Capacitor online health monitoring system and method based on double-path differential sensing and working condition self-adaption

Abstract

The invention provides a capacitor on-line health monitoring system and method based on double-path differential sensing and working condition self-adaption, and the system employs a double-path differential current sensing architecture, synchronously measures the current of a total bus and the total current of other parallel branches except a to-be-measured capacitor, and carries out the real-time differential operation. Net ripple current flowing through a specific capacitor bank to be measured is accurately reconstructed, and weak signal extraction with a high signal-to-noise ratio is realized under strong common-mode interference. According to the system, a working condition self-adaptive identification module is introduced to analyze macroscopic electrical parameters in real time and classify operation states, and it is ensured that a high-precision parameter identification algorithm is only triggered under a stable steady-state working condition. In the parameter identification stage, the system uses switch harmonic waves of a converter as a natural excitation source, the equivalent series resistance, capacitance and equivalent series inductance of a capacitor are synchronously solved through multi-frequency-point impedance spectroscopy measurement and nonlinear and least square fitting, and finally online health state evaluation and predictive maintenance are achieved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of power electronic device state monitoring and predictive maintenance, and particularly relates to a system and method for online health state monitoring of a DC bus parallel capacitor bank in a power electronic converter, and especially to a non-intrusive monitoring scheme based on a dual-path differential current sensing technology and adaptive identification of operating conditions. BACKGROUND

[0002] DC bus capacitors are core energy storage and filtering elements in power electronic converters (such as inverters, converters, etc.), and their performance state is directly related to the stability, reliability of the converter operation and the quality of the output power. In the long-term operation process, capacitors will undergo electrochemical aging due to bearing high ripple current, high voltage stress and periodic temperature changes, and the main macroscopic manifestations are the increase of equivalent series resistance (ESR) and the decrease of capacitance (C). The attenuation of capacitance will lead to an increase in bus voltage ripple, thereby reducing the key performance indicators of the converter; while the increase of ESR will exacerbate the power loss and heating of the capacitor itself, forming a vicious cycle, accelerating the aging process, and in extreme cases, may cause capacitor bulging, liquid leakage and even explosion, thereby causing equipment downtime, system paralysis and even serious safety accidents.

[0003] Currently, the monitoring methods for capacitor health state mainly include offline detection and online detection: 1. Offline detection: This method requires the capacitor to be removed from the system after the device is shut down, and then measured using special precision instruments such as LCR bridges. Although offline detection has high accuracy, its disadvantages are obvious: it not only interrupts the normal operation of the device, significantly increasing the operation and maintenance cost and work complexity, but more importantly, it cannot provide continuous state data, thus cannot achieve state-based predictive maintenance.

[0004] 2. Online detection: To overcome the shortcomings of offline detection, various online detection technologies have emerged. However, the existing mainstream online schemes have one or more inherent defects: Based on active signal injection: This method measures capacitor impedance by injecting a specific frequency excitation signal into the system. The main problems are: first, the additional injected signal may interfere with the switching frequency of the converter itself, introducing electromagnetic compatibility (EMC) risks. Second, to ensure safety, the amplitude of the injected signal is usually weak, and in the presence of strong main circuit current background noise, the signal-to-noise ratio (SNR) is not ideal, affecting the measurement accuracy.

[0005] Complex algorithm-based solutions: Some methods attempt to use artificial intelligence or advanced signal processing algorithms to indirectly infer the capacitor state by analyzing system macro electrical quantities. Such methods often require massive data for model training and have a large amount of computation, making it difficult to achieve real-time and efficient monitoring on resource-constrained embedded platforms. For example, the patent document "Capacitor device online monitoring method and system based on artificial intelligence" (publication number CN202311758289.9) describes.

[0006] Lack of working condition adaptability: Most existing solutions ignore the serious interference of transformer dynamic operating conditions (such as sudden changes in load power, start-stop impact, etc.) on impedance measurement. In non-steady-state conditions, the transient changes in bus voltage and current are not determined by the capacitor itself, and parameter identification at this time will produce large errors, leading to distorted results and failing to truly reflect the health of the capacitor.

[0007] Therefore, there is an urgent need in the art for a new type of capacitor online health monitoring technology that can simultaneously meet multiple stringent requirements such as non-invasiveness (no shutdown, no additional signal injection), high precision, strong robustness (anti-working condition interference), and real-time performance. SUMMARY

[0008] The purpose of the present application is to address the limitations of existing technology by proposing a capacitor online health monitoring system and method based on dual-path differential sensing and working condition adaptation.

[0009] The present application employs the following technical solutions: A capacitor online health monitoring system based on dual-path differential sensing and working condition adaptation, characterized by: the system includes a sensing module, a data acquisition module, and a processor module, wherein the sensing module includes a first current sensor for measuring the total current including the capacitor to be measured, other parallel capacitors, and downstream loads, a second current sensor for measuring the total current flowing to all parallel branches except the capacitor group to be measured, a voltage sensor for synchronously measuring the bus voltage, and a multi-point temperature sensing unit for capturing the temperature gradient of the capacitor group due to layout and air duct effects; The data acquisition module is connected to all sensors and is responsible for high-precision, synchronous sampling and analog-to-digital conversion of the output signals of each sensor at a predetermined high sampling rate; The processor module is used to execute software algorithms.

[0010] Preferably, the first current sensor is arranged on the DC bus total current path upstream of the capacitor group to be measured, the second current sensor is arranged on the branch path upstream of other parallel capacitor groups and loads downstream of the capacitor group to be measured, and the voltage sensor is connected across the DC bus.

[0011] The application discloses a capacitor online health monitoring method based on double-path differential sensing and working condition self-adaption.

[0012] Preferably, the macroscopic electrical characteristic quantity comprises one or more of effective values of input / output voltage / current, active / reactive power, power change rate (dP / dt), total harmonic distortion (THD), and the working condition type comprises light load steady state, heavy load steady state and dynamic impact.

[0013] Preferably, the differential current reconstruction comprises: acquiring path A current i_A(t) and path B current i_B(t) synchronously sampled by a data acquisition module, calculating i_dut(t) = i_A(t) - i_B(t), and constructing a net ripple current signal i_dut(t) flowing through the capacitor group to be measured.

[0014] Preferably, the spectrum analysis comprises applying a fast Fourier transform (FFT) algorithm to the reconstructed capacitor net current i_dut(t) and the synchronously sampled bus voltage v_bus(t) to obtain complex spectra I_dut(f) and V_bus(f) of the capacitor net current and the bus voltage in a frequency domain.

[0015] Preferably, the multi-frequency point impedance calculation comprises using a switching harmonic generated during operation of the transformer as an excitation source, and calculating complex impedances of the capacitor to be measured at at least two harmonic frequency points f_m with energy greater than a predetermined threshold according to Ohm's law: Z(f_m) = V_bus(f_m) / I_dut(f_m).

[0016] Preferably, the harmonic frequency points f_m comprise a switching frequency fundamental wave, a second harmonic and a third harmonic.

[0017] Preferably, the parameter identification includes taking the calculated multi-frequency point complex impedance data (f_m, Z(f_m)) as input, using a numerical optimization algorithm such as nonlinear least squares method to curve fit the multi-frequency point impedance data, and solving the equivalent series resistance (ESR), capacitance (C) and equivalent series inductance (ESL) of the capacitor under test.

[0018] The health assessment and output module stores or establishes a multi-dimensional health baseline database through online learning, which records the standard reference values of ESR and C of the capacitor in a healthy state under different steady-state working conditions (such as different load rates) and different working temperatures; the state assessment and aging warning module matches the latest calculated ESR and C values of the parameter identification module with the working condition type identified by the current working condition classification module and the average / maximum temperature measured by the temperature sensing unit, calls the corresponding health baseline from the database, and outputs relevant information according to the evaluation result.

[0019] Compared with the prior art, the system and method have one or more of the following significant beneficial effects: Compared with the prior art, the system and method have one or more of the following significant beneficial effects: 1. High signal-to-noise ratio and precise positioning measurement: The inventive dual-path differential current sensing architecture cleverly cancels the strong common-mode load current flowing to other parallel branches as interference through real-time operation of i_A(t) - i_B(t), thereby realizing precise extraction and reconstruction of weak ripple current in a specific capacitor branch under test. This fundamentally solves the "weak signal extraction" problem in parallel capacitor online monitoring, and the signal-to-noise ratio is improved by orders of magnitude compared with the traditional single sensor scheme.

[0020] 2. Pure non-invasiveness and high economy: This scheme completely uses the harmonics generated by the switching action of the power electronic converter itself as the "natural excitation source" for impedance spectrum analysis, without any additional signal injection circuit. This not only eliminates the electromagnetic compatibility (EMC) risk that may be brought by external signal injection from the source, but also greatly simplifies the system hardware design, reduces the implementation cost and complexity.

[0021] 3. Excellent working condition robustness and high reliability: The application innovatively introduces a working condition classification module based on macroscopic electrical parameters, realizing an intelligent gating mechanism of "steady-state measurement, dynamic shielding". This mechanism ensures that the parameter identification algorithm is only executed in the window period of stable and reliable data, effectively eliminating the fatal interference of load dynamic changes on the diagnosis results, thereby significantly improving the accuracy and reliability of the monitoring results in the real and variable operating environment.

[0022] 4. Comprehensive parameter identification and fine health assessment: By using the multi-frequency point impedance spectrum fitting technology, the application can solve the three key parameters of ESR, C and ESL, which reflect different aging dimensions of the capacitor, with high precision at one time. Further, combined with the working condition and temperature adaptive health baseline comparison strategy, a multi-dimensional health state assessment model is constructed, so that the aging assessment is no longer a static and isolated parameter comparison, but a dynamic and closely related comprehensive diagnosis with the operating environment, thereby providing comprehensive and reliable data support for the fine health management and predictive maintenance of the capacitor.

[0023] 5. In the parallel capacitor group, the ripple current flowing through the capacitor group is much smaller than the total system ripple current, and is overwhelmed by the strong common-mode load current. The traditional single current sensor scheme cannot solve this "weak signal extraction" problem. The dual-path differential architecture of the application is proposed to solve the core pain point of accurately separating the specific capacitor current in the parallel branch. And through the working condition adaptive gating measurement mechanism, the problem of variable working condition interference is solved.

[0024] In order to further understand the features and technical contents of the application, please refer to the following detailed description and drawings of the application. However, the provided drawings are only used for reference and illustration, and are not used to limit the application. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a hardware block diagram of the monitoring system of the application and a schematic diagram of the installation position of the sensor; Figure 2 is a schematic diagram of the installation position of the temperature sensor on the capacitor group; Figure 3 is a general flowchart of the monitoring algorithm of the application; Figure 4 is a principle flowchart of the working condition classification algorithm in Figure 3 is a detailed flowchart of the parameter identification algorithm steps in Figure 5 Figure 3 Figure 6 is a result waveform diagram of the light load steady-state simulation of the working condition; Figure 7 is a result waveform diagram of the ordinary steady-state simulation of the working condition;​​ Figure 8 is the result waveform diagram of the working condition of heavy load steady-state simulation; Figure 9 is the result waveform diagram of the working condition of dynamic impact simulation; Figure 10 is the simulation verification diagram of differential current reconstruction; Figure 11 is the simulation verification diagram of parameter identification result; Figure 12 is the capacitor physical parameter simulation degradation curve diagram; Figure 13 is the relationship diagram of capacitor health state and cycle number. DETAILED DESCRIPTION

[0026] The following is to illustrate the embodiments of the present application by specific embodiments, and the advantages and effects of the present application can be understood by the person skilled in the art from the disclosure of the present specification. The present application can be implemented or applied by other different embodiments, and various modifications and changes can be made to the details in the present specification based on different views and applications without departing from the spirit of the present application. In addition, the drawings of the present application are only simple schematic illustrations, not actual size drawings, and it is declared in advance. The following embodiments will further illustrate the related technical content of the present application, but the disclosed content is not used to limit the protection scope of the present application.

[0027] Example 1: refer to the attached Figure 1 (system hardware block diagram), a capacitor online health monitoring system based on double-path differential sensing and working condition self-adaptation, characterized in that: the system comprises a sensing module, a data acquisition module and a processor module, wherein the sensing module comprises a first current sensor for measuring the total current including the to-be-measured capacitor, other parallel capacitors and downstream loads, a second current sensor for measuring the total current flowing to all parallel branches except the to-be-measured capacitor group, a voltage sensor for synchronously measuring the bus voltage, and a multi-point temperature sensing unit for capturing the temperature gradient of the capacitor group due to the influence of layout and air duct; The data acquisition module is connected with all the sensors, and is responsible for high-precision and synchronous sampling and analog-digital conversion of the output signals of the sensors at a preset high sampling rate; The processor module is used for executing software algorithms.

[0028] Preferably, the first current sensor is arranged on the DC bus total current path upstream of the to-be-measured capacitor group, the second current sensor is arranged on the branch path upstream of other parallel capacitor groups and loads downstream of the to-be-measured capacitor group, and the voltage sensor is connected to both ends of the DC bus.

[0029] The first current sensor and the second current sensor constitute a double-path differential current sensing unit, specifically: Double-path differential current sensing unit: First current sensor (CS_A): as shown, CS_A is installed in series on the DC bus main line before the parallel capacitor group and the downstream load. Its function is to measure the total current i_A(t) flowing through the main line, which is the sum of the current of the capacitor group to be measured and the current of all other parallel branches (including the load). Figure 1

[0030] Second current sensor (CS_B): as shown, CS_B is installed in series on the branch downstream of the capacitor group to be measured (C_dut) and leads to the load such as a switching device. Its function is to accurately measure the total current i_B(t) flowing to all other parallel branches except C_dut. Figure 1

[0031] Voltage sensor (VS_A) of the embodiment is as shown, VS_A is connected in parallel across the DC bus, used for synchronous measurement of the bus voltage v_bus(t) on the capacitor group to be measured. Figure 1

[0032] The multi-point temperature sensor of the embodiment includes a first temperature sensor, a second temperature sensor, and a third temperature sensor, specifically: Multi-point temperature sensing unit: Referring to the attached Figure 2 (Temperature sensor installation schematic diagram), in order to accurately capture the internal temperature gradient of the capacitor group due to forced air cooling, the embodiment deploys three temperature sensors: First temperature sensor (TS_A): installed on the top side of the capacitor shell on the windward side of the capacitor group, used to measure the area temperature T_A(t) with the best cooling effect.

[0033] Second temperature sensor (TS_B): installed on the top side of the capacitor shell at the center position of the capacitor group, which is usually poor in heat dissipation and is a potential "hot spot", used to measure the core area temperature T_B(t).

[0034] Third temperature sensor (TS_C): installed on the top side of the capacitor shell on the leeward side of the capacitor group, used to measure the area temperature T_C(t) affected by the airflow heated by the upstream capacitor.

[0035] ​​​In this embodiment, the analog output signals of all sensors (CS_A, CS_B, VS_A, TS_A, TS_B, TS_C) are connected to the data acquisition module (DAQ). The DAQ is responsible for high-precision, synchronous analog-to-digital conversion of each channel signal, especially strict synchronous sampling of the three signals i_A(t), i_B(t) and v_bus(t) to ensure the phase accuracy of subsequent impedance calculation.

[0036] The processor module (CPU) of this embodiment is the operation core of the system, usually a digital signal processor (DSP) or a microcontroller (MCU) with a floating point operation unit. It receives the digital signal stream from the DAQ and executes the preset software algorithm.

[0037] In this embodiment, the ripple current flowing through the capacitor bank is much smaller than the total system ripple current, and is overwhelmed by the strong common-mode load current. The traditional single-current sensor solution cannot solve this "weak signal extraction" problem. The dual-path differential architecture of this embodiment is proposed to solve the core pain point of accurately separating the specific capacitor current in parallel branches. And through the self-adaptive gating measurement mechanism, the problem of variable working condition interference is solved.

[0038] Embodiment two: this embodiment should be understood as at least containing all the features of any one of the preceding embodiments, and further improving on the basis thereof; The embodiment provides a capacitor online health monitoring method based on dual-path differential sensing and working condition self-adaptation, characterized by being realized based on a processor module of the capacitor online health monitoring system, and comprising the following steps: using a working condition classification module to analyze the macro state of the transformer in real time, comparing the calculated macro electrical characteristic quantity with a preset threshold, and classifying the current running state of the transformer into a predefined working condition type in real time; a parameter identification module is triggered to perform differential current reconstruction, spectrum analysis, multi-frequency point impedance calculation and parameter identification when the working condition classification module judges that the current working condition is a steady state; and a health evaluation and output module analyzes and evaluates the parameters identified by the parameter identification module, and outputs relevant information according to the evaluation result.

[0039] Specifically, the software algorithm running in the processor module (CPU) is the key to realizing the function of the present application. The complete monitoring method is shown in the attached Figure 3 (algorithm main flowchart) and Figure 5 (detailed flowchart of parameter identification algorithm steps), and the specific steps are as follows: Step S101: macro electrical parameter acquisition and working condition classification.

[0040] The working condition classification algorithm in the processing module continuously acquires input / output voltage and current signals from the low-speed acquisition module of the converter controller. By calculating the effective value (RMS), active power (P), total harmonic distortion (THD_i), power change rate (dP / dt), and other parameters within a time window (for example, 100 ms), a working condition feature vector F is formed. F(t)=[Vrms Irms P THD_i dP / dt]; wherein Vrms is the voltage effective value, and Irms is the current effective value.

[0041] The working condition classification algorithm flow chart is as shown in Figure 4 The method includes the following steps: 1. Signal acquisition.

[0042] Voltage and current signals are acquired from the converter controller, and the sampling frequency is kHz level.

[0043] 2. Feature extraction.

[0044] Within a fixed time window (for example, 100 ms), the following characteristic quantities are calculated: Voltage and current effective values: and ; Active power: wherein N is the number of sampling points, v(n) and I(n) are the voltage and current values (discrete sampling signals) of the nth sampling point, respectively; Power factor: ; Total harmonic distortion: wherein I h is the effective value of the hth harmonic current, and I1 is the effective value of the fundamental current.

[0045] Power variance: wherein P(n) is the active power, and is the average value of the active power.

[0046] Instantaneous power change rate:

[0047] 3. Threshold setting.

[0048] Light load and heavy load thresholds: 20% and 80% of the rated power, respectively; Dynamic impact threshold: wherein μ,σ are the sliding mean and standard deviation corresponding to the dP / dt statistical quantity, μ is the mean value of dP / dt (usually close to 0, because the power change rate fluctuates up and down when in steady state); The standard deviation of σ = dP / dt is used to measure the "amplitude" of change. k = 2~3, usually 3, is used as a "multiplier factor" to determine when the rate of change exceeds normal fluctuations.

[0049] Setting the threshold in this way is more reasonable than using a fixed constant, because it can adapt to different operating conditions.

[0050] 4. Logical classification.

[0051] Define light load, normal load, and heavy load conditions for power P; Below 20% of the rated power is considered a light load. A load exceeding 80% of the rated power is considered heavy load. The others are considered normal operating conditions.

[0052] 5. Enhanced dynamic impact discrimination.

[0053] If the dynamic impact threshold condition is met, it is preferentially determined to be a dynamic impact: significantly greater than the steady-state reference value.

[0054] 6. PF (Power Factor) – an auxiliary correction factor.

[0055] Under steady-state conditions: If PF ≥ 0.8, it is marked as "ideal load"; If PF < 0.8, it is marked as "distorted load".

[0056] 7. Output the results.

[0057] The final output operating condition category includes, for example: Figure 6 The “light load steady state (ideal / distorted)” shown is as follows: Figure 7 The “ordinary steady state (ideal / distorted)” shown, such as Figure 8 The “heavy load steady state (ideal / distorted)” shown and as Figure 9 The "dynamic impact" shown.

[0058] Step 102: Steady-state judgment and gating trigger.

[0059] The system determines whether the current operating condition belongs to the "steady state" set. If it is a "dynamic shock", the process is suspended, waiting for the next judgment cycle. If it is any kind of "steady state", a trigger signal is generated to start the subsequent fine measurement process.

[0060] Step S103: High-frequency synchronous data acquisition.

[0061] Upon receiving the trigger signal, the data acquisition module (DAQ) immediately and synchronously acquires a snapshot of the bus voltage v_bus(t), i_A(t), and i_B(t) at a high sampling rate (e.g., 1 MSPS), covering a duration of dozens of switching cycles.

[0062] Step S104: Differential current reconstruction.

[0063] The parameter identification module calculates the collected data as i_dut(t) = i_A(t) - i_B(t), thus obtaining the time-domain waveform of the current flowing purely through the capacitor bank C_dut under test.

[0064] Step S105: Spectrum analysis.

[0065] Apply a window function (such as the Hanning window) to the AC component of v_bus(t) and i_dut(t) and then perform an FFT to obtain the complex spectrum V_bus(f) and I_dut(f).

[0066] Step S106: Calculation of complex impedance at multiple frequencies.

[0067] The system automatically searches the frequency f_sw and the precise peak position f_m of its m harmonics (m=1, 2, ..., M) in the frequency spectrum. At each f_m, the complex values ​​V_bus(fm) and I_dut(fm) are extracted, and M complex impedance measurements are calculated according to the formula Z(fm) = V_bus(fm) / I_dut(fm).

[0068] Step S107: Parameter identification.

[0069] This step is performed by the parameter identification module, which uses a strategy that combines step-by-step and fitting methods to solve for the three parameters in the capacitance equivalent model with high accuracy.

[0070] ESR Estimation: This invention employs a weighted average method. Considering that the signal-to-noise ratio of measurement data at different frequency points may vary, the real parts Re{Z(f_m)} of the M complex impedances obtained in the preceding steps are weighted and averaged to obtain the estimated ESR value ESR_est. The weighting strategy can be determined based on the current amplitude or signal-to-noise ratio at each frequency point.

[0071] C and ESL estimation: Since the equivalent series resistance ESR mainly contributes to the real part of the impedance and changes little with frequency, it has been estimated using a weighted average method. In the C and ESL fitting, to simplify the optimization model and improve convergence speed and stability, this invention establishes a nonlinear least-squares cost function with capacitance C and inductance ESL as optimization variables, the mathematical expression of which is: Where, Im{Z(f_m)} represents the imaginary part of the complex impedance Z(f_m), f m is the mth harmonic frequency point, ESL is the equivalent series inductance, C is the capacitance of the capacitor C, ESL is the equivalent series inductance, M is the total number of harmonic frequency points used, and J(C, ESL) is a nonlinear least squares cost function (i.e., an optimization objective function).

[0072] Subsequently, a Levenberg-Marquardt or other nonlinear optimization algorithm is used to take the nominal value of the capacitor as the initial value of iteration, and by minimizing the above cost function J, the final solution of the capacitance estimate C_est and the equivalent series inductance estimate ESL_est that best matches the model and the measured data is obtained.

[0073] Step S108: Health status assessment.

[0074] The storage and output module retrieves the health baseline {ESR_initial_heavy (initial value of ESR under heavy load), C_initial_heavy (initial value of C under heavy load)} at the corresponding temperature from the database according to the current operating condition (e.g., "heavy load steady state") obtained in step S101. By comparing the increments of ESR_est and ESR_initial_heavy and the decrements of C_est and C_initial_heavy, a comprehensive health index (SOH) is calculated, and a pre-warning of aging is issued according to a pre-set threshold value.

[0075] The calculation formula of the comprehensive health index (SOH) is: The health index (SOH_C) based on the capacitance SOH_C (%) = [ (C_current - C_EOL) / (C_initial - C_EOL) ] * 100% C_current: the capacitance value estimated online at present.

[0076] C_initial: the capacitance value of the capacitor in the initial state (when new), i.e., the "health baseline".

[0077] C_EOL: the capacitance threshold value of the capacitor at the end of life (End of Life).

[0078] The health index (SOH_ESR) based on the equivalent series resistance SOH_ESR (%) = [ (ESR_EOL - ESR_current) / (ESR_EOL - ESR_initial) ] *100% ESR_current: Current online estimated ESR value.

[0079] ESR_initial: Initial ESR value of the capacitor (when brand new), i.e. the "healthy baseline".

[0080] ESR_EOL: ESR threshold value when the capacitor health is terminated.

[0081] Final overall health index (SOH_Total) SOH_Total = min( SOH_C, SOH_ESR ) This SOH_Total index intuitively reflects the "remaining life percentage" of the capacitor relative to its brand new state, when SOH_Total drops to 0%, the capacitor is considered to have reached End of Life (EOL) state.

[0082] Method to determine the capacitor C_initial and ESR_initial (healthy baseline): Device commissioning phase self-learning: This is the most accurate method. When your device (converter) is first powered on for commissioning, run a capacitor parameter detection once, and store the measured initial values as the initial values specific to this device into non-volatile memory.

[0083] Method to determine the capacitor C_EOL and ESR_EOL (healthy termination threshold): Follow industry standards / manufacturer recommendations: Generally, the health termination standards for electrolytic capacitors are: C_EOL = 80% of the initial nominal value (i.e. 20% drop); ESR_EOL = 200% ~ 300% of the initial upper limit value (i.e. 2 to 3 times the original value).

[0084] Preferably, the macroscopic electrical characteristic quantities include one or more of the effective values of input / output voltage / current, active / reactive power, power change rate (dP / dt), total harmonic distortion (THD), and the working condition types include light load steady state, heavy load steady state, dynamic impact.

[0085] Preferably, the differential current reconstruction includes: obtaining the path A current i_A(t) and path B current i_B(t) synchronously sampled by the data acquisition module, calculating i_dut(t) = i_A(t) - i_B(t), and constructing the net ripple current signal i_dut(t) that purely flows through the capacitor group to be measured.

[0086] Preferably, the spectrum analysis includes applying a Fast Fourier Transform (FFT) algorithm to the reconstructed DC under test current idut(t) and the synchronously sampled bus voltage v bus(t) to obtain their complex spectra in the frequency domain, Idut(f) and V bus(f). Idut(f) can also be understood as the frequency domain representation of the DC under test current.

[0087] Preferably, the multi-frequency point impedance calculation includes using the switching harmonics generated by the operation of the converter itself as an excitation source, and calculating the complex impedance of the DC under test capacitor at at least two harmonic frequency points f m (harmonic frequency points with frequency values f m) with energy greater than a predetermined threshold according to Ohm's law: Z(f m) = V bus(f m) / Idut(f m).

[0088] Preferably, the harmonic frequency points f m include the switching frequency fundamental wave, the second harmonic, and the third harmonic.

[0089] Preferably, the parameter identification includes taking the calculated multi-frequency point complex impedance data (f m, Z(f m)) as input, using a numerical optimization algorithm such as a nonlinear least squares method to curve fit the multi-frequency point impedance data, and solving for the equivalent series resistance (ESR), capacitance (C), and equivalent series inductance (ESL) of the DC under test capacitor.

[0090] The health assessment and output module stores or establishes a multi-dimensional health baseline database through online learning, which records the standard reference values of ESR and C of the capacitor in a healthy state under different steady-state operating conditions (such as different load rates) and different operating temperatures; the state assessment and aging warning module matches the latest calculated ESR and C values from the parameter identification module with the operating condition type identified by the operating condition classification module and the average / maximum temperature measured by the temperature sensing unit, calls the corresponding health baseline from the database, and outputs relevant information according to the assessment results.

[0091] In this embodiment, the software algorithm running inside the processor module (CPU) is the key to realizing the functions of the application, and the complete monitoring method is as shown in the attached Figure 3 (flowchart of the main algorithm). Among them, S101-S102 are realized by the operating condition classification module, S103-S107 are realized by the parameter identification module, and S108 is realized by the health assessment and output module.

[0092] The embodiment introduces a working condition adaptive intelligent gating mechanism. By analyzing macro electrical parameters (such as power, rate of change dP / dt, harmonic distortion THD) in real time, the system can accurately classify the operating state of the converter into predefined types such as "light load steady state", "heavy load steady state" or "dynamic impact". This mechanism ensures that the high-precision parameter identification algorithm is only triggered to execute in the "steady state" time window with stable and reliable data, and is automatically shielded in the "dynamic impact" working condition with severe load changes. This effectively eliminates the fatal interference of non-steady state working conditions on the diagnosis results, greatly improves the robustness and reliability of the monitoring system in real industrial environments, and avoids false alarms.

[0093] Embodiment three: the embodiment should be understood as at least containing all the features of any one of the preceding embodiments; The embodiment provides a simulation verification method for a working condition classification module. To verify the working condition classification module proposed in the application, different operating states of power electronic converters can be effectively distinguished, so that the adaptive triggering of the diagnosis process is realized. The application uses MATLAB software environment to simulate and verify the function of the module. The simulation aims to prove that through real-time analysis of macro electrical parameters, the application can accurately identify "steady state" working conditions suitable for precise diagnosis and effectively avoid "dynamic impact" working conditions that may introduce large errors.

[0094] The simulation verification mainly includes the following steps: Step (S201): Construct a typical working condition signal model In the simulation environment, a reference sinusoidal voltage signal V(t) is first defined. Then, four current signals I(t) representing typical operating states of the converter are constructed, including: "Light load steady state" working condition: simulate a low-amplitude, waveform-stable sinusoidal current.

[0095] "Heavy load steady state" working condition: simulate a high-amplitude, waveform-stable sinusoidal current.

[0096] "Normal steady state" working condition: simulate a sinusoidal current between light load and heavy load, with stable waveform.

[0097] "Dynamic impact" working condition: simulate a non-stationary current with a sharp step change in current amplitude within a sampling window, representing a scenario of sudden load increase or decrease.

[0098] Step (S202): Macro electrical feature extraction The simulation program simulates the working condition classification module of the present application. In a preset time window (for example, 0.1 seconds), the voltage V(t) and current I(t) signals in each working condition described above are calculated in real time to obtain a series of macro electrical characteristic parameters that can represent the running state. These parameters mainly include: Power-related parameters: Calculate the current effective value (Irms) and active power (P), and compare them with the preset "light load power threshold" (Threshold_Light) and "heavy load power threshold" (Threshold_Heavy) based on the rated current. This is the core basis for determining the load level.

[0099] Dynamic-related parameters: Calculate the instantaneous power Pinst(t) = V(t) * I(t), and extract its rate of change with time (dP / dt) and its statistical variance (SigmaP). These two parameters are used as key indicators to determine whether the working condition is stable. A dP / dt peak far exceeding the statistical average or a large power variance indicates that the system is in dynamic change.

[0100] Waveform distortion-related parameters: Calculate the total harmonic distortion (THDi) of the current by fast Fourier transform (FFT) to distinguish between ideal linear loads and nonlinear loads.

[0101] Step (S203): Rule-based working condition classification The simulation program internally incorporates a set of expert decision-making rules based on thresholds that are the same as the working condition classification module. The logical priority of the rules is as follows: First, determine the dynamic impact: First, check the dynamic-related parameters. If dP / dt or SigmaP exceeds the adaptively set dynamic threshold (Threshold_Dynamic), the current working condition is immediately classified as "dynamic impact" regardless of the power size.

[0102] Second, determine the load level: If the working condition is stable, the working condition is preliminarily classified as "light load steady state", "heavy load steady state", or "ordinary steady state" according to the comparison results of the active power P and Threshold_Light and Threshold_Heavy.

[0103] Refine the load type: Based on the steady state, the working condition can be further described according to the size of the power factor (PF) or THDi, such as "heavy load steady state (ideal load)" or "heavy load steady state (distorted load)".

[0104] Simulation data output: Working Condition 1: Light Load Steady State Vrms=220.00 V, Irms=2.00 A, P=437.80 W, PF=1.00, THDi=59.81 % dP / dt=27.64, threshold=58.69, sigma 2 P=96896.90 Case 2: Normal steady state Vrms=220.00 V, Irms=10.00 A, P=2175.30 W, PF=0.99, THDi=60.06 % dP / dt=138.20, threshold=293.46, sigma 2 P=2422422.42 Case 3: Heavy steady state Vrms=220.00 V, Irms=20.00 A, P=4312.29 W, PF=0.98, THDi=60.42 % dP / dt=276.36, threshold=586.89, sigma 2 P=9689689.69 Case 4: Dynamic impact Vrms=220.00 V, Irms=14.58 A, P=2703.40 W, PF=0.84, THDi=94.78 % dP / dt=276.36, threshold=427.74, sigma 2 P=7738779.26 Simulation waveform output: The simulation results show that the proposed working condition classification method is completely effective and reliable.

[0105] For the first three steady state conditions, the simulation program can accurately classify them as "light load steady state", "normal steady state" and "heavy load steady state" according to the power size.

[0106] Especially critical is that for the fourth "dynamic impact" condition, although its power also spans the light load and heavy load intervals, due to the dramatic change in current, dP / dt and SigmaP parameters instantaneously increase and break through the dynamic threshold, the simulation program successfully and preferentially identifies it as "dynamic impact".

[0107] The embodiment realizes multi-parameter, high-precision synchronous identification and comprehensive health evaluation. In the parameter identification stage, the algorithm uses the switch harmonic of the transformer itself as a "natural excitation source", and solves the three key health parameters of equivalent series resistance (ESR), capacitance (C) and equivalent series inductance (ESL) through multi-frequency point impedance spectrum measurement and nonlinear and least square fitting. In the health evaluation stage, the comprehensive health index (SOH_Total) is innovatively used, which takes the minimum value of SOH_C and SOH_ESR, and ensures that the evaluation result always reflects the "short board" dimension of the first performance degradation. Combined with the working condition and temperature adaptive health baseline database, dynamic and multi-dimensional fine health state evaluation is realized, which provides comprehensive and reliable data support for predictive maintenance.

[0108] Embodiment four: the embodiment should be understood as at least containing all the features of any one of the preceding embodiments; The embodiment provides a simulation verification method of a parameter identification process. In order to further verify the feasibility and accuracy of the technical solutions proposed in the application, the application uses MATLAB scientific calculation software to build a simulation platform, and simulates and verifies the core identification process of the application. The simulation verification process is as follows: Step (S301): first, a signal model consistent with the actual working condition is constructed in the simulation environment. A direct current bus voltage signal with a sampling rate of 1 MSPS is generated, which contains a direct current component of 400V and superimposes an alternating current ripple component corresponding to the 20kHz switching frequency and its 1st to 5th harmonics. In order to simulate the "double path" architecture of the application, according to a capacitor model with a preset aging parameter (for example, C_true = 4250uF, ESR_true = 25mOhm), the theoretical true current i_dut_true(t) is calculated. At the same time, a strong common mode current i_common(t) with an amplitude much larger than i_dut_true(t) is generated to simulate the inverter noise. Finally, the signal i_A(t) measured by sensor A is synthesized, i.e. i_A(t) = i_dut_true(t) + i_common(t) + noise(t), and the signal i_B(t) measured by sensor B is synthesized, i.e. i_B(t) = i_common(t) + noise(t).

[0109] Step (S302): the simulation program performs difference operation on the collected i_A(t) and i_B(t) to obtain the reconstructed capacitor current i_dut_reconstructed(t). The simulation results show that the reconstructed current is highly consistent with the theoretical true current i_dut_true(t), which proves that the difference method of the application can effectively suppress strong common mode interference.

[0110] Steps (S303 and S304): Next, the simulation program applies a Hanning window function to the reconstructed current i_dut_reconstructed(t) and the voltage ripple signal and performs a Fast Fourier Transform (FFT). At the pre-set harmonic frequency points (20 kHz, 40 kHz,..., 100 kHz), the complex voltage and complex current spectral values are extracted, and a plurality of discrete complex impedance measurements Z_measured(f_m) are calculated through complex division.

[0111] Step (corresponding to S305): Finally, the parameter identification phase is entered. By averaging the real part of Z_measured(f_m), the estimated value of ESR is obtained. Then, a non-linear least squares optimization algorithm (e.g., Levenberg-Marquardt algorithm) is used to curve fit the imaginary part of Z_measured(f_m), thereby solving the estimated values of capacitance C and equivalent series inductance ESL at one time.

[0112] Simulation data output: --- True parameters (Aged Capacitor) --- C_true = 4250.0 uF, ESR_true = 25.0 mOhm, ESL_true = 15.0 nH Step S301: Simulate DAQ complete. Step S302: Differential current reconstruction: i_dut(t) = i_A(t) - i_B(t). Step S303: Window function processing and Fast Fourier Transform complete. Step S304: Complex impedance s calculated at harmonic frequencies. Step S305: Parameter identification complete. --- Estimated results --- ESR Estimated: 25.0 mOhm (Error: -0.01%) C Estimated: 4102.2 uF (Error: -3.48%) ESL Estimated: 15.1 nH (Error: 0.71%) Figure 10The simulation results of the differential current reconstruction of the application are shown. In the figure, the total current i_A(t) (gray curve) has a large amplitude and is submerged by noise, while the current i_dut_reconstructed(t) (black dotted line) reconstructed after the method of the application almost perfectly covers the theoretical true current i_dut_true(t) (red solid line), proving the excellent noise suppression capability of the application.

[0113] Figure 11 The final parameter identification results are shown. In the figure, the discrete marker points (circles and squares) represent the impedance measurement values containing errors calculated from the simulation signals. The solid line represents the theoretical impedance curve drawn according to the finally estimated ESR and C, ESL parameters according to the method of the application. As shown in the figure, the theoretical curve accurately passes through all the measurement points, and the error of the finally estimated parameter value compared with the preset true value is less than 1%.

[0114] In the estimation results, the error of C (-3.48%) is slightly larger than ESR and ESL, which is mainly due to the fact that the impedance modulus of the capacitor at low frequency is extremely large, resulting in a weak current signal flowing through and a relatively low signal-to-noise ratio, which poses a certain challenge to the accuracy of FFT calculation. The error is within the acceptable range of engineering, proving the effectiveness of the method.

[0115] The above simulation results strongly prove that the system and method proposed by the application can realize high-precision and high-robustness online identification of the health state parameters of the capacitor under strong noise background, and has significant technical advantages and industrial practical value.

[0116] The detailed MATLAB simulation verification of the embodiment strongly proves the feasibility and accuracy of the whole technical scheme. The simulation results show that even under the submersion of strong common-mode interference noise, the differential current reconstruction algorithm can almost perfectly restore the real capacitor current waveform. The final multi-parameter identification result error is less than 1% (ESR error -0.01%, ESL error 0.71%), and the capacitance estimation error -3.48% is also within the completely acceptable range of engineering. The simulation not only verifies the correctness of the theoretical model, but also highlights the excellent anti-noise capability and engineering practical value of the scheme, providing strong evidence for its deployment in actual industrial products.

[0117] Embodiment five: this embodiment should be understood as at least containing all the features of any one of the preceding embodiments; this embodiment provides a simulation verification method of a health index algorithm, in order to verify the effectiveness of the comprehensive state of health (SOH) evaluation model proposed in the present application, the model aims to convert multiple physical parameters such as identified capacitance (C) and equivalent series resistance (ESR) into a single, standardized percentage health index. The present application uses MATLAB software environment to simulate and verify the calculation logic and "short board effect" principle of the SOH algorithm.

[0118] The simulation verification mainly includes the following steps: Step S401: define the health index (SOH) calculation model In the simulation program, the SOH calculation formula of the present application is first solidified. The model is based on linear normalization method, which maps the aging process of capacitance to the health interval of 100% to 0%.

[0119] Capacitance-based health index (SOH_C): SOH_C = [(C_current - C_EOL) / (C_initial - C_EOL)] * 100%.

[0120] Where C_current is the current measurement value, C_initial is the initial health baseline, and C_EOL is the preset health termination threshold. The formula represents the remaining health of the capacitance.

[0121] ESR-based health index (SOH_ESR): SOH_ESR = [(ESR_EOL - ESR_current) / (ESR_EOL - ESR_initial)] * 100%.

[0122] Where ESR_current is the current measurement value, ESR_initial is the initial health baseline, and ESR_EOL is the health termination threshold. The formula represents the remaining health of the ESR.

[0123] Final comprehensive health index (SOH_Total): SOH_Total = min(SOH_C, SOH_ESR).

[0124] The min function is used to follow the "wooden barrel theory" or "weakest link" principle in engineering safety, ensuring that SOH can reflect the first deteriorated performance dimension of the capacitor.

[0125] Step S402: Setting baseline parameters and simulating aging process At the beginning of the simulation, a set of reasonable initial health baseline (e.g., C_initial = 4700uF, ESR_initial = 10mOhm) and health termination threshold (e.g., C_EOL is 80% of the initial value, ESR_EOL is 300% of the initial value) are set, which are in line with industry standards.

[0126] Subsequently, an iterative loop is used to simulate the natural aging process of the capacitor over thousands of working cycles. In each cycle, the value of C_current decreases linearly at a slower rate, while the value of ESR_current increases linearly at a relatively faster rate. This uneven aging rate setting aims to simulate the common scenario in practical applications where the ESR deteriorates faster than the capacitance due to the intensification of internal losses.

[0127] Step S403: Dynamically calculating and recording SOH evolution In each simulated working cycle, the simulation program calls the SOH calculation model defined in Step 1 to calculate SOH_C, SOH_ESR, and the final SOH_Total in real time based on the current simulated C_current and ESR_current values, and records the trajectory of its evolution over time.

[0128] Simulation data output: --- Simulation parameters --- Initial state: C = 4700 uF, ESR = 10.0 mOhm Health termination threshold: C <= 3760 uF, ESR >= 30.0 mOhm --- Simulation results --- Capacitor failure EOL (SoH <= 0%) cycle number: 2748 EOL state: C = 4211 uF, ESR = 30.0 mOhm Simulation waveform output: Figure 12 The simulation of the degradation curve of the capacitor physical parameters is shown, Figure 13 The results of this SOH algorithm simulation are shown.

[0129] In the figure, the curve representing SOH_C (e.g., blue dashed line) shows a relatively flat decline in health due to the slow decline in capacitance.

[0130] The curve representing SOH_ESR (e.g., orange dotted line) shows a faster decline due to the faster ESR degradation rate.

[0131] The most critical result is embodied in the curve representing the final overall health index SOH_Total (e.g., bold black solid line). As shown, this SOH_Total curve completely follows and covers the SOH_ESR curve, which declines at a faster rate.

[0132] This embodiment focuses on the verification and aging prediction of the health index (SOH) algorithm. The simulation successfully reproduces the real failure mode by simulating the unbalanced aging process of the capacitor (rapid rise of ESR and slow decline of C). The simulation results clearly show that the overall health index SOH_Total curve based on the "weakest link" principle completely follows and covers the SOH_ESR curve, which declines at a faster rate. This proves that the algorithm can accurately capture the performance "short board" of the system and give a more conservative and safer health status evaluation, effectively avoiding the problem that potential risks may be covered due to parameter averaging, providing a core algorithm guarantee for accurate end-of-life (EOL) prediction and predictive maintenance.

[0133] The simulation results strongly prove the rationality and advancement of the overall health index SOH algorithm proposed in the present application. By using the min() function, SOH_Total can accurately capture the "short board" of the capacitor performance (rapid degradation of ESR in this case) and give a more conservative and reliable health status evaluation that conforms to the engineering safety reality. This method avoids the potential covering of potential risks due to parameter averaging, providing a solid data foundation for accurate predictive maintenance.

[0134] The above disclosed content is only the preferred feasible embodiment of the present application, and does not limit the protection scope of the present application, so any equivalent technical changes made according to the content of the present application specification and drawings are included in the protection scope of the present application, and in addition, the elements can be updated as technology develops.

Claims

1. A capacitance online health monitoring system based on dual-path differential sensing and adaptive operating conditions, characterized in that: The system includes a sensing module, a data acquisition module, and a processor module. The sensing module includes a first current sensor for measuring the total current including the capacitor under test, other parallel capacitors, and downstream loads; a second current sensor for measuring the total current flowing to all parallel branches except the capacitor bank under test; a voltage sensor for synchronously measuring the bus voltage; and a multi-point temperature sensing unit for capturing the temperature gradient of the capacitor bank caused by its layout and air duct. The data acquisition module is connected to all sensors and is responsible for high-precision, synchronous sampling and analog-to-digital conversion of the output signals of each sensor at a preset high sampling rate. The processor module is used to execute software algorithms.

2. The online capacitance health monitoring system based on dual-path differential sensing and adaptive operating conditions as described in claim 1, characterized in that, The first current sensor is installed on the DC bus total current path upstream of the capacitor bank under test, the second current sensor is installed downstream of the capacitor bank under test and on the upstream branch path of other parallel capacitor banks and load, and the voltage sensor is connected to both ends of the DC bus.

3. A method for online capacitance health monitoring based on dual-path differential sensing and adaptive operating conditions, characterized in that, Based on the processor module of the capacitor online health monitoring system of claim 2, the method includes using the operating condition classification module to analyze the macroscopic state of the converter in real time, comparing the calculated macroscopic electrical characteristic quantities with a preset threshold, and classifying the current operating state of the converter into a predefined operating condition type in real time. When the operating condition classification module determines that the current operating condition is steady state, the parameter identification module is triggered to perform differential current reconstruction, spectrum analysis, multi-frequency impedance calculation and parameter identification. The health assessment and output module analyzes and evaluates the parameters identified by the parameter identification module and outputs relevant information based on the assessment results.

4. The online capacitance health monitoring method based on dual-path differential sensing and adaptive operating conditions as described in claim 3, characterized in that, The macroscopic electrical characteristics include one or more of the following: effective values ​​of input / output voltage / current, active / reactive power, power change rate, and total harmonic distortion. The operating conditions include light-load steady state, heavy-load steady state, and dynamic impact.

5. The online capacitance health monitoring method based on dual-path differential sensing and adaptive operating conditions as described in claim 4, characterized in that, The differential current reconstruction includes: acquiring the path A current i_A(t) and path B current i_B(t) synchronously sampled by the data acquisition module, calculating i_dut(t) = i_A(t) - i_B(t), and constructing the net ripple current signal i_dut(t) flowing through the capacitor bank under test.

6. The online capacitance health monitoring method based on dual-path differential sensing and adaptive operating conditions as described in claim 5, characterized in that, Spectrum analysis involves applying the Fast Fourier Transform algorithm to the reconstructed net current i_dut(t) of the capacitor under test and the synchronously sampled bus voltage v_bus(t) to obtain their complex spectra I_dut(f) and V_bus(f) in the frequency domain.

7. The online capacitance health monitoring method based on dual-path differential sensing and adaptive operating conditions as described in claim 6, characterized in that, Multi-frequency impedance calculation involves using the switching harmonics generated during the operation of the converter as the excitation source, and calculating the complex impedance of the capacitor under test at at least two harmonic frequency points f_m with energy greater than a predetermined threshold according to Ohm's law: Z(f_m) = V_bus(f_m) / I_dut(f_m).

8. The online capacitance health monitoring method based on dual-path differential sensing and adaptive operating conditions as described in claim 7, characterized in that, The harmonic frequency point f_m includes the fundamental frequency, the second harmonic, and the third harmonic.

9. The online capacitance health monitoring method based on dual-path differential sensing and adaptive operating conditions as described in claim 8, characterized in that, The parameter identification includes a second-order equivalent circuit model of a capacitor, Z(f) = ESR + j(2πf·ESL - 1 / (2πf·C)), where ESR is the equivalent series resistance of the capacitor under test, C is the capacitance of the capacitor under test, ESL is the equivalent series inductance, f is the frequency, and j represents the imaginary part. The calculated multi-frequency impedance data (f_m, Z(f_m)) are used as input, and numerical optimization algorithms such as nonlinear least squares are used to perform curve fitting on the multi-frequency impedance data to solve for the three key health parameters of the capacitor under test: equivalent series resistance, capacitance, and equivalent series inductance.

10. The online capacitance health monitoring method based on dual-path differential sensing and adaptive operating conditions as described in claim 9, characterized in that, The health assessment and output module first stores or establishes a multi-dimensional health baseline database through online learning. This database records the standard reference values ​​of the ESR and C values ​​of the capacitor under different steady-state operating conditions and different operating temperatures when it is in a healthy state. The condition assessment and aging early warning module matches the latest ESR and C values ​​calculated by the parameter identification module with the operating condition type identified by the current operating condition classification module and the average / maximum temperature measured by the temperature sensing unit. It then retrieves the corresponding health baseline from the database and outputs relevant information based on the assessment results.

Images