A method and system for online monitoring of insulation faults of a medical IT system control cabinet

Abstract

This invention relates to the field of online monitoring technology, specifically to an online monitoring method and system for insulation faults in a medical IT system control cabinet. The method includes: selecting a non-integer harmonic detection frequency based on the equivalent distributed capacitance to ground from the previous monitoring cycle, and superimposing it with the fundamental frequency to form a composite signal injection system; acquiring signals and calculating the gradient of the instantaneous harmonic distortion rate of the voltage, and selecting a time window to window the leakage current accordingly; calculating the insulation resistance and distributed capacitance based on the discreteness estimation signal-to-noise ratio weighted fusion of the current vector; constructing a four-dimensional original feature vector composed of insulation parameters, phase difference, and intermodulation component energy; constructing a diagonally weighted matrix transformation feature vector based on the resistance-capacitance ratio and gradient, and determining the fault attribution by calculating the Mahalanobis distance with a fault template library. In other words, the solution of this invention can avoid harmonic interference, improve computational stability, quickly respond to instantaneous changes in insulation, and accurately distinguish different fault types to guide maintenance.

Description

Technical Field

[0001] This invention relates to the field of online monitoring technology. More specifically, this invention relates to a method and system for online monitoring of insulation faults in a medical IT system control cabinet. Background Technology

[0002] Medical IT systems typically employ ungrounded (IT) power supply, which offers the advantage of maintaining power even during an initial single-phase ground fault, ensuring the normal operation of medical equipment. However, this method also carries a potential risk: if the initial ground fault is not detected and addressed promptly, a subsequent second non-phase ground fault can create a phase-to-phase short circuit, leading to a power outage. Therefore, to prevent escalation of faults and subsequent power outages, real-time monitoring of the system's insulation status is essential.

[0003] To address the aforementioned issues, existing insulation fault monitoring technologies typically employ low-frequency signal injection. This method injects a probe signal of a specific frequency into the IT system and measures the magnitude and phase of the system's leakage current to ground, thereby calculating the system's equivalent insulation resistance to ground and distributed capacitance. Based on these parameters, the system can determine the current insulation level, thus achieving a certain degree of insulation fault monitoring.

[0004] However, due to the presence of numerous power electronic devices in medical environments, which generate significant harmonic and noise interference during operation, existing single-frequency injection methods are susceptible to interference from the same or near frequencies, leading to a significant reduction in the signal-to-noise ratio of the measured signal and potentially causing false alarms or missed alarms. Furthermore, the evolution of insulation faults is diverse, manifesting as slow degradation or sudden changes. Existing technologies, which use data processing windows for signal analysis, struggle to simultaneously address both rapid response to transient faults and accurate measurement of steady-state signals. Moreover, monitoring methods rely solely on insulation resistance values, failing to effectively differentiate between purely resistive degradation, capacitive changes, or parallel resistance-capacitance fault modes. This results in limited fault diagnosis information, hindering maintenance personnel from locating and troubleshooting faults. Summary of the Invention

[0005] The purpose of this invention is to provide an online monitoring method and system for insulation faults in medical IT system control cabinets, in order to solve the problem that existing technologies cannot effectively distinguish fault types; to this end, this invention provides solutions in the following two aspects.

[0006] In a first aspect, the present invention provides an online monitoring method for insulation faults in a medical IT system control cabinet, comprising:

[0007] Based on the equivalent ground capacitance of the previous monitoring period, the detection frequency is selected from the preset set of non-integer harmonic frequencies. , baseband The sinusoidal signal and the detection frequency The sinusoidal signals are superimposed to form a composite detection signal and injected into the medical IT system; the leakage current to ground and the bus voltage signal are collected, the instantaneous harmonic distortion rate change gradient of the bus voltage is calculated, and the leakage current signal is windowed according to the change gradient between the first time window parameter and the second time window parameter; the frequency within each time window is calculated. and The leakage current vector at a given location is used to weight and fuse the leakage current vectors of each time window based on the signal-to-noise ratio estimated by the dispersion. The equivalent insulation resistance to ground for the current cycle is then calculated based on this weighted fusion. and equivalent distributed capacitance to ground ; Construct by , Impedance at frequency and The phase difference at the point, and the leakage current signal from and The four-dimensional original eigenvector is composed of the energy norms of the generated intermodulation components; based on and Based on the ratio and the gradient of change, a fourth-order diagonal weighted matrix is ​​constructed, and the original feature vector is transformed to obtain a weighted fault feature vector. The weighted fault feature vector is then compared with a preset fault template library to calculate the Mahalanobis distance and determine the fault attribution of the current insulation state.

[0008] Preferably, the step of selecting the detection frequency from a preset set of non-integer multiples of the frequency range... This includes: assigning each candidate frequency in the set to... Substitute into the impedance prediction model Calculate the predicted impedance magnitude at each candidate frequency, where The insulation resistance of the previous cycle. Given the distributed capacitance of the previous cycle; calculate the minimum frequency difference between each candidate frequency and the 2nd to 5th harmonic frequencies of the fundamental frequency; define the cost function. ,in =2,3,4,5 and Given preset weighting coefficients, the candidate frequency that minimizes the cost function is selected as the probe frequency. .

[0009] Preferably, the calculation of the instantaneous harmonic distortion rate change gradient of the bus voltage includes: performing a short-time Fourier transform on the acquired bus voltage signal sequence to obtain a spectrum that changes over time; calculating the total harmonic distortion rate (THD) at each time point based on the spectrum to obtain a THD time series; and calculating the difference between two adjacent THD values ​​in the THD time series to obtain the change gradient of the instantaneous harmonic distortion rate.

[0010] Preferably, the step of selecting between a first time window parameter and a second time window parameter to window the leakage current signal based on the changing gradient includes: if the absolute value of the instantaneous harmonic distortion rate change gradient of the bus voltage is less than a preset gradient discrimination threshold, then the first time window parameter is used, with a Hamming window type, a window length of 200ms, and an overlap rate of 50% for windowing; if the absolute value of the changing gradient is greater than or equal to the preset gradient discrimination threshold, then the second time window parameter is used, with a Hamming window type, a window length of 40ms, and an overlap rate of 75% for windowing.

[0011] Preferably, the step of weighting and fusing the leakage current vectors of each time window based on the dispersion-based signal-to-noise ratio estimation as weight includes: calculating the mean vector of the leakage current vectors; calculating the first... The square of the Euclidean distance between the leakage current vector and the mean vector in the nth time window is taken as the nth time window. The dispersion of the nth time window; the reciprocal of the dispersion is used as the nth time window. The signal-to-noise ratio estimate for the nth time window is calculated and normalized. The weights of each time window are calculated; the leakage current vectors of each time window are weighted and fused using the calculated weights to obtain the fused leakage current vector.

[0012] Preferably, the energy norm of the intermodulation component is calculated as follows: a fast Fourier transform is performed on the acquired leakage current signal to obtain the signal spectrum; the intermodulation frequency point is determined as... and Extract the amplitude of all spectral components within a 20Hz bandwidth centered at the intermodulation frequency point; take the square root of the sum of the squares of the amplitudes of all spectral components within the bandwidth to obtain the energy norm of the intermodulation component.

[0013] Preferably, the one based on and Based on the ratio and the gradient of change, a fourth-order diagonal weighted matrix is ​​constructed, including: calculating the ratio. and gradient of change Construct a fourth-order diagonal weighted matrix, where the diagonal elements are: ; ; ; ;in, and They are respectively and Feature weights The weights for the phase difference features, The weights are the energy norm features of the intermodulation components.

[0014] Preferably, the transformation of the original feature vector to obtain the weighted fault feature vector specifically involves: left-multiplying the four-dimensional original feature vector by the fourth-order diagonal weighting matrix to obtain the weighted fault feature vector, so as to... and Increase when the ratio is small and The weights of the intermodulation component energy norms are increased when the gradient is large.

[0015] Preferably, the step of calculating the Mahalanobis distance between the weighted fault feature vector and a preset fault template library to determine the fault attribution of the current insulation state includes: the preset fault template library contains three categories: normal, metallic grounding fault, and capacitive grounding fault; each category is defined by the mean and covariance matrix of the feature vector obtained through statistical analysis of a large amount of experimental data; the weighted fault feature vector of the current period is substituted into the Mahalanobis distance formulas of the three categories to calculate three Mahalanobis distance values; the category corresponding to the smallest Mahalanobis distance value is selected as the fault attribution of the current insulation state.

[0016] In a second aspect, an online monitoring system for insulation faults in a medical IT system control cabinet includes:

[0017] The processor; a memory storing computer instructions for online monitoring of insulation faults in a medical IT system control cabinet, wherein when the computer instructions are executed by the processor, the system performs the aforementioned method for online monitoring of insulation faults in a medical IT system control cabinet.

[0018] The beneficial effects of this invention are as follows: By injecting composite detection signals with non-integer multiples of frequencies and combining them with weighted fusion processing based on signal-to-noise ratio estimation using signal discreteness, this invention avoids harmonic interference in the medical environment, improves the signal-to-noise ratio of leakage current signals, and thus enhances the stability of insulation resistance and distributed capacitance calculations. By utilizing the gradient of voltage harmonic distortion rate changes to select different analysis time windows, it enables rapid response to instantaneous changes in insulation state. A fault feature vector containing multi-dimensional information such as insulation parameters, phase difference, and intermodulation components is constructed. Using weighted transformation and pattern matching based on Mahalanobis distance, it can not only determine the occurrence of insulation faults but also distinguish between purely resistive and purely capacitive fault types, providing fault diagnosis information and guidance for fault investigation and maintenance. Attached Figure Description

[0019] Figure 1 This schematically illustrates the steps of the online monitoring method for insulation faults in the control cabinet of a medical IT system in this embodiment;

[0020] Figure 2 The schematic diagram illustrates the structural block diagram of the online monitoring system for insulation faults in the medical IT system control cabinet of this embodiment. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0022] like Figure 1 As shown in this embodiment, an online monitoring method for insulation faults in a medical IT system control cabinet includes the following steps:

[0023] Step S1: Select a detection frequency from a preset set of non-integer harmonic frequencies based on the equivalent ground distributed capacitance of the previous monitoring period. , baseband The sinusoidal signal and the detection frequency The sinusoidal signals are superimposed to form a composite detection signal and injected into the medical IT system; the leakage current to ground and the bus voltage signal are collected, the instantaneous harmonic distortion rate change gradient of the bus voltage is calculated, and the leakage current signal is windowed according to the change gradient between the first time window parameter and the second time window parameter.

[0024] Set base frequency For a low-frequency setting, such as 10Hz, a set of non-integer multiples of frequencies is predefined, such as 13Hz, 17Hz, 23Hz, and 29Hz. Based on the value of the distributed capacitance to ground calculated in the previous cycle, and combined with the known equivalent inductance, the impedance to ground at each frequency point in the above set is estimated. The frequency with the smallest impedance to ground is selected as the detection frequency for the current cycle. To obtain the maximum leakage current response, a digital signal processor is used. and Two sinusoidal digital signals of different frequencies are digitally superimposed and then injected into the neutral point and protective ground of the isolation transformer of the medical IT system through a transformer via a digital-to-analog converter and a power amplifier.

[0025] The total leakage current signal to ground on the protective grounding line and the voltage signal on the L / N bus are synchronously acquired using current and voltage sensors and digitized by an analog-to-digital converter. A short-time Fourier transform is performed on the acquired bus voltage signal sequence to obtain its time-varying spectrum. At each time point, the total harmonic distortion (THD) is calculated based on the spectrum, resulting in a THD time series. The gradient of the instantaneous harmonic distortion is obtained by calculating the difference between two adjacent THD values ​​in this time series. A gradient threshold is set, for example, 0.01% / ms. If the absolute value of the gradient is greater than this threshold, it indicates a possible abrupt change in the state, and a first time window parameter, such as a short time window of 200ms, is selected for rapid detection of the change; if it is less than or equal to the threshold, a second time window parameter, such as a long time window of 1000ms, is selected to obtain higher measurement accuracy under steady-state conditions. Based on the selected time window length, the continuous leakage current data stream is divided into data segments.

[0026] To select an optimal detection frequency that achieves both high measurement sensitivity (i.e., low impedance amplitude) and avoids harmonic interference, in one optional embodiment, the detection frequency is selected from a preset set of non-integer harmonic frequencies. This includes: assigning each candidate frequency in the set to... Substitute into the impedance prediction model Calculate the predicted impedance magnitude at each candidate frequency, where The insulation resistance of the previous cycle. Given the distributed capacitance of the previous cycle; calculate the minimum frequency difference between each candidate frequency and the 2nd to 5th harmonic frequencies of the fundamental frequency; define the cost function. ,in =2,3,4,5 and Given preset weighting coefficients, the candidate frequency that minimizes the cost function is selected as the probe frequency. .

[0027] Specifically, the impedance prediction model is a parallel RC circuit model that can predict the current impedance based on the state of the previous cycle. Assuming the fundamental frequency... The frequency is 50Hz, the insulation resistance of the previous cycle is 100kΩ, and the distributed capacitance is 1μF. The preset set of non-integer harmonic frequencies is {90, 130, 170, 210}Hz, with weighting coefficients... and All are set to 1. The 2nd to 5th harmonic frequencies of the fundamental frequency are 100, 150, 200, and 250 Hz, respectively.

[0028] by candidate frequency Taking 130Hz as an example, the predicted impedance amplitude |Z(130)| is calculated to be 1221Ω. The frequency differences between this frequency and each harmonic are calculated to be 30Hz, 20Hz, 70Hz, and 120Hz, with the minimum value being 20Hz. The cost function value for 130Hz is approximately 0.0508. The same calculation process is performed on all other candidate frequencies in the set, and the cost function values ​​are compared. For example, if the calculation result for 90Hz is 0.0550, the result for 170Hz is 0.0515, and the result for 210Hz is 0.0521, since the cost function value corresponding to 130Hz is the smallest, 130Hz is selected as the detection frequency for this measurement. .

[0029] To adjust the signal processing window parameters according to the stable state of the power grid to balance frequency resolution and time resolution, in an optional embodiment, the leakage current signal is windowed based on a selection between a first time window parameter and a second time window parameter. This includes: if the absolute value of the gradient of the instantaneous harmonic distortion rate change of the bus voltage is less than a preset gradient discrimination threshold, then the first time window parameter is used, with a Hamming window type, a window length of 200ms, and an overlap rate of 50%; if the absolute value of the gradient change is greater than or equal to the preset gradient discrimination threshold, then the second time window parameter is used, with a Hamming window type, a window length of 40ms, and an overlap rate of 75%. In this embodiment, the gradient discrimination threshold is set to 0.05.

[0030] When the power grid is stable, the instantaneous harmonic distortion rate of the bus voltage changes gradually, and the absolute value of the gradient is less than the preset gradient discrimination threshold. For example, assuming the currently calculated gradient is 0.02, since 0.02 is less than 0.05, the current condition is determined to be steady-state. At this time, the first time window parameter is selected to process the leakage current signal. The processing procedure is as follows: a 200ms segment of signal data is extracted, a Hamming window function is applied to the data, and the window is moved forward by 100ms, i.e., the 200ms × 50% overlap portion, to extract the next signal segment. The long time window and low overlap setting provides high frequency resolution, suitable for analyzing the spectrum of steady-state signals.

[0031] When a switching action or transient fault occurs in the power grid, the harmonic distortion rate of the bus voltage changes drastically, causing a sharp increase in the absolute value of the gradient. For example, assuming the currently calculated gradient is 0.1, since 0.1 ≥ 0.05, the current condition is determined to be transient. At this point, switching to the second time window parameter, the processing becomes: capturing a 40ms segment of signal data, applying a Hamming window function, sliding the window forward by 10ms (i.e., the 40ms × 75% overlap), and capturing the next signal segment. While this short time window and high overlap setting sacrifices frequency resolution, it provides high temporal resolution, enabling the detection of the instantaneous occurrence of a fault or disturbance and ensuring that the long time window does not average out transient characteristics.

[0032] Step S2: Calculate the intrinsic frequency within each time window. and The leakage current vector at a given location is used to weight and fuse the leakage current vectors of each time window based on the signal-to-noise ratio estimated by the dispersion. The equivalent insulation resistance to ground for the current cycle is then calculated based on this weighted fusion. and equivalent distributed capacitance to ground .

[0033] Perform a Fast Fourier Transform on the leakage current data segment after each windowing to extract the frequency. and The complex components at that point, namely amplitude and phase, constitute the leakage current vector for that time window.

[0034] The signal-to-noise ratio is estimated based on the dispersion and used as a weight to perform weighted fusion of the leakage current vectors for each time window, including: calculating the mean vector of the leakage current vectors; calculating the first... The square of the Euclidean distance between the leakage current vector and the mean vector in the nth time window is taken as the nth time window. The dispersion of the nth time window; the reciprocal of the dispersion is used as the nth time window. The signal-to-noise ratio estimate for the nth time window is calculated and normalized. The weights of each time window are calculated; the leakage current vectors of each time window are weighted and fused using the calculated weights to obtain the fused leakage current vector.

[0035] Specifically, N time windows are obtained through windowing, and N leakage current vectors are calculated. Suppose that four time windows are obtained through windowing, and four leakage current vectors are calculated from them. Each vector contains a real part and an imaginary part, for example... , , , and a vector contaminated by strong noise. The mean vector of the above four vectors can be calculated. .

[0036] Calculate the dispersion of each vector from the mean vector, which is the square of the Euclidean distance. For example, the dispersion of the first time window... The dispersion of the fourth time window affected by noise pollution It can be seen that the dispersion of outliers is much greater than that of normal points. The reciprocal of the dispersion is used as an estimate of the signal-to-noise ratio (SNR), and weights are calculated based on this. Assume the sum of all SNR estimates is... Then the first The weight of each time window is In the example above, It is very large, therefore it leads to The weight is lower than , and The leakage current vector is obtained through weighted fusion. The result will be mainly determined by the first three high-quality vectors, which reduces the impact of the fourth noisy data point.

[0037] Step S3, construct from , Impedance at frequency and The phase difference at the point, and the leakage current signal from and The four-dimensional original eigenvector is composed of the energy norms of the generated intermodulation components.

[0038] The above calculations and The first two components of the eigenvector; using the leakage current vector and the known injection voltage vector, the following are calculated: and complex impedance to ground at the location and The phase angles of each component are extracted, and the difference between the phase angles is taken as the third component of the eigenvector; the energy norm of the intermodulation component is taken as the fourth component of the eigenvector. Then, the above four components are combined in sequence to form a four-dimensional original eigenvector.

[0039] To represent the insulation state from multiple dimensions and obtain an information-rich feature vector, the energy norm of the intermodulation components is calculated as follows: A fast Fourier transform is performed on the acquired leakage current signal to obtain the signal spectrum; the intermodulation frequency point is then determined. and Extract the amplitude of all spectral components within a 20Hz bandwidth centered at the intermodulation frequency point; take the square root of the sum of the squares of the amplitudes of all spectral components within the bandwidth to obtain the energy norm of the intermodulation component.

[0040] For example, assume that the insulation resistance has been obtained through prior calculations. 50kΩ, distributed capacitance The impedance is 2 μF, and the phase difference between the impedance at the fundamental frequency of 50 Hz and the detection frequency of 130 Hz is -25°. To calculate the energy norm of the intermodulation components, a Fast Fourier Transform (FFT) is performed on the acquired leakage current signal to obtain a detailed spectrum. Based on the fundamental frequency... Hz and detection frequency The intermodulation frequencies are determined to be 80Hz and 180Hz. An analysis bandwidth of 20Hz is defined centered on each of these two frequencies. For the 80Hz point, the analysis bandwidth is 70 to 90Hz; for the 180Hz point, the analysis bandwidth is 170 to 190Hz. The amplitudes of all spectral components within these two bandwidths are extracted. Assuming that within the 70 to 90Hz bandwidth, three components with amplitudes of 0.05, 0.2, and 0.08 are extracted; and within the 170 to 190Hz bandwidth, three components with amplitudes of 0.04, 0.15, and 0.06 are extracted, the sum of the squares of all amplitudes is taken as the square root, and the energy norm of the intermodulation component is calculated to be 0.276. Based on these four components, a four-dimensional original feature vector can be constructed as [50000, 2E-6, -25, 0.276].

[0041] Step S4, based on and Based on the ratio and the gradient of change, a fourth-order diagonal weighted matrix is ​​constructed, and the original feature vector is transformed to obtain a weighted fault feature vector. The weighted fault feature vector is then compared with a preset fault template library to calculate the Mahalanobis distance and determine the fault attribution of the current insulation state.

[0042] To adjust the importance of each component in the four-dimensional feature vector according to real-time operating conditions and highlight the current fault characteristics, the steps to construct a fourth-order diagonal weighted matrix include: calculating the ratio. and gradient of change Construct a fourth-order diagonal weighted matrix, where the diagonal elements are: ; ; ; ;in, and They are respectively and Feature weights The weights for the phase difference features, The weights are the energy norm features of the intermodulation components.

[0043] Specifically, assuming we use the original four-dimensional feature vector [50000, 2E-6, -25, 0.276] obtained in the previous step, and it is currently in a stable state, the corresponding gradient of the bus voltage harmonic distortion rate change is... If the ratio is 0.02, then the ratio is... This ratio is very large, indicating strong resistance. The four diagonal elements of the fourth-order diagonal weighted matrix are calculated according to the formula, where the first weight... The result approaches 0; the second weight and Similarly, it also approaches 0. When the insulation resistance is extremely high, the sensitivity to specific values ​​decreases, and noise can be reduced by suppressing the influence of weights; the third weight The value is always 1.0, indicating that the weight of the phase difference feature remains unchanged; the fourth weight... This indicates that the feature weights are slightly enhanced under steady-state conditions. That is, the constructed fourth-order diagonal weighted matrix is ​​a matrix with diagonal elements [0,0,1.0,1.1]. Multiplying the original four-dimensional feature vector on the left by this fourth-order diagonal weighted matrix yields a weighted feature vector [0,0,-25,0.3036]. This vector will be used for fault classification, equivalent to multiplying each component of the original feature vector by its corresponding weight, thus obtaining a weighted fault feature vector that better highlights the characteristics of the current fault. If in a transient state... As the value increases, This will increase, thereby amplifying the role of this feature in fault diagnosis.

[0044] The weighted fault feature vector is compared with a preset fault template library to calculate the Mahalanobis distance to determine the fault attribution of the current insulation state. This includes: the preset fault template library contains three categories: normal, metallic grounding fault, and capacitive grounding fault; each category is defined by the mean and covariance matrix of the feature vectors obtained through statistical analysis of a large amount of experimental data; the weighted fault feature vector of the current period is substituted into the Mahalanobis distance formulas of the three categories to calculate three Mahalanobis distance values; the category corresponding to the smallest Mahalanobis distance value is selected as the fault attribution of the current insulation state.

[0045] Specifically, fault classification is achieved by comparing current characteristics with predefined standard fault modes. The system internally has a fault template library, built through learning and statistical analysis of a large amount of historical data. The library contains at least three categories, such as normal state, metallic grounding fault, and capacitive grounding fault, each represented by a four-dimensional mean vector. and covariance matrix They are defined as representing the central position and distribution pattern of the weighted feature vectors under that category, respectively.

[0046] Assuming the weighted fault feature vector calculated in the current cycle is x=[0,0,-25,0.287], we calculate the Mahalanobis distance between this vector and each category in the template library. The advantage of Mahalanobis distance is that it considers the correlation between features and performs scale normalization. The formula for Mahalanobis distance is... ,in Indicates transpose. This represents matrix inversion. It calculates the inversion of the current vector sequentially. Mahalanobis distance to the center of the normal state class The Mahalanobis distance to the center of a metallic ground fault And the Mahalanobis distance to the center of the capacitive ground fault class. After the calculation is complete, three specific distance values ​​will be obtained. Let's assume the calculation results are as follows: , , The three distance values ​​are compared. In this example, 1.2 is the minimum value, which corresponds to the metallic grounding fault category. Therefore, the current insulation condition is determined to be a metallic grounding fault, and an alarm is issued or corresponding protective action is performed accordingly.

[0047] This invention also provides an online monitoring system for insulation faults in a medical IT system control cabinet. For example... Figure 2 As shown, the system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the online monitoring method for insulation faults in the medical IT system control cabinet according to the present invention.

[0048] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0049] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0050] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0051] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for online monitoring of insulation faults of a medical IT system control cabinet, characterized in that, include: Based on the equivalent ground capacitance of the previous monitoring period, the detection frequency is selected from the preset set of non-integer harmonic frequencies. , baseband The sinusoidal signal and the detection frequency The sinusoidal signals are superimposed to form a composite detection signal and injected into the medical IT system; the leakage current to ground and the bus voltage signal are collected, the instantaneous harmonic distortion rate change gradient of the bus voltage is calculated, and the leakage current signal is windowed according to the change gradient between the first time window parameter and the second time window parameter. Calculate the intrinsic frequency of each time window and The leakage current vector at a given location is weighted and fused based on the signal-to-noise ratio estimated using the dispersion as a weight to obtain the leakage current vector for each time window, including: Calculate the mean vector of the leakage current vector; calculate the first... The square of the Euclidean distance between the leakage current vector and the mean vector for the nth time window is taken as the nth time window. The dispersion of each time window; The reciprocal of the dispersion is taken as the first... The signal-to-noise ratio estimate for the nth time window is calculated and normalized. The weights of each time window are calculated; the leakage current vectors of each time window are weighted and fused using the calculated weights to obtain the fused leakage current vector. Based on this, the equivalent insulation resistance to ground for the current cycle is calculated. and equivalent distributed capacitance to ground ; Build by , Impedance at frequency and The phase difference at the point, and the leakage current signal from and The four-dimensional original eigenvector is composed of the energy norms of the generated intermodulation components; based on and Based on the ratio and the gradient of change, a fourth-order diagonal weighted matrix is ​​constructed, including: Calculate the ratio and gradient of change ; Construct a fourth-order diagonal weighted matrix, where the diagonal elements are: ; ; ; ;in, and They are respectively and Feature weights The weights for the phase difference features, Weights for the energy norm characteristics of the intermodulation components; The original feature vector is transformed to obtain a weighted fault feature vector; the weighted fault feature vector is then compared with a preset fault template library to calculate the Mahalanobis distance and determine the fault attribution of the current insulation state.

2. The online monitoring method for insulation faults in a medical IT system control cabinet according to claim 1, characterized in that, The detection frequency is selected from a preset set of non-integer multiples of the frequency. ,include: Each candidate frequency in the set Substitute into the impedance prediction model Calculate the predicted impedance magnitude at each candidate frequency, where The insulation resistance of the previous cycle. This represents the distributed capacitance of the previous cycle; Calculate the minimum frequency difference between each candidate frequency and the 2nd to 5th harmonic frequencies of the fundamental frequency; Define the cost function ,in =2,3,4,5 and Given preset weighting coefficients, the candidate frequency that minimizes the cost function is selected as the probe frequency. .

3. The online monitoring method for insulation faults in a medical IT system control cabinet according to claim 1, characterized in that, The calculation of the instantaneous harmonic distortion rate gradient of the bus voltage includes: The acquired bus voltage signal sequence is subjected to a short-time Fourier transform to obtain the spectrum that varies with time. Calculate the total harmonic distortion (THD) at each time point based on the spectrum to obtain the THD time series. The difference between two adjacent THD values ​​in the THD time series is calculated to obtain the gradient of the instantaneous harmonic distortion rate.

4. The online monitoring method for insulation faults in the control cabinet of a medical IT system according to claim 1, characterized in that, The step of windowing the leakage current signal by selecting between a first time window parameter and a second time window parameter based on the changing gradient includes: If the absolute value of the instantaneous harmonic distortion rate change gradient of the bus voltage is less than the preset gradient discrimination threshold, then the first time window parameters are used, with the window function type being Hamming window, the window length being 200ms, and the overlap rate being 50% for windowing. If the absolute value of the changing gradient is greater than or equal to the preset gradient discrimination threshold, then the second time window parameters are used, with the window function type being Hamming window, the window length being 40ms, and the overlap rate being 75% for windowing.

5. The online monitoring method for insulation faults in a medical IT system control cabinet according to claim 1, characterized in that, The energy norm of the intermodulation component is calculated as follows: The collected leakage current signal is subjected to a fast Fourier transform to obtain the signal spectrum; Determine the intermodulation frequency point as and ; Extract the amplitude of all spectral components within a 20Hz bandwidth centered at the intermodulation frequency point; The energy norm of the intermodulation component is obtained by taking the square root of the sum of the squares of the amplitudes of all spectral components within the bandwidth.

6. The online monitoring method for insulation faults in a medical IT system control cabinet according to claim 1, characterized in that, The original feature vector is transformed to obtain a weighted fault feature vector, specifically as follows: Multiplying the fourth-order diagonal weighted matrix on the left by the four-dimensional original feature vector yields the weighted fault feature vector, in order to... and Increase when the ratio is small and The weights of the intermodulation component energy norms are increased when the gradient is large.

7. The online monitoring method for insulation faults in a medical IT system control cabinet according to claim 1, characterized in that, The step of calculating the Mahalanobis distance between the weighted fault feature vector and a preset fault template library to determine the fault attribution of the current insulation state includes: The preset fault template library includes three categories: normal, metallic grounding fault, and capacitive grounding fault; each category is defined by the mean and covariance matrix of the feature vectors obtained through statistical analysis of a large amount of experimental data. The weighted fault feature vector of the current period is substituted into the Mahalanobis distance formula for the three categories to calculate the three Mahalanobis distance values. The category corresponding to the smallest Mahalanobis distance is selected as the fault classification for the current insulation state.

8. An online monitoring system for insulation faults in a medical IT system control cabinet, characterized in that, include: processor; A memory storing computer instructions for online monitoring of insulation faults in a medical IT system control cabinet, wherein when the computer instructions are executed by the processor, the system performs the online monitoring method for insulation faults in a medical IT system control cabinet according to any one of claims 1-7.

Images