corresponding confidence level.

Images