Self-discharge diagnosis method for large-scale lithium battery pack

Abstract

The application discloses a self-discharge diagnosis method and system for large-scale lithium battery packs. The battery cell voltage, battery pack current, SOC and equalization state are collected, and the effective diagnosis time after the near static or small current, the smooth SOC change, the equalization closing or the equalization recovery completion is screened. The median and low quantile boundary of all cell voltages are calculated at each effective time, the difference between the two is taken as a normalized scale, and a low quantile boundary normalized deviation feature of the cell relative to the low voltage boundary of the battery pack is constructed. Based on all cell deviation features at the same time, a dynamic under-deviation threshold is generated, and then the cell deviation feature is clustered by DBSCAN. The cell meets the two conditions of exceeding the dynamic under-deviation threshold and the few cluster samples at the same time, which is judged as a suspected self-discharge anomaly, and the diagnosed abnormal cell is output. The application can weaken the false alarm caused by the charge-discharge polarization, equalization action, cell initial difference and instantaneous noise, and is suitable for online self-discharge diagnosis of large-scale energy storage lithium battery packs with more than 300 strings.

Description

Technical Field

[0001] This invention belongs to the field of electrochemical energy storage system operation status monitoring and fault diagnosis technology, specifically relating to an online detection method and system for self-discharge fault cells in large-capacity lithium-ion battery packs. Background Technology

[0002] Lithium-ion batteries are commonly used in large-scale battery packs in energy storage power stations, industrial and commercial energy storage, and grid frequency regulation. For high-capacity cells of 300Ah or higher, factors such as manufacturing inconsistency, internal resistance distribution, micro-short circuits, separator defects, and electrode burrs can lead to differences in the self-discharge rate of different cells. Abnormal self-discharge typically manifests as: under the same state of charge or similar operating conditions, the voltage of individual cells continuously deviates from the collective trend and gradually increases. In severe cases, this can lead to a decrease in the usable capacity of the battery pack, an increase in the frequency of equalization, an increase in thermal risk, and even triggering system protection shutdown.

[0003] In existing technologies, common methods for diagnosing self-discharge include:

[0004] (1) Open circuit voltage decay method under static conditions: requires long-term shutdown or strict static conditions, which is difficult to meet the online operation requirements of energy storage systems;

[0005] (2) Comparison of energy loss based on SOC estimation: It is greatly affected by model error, temperature and aging coupling;

[0006] (3) Simple threshold method or mean difference method: sensitive to outliers and easily affected by measurement noise, transient conditions and communication jitter;

[0007] (4) Supervised classification based on machine learning: It relies on a large number of labeled fault samples, which are difficult to obtain in actual engineering, and the model transferability is insufficient.

[0008] (5) Data-driven clustering method: Although existing data-driven clustering methods can identify anomalies using algorithms such as voltage deviation and DBSCAN, they are usually aimed at general faults or potential thermal runaway of power batteries. The diagnostic features are mostly fault degree indicators such as voltage offset increment and cumulative number of deviations within the historical window, and do not constrain the mechanism of "continuous negative deviation under near-static conditions" of self-discharge anomalies.

[0009] For large-scale energy storage battery packs, the large number of individual cells and their inherent dispersion are significant factors. Furthermore, voltage is significantly affected by charge / discharge polarization, equalization processes, communication jitter, and temperature differences. Without first screening for suitable operating conditions for self-discharge identification, short-term voltage differences caused by normal current disturbances or equalization may be misjudged as self-discharge. Relying solely on clustering results may lead to the identification of very small fluctuations as anomalies. Relying solely on thresholds may be susceptible to transient noise. Therefore, an online diagnostic solution addressing the self-discharge mechanism is urgently needed, possessing dynamic threshold bias, minority detection, and time consistency verification capabilities. Summary of the Invention

[0010] The purpose of this invention is to provide a self-discharge diagnosis method and system for large-scale lithium battery packs, which solves the problems in the prior art that it does not diagnose the effective self-discharge conditions and the mechanism of continuous negative deviation, is easily affected by charge-discharge polarization and equalization, and has a high false alarm rate due to simple threshold or simple clustering.

[0011] This invention discloses a self-discharge diagnosis method for large-scale lithium battery packs, which includes the following three steps:

[0012] Step 1: Obtain the voltage of each cell in the lithium battery pack, perform preprocessing, and filter out the valid sampling times and corresponding base voltages. The specific process includes:

[0013] Step 11: First, collect data such as individual cell voltage and current from the BMS, acquire individual cell voltages, and construct voltage vectors for each sampling time; The individual cell voltage collected at each moment Recorded as:

[0014] ;

[0015] Then, for each sampling time Constructing voltage vector :

[0016] ;

[0017] Where M represents the number of individual cells in the battery pack.

[0018] Step 12: Construct a voltage vector for each sampling time and perform denoising and smoothing processing;

[0019] First, in handling communication gaps, short gaps are handled using linear interpolation or forward hold, while long gaps are deleted at that point in time.

[0020] Then, for satisfying The point with the abrupt change value is marked as an anomaly and replaced with the neighborhood median. The threshold is set by the sampling period and the rate of physical change;

[0021] Secondly, align with the system master clock to ensure consistency. All individual unit voltages are sampled from the same sampling window to avoid inconsistent timestamps;

[0022] Finally, to suppress measurement noise, a moving median filter or a moving average filter can be applied to each individual sequence:

[0023] ;

[0024] in, This indicates that the i-th single entity at time t k Filtered voltage value; This indicates that the median of all data within the parentheses is to be calculated. This represents the half-width of the filtering window; the voltage value after noise reduction and smoothing. Replace the original single-unit voltage .

[0025] Step 13: Screen the effective operating conditions for self-discharge diagnosis after noise reduction and smoothing. The effective operating condition screening includes the following conditions:

[0026] Condition 1: The battery pack current meets the requirements. , The near-static current threshold, Let time t k Current;

[0027] Condition 2: The battery pack's SOC change rate is lower than the threshold;

[0028] Condition 3: All or individual cells are in the equalization shutdown window;

[0029] The set of times when any activation condition is met is denoted as .

[0030] Step 2: Construct the low quantile boundary normalized deviation characteristics based on the baseline voltage at the effective sampling time, and calculate the statistical scale estimate of the low quantile boundary normalized deviation characteristics of all individual cells at the effective sampling time, including the following steps:

[0031] Step 21: First, for each valid sampling time... Calculate the median voltage of all individual cells:

[0032]

[0033] in, The median of all voltages. This indicates that the median of all data within the parentheses is calculated.

[0034] Step 22: Next, calculate the lower quantile boundary of all individual cell voltages:

[0035]

[0036] in, Indicates time The lower boundary of all individual unit voltages; It is a low quantile proportion, and ;

[0037] Step 23: Then, construct a normalized scale based on the median voltage of all individual cells and the low quantile boundary:

[0038]

[0039] in, Indicates time The low quantile normalization scale; It is a positive number, used to prevent the scale from being zero;

[0040] Step 24: Then, based on the low quantile boundary and the normalized scale, calculate the... Each individual at the effective sampling time Low quantile boundary normalization bias characteristics:

[0041]

[0042] in, Used to characterize the The degree of deviation of each individual cell from the low voltage boundary of the current battery pack; when When the value is less than zero, it indicates that the cell is below the low quantile boundary of the current battery pack;

[0043] Step 25: Finally, model the dispersion of the normalized bias characteristic distribution of the low quantile boundary using statistical scaling estimation:

[0044]

[0045] in, For a moment Statistical scale estimates of the normalized bias characteristics of all individual low quantile boundaries; ; It is a positive number, used to prevent the scale from being zero.

[0046] Step 3: By combining dynamic threshold discrimination with DBSCAN clustering results, accurate identification of self-discharge faulty cells is achieved, including the following steps:

[0047] Step 31: Define the dynamic threshold for the effective sampling time based on the normalized bias characteristics of the low quantile boundary of the single entity and its statistical scale estimate at the effective sampling time:

[0048] ;

[0049] ;

[0050] in, This represents the mean of all individual eigenvalues ​​at that moment; when the individual eigenvalues ​​satisfy... It is assumed that the single entity meets the dynamic threshold condition at that moment.

[0051] Step 32: Perform DBSCAN clustering based on the normalized bias features of the low quantile boundaries of all monomers at the effective sampling time: Perform DBSCAN clustering on the normalized bias features of the low quantile boundaries of monomers at each time point, using DBSCAN parameters. Pair with minimum sample size MinPts Clustering yields the category label for each time step:

[0052] ;

[0053] Among them, the output categories of DBSCAN are and noise ;

[0054] The first is considered to be true when any of the following conditions are met. Each individual satisfies the DBSCAN clustering minority judgment result:

[0055]

[0056] or:

[0057]

[0058] or:

[0059]

[0060] in, Indicates the first The number of samples in the cluster to which each individual belongs. This is the proportional threshold. This represents the main class with the largest number of samples.

[0061] Step 33: Determine that a single cell that simultaneously meets the dynamic threshold condition and the DBSCAN clustering judgment result is a suspected self-discharge fault cell, and output the final fault cell after continuous confirmation through a sliding window:

[0062] Specifically, if both the dynamic threshold and the DBSCAN clustering result are met at any given time, it is judged as a single entity. At any moment Suspected malfunction ;

[0063] If at the end The number of times a suspected fault is met in a valid sampling time reaches Then the final faulty unit will be output:

[0064]

[0065] in, For the most recent A set of valid sampling times Here, r is the window length and r is the preset number of failures. Monomer Final failure result.

[0066] The beneficial effects achieved by this invention are:

[0067] This invention eliminates interference from high-current charge-discharge polarization, rapid changes in SOC, and equalization actions on individual cell voltage differences by effectively screening operating conditions, making the diagnostic target closer to the fault mechanism of slow self-discharge and continuous negative deviation.

[0068] This invention uses the low percentile boundary of all individual cell voltages as the low voltage reference benchmark, which can pay more attention to whether the low voltage end cells are further separated from the normal low voltage group boundary, thereby enhancing the characterization ability of cells with abnormal self-discharge.

[0069] This invention constructs a normalized scale based on the distance between the low quantile boundary and the median voltage, which makes the characteristics of different SOC platforms, different battery packs, and different operating stages more consistent, avoiding the problem that fixed voltage thresholds are difficult to adapt to multi-condition operating data.

[0070] This invention uses the low quantile boundary normalized deviation feature combined with the statistical scale estimate of the deviation feature distribution at the current moment to generate a dynamic downward deviation threshold. It can adaptively adjust the anomaly discrimination boundary according to the current overall discrete state of the battery pack, reducing the risk of misjudgment caused by the aging degree of the battery pack, the difference in the consistency of individual cells and the slow change of voltage in the plateau area.

[0071] This invention only identifies negatively anomalous cells below the low quantile boundary of the battery pack and that meet the dynamic underbias threshold, thus avoiding misjudging positive anomalies, bidirectional fluctuations, or inconsistencies of ordinary cells that are not self-discharge characteristics as self-discharge faults.

[0072] This invention limits DBSCAN clustering to minority verification and performs logical AND judgment with dynamic threshold conditions, which can simultaneously reduce false alarms caused by simple threshold judgment and simple clustering judgment, thereby improving the accuracy of large-scale lithium battery pack self-discharge anomaly identification.

[0073] This invention uses a sliding window for continuous verification, which conforms to the fault evolution characteristics of slow self-discharge and continuous development, and can effectively suppress instantaneous false alarms caused by communication jitter, sampling noise and short-term disturbances. Attached Figure Description

[0074] Figure 1 This is a schematic diagram of the self-discharge diagnosis method for large-scale lithium battery packs in an embodiment of the present invention. Detailed Implementation

[0075] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer as a result. However, these embodiments are merely exemplary and do not constitute any limitation on the scope of the present invention. Those skilled in the art should understand that modifications or substitutions can be made to the details and form of the technical solutions of the present invention without departing from the spirit and scope of the present invention, but all such modifications and substitutions fall within the protection scope of the present invention.

[0076] like Figure 1 As shown, the present invention provides a self-discharge diagnosis method for large-scale lithium battery packs, which includes the following steps:

[0077] Step 1: Data processing, which includes the following steps.

[0078] Step 11: Data Integrity and Synchronization;

[0079] Assume the battery pack consists of Composed of individual monomers, suitable for The sampling time sequence is as follows: The collected individual cell voltages are recorded as follows:

[0080]

[0081] Construct a voltage vector for each sampling time:

[0082]

[0083] If there are communication gaps, abnormal jumps, or inconsistent timestamps, the following rules will be applied:

[0084] Missing values: For missing values, linear interpolation or forward hold is used (configurable in engineering); for long missing values, the missing value is deleted.

[0085] Jump value: If satisfied, mark the point as an anomaly and replace it with the neighborhood median;

[0086]

[0087] in, The threshold can be set by the sampling period and the physical rate of change;

[0088] Synchronization processing: Align with the system master clock to ensure consistency. All individual cell voltages are obtained from the same sampling window.

[0089] Step 12: Noise reduction and smoothing.

[0090] To suppress measurement noise, a moving median filter or a moving average filter can be applied to each individual sequence. Taking moving median filtering as an example:

[0091]

[0092] in, This indicates that the i-th single entity is at time t. k Filtered voltage value; This indicates that the median of all data within the parentheses is to be calculated. This indicates the half-width of the filtering window, meaning that n points are selected to the left and right of the current sampling point for filtering. (Window half-width) Desirable Each sampling point was selected, with the principle of not masking the slow self-discharge trend. Subsequent sampling points were... Replace the original .

[0093] Step 13: Screening of valid operating conditions.

[0094] Self-discharge faults typically manifest as a slow, continuous negative voltage deviation. During high-current charging and discharging, rapid changes in SOC, or equalization processes, the voltage difference between individual cells is easily affected by polarization voltage, internal resistance voltage drop, and equalization current. Therefore, this invention sets effective operating condition screening conditions, preferably enabling diagnosis under the following conditions:

[0095] Condition 1: The battery pack current meets the requirements. (Near static or low current region);

[0096] Condition 2: The rate of change of SOC is lower than the threshold;

[0097] Condition 3: All or target units are in the equalization off window, or the recovery time after equalization ends exceeds the preset recovery time.

[0098] The set of times that meet the preset combination activation conditions is denoted as:

[0099]

[0100] in, This is the set of valid sampling times for self-discharge diagnosis, only for... Subsequent feature extraction and fault diagnosis are performed at the sampling time.

[0101] Step 2: Feature extraction.

[0102] This invention constructs a low-quantile boundary normalized deviation feature based on the low-quantile boundary of all individual cell voltages. This feature characterizes whether a target cell has fallen below the boundary of the normal low-voltage group of the current battery pack, and the relative degree to which it has fallen below this boundary, consistent with the fault behavior of a self-discharge cell with continuous negative deviation. Specifically, it includes the following steps:

[0103] Step 21: Median reference construction.

[0104] For each valid sampling time Calculate the median voltage of all individual cells:

[0105]

[0106] in, The median of all voltages. This indicates that the median of all data within the parentheses is calculated.

[0107] Step 22: Constructing low-quantile boundaries.

[0108] Calculate the lower quantile boundary for all individual cell voltages:

[0109]

[0110] in, Indicates time The lower boundary of all individual unit voltages; It is a low quantile proportion, and In practice, p can be 0.05, 0.10, or other low quantiles suitable for the battery pack size and consistency level.

[0111] Step 23: Normalized scale construction.

[0112] A normalized scale is constructed based on the median voltage of all individual cells and the low quantile boundary:

[0113]

[0114] in, Indicates time The low quantile normalization scale; It is a positive number, used to prevent the scale from being zero.

[0115] Step 24: Construction of low quantile boundary normalization bias characteristics.

[0116] Based on the low quantile boundary and normalized scale, the calculation of the first... Each individual at the effective sampling time Low quantile boundary normalization bias characteristics:

[0117]

[0118] in, Used to characterize the The degree of deviation of each individual cell from the low voltage boundary of the current battery pack; when When the value is less than zero, it indicates that the cell is below the low quantile boundary of the current battery pack;

[0119] Step 25: Statistical scale estimation.

[0120] The dispersion of the normalized bias characteristic distribution at the lower quantile boundary is modeled using statistical scaling estimation:

[0121]

[0122] in, For a moment Statistical scale estimates of the normalized bias characteristics of all individual low quantile boundaries; ; It is a positive number, used to prevent the scale from being zero.

[0123] Step 3: Diagnostic strategy.

[0124] This invention employs a diagnostic strategy combining dynamic thresholding, DBSCAN clustering for minority cases, and time consistency. Dynamic thresholding is used to determine if a target monomer exhibits a significant negative deviation; DBSCAN clustering is used to determine if the monomer belongs to a minority anomalous state; and time consistency is used to confirm whether the anomalous state is persistent. This reduces false alarms caused by simple thresholding, simple clustering, and instantaneous judgment. Specifically, it includes the following steps:

[0125] Step 31: Condition A: Lower bias threshold criterion.

[0126] At each valid sampling time For eigenvalues Calculate its mean :

[0127]

[0128]

[0129] Define a dynamic down-biased threshold:

[0130]

[0131] in, This represents the mean of all individual eigenvalues ​​at that moment;

[0132] When the monomer satisfies:

[0133]

[0134] The i-th individual is considered to satisfy the dynamic threshold condition at that moment. This condition indicates that the target individual is not only below the low quantile boundary, but also shows a significant negative deviation relative to the current population characteristic distribution.

[0135] Step 32: Condition B: Minority class discrimination based on DBSCAN.

[0136] To leverage the prior knowledge that "faulty cells are typically a minority among 416 cells," this invention performs DBSCAN clustering on the normalized bias features of the low quantile boundaries of cells at each time step. A one-dimensional sample set is constructed as follows:

[0137]

[0138] Using DBSCAN parameters Pair with MinPts (minimum number of samples) Clustering. Obtain the category label for each time step:

[0139]

[0140] DBSCAN's output can include categories. and noise .

[0141] Minority or outlier states can be defined using any of the following rules or a combination thereof:

[0142] Rule B1: If:

[0143]

[0144] Then it is believed that the first Each individual belongs to a DBSCAN noise point, satisfying the DBSCAN clustering minority judgment result;

[0145] Rule B2: If the first The number of samples in the cluster to which each individual belongs satisfies:

[0146]

[0147] Then it is believed that the first Several monomers belong to small clusters, satisfying the DBSCAN clustering minority judgment result. This is the proportional threshold;

[0148] Rule B3: Define the cluster with the largest number of samples as the primary cluster. ,like:

[0149]

[0150] Then it is believed that the first Each individual belongs to a non-primary cluster and satisfies the DBSCAN clustering minority judgment result.

[0151] Step 33: Joint discrimination based on dynamic threshold, DBSCAN minority and time consistency.

[0152] At any valid sampling time If the first If a single entity simultaneously satisfies both the dynamic threshold condition and the DBSCAN clustering minority judgment result, then the first individual is determined to be... At that moment, each individual cell was suspected of being a faulty cell.

[0153]

[0154] That is:

[0155]

[0156] And the When a single individual satisfies the DBSCAN clustering minority judgment result, let:

[0157]

[0158] otherwise:

[0159]

[0160] Considering the slow and continuous development of self-discharge faults, and that anomalies at a single sampling moment may arise from communication jitter, measurement noise, or short-term disturbances, this invention further introduces a sliding window persistent confirmation. Let... For the most recent A set of valid sampling times, if in the most recent In the effective sampling time, the first The number of times a single unit meets the suspected fault criteria reaches the preset fault count. Then the final faulty unit will be output:

[0161]

[0162] in, For window length, To preset the number of failures, Indicates the first One cell was identified as the final self-discharge fault cell.

[0163] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the scope of protection of the present invention; all technical solutions formed by equivalent transformations or equivalent substitutions fall within the scope of protection of the present invention; the parts of the present invention not described in detail are well-known technologies to those skilled in the art.

Claims

1. A self-discharge diagnosis method for large-scale lithium battery packs, characterized in that, The self-discharge diagnosis method for large-scale lithium battery packs includes the following steps: Step 1: Obtain the voltage of each cell in the lithium battery pack, perform preprocessing, and filter out the valid sampling time and the corresponding base voltage; Step 2: Construct the low quantile boundary normalized deviation characteristics based on the base voltage at the effective sampling time, and calculate the statistical scale estimate of the low quantile boundary normalized deviation characteristics of all individual cells at the effective sampling time. Step 3: By combining dynamic threshold discrimination with DBSCAN clustering results, accurate identification of self-discharge faulty cells can be achieved.

2. The self-discharge diagnosis method for large-scale lithium battery packs according to claim 1, characterized in that, In step 1, data such as individual cell voltage and current are first collected from the BMS. The individual cell voltage collected at each moment Recorded as: ； For each sampling time Constructing voltage vector : ； Where M is the number of individual cells in the battery pack; Then, the data is preprocessed, and for missing communication data: short missing data is handled using linear interpolation or forward hold, while long missing data is deleted at that time point; for data that satisfies... The point with the abrupt change value is marked as an anomaly and replaced with the neighborhood median. The threshold is set by the sampling period and the rate of physical change; and it is aligned with the system master clock to ensure consistency. All individual cell voltages are sampled from the same sampling window to avoid inconsistent timestamps; To suppress measurement noise, a moving median filter or a moving average filter can be applied to each individual sequence: ； in, This indicates that the i-th single entity is at time t. k Filtered voltage value; This indicates that the median of all data within the parentheses is to be calculated. This represents the half-width of the filtering window; then, the voltage value after denoising and smoothing is used. Replace the original single-unit voltage ; Finally, the effective operating conditions are screened, specifically including the following criteria: Condition 1: The battery pack current meets the requirements. , The near-static current threshold, Let time t k Current; Condition 2: The battery pack's SOC change rate is lower than the threshold; Condition 3: All or individual cells are in the equalization shutdown window; The set of times when any activation condition is met is denoted as .

3. The self-discharge diagnosis method for large-scale lithium battery packs according to claim 1, characterized in that, In step 2, firstly for each valid sampling time... Calculate the median voltage of all individual cells: ； in, The median of all voltages. This indicates that the median of all data within the parentheses is calculated. Secondly, calculate the lower quantile boundary of all individual cell voltages: ； in, Indicates time The lower boundary of all individual unit voltages; It is a low quantile proportion, and ; Then, a normalized scale is constructed based on the median voltage of all individual cells and the low quantile boundary: ； in, Indicates time The low quantile normalization scale; It is a positive number, used to prevent the scale from being zero; Then, based on the low quantile boundary and the normalized scale, the 1st... Each individual at the effective sampling time Low quantile boundary normalization bias characteristics: ； in, Used to characterize the The degree of deviation of each individual cell from the low voltage boundary of the current battery pack; when When the value is less than zero, it indicates that the cell is below the low quantile boundary of the current battery pack; Finally, the dispersion of the normalized bias characteristic distribution at the lower quantile boundary is modeled using statistical scaling estimation: ； in, For a moment Statistical scale estimates of the normalized bias characteristics of all individual low quantile boundaries; ; It is a positive number, used to prevent the scale from being zero.

4. The self-discharge diagnosis method for large-scale lithium battery packs according to claim 1, characterized in that, Step 3 includes the following steps: Step 31: Define the dynamic threshold of the effective sampling time based on the normalized bias characteristics of the low quantile boundary of the single entity and its statistical scale estimate in the effective sampling time. Step 32: Perform DBSCAN clustering based on the low quantile boundary normalization bias characteristics of all monomers in the effective sampling time. Step 33: Determine that a single cell that simultaneously meets the dynamic threshold condition and the DBSCAN clustering minority judgment result is a suspected self-discharge fault cell, and output the final fault cell after continuous confirmation through a sliding window.

5. The self-discharge diagnosis method for large-scale lithium battery packs according to claim 4, characterized in that, In step 31, the dynamic threshold is: ； ； in, This represents the mean of all individual eigenvalues ​​at that moment; when the individual eigenvalues ​​satisfy... It is assumed that the single entity meets the dynamic threshold condition at that moment.

6. The self-discharge diagnosis method for large-scale lithium battery packs according to claim 1, characterized in that, In step 32, DBSCAN clustering is performed on the individual bias characteristics at each time point, using DBSCAN parameters. Pair with minimum sample size MinPts Clustering yields the category label for each time step: ； Among them, the output category of DBSCAN is and noise ; The first is considered to be true when any of the following conditions are met. Each individual satisfies the DBSCAN clustering minority judgment result: ； or: ； or: ； in, Indicates the first The number of samples in the cluster to which each individual belongs. This is the proportional threshold. This represents the main class with the largest number of samples.

7. The self-discharge diagnosis method for large-scale lithium battery packs according to claim 1, characterized in that, In step 33, if both the dynamic threshold and the DBSCAN clustering result are satisfied at any given time, it is determined to be a single entity. At any moment Suspected malfunction ; If at the end The number of times a suspected fault is met in a valid sampling time reaches Then the final faulty unit will be output: ； in, For the most recent A set of valid sampling times Here, r is the window length and r is the preset number of failures. Monomer Final failure result.

Images