Filter cartridge breakage and leakage ai discrimination method and system based on dust concentration mutation and pressure difference steady state characteristics

Abstract

The application provides an AI discrimination method for filter cartridge damage and leakage based on dust concentration mutation and differential pressure steady-state characteristics, comprising the following steps: S1, collecting multi-source data; S2, data preprocessing; S3, extracting double characteristic parameters; S4, opposite coupling AI discrimination; S5, fault early warning; S6, closed-loop operation and maintenance; and a system comprising a multi-source data collection module, a data preprocessing module, a double characteristic parameter extraction module, an opposite coupling AI discrimination module, a fault early warning module and an operation and maintenance closed-loop module. The application adopts opposite coupling discrimination logic of differential pressure steady state and dust concentration mutation, accurately captures the hidden leakage that cannot be identified by traditional differential pressure monitoring, accurately distinguishes filter cartridge blockage and small leakage, realizes intelligent discrimination of filter cartridge small damage and hidden leakage fault, reduces the probability of false negatives and false positives, adapts to changes in the field such as equipment aging and working condition deviation, and is suitable for various dust removal equipment.

Description

Technical Field

[0001] This invention relates to the fields of intelligent fault diagnosis, IoT monitoring and environmental protection operation and maintenance technology for industrial dust removal equipment, specifically to an AI-based method and system for identifying filter cartridge damage and leakage based on dust concentration mutations and pressure differential steady-state characteristics. Background Technology

[0002] As the core filtration component of industrial dust collectors, the integrity of the filter cartridge directly determines the compliance of exhaust gas emissions and the stability of equipment operation. Currently, dust collector filter cartridge fault monitoring technologies in the industry generally suffer from technical deficiencies such as simplistic identification logic, inability to detect minute leaks, and high false alarm rates. Specific problems are as follows: First, traditional differential pressure monitoring has a significant blind spot. Existing equipment mainly relies on changes in inlet and outlet pressure differences to determine filter cartridge malfunctions, which can only identify severe malfunctions such as severe blockage or large-area damage to the filter cartridge. When the filter cartridge experiences minor damage such as micropore damage, localized fiber cracking, or pinhole leakage, the overall air permeability resistance of the filter cartridge remains almost unchanged, and the system differential pressure remains steady without significant fluctuations. Traditional differential pressure monitoring methods are completely unable to identify these types of hidden leakage malfunctions.

[0003] Second, single dust concentration threshold monitoring suffers from strong lag and high false alarm rate. Traditional dust monitoring only sets a fixed threshold for exceeding the standard, and can only trigger an alarm when the filter cartridge is severely damaged or a large amount of dust leaks, causing the concentration to exceed the standard. For early, minor leaks, the increase in dust concentration is small and does not reach the threshold for exceeding the standard, so traditional monitoring cannot detect them. This results in hidden leaks persisting for a long time, continuously damaging equipment and creating environmental hazards.

[0004] Third, it cannot distinguish between different fault conditions, leading to serious misjudgments and interference. During the operation of the dust collector, factors such as airflow disturbances, fluctuations in inlet dust load, and instantaneous dust emission during cleaning can all cause short-term fluctuations in dust concentration. Traditional technologies cannot distinguish between "concentration changes caused by fluctuations in operating conditions" and "continuous sudden changes in concentration caused by filter cartridge damage," which easily leads to false alarms or missed alarms, making operation and maintenance judgment difficult.

[0005] Fourth, there is no targeted coupling discrimination logic, and there is no dedicated identification mechanism for minor leaks. Existing technologies do not utilize the unique characteristics of minor leaks, such as "pressure differential steady state and dust micro-mutation", and lack a two-parameter opposing coupling discrimination mechanism. They cannot accurately isolate early filter cartridge damage and leakage faults from complex operating conditions, and can only passively deal with severe faults, failing to achieve proactive and refined fault prevention and control.

[0006] In summary, existing filter cartridge fault monitoring technologies suffer from poor adaptability, lack of latent fault identification capabilities, and rudimentary judgment logic, failing to meet the intelligent needs of industrial enterprises for zero environmental violations and predictive maintenance of equipment. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by proposing an AI-based method and system for identifying filter cartridge damage and leakage based on the characteristics of sudden changes in dust concentration and steady-state pressure difference. Utilizing the contradictory coupling relationship between the characteristics of sudden changes in outlet dust concentration and the lack of change in the steady-state system pressure difference, an AI-powered intelligent identification and precise early warning system for minor filter cartridge damage and hidden leakage faults is achieved. This system is applicable to various industrial cartridge and baghouse dust collectors for refined fault identification and environmental risk prevention and control.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: An AI-based method for detecting filter cartridge damage and leakage based on abrupt changes in dust concentration and steady-state characteristics of pressure difference includes the following steps: S1: Collect data from multiple sources; Real-time synchronous acquisition of differential pressure data at the inlet and outlet of the dust collector, dust concentration data at the equipment outlet, equipment operating load data, and pulse cleaning condition data to obtain raw time-series data; S2: Data preprocessing; The raw time series data is filtered, dust interference is removed, and steady-state operating conditions are screened to obtain effective monitoring data under steady-state operation of the equipment. S3: Extract dual feature parameters; Based on effective monitoring data, under steady-state operating conditions of the equipment, the steady-state characteristics of pressure difference and the sudden change characteristics of dust concentration are extracted simultaneously. S4: Opposing Coupling AI Judgment; Based on dual feature parameters, AI intelligently distinguishes between filter cartridge damage and leakage faults and normal operating condition fluctuations, and classifies the level of filter cartridge damage and leakage faults. S5: Fault warning; The warning system is divided into three levels: minor, moderate, and severe, and warning information and maintenance suggestions are pushed to the corresponding filter cartridge damage and leakage fault level. S6: Closed-loop operation and maintenance; Record fault identification data and maintenance records, compare the changes in dual parameters before and after maintenance, and iteratively optimize the AI ​​identification model.

[0009] Preferably, the equipment outlet dust concentration time series data includes dust concentration value and timestamp; the equipment outlet dust concentration time series data includes dust concentration value and timestamp; the equipment operating load data includes timestamp and fan frequency; the pulse cleaning condition data includes cleaning signal and timestamp, and the cleaning signal includes no cleaning and cleaning in progress; all data have the same sampling frequency and the timestamps are aligned.

[0010] Preferably, step S2 includes: S21: Filtering process; First-order low-pass filtering was used to process the time series data of pressure difference between the inlet and outlet systems of the dust collector, and median filtering was used to process the time series data of dust concentration at the equipment outlet. After processing, pressure difference sequence and dust concentration sequence were obtained respectively. S22: Eliminate interference from dust removal; The preset buffer time after dust removal is used to remove all pressure difference sequences and dust concentration sequences during and after the dust removal process, based on the dust removal signal. S23: Screening steady-state operating conditions; Preset the rated load range of the equipment, determine whether the equipment is operating in a steady state, and eliminate the pressure difference sequence and dust concentration sequence corresponding to non-steady state operation; Define a steady-state judgment threshold for differential pressure, calculate the absolute value of the difference between two adjacent differential pressures within the differential pressure sequence, and determine whether it is a steady-state condition based on the fact that the absolute value of the continuous difference is less than the steady-state judgment threshold for differential pressure. Retain the differential pressure sequence and dust concentration sequence corresponding to the steady-state condition to obtain effective monitoring data.

[0011] Preferably, step S3 includes: S31: Extract steady-state characteristics of pressure difference; Based on effective monitoring data, the pressure difference sequence during the steady-state operating period is obtained, the mean of the pressure difference sequence is calculated, and the maximum absolute value of the difference between each pressure difference value and the mean is extracted to obtain the maximum fluctuation. A preset fluctuation threshold is used to compare the maximum fluctuation with the fluctuation threshold to obtain the pressure difference characteristics, which include 0 and 1. The maximum fluctuation and pressure difference characteristics are extracted as the steady-state characteristics of the pressure difference. S32: Extracting characteristics of sudden changes in dust concentration; S321: Obtain the slope of the time-series change in dust concentration; The dust concentration sequence during steady-state operating conditions was obtained, and the slope of the time series change of dust concentration was obtained by least squares linear fitting. S322: Obtain the cumulative increment; The cumulative increment is obtained based on the difference between the dust concentration value at the end of the dust concentration sequence and the first dust concentration value. S323: Preset threshold for detecting concentration mutations; The thresholds for determining concentration mutations include the minimum slope K and the minimum cumulative increment. ; S324: Obtain concentration mutation characteristics; Compare the slope and cumulative increment of dust concentration over time with the minimum slope K and minimum cumulative increment. The size of the concentration mutation feature is used to obtain the concentration mutation characteristics, which include 0 and 1. S325: Obtain the characteristics of sudden changes in dust concentration; The obtained dust concentration temporal change slope, cumulative increment, and concentration mutation features are extracted as dust concentration mutation features for the steady-state operating period.

[0012] Preferably, step S4 includes: S41: Initial determination of opposing coupling; The pressure difference characteristics and concentration abrupt change characteristics are obtained based on the steady-state characteristics of pressure difference and the abrupt change characteristics of dust concentration. If the pressure difference characteristic is 0 and the concentration change characteristic is also 0, then the operating condition is considered normal. If the differential pressure characteristic is 1 and the concentration change characteristic is also 1, or if the differential pressure characteristic is 1 and the concentration change characteristic is 0, it is determined to be an operating condition disturbance or blockage abnormality. If the differential pressure characteristic is 0 and the concentration change characteristic is 1, it is determined to be a minor leak in the filter media. S42: Classify leakage fault levels; The pressure differential steady-state characteristics and dust concentration abrupt change characteristics of the initially identified minor filter material damage and leakage fault are obtained. The slope, cumulative increment and maximum fluctuation of dust concentration over time are normalized, the comprehensive leakage score is calculated, and the leakage fault level is classified.

[0013] Preferably, step S42 includes: S421: Obtain the normalized value of the slope of the time-series change in dust concentration; The slope of the time-series change in dust concentration is normalized and mapped to the interval [0,1]. S422: Obtain the normalized value of the cumulative increment; Map the cumulative increment within the interval [0,1]. S423: Obtain the maximum fluctuation normalized value; Map the maximum fluctuation within the interval [0,1]. S424: Calculate the overall fault score; The weights of the normalized value of the slope of the dust concentration change over time, the normalized value of the cumulative increment, and the normalized value of the maximum fluctuation are set, and the comprehensive fault score S is obtained by weighted calculation. S425: Classify fault levels; Set a level judgment threshold, compare the comprehensive fault score S with the judgment threshold, and classify the fault level into minor leakage, general leakage and serious leakage.

[0014] Preferably, in step S5: Minor warnings are matched with minor leaks, and pop-up notifications are pushed out. Maintenance suggestions are provided, including routine inspections, tracking and monitoring, and planned shutdowns for maintenance. A general early warning system will be issued for general leaks, pushing alarm messages and suggesting maintenance, requiring shutdown and maintenance at an opportune time, and replacement of the filter cartridge. A severe warning is issued in conjunction with a severe leak, accompanied by audible and visual alarms and a push notification. The maintenance recommendations include immediate shutdown for investigation, emergency replacement of the filter cartridge, and verification of exhaust emissions.

[0015] Preferably, step S6 includes: S61: Archive fault data; Fault identification data includes steady-state characteristics of differential pressure and sudden changes in dust concentration, fault level, and early warning level. Maintenance records include maintenance time and maintenance results. All data are stored in a unified database to establish a fault data archive. S62: Compare the changes in two parameters before and after maintenance; Obtain the steady-state characteristics of differential pressure and the sudden change characteristics of dust concentration after maintenance. Compare the maximum fluctuation, the slope of the time-series change of dust concentration, and the cumulative increment before and after maintenance. If the differential pressure characteristic is 0 after maintenance and the sudden change characteristics of concentration are restored to 0, the fault is determined to be eliminated. Otherwise, it is marked as another fault and re-inspected until the fault is eliminated. S63: Iterative optimization of the parameters of the discriminant model; After the fault has been eliminated, samples of normal operating conditions are collected, and the parameters of the discrimination model are iteratively optimized.

[0016] Preferably, step S63 includes: S631: Collect samples under normal operating conditions and obtain the maximum value of the index; After determining that the fault has been eliminated, the time series data of the pressure difference between the inlet and outlet systems of the dust collector and the time series data of the dust concentration at the equipment outlet are obtained under multiple steady-state operating conditions. Multiple sets of maximum fluctuation, dust concentration time series change slope, and cumulative increment are extracted to obtain the maximum value of the maximum fluctuation, the maximum value of the dust concentration time series change slope, and the maximum value of the cumulative increment. S632: Calculate the temporary new thresholds respectively; The temporary new threshold is obtained by multiplying the maximum fluctuation value, the maximum slope of the dust concentration change over time, and the maximum cumulative increment by a safety factor.

[0017] S633: Weighted fusion yields a new threshold; Set the weights of the current old threshold and the temporary new threshold, and calculate the new threshold by weighting them. S634: Upper and lower limit amplitude processing updates the final new threshold; Set upper and lower limits for the threshold. If the threshold exceeds the upper or lower limit, the value of the excess upper or lower limit is used as the final new threshold. Otherwise, the calculated new threshold is used as the new threshold, and the current old threshold is updated and replaced.

[0018] The system based on the AI-based detection method for filter cartridge damage and leakage characteristics of dust concentration change and pressure difference steady state characteristics includes a multi-source data acquisition module, a data preprocessing module, a dual feature parameter extraction module, an opposing coupled AI detection module, a fault early warning module, and an operation and maintenance closed-loop module. The multi-source data acquisition module is used to acquire raw time-series data in real time. The data preprocessing module is used to filter, remove interference, and screen steady-state conditions from the raw time-series data. The dual feature parameter extraction module is used to simultaneously extract the steady-state features of pressure difference and the sudden change features of dust concentration under the steady-state operating conditions of the equipment. The opposing coupled AI discrimination module is used to distinguish between minor damage and leakage faults of filter media and fluctuations in normal operating conditions, and to classify minor damage and leakage faults of filter media into different levels. The fault early warning module is used to push early warning information and maintenance suggestions in a tiered manner; The operation and maintenance closed-loop module is used to record fault identification data and maintenance records, compare the changes in dual parameters before and after maintenance, and iteratively optimize the AI ​​identification model.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention adopts the opposing coupling discrimination logic of differential pressure steady state and dust concentration change. It utilizes the unique characteristics of constant filter cartridge ventilation resistance and steady system differential pressure under micro-leakage conditions, but continuous dust overflow causing the concentration to gradually increase. It constructs a dual-parameter coupling discrimination model to accurately capture the hidden leaks that traditional differential pressure monitoring cannot identify, eliminate the blind zone of traditional differential pressure monitoring, and fill the technical gap in micro-leakage fault monitoring. (2) This invention abandons the instantaneous threshold judgment method and adopts AI time-series feature algorithm to identify the continuous and trend-based micro-changes in dust concentration. It can capture the gradual concentration increase characteristics in the early stage of leakage, realize the early prediction of faults, greatly reduce the probability of missed alarms, avoid the risk of environmental protection exceeding standards and equipment damage in advance, get rid of the limitation of fixed concentration thresholds, and solve the problems of monitoring lag and high missed alarm rate. (3) This invention eliminates interference factors such as instantaneous fluctuations in dust removal, changes in production line load, and airflow disturbances through working condition screening. It only performs feature discrimination under stable equipment operation conditions, completely avoids false changes in dust concentration caused by fluctuations in conventional working conditions, greatly reduces the fault misjudgment and false alarm rate, reduces false alarm phenomenon from the root, improves the reliability of monitoring results, and significantly improves the discrimination accuracy. (4) This invention achieves accurate fault classification through dual-parameter coupling logic: synchronous fluctuation of differential pressure and concentration change are judged as blockage and working condition disturbance; constant differential pressure and continuous concentration change are judged as filter material damage and leakage, which completely solves the industry pain point that traditional technology cannot distinguish between filter cartridge blockage and micro-leakage. Clearly classify different fault types to provide accurate basis for on-site operation and maintenance. (5) This invention archives and retains fault data and maintenance records in a unified manner. By comparing the changes in pressure difference and dust concentration characteristics before and after maintenance, the fault elimination status is verified. At the same time, the judgment threshold is dynamically updated based on normal working condition samples. With the help of the amplitude limiting protection mechanism, the threshold jump is avoided, so that the AI ​​judgment model can adapt to on-site changes such as equipment aging and working condition deviation in the long term, ensuring the long-term stable and high-precision operation of the system. (6) This invention quantifies and scores the identified filter cartridge micro-leakage faults, classifies them into three levels of faults: minor, general, and serious, and matches them with corresponding early warning strategies and maintenance suggestions, changing the traditional technology mode of only making a single fault alarm. Maintenance personnel can reasonably arrange maintenance plans according to the severity of the fault, transforming from passive fault handling to proactive predictive maintenance, taking into account both equipment operation safety and operation and maintenance efficiency. Attached Figure Description

[0020] Figure 1 This is a flowchart of the steps of the AI-based filter cartridge damage and leakage discrimination method based on the dust concentration abrupt change and pressure difference steady-state characteristics of the present invention. Detailed Implementation

[0021] To provide a further understanding of the purpose, structure, features, and functions of the present invention, detailed descriptions are provided below with reference to specific embodiments. Example

[0022] Please refer to the reference. Figure 1 This invention provides an AI-based method for identifying filter cartridge damage and leakage based on abrupt changes in dust concentration and steady-state characteristics of pressure difference, comprising the following steps: S1: Collect data from multiple sources; Real-time synchronous acquisition of differential pressure data at the inlet and outlet of the dust collector, dust concentration data at the equipment outlet, equipment operating load data, and pulse cleaning condition data to obtain raw time-series data; all data are sampled at the same frequency and timestamps are aligned.

[0023] The time-series data of pressure difference between the inlet and outlet systems of a dust collector refers to the sequence value of the gas static pressure difference between the inlet and outlet pipes of the dust collector as a function of time, including the pressure difference value and timestamp.

[0024] The time series data of dust concentration at the equipment outlet refers to the sequence value of the mass concentration of dust in the exhaust pipe after the dust collector is purified, which changes over time, including dust concentration value, timestamp, etc. Equipment operating load data are parameters that characterize the actual operating conditions of the dust collector and the front-end production line, as well as the processing air volume / dust handling capacity; including timestamps and fan frequency.

[0025] The pulse cleaning operation data records the timing of the filter cartridge pulse backflushing cleaning system's actions, including cleaning signals and timestamps. A cleaning signal of 0 indicates no cleaning, and 1 indicates cleaning in progress.

[0026] It not only collects the core parameters of the fault, but also incorporates operating condition data such as load and dust removal, providing a complete data source for distinguishing between normal disturbances and real faults.

[0027] By deploying various types of sensing units on the dust removal equipment, including differential pressure sensors, dust concentration sensors, load sensors, and pulse signal acquisition units; A fixed sampling frequency is uniformly set, and the sampling frequency and sampling duration of all sensing units are completely consistent. Each set of collected data is stamped with a uniform timestamp to ensure that the differential pressure value, dust concentration value, fan frequency, and dust removal signal timing are completely aligned to avoid data misalignment.

[0028] S2: Data preprocessing; The original time-series data is filtered, dust interference is removed, and steady-state operating conditions are screened to obtain effective monitoring data under steady-state operation of the equipment.

[0029] S21: Filtering process; First-order low-pass filtering is used to process the time-series data of the pressure difference between the inlet and outlet systems of the dust collector. For the time-series data of dust concentration at the equipment outlet, a median filtering method is used: A sliding window is formed by taking three points before and after the current sampling point (a total of seven points). The concentration values ​​within the window are sorted by magnitude, and the median is taken as the filtered output for that point. This avoids instantaneous noise being misidentified as sudden concentration changes or pressure fluctuations, thus improving the accuracy of the original data.

[0030] After processing, the smoothed pressure difference sequence and dust concentration sequence were obtained respectively.

[0031] S22: Eliminate interference from dust removal; The preset buffer time after dust removal is 20 seconds. Based on the dust removal signal, when the dust removal signal is 1, the period is determined to be in the process of dust removal. All pressure difference sequences and dust concentration sequences within this period are directly discarded, and the timestamp of the last dust removal signal being 1 is obtained. All pressure difference sequences and dust concentration sequences within the buffer time after the obtained timestamp are then discarded. This completely avoids false increases in concentration caused by dust generation during dust removal. It distinguishes between fluctuations in dust removal conditions and true concentration increases caused by filter cartridge leakage, solving the problem of frequent false alarms due to dust removal interference in traditional monitoring.

[0032] S23: Screening steady-state operating conditions; The system presets the rated load range of the equipment and compares the fan frequency of the load data with the rated load range of the equipment. If not, it determines that the equipment is operating in an unsteady state and removes the corresponding differential pressure sequence and dust concentration sequence.

[0033] Define a steady-state pressure difference threshold. Calculate the absolute value of the difference between two adjacent pressure differences within a pressure difference sequence. If the absolute value of at least 20 consecutive differences is less than the steady-state pressure difference threshold, the continuous period is determined to be a steady-state condition, and the corresponding steady-state pressure difference sequence and dust concentration sequence are retained. Otherwise, it is determined to be a non-steady-state condition, and the non-steady-state pressure difference sequence and dust concentration sequence are discarded. Traverse the pressure difference sequences, retaining all steady-state pressure difference sequences and dust concentration sequences to obtain valid monitoring data.

[0034] By screening operating conditions, interference factors such as instantaneous fluctuations in dust removal, changes in production line load, and airflow disturbances are eliminated. Feature identification is only performed when the equipment is operating stably, which completely avoids false changes in dust concentration caused by fluctuations in normal operating conditions, greatly reduces the rate of fault misjudgment and false alarm, and significantly improves the accuracy of identification.

[0035] S3: Extract dual feature parameters; Based on effective monitoring data, under steady-state operating conditions of the equipment, the steady-state characteristics of pressure difference and the abrupt change characteristics of dust concentration are extracted simultaneously.

[0036] S31: Extract steady-state characteristics of pressure difference; Calculate and extract the maximum fluctuation and pressure difference characteristics; Based on valid monitoring data, the differential pressure sequence during steady-state operating conditions is obtained, and the mean of the differential pressure sequence is calculated. and maximum fluctuation ;according to Where n is the total number of differential pressure sequences during the current steady-state operating period; It is the i-th differential pressure value in the differential pressure sequence during the current steady-state operating period.

[0037] Each pressure difference value in the pressure difference series and the mean absolute value of the difference Obtain the differential pressure fluctuation sequence; then according to the formula Extract each pressure difference value and the mean from the pressure difference sequence. The maximum absolute value of the difference yields the maximum fluctuation. ; Preset fluctuation threshold ,like If the pressure difference is within a certain range, the pressure difference feature is marked as 0, indicating that the pressure difference is in a steady state throughout, with no significant rise or fall. Otherwise, the pressure difference feature is marked as 1, indicating that the pressure difference is not in a steady state. The maximum fluctuation and pressure difference feature are extracted as the steady-state features of the pressure difference.

[0038] The quantitative characterization of differential pressure fluctuations, with the differential pressure characteristics being standardized binary labels, provides a core basis for subsequent discrimination of opposing couplings, thus overcoming the shortcomings of traditional differential pressure monitoring which can only identify severe faults and cannot determine steady-state conditions.

[0039] S32: Extracting characteristics of sudden changes in dust concentration; Calculate and extract the slope of the time-series change in dust concentration, the cumulative increment, and the characteristics of concentration abrupt changes.

[0040] S321: Obtain the slope of the time-series change in dust concentration; The dust concentration sequence during steady-state operating conditions was obtained, and the slope of the time-series change in dust concentration was obtained by least-squares linear fitting; according to the formula: Obtain the slope of the time-series change in dust concentration; where n is the total number of sampling points in the dust concentration sequence, and t i Let t be the time sequence number of the i-th sampling point in the dust concentration sequence. i =1,2,…,n. This represents the preprocessed dust concentration value at the i-th sampling point in the dust concentration sequence. This is the sum of all time indices in the dust concentration sequence. It is the sum of all dust concentration values ​​in the dust concentration sequence. It is the sum of squares of all time indices in the dust concentration sequence. It captures the continuous trend of dust concentration changes, no longer relying on single-point instantaneous concentration values, and can identify small leaks with early, slow increases.

[0041] S322: Obtain the cumulative increment; According to the formula The cumulative increment is obtained by calculating the concentration difference between the end and the beginning of the dust concentration sequence; where... It is the dust concentration value at the last sampling point in the dust concentration sequence; This is the dust concentration value of the first sampling point in the dust concentration sequence. It quantifies the overall increase in dust concentration over a period of time, and combines the slope to determine persistent abrupt changes, distinguishing between instantaneous random fluctuations and gradual leakage, thus reducing the probability of missed detections.

[0042] S323: Preset threshold for detecting concentration mutations; The thresholds for determining concentration mutations include the minimum slope K and the minimum cumulative increment. Initially, on-site calibration can be performed by collecting one hour of normal operating data under known leak-free conditions. A 5-minute sliding window is used to calculate the time-series slope and cumulative increment of dust concentration for each window, obtaining the maximum value of the time-series slope and the maximum value of the cumulative increment. The safety factor α is set to 1.2-1.5. The minimum slope K = maximum value of the time-series slope of dust concentration * α; minimum cumulative increment... = Maximum cumulative increment * α.

[0043] Thresholds are calibrated based on actual normal operating data of the equipment to adapt to the on-site working conditions of different dust collectors and production lines. Compared with general fixed thresholds, the judgment standard is more in line with the actual on-site conditions, and has higher versatility and accuracy.

[0044] S324: Obtain concentration mutation characteristics; Compare the slope of the time-series change in dust concentration, the cumulative increase, and the threshold for determining abrupt concentration changes; if the slope of the time-series change in dust concentration... Minimum slope K, and cumulative increment Minimum cumulative increment If the concentration change feature is positive, it is labeled as 1, indicating a continuous micro-change in concentration and a sustained increase; otherwise, it is labeled as 0, indicating a stable fluctuation in concentration. Converting the concentration trend and increase into standardized binary labels, and pairing them with the pressure difference feature as input, facilitates machine recognition.

[0045] S325: Obtain the characteristics of sudden changes in dust concentration; The obtained dust concentration temporal change slope, cumulative increment, and concentration mutation features are extracted as dust concentration mutation features for the steady-state operating period.

[0046] By using AI time-series algorithms to identify the continuous and trend-based micro-sudden increases in dust concentration, which differs from instantaneous random fluctuations, it can accurately identify the gradual deterioration trend of early minor leaks, enabling ultra-early fault prediction, far earlier than traditional threshold alarm timing.

[0047] S4: Opposing Coupling AI Judgment; Based on dual feature parameters, AI intelligently distinguishes between filter cartridge damage and leakage faults and fluctuations in normal operating conditions, and classifies the level of filter cartridge damage and leakage faults.

[0048] S41: Initial determination of opposing coupling; The pressure difference characteristics and concentration abrupt change characteristics are obtained based on the steady-state characteristics of pressure difference and the abrupt change characteristics of dust concentration. If the pressure difference characteristic is 0 and the concentration change characteristic is also 0, and there is no continuous change in the pressure difference steady state and dust concentration, then the operating condition is judged to be normal. If the differential pressure characteristic is 1 and the concentration change characteristic is also 1, or if the differential pressure characteristic is 1 and the concentration change characteristic is 0, and the differential pressure fluctuates significantly synchronously, with or without concentration changes, it is judged as an operating condition disturbance or blockage abnormality. If the differential pressure characteristic is 0 and the concentration mutation characteristic is 1, the differential pressure remains steady and without fluctuation throughout the process, but the dust concentration continues to mutate, then it is determined to be a minor damage and leakage fault of the filter material.

[0049] Breaking away from traditional single-parameter and same-parameter change-based discrimination methods, this system is the first to employ a dual-parameter, opposing feature coupling discrimination mechanism. Addressing the unique characteristics of micro-damage leaks in filter cartridges—small pore size, no impact on overall ventilation resistance, and no significant change in differential pressure, but with continuous penetration of trace dust and a continuous, abrupt increase in concentration—a dedicated discrimination model is constructed. This model accurately captures hidden leaks that traditional technologies cannot identify, filling a technological gap in the industry. Through dual-parameter coupling logic, precise fault classification is achieved: synchronous fluctuations in differential pressure and changes in concentration are identified as blockage or operational disturbances; a steady-state differential pressure combined with continuous, abrupt changes in concentration is identified as filter media damage and leakage. This completely solves the industry pain point of traditional technologies being unable to distinguish between filter cartridge blockage and micro-leakage.

[0050] S42: Classify leakage fault levels; The pressure differential steady-state characteristics and dust concentration abrupt change characteristics of the initially identified minor filter material damage and leakage fault are obtained. The slope, cumulative increment and maximum fluctuation of dust concentration over time are normalized, the comprehensive leakage score is calculated, and the leakage fault level is classified.

[0051] S421: Obtain the normalized value of the slope of the time-series change in dust concentration; According to the formula The normalized slope of the time-series change in dust concentration is calculated and mapped to the interval [0,1]. Here, k is the slope of the time-series change in dust concentration during the current steady-state operating period, and K is the preset minimum slope; k max To determine the maximum slope of minor leaks, an initial value is set. Based on the judgment of minor leaks caused by filter media damage, the slope of dust concentration over time is compared with the initial value. If the initial value is greater, the maximum slope of minor leaks is updated to the slope of dust concentration over time; otherwise, it remains unchanged, and k is iteratively updated.max .

[0052] S422: Obtain a cumulative increment normalized value; According to the formula calculate and obtain the cumulative increment normalized value, which maps the cumulative increment to the interval [0,1]. Where is the cumulative increment in the current steady-state working condition period, is the preset minimum cumulative increment, is the maximum cumulative increment of micro-leakage, which is obtained by presetting an initial value, comparing the cumulative increment with the initial value according to the judgment result of the filter material micro-damage leakage fault, if it is greater, updating the maximum micro-leakage slope to the cumulative increment, otherwise keeping it unchanged, and performing iterative updating .

[0053] S423: Obtain a maximum fluctuation normalized value; According to the formula calculate and obtain the maximum fluctuation normalized value, S P is within the interval [0,1]; where is the maximum fluctuation in the current steady-state working condition period, is the fluctuation threshold.

[0054] The normalization process eliminates the difference in parameter units, and provides a unified calculation basis for comprehensive scoring.

[0055] S424: Calculate a comprehensive fault score; Set the weights of the normalized value of the time-series change slope of dust concentration, the cumulative increment normalized value and the maximum fluctuation normalized value, and calculate the comprehensive fault score ; where w1, w2 and w3 are preset weights, w1+w2+w3=1; set w1=0.4; w2=0.4; w3=0.2. The weighted algorithm integrates multiple features to quantify the leakage severity.

[0056] S425: Divide fault levels; Set level judgment thresholds, compare the magnitude of the comprehensive fault score S with the judgment thresholds, and divide the fault levels into slight leakage, general leakage and severe leakage.

[0057] Specifically: if 0<S≤0.3, the fault level is determined as slight leakage; it means that the filter cartridge has tiny holes, the leakage amount is extremely small, and there is no short-term environmental risk.

[0058] If 0.3<S≤0.7, the fault level is determined as general leakage; it means that the filter cartridge is partially cracked, the leakage continues to increase, and there is a potential risk.

[0059] If 0.7<S≤1.0, the fault level is determined as severe leakage. It means that the damage range of the filter cartridge is expanded, the leakage is obvious, and it is close to the exceeding standard threshold.

[0060] Minor leaks can be classified as light, medium, or severe, which intuitively reflects the development trend of the leak, provides a basis for differentiated operation and maintenance, and enables predictive maintenance.

[0061] S5: Fault warning; The warning system is divided into three levels: minor, moderate, and severe, based on the severity of the problem. Warning information and maintenance suggestions are pushed to the corresponding level of filter cartridge damage and leakage.

[0062] For minor leaks, a warning will be issued, and a pop-up notification will be sent. Maintenance suggestions will be provided, including routine inspections, monitoring, and planned shutdowns for maintenance.

[0063] A general early warning system will detect a general leak and send an alarm message. Maintenance recommendations will be sent, requiring the system to be shut down for maintenance and the filter cartridge to be replaced.

[0064] A severe warning is issued in conjunction with a severe leak, accompanied by audible and visual alarms and a push notification. The maintenance recommendations include immediate shutdown for investigation, emergency replacement of the filter cartridge, and verification of exhaust emissions.

[0065] Tiered push notifications ensure that maintenance personnel can intuitively grasp the risk level, rationally arrange maintenance times, and improve maintenance efficiency.

[0066] S6: Closed-loop operation and maintenance; Record fault identification data and maintenance records, compare the changes in dual parameters before and after maintenance, and iteratively optimize the AI ​​identification model.

[0067] S61: Archive fault data.

[0068] Fault identification data includes steady-state characteristics of differential pressure and sudden changes in dust concentration, fault level, early warning level, etc., and maintenance records include maintenance time, maintenance results, etc.; all are uniformly stored in the database to establish a fault data archive for easy source tracing.

[0069] S62: Compare the changes in two parameters before and after maintenance; After maintenance, collect time-series data on the differential pressure between the dust collector's inlet and outlet systems and the dust concentration at the equipment outlet under steady-state conditions. Obtain the maximum fluctuation, differential pressure characteristics, the slope of the dust concentration change over time, the cumulative increment, and the concentration mutation characteristics. Compare the maximum fluctuation, the slope of the dust concentration change over time, and the cumulative increment before and after maintenance. If the differential pressure characteristic is 0 and the concentration mutation characteristic returns to 0 after maintenance, it indicates that the concentration trend is stable, and the fault is determined to be eliminated. Otherwise, mark it as another fault and re-maintain until the fault is eliminated. Verify the maintenance effect to confirm whether the fault has been completely eliminated; at the same time, distinguish filter cartridge leakage from other types of faults.

[0070] S63: Iterative optimization of the parameters of the discriminant model.

[0071] After the fault has been eliminated, samples of normal operating conditions are collected, and the parameters of the discrimination model are iteratively optimized.

[0072] S631: Collect samples under normal operating conditions and obtain the maximum value of the index; After determining that the fault has been eliminated, obtain the time-series data of the pressure difference between the inlet and outlet systems of the dust collector and the time-series data of the dust concentration at the equipment outlet under N steady-state operating conditions. Extract N sets of maximum fluctuation, dust concentration time-series change slope, and cumulative increment respectively; N is 20. Obtain the maximum values ​​of the maximum fluctuation, dust concentration time-series change slope, and cumulative increment respectively, and set them as the maximum value of the maximum fluctuation. Maximum slope of dust concentration over time and the maximum cumulative increment The baseline values ​​are calculated based on the latest normal operating conditions of the equipment, and are adapted to long-term changes in the field, such as filter cartridge aging and slow changes in operating conditions.

[0073] S632: Calculate the temporary new thresholds respectively; Use the maximum fluctuation value Maximum slope of dust concentration over time and the maximum cumulative increment The results are obtained by multiplying each by a safety factor α.

[0074] According to the formula: Temporary threshold for volatility Temporary minimum slope Temporary minimum cumulative increment Where α is the safety factor.

[0075] S633: Weighted fusion yields a new threshold; Using a 9:1 weighting, the new threshold is the sum of the current old threshold multiplied by 0.9 and the temporary new threshold multiplied by 0.1, i.e.: ;in The current fluctuation threshold, This is a temporary threshold for fluctuation.

[0076] Where K is the current minimum slope, This is the temporary minimum slope.

[0077] ;in This is the current minimum cumulative increment. This is the temporary minimum cumulative increment.

[0078] The original threshold is used as the primary reference, supplemented by new sample data, to make smooth updates, avoiding large jumps in the threshold caused by single operating condition fluctuations, and ensuring a smooth transition of the judgment logic.

[0079] S634: Upper and lower limit amplitude processing updates the final new threshold.

[0080] Set upper and lower limits for the threshold, and compare the new threshold with the old threshold to see if the new threshold exceeds these limits. Specifically, check if the new threshold is higher than 130% of the old threshold or lower than 70% of the old threshold. If it exceeds these limits, the excess value is used as the final new threshold; otherwise, the calculated new threshold is used to update and replace the old threshold. Setting upper and lower limits for threshold fluctuation strictly constrains the threshold adjustment range and ensures operational stability.

[0081] Example 2 The filter cartridge damage and leakage AI discrimination system based on dust concentration change and pressure difference steady-state characteristics includes a multi-source data acquisition module, a data preprocessing module, a dual feature parameter extraction module, an opposing coupled AI discrimination module, a fault early warning module, and an operation and maintenance closed-loop module.

[0082] The multi-source data acquisition module is used to collect real-time data of differential pressure between the inlet and outlet systems of the dust collector, time-series data of dust concentration at the equipment outlet, equipment operating load data, and pulse cleaning condition data; ensuring that the time sequence of the dual-parameter data is synchronized and the sampling frequency is consistent; The data preprocessing module is used to filter the raw time series data, remove dust cleaning interference, and screen steady-state operating conditions. It removes invalid interference data caused by instantaneous dust during dust cleaning, airflow disturbance, and load fluctuation, and retains only the effective monitoring data under steady-state operation of the equipment. The dual feature parameter extraction module is used to simultaneously extract the steady-state features of pressure difference and the sudden change features of dust concentration under the steady-state operation of the equipment, determine that the fluctuation range of system pressure difference is within the preset steady-state range and there is no obvious rise or fall, and at the same time extract the change features of continuous time-series micro-sudden changes and continuous small increases in outlet dust concentration.

[0083] The AI ​​discrimination module for opposing coupling has a built-in dual-parameter opposing coupling discrimination logic. It is used to combine the exclusive correlation characteristics of "no change in steady-state pressure difference and continuous abrupt changes in dust concentration" to intelligently distinguish between minor damage and leakage faults of filter media and fluctuations in normal operating conditions. If the dust concentration fluctuation is accompanied by a synchronous change in pressure difference, it is judged as an operating condition disturbance or blockage abnormality. If the dust concentration continuously changes abruptly and the pressure difference remains steady, it is judged as a minor damage and leakage fault of filter media. The minor damage and leakage fault of filter media is then classified into different levels.

[0084] The fault early warning module is used to classify the warning into three levels—minor, moderate, and severe—based on the severity of filter cartridge damage and leakage, and to push warning information and maintenance suggestions according to the corresponding fault level. The operation and maintenance closed-loop module is used to record fault identification data and maintenance records, compare the changes in dual parameters before and after maintenance, iteratively optimize the AI ​​identification model, and form a closed-loop process of fault identification, early warning, maintenance and model optimization.

[0085] The present invention has been described in the above-described embodiments; however, these embodiments are merely examples for implementing the present invention. It must be noted that the disclosed embodiments do not limit the scope of the present invention. Conversely, any modifications and refinements made without departing from the spirit and scope of the present invention are within the scope of patent protection of the present invention.

Claims

1. An AI-based method for detecting filter cartridge damage and leakage based on sudden changes in dust concentration and steady-state characteristics of pressure difference, characterized in that: Includes the following steps: S1: Collect data from multiple sources; Real-time synchronous acquisition of differential pressure data at the inlet and outlet of the dust collector, dust concentration data at the equipment outlet, equipment operating load data, and pulse cleaning condition data to obtain raw time-series data; S2: Data preprocessing; The raw time series data is filtered, dust removal interference is eliminated, and steady-state operating conditions are screened to obtain effective monitoring data under steady-state operation of the equipment. S3: Extract dual feature parameters; Based on effective monitoring data, under steady-state operating conditions of the equipment, the steady-state characteristics of pressure difference and the sudden change characteristics of dust concentration are extracted simultaneously. S4: Opposing Coupling AI Judgment; Based on dual feature parameters, AI intelligently distinguishes between filter cartridge damage and leakage faults and normal operating condition fluctuations, and classifies the level of filter cartridge damage and leakage faults. S5: Fault warning; The warning system is divided into three levels: minor, moderate, and severe, and the warning information and maintenance suggestions are pushed out according to the corresponding level of filter cartridge damage and leakage. S6: Closed-loop operation and maintenance; Record fault identification data and maintenance records, compare the changes in dual parameters before and after maintenance, and iteratively optimize the AI ​​identification model.

2. The filter cartridge damage and leakage AI discrimination method based on dust concentration abrupt change and pressure difference steady-state characteristics as described in claim 1, characterized in that: The equipment outlet dust concentration time series data includes dust concentration value and timestamp; the equipment operating load data includes timestamp and fan frequency; the pulse cleaning condition data includes cleaning signal and timestamp, and the cleaning signal includes no cleaning and cleaning in progress; all data are sampled at the same frequency and the timestamps are aligned.

3. The filter cartridge damage and leakage AI discrimination method based on dust concentration abrupt change and pressure difference steady-state characteristics as described in claim 1, characterized in that: Step S2 includes: S21: Filtering process; First-order low-pass filtering was used to process the time series data of pressure difference between the inlet and outlet systems of the dust collector, and median filtering was used to process the time series data of dust concentration at the equipment outlet. After processing, pressure difference sequence and dust concentration sequence were obtained respectively. S22: Eliminate interference from dust removal; The preset buffer time after dust removal is used to remove all pressure difference sequences and dust concentration sequences during and after the dust removal process, based on the dust removal signal. S23: Screening steady-state operating conditions; Preset the rated load range of the equipment, determine whether the equipment is operating in a steady state, and eliminate the pressure difference sequence and dust concentration sequence corresponding to non-steady state operation; Define a steady-state judgment threshold for differential pressure, calculate the absolute value of the difference between two adjacent differential pressures within the differential pressure sequence, and determine whether it is a steady-state condition based on the fact that the absolute value of the continuous difference is less than the steady-state judgment threshold for differential pressure. Retain the differential pressure sequence and dust concentration sequence corresponding to the steady-state condition to obtain effective monitoring data.

4. The filter cartridge damage and leakage AI discrimination method based on dust concentration abrupt change and pressure difference steady-state characteristics as described in claim 1, characterized in that: Step S3 includes: S31: Extract steady-state characteristics of pressure difference; Based on effective monitoring data, the pressure difference sequence during the steady-state operating period is obtained, the mean of the pressure difference sequence is calculated, and the maximum absolute value of the difference between each pressure difference value and the mean is extracted to obtain the maximum fluctuation. A preset fluctuation threshold is used to compare the maximum fluctuation with the fluctuation threshold to obtain the pressure difference characteristics, which include 0 and 1. The maximum fluctuation and pressure difference characteristics are extracted as the steady-state characteristics of the pressure difference. S32: Extracting characteristics of sudden changes in dust concentration; S321: Obtain the slope of the time-series change in dust concentration; The dust concentration sequence during steady-state operating conditions was obtained, and the slope of the time series change of dust concentration was obtained by least squares linear fitting. S322: Obtain the cumulative increment; The cumulative increment is obtained based on the difference between the dust concentration value at the end of the dust concentration sequence and the first dust concentration value. S323: Preset threshold for detecting concentration mutations; The thresholds for determining concentration mutations include the minimum slope K and the minimum cumulative increment. ; S324: Obtain concentration mutation characteristics; Compare the slope and cumulative increment of dust concentration over time with the minimum slope K and minimum cumulative increment. The size of the concentration mutation feature is used to obtain the concentration mutation characteristics, which include 0 and 1. S325: Obtain the characteristics of sudden changes in dust concentration; The obtained dust concentration temporal change slope, cumulative increment, and concentration mutation features are extracted as dust concentration mutation features for the steady-state operating period.

5. The filter cartridge damage and leakage AI discrimination method based on dust concentration mutation and pressure difference steady-state characteristics as described in claim 1, characterized in that: Step S4 includes: S41: Initial determination of opposing coupling; The pressure difference characteristics and concentration abrupt change characteristics are obtained based on the steady-state characteristics of pressure difference and the abrupt change characteristics of dust concentration. If the pressure difference characteristic is 0 and the concentration change characteristic is also 0, then the operating condition is judged to be normal. If the pressure difference characteristic is 1 and the concentration change characteristic is also 1, or if the pressure difference characteristic is 1 and the concentration change characteristic is 0, it is determined to be an operating condition disturbance or blockage abnormality. If the differential pressure characteristic is 0 and the concentration change characteristic is 1, it is determined to be a minor leak fault due to minor damage to the filter material. S42: Classify leakage fault levels; The pressure differential steady-state characteristics and dust concentration abrupt change characteristics of the initially identified minor filter material damage and leakage fault are obtained. The slope, cumulative increment and maximum fluctuation of dust concentration over time are normalized, the comprehensive leakage score is calculated, and the leakage fault level is classified.

6. The filter cartridge damage and leakage AI discrimination method based on dust concentration abrupt change and pressure difference steady-state characteristics as described in claim 5, characterized in that: Step S42 includes: S421: Obtain the normalized value of the slope of the time-series change in dust concentration; The slope of the time-series change in dust concentration is normalized and mapped to the interval [0,1]. S422: Obtain the normalized value of the cumulative increment; Map the cumulative increment within the interval [0,1]. S423: Obtain the maximum fluctuation normalized value; Map the maximum fluctuation within the interval [0,1]. S424: Calculate the overall fault score; The weights of the normalized value of the slope of the time-series change of dust concentration, the normalized value of the cumulative increment, and the normalized value of the maximum fluctuation are set, and the comprehensive fault score S is obtained by weighted calculation. S425: Classify fault levels; Set a level judgment threshold, compare the comprehensive fault score S with the judgment threshold, and classify the fault level into minor leakage, general leakage and serious leakage.

7. The filter cartridge damage and leakage AI discrimination method based on dust concentration abrupt change and pressure difference steady-state characteristics as described in claim 6, characterized in that: In step S5: Minor warnings are matched with minor leaks, and pop-up notifications are pushed out. Maintenance suggestions are provided, including routine inspections, tracking and monitoring, and planned shutdowns for maintenance. A general early warning system will be issued for general leaks, pushing alarm messages and suggesting maintenance shutdowns for inspection and replacement of filter cartridges. A severe warning is issued in conjunction with a severe leak, accompanied by audible and visual alarms and a push notification. The maintenance recommendations include immediate shutdown for investigation, emergency replacement of the filter cartridge, and verification of exhaust emissions.

8. The filter cartridge damage and leakage AI discrimination method based on dust concentration abrupt change and pressure difference steady-state characteristics as described in claim 1, characterized in that: Step S6 includes: S61: Archive fault data; Fault identification data includes steady-state characteristics of differential pressure and sudden changes in dust concentration, fault level, and early warning level. Maintenance records include maintenance time and maintenance results. All data are stored in a unified database to establish a fault data archive. S62: Compare the changes in two parameters before and after maintenance; Obtain the steady-state characteristics of differential pressure and the sudden change characteristics of dust concentration after maintenance. Compare the maximum fluctuation, the slope of the time-series change of dust concentration, and the cumulative increment before and after maintenance. If the differential pressure characteristic is 0 after maintenance and the sudden change characteristics of concentration are restored to 0, the fault is determined to be eliminated. Otherwise, it is marked as another fault and re-inspected until the fault is eliminated. S63: Iterative optimization of the parameters of the discriminant model; After the fault has been eliminated, samples of normal operating conditions are collected, and the parameters of the discrimination model are iteratively optimized.

9. The filter cartridge damage and leakage AI discrimination method based on dust concentration mutation and pressure difference steady-state characteristics as described in claim 8, characterized in that: Step S63 includes: S631: Collect samples under normal operating conditions and obtain the maximum value of the index; After determining that the fault has been eliminated, the time series data of the pressure difference between the inlet and outlet systems of the dust collector and the time series data of the dust concentration at the equipment outlet are obtained under multiple steady-state operating conditions. Multiple sets of maximum fluctuation, dust concentration time series change slope, and cumulative increment are extracted to obtain the maximum value of the maximum fluctuation, the maximum value of the dust concentration time series change slope, and the maximum value of the cumulative increment. S632: Calculate the temporary new thresholds respectively; The temporary new threshold is obtained by multiplying the maximum fluctuation value, the maximum slope of the dust concentration change over time, and the maximum cumulative increment by a safety factor, respectively. S633: Weighted fusion yields a new threshold; Set the weights of the current old threshold and the temporary new threshold, and calculate the new threshold by weighting them. S634: Upper and lower limit amplitude processing updates the final new threshold; Set upper and lower limits for the threshold. If the threshold exceeds the upper or lower limit, the value of the excess upper or lower limit is used as the final new threshold. Otherwise, the calculated new threshold is used as the new threshold, and the current old threshold is updated and replaced.

10. The system for AI-based filter cartridge damage and leakage discrimination method based on dust concentration abrupt changes and pressure differential steady-state characteristics as described in any one of claims 1-9, characterized in that: It includes a multi-source data acquisition module, a data preprocessing module, a dual-feature parameter extraction module, an opposing and coupled AI discrimination module, a fault early warning module, and an operation and maintenance closed-loop module; The multi-source data acquisition module is used to acquire raw time-series data in real time. The data preprocessing module is used to filter, remove interference, and screen steady-state conditions from the raw time-series data. The dual feature parameter extraction module is used to simultaneously extract the steady-state features of pressure difference and the sudden change features of dust concentration under the steady-state operating conditions of the equipment. The opposing coupled AI discrimination module is used to distinguish between minor damage and leakage faults of filter media and fluctuations in normal operating conditions, and to classify minor damage and leakage faults of filter media into different levels. The fault early warning module is used to push early warning information and maintenance suggestions in a tiered manner; The operation and maintenance closed-loop module is used to record fault identification data and maintenance records, compare the changes in dual parameters before and after maintenance, and iteratively optimize the AI ​​identification model.

Images