Data acquisition method and predictive maintenance system of edge gateway

Abstract

The invention relates to the technical field of edge gateway health management and predictive maintenance, in particular to an edge gateway data collection method and predictive maintenance system.The edge gateway data collection method comprises the steps that multi-dimensional state data of an edge gateway are collected, and the multi-dimensional state data comprise working condition and environment data and key component degradation data; calculating the comprehensive degradation rate of the capacitor, and determining the health degree of the capacitor and the health degree of the flash memory; calculating a basic health degree; determining an environment penalty factor; a gateway comprehensive health degree score is determined through fusion processing of a preset multiplication penalty model; estimating the instantaneous degradation rate through a preset state space model, and determining the remaining service life; a corrected data acquisition frequency is calculated through load adaptive closed-loop control logic, and an adaptive control instruction is output; according to the method, traditional passive response is converted into active predictive maintenance, and production line shutdown and data interruption caused by unexpected failure are effectively avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of edge gateway health management and predictive maintenance, in particular to a data acquisition method of an edge gateway and a predictive maintenance system. BACKGROUND

[0002] As the core equipment of industrial data acquisition, the running stability of the edge gateway is crucial to ensure the continuity of the industrial process. The edge gateway is usually deployed in harsh working conditions such as high temperature, vibration, and electrical stress fluctuation. It faces the risks of chronic degradation of key components such as capacitors and flash memory, and acute impact of external environment.

[0003] Existing maintenance strategies are mostly passive responses or rely on single-dimensional state monitoring. These methods are difficult to integrate multi-source heterogeneous data, cannot accurately assess the comprehensive health status of the gateway under multi-stress coupling, and cannot effectively predict its remaining useful life. Therefore, when the gateway fails unexpectedly, it is easy to cause data interruption and production line downtime, which poses a serious safety hazard and economic loss. SUMMARY

[0004] To solve the above technical problems, the present application provides a data acquisition method of an edge gateway and a predictive maintenance system. Specifically, the technical solution of the present application is:

[0005] The data acquisition method of the edge gateway comprises:

[0006] S1, acquiring multi-dimensional state data of the edge gateway, the multi-dimensional state data comprising working condition and environmental data, and key component degradation data;

[0007] S2, based on the working condition and environmental data and a preset physical failure model, calculating the comprehensive degradation rate of the capacitor and determining the capacitor health degree; and based on the key component degradation data and a preset data-driven model, determining the flash memory health degree;

[0008] S3, combining the capacitor health degree and the flash memory health degree, and calculating the basic health degree according to a preset component importance weight; and based on the working condition and environmental data and a pre-trained anomaly detection model, determining an environmental penalty factor;

[0009] S4, fusing and processing the basic health degree and the environmental penalty factor through a preset multiplicative penalty model to determine the gateway comprehensive health degree score; and based on the time series of the gateway comprehensive health degree score, estimating the instantaneous degradation rate through a preset state space model to determine the remaining useful life;

[0010] S5, based on the real-time on-board temperature in the working condition and environment data, the preset safety temperature threshold, the preset load-temperature control coefficient, the preset minimum performance factor and the original data acquisition frequency, calculating the corrected data acquisition frequency through the load adaptive closed-loop control logic, and outputting the adaptive control instruction.

[0011] Preferably, the working condition and environment data include: CPU occupancy; network throughput; on-board temperature; environmental humidity; three-axis vibration acceleration; and power input voltage fluctuation.

[0012] Preferably, the key component degradation data includes: cumulative erase count of NANDFlash; ECC error correction count; number of bad block growth; and output voltage ripple of power module.

[0013] Preferably, the comprehensive degradation rate of the capacitor is calculated by a physical failure model, including:

[0014] Based on the Arrhenius model and the on-board temperature, the basic thermal aging rate is calculated;

[0015] Combined with the basic thermal aging rate, the CPU occupancy and the output voltage ripple, and introducing the preset load and voltage stress sensitivity coefficient, the comprehensive degradation rate of the capacitor is calculated.

[0016] Preferably, the environmental penalty factor is determined, including:

[0017] Taking the three-axis vibration acceleration, the environmental humidity and the power input voltage fluctuation as the input of the abnormality detection model, the vibration real-time abnormality score, the humidity real-time abnormality score and the power real-time abnormality score are calculated respectively;

[0018] Combined with the vibration real-time abnormality score, the humidity real-time abnormality score, the power real-time abnormality score and the preset environmental stress penalty weight, the environmental penalty factor is determined through minimum amplitude limiting processing.

[0019] Preferably, the remaining useful life is determined, including:

[0020] Taking the time series of the gateway comprehensive health score as the observation value, the state space model of the particle filtering or Kalman filtering framework is applied to estimate the instantaneous degradation rate;

[0021] Based on the instantaneous degradation rate, the current gateway comprehensive health score and the preset failure threshold, the remaining useful life is calculated through linear extrapolation.

[0022] Preferably, it further includes:

[0023] In response to the gateway comprehensive health score being less than the preset health threshold, or the remaining useful life being less than the preset time threshold, triggering the predictive maintenance alarm and automatically generating the maintenance work order.

[0024] The predictive maintenance system of the edge gateway comprises:

[0025] A data acquisition module is configured to acquire multi-dimensional state data of the edge gateway, wherein the multi-dimensional state data comprises working condition and environment data and key component degradation data.

[0026] A predictive maintenance module comprises:

[0027] A health degree calculation unit is configured to calculate a comprehensive degradation rate of the capacitor based on the working condition and environment data and a preset physical failure model, and determine a capacitor health degree; and determine a flash memory health degree based on the key component degradation data and a preset data-driven model.

[0028] A fusion evaluation unit is configured to combine the capacitor health degree and the flash memory health degree, and calculate a basic health degree according to a preset component importance weight; and determine an environment penalty factor based on the working condition and environment data and a pre-trained anomaly detection model.

[0029] An integrated processing unit is configured to fuse and process the basic health degree and the environment penalty factor through a preset multiplication penalty model to determine a gateway comprehensive health score.

[0030] A life prediction unit is configured to estimate an instantaneous degradation rate based on a time series of the gateway comprehensive health score through a preset state space model to determine a remaining useful life.

[0031] A closed-loop control unit is configured to calculate a corrected data acquisition frequency through load adaptive closed-loop control logic based on real-time on-board temperature in the working condition and environment data, a preset safety temperature threshold, a preset load-temperature control coefficient, a preset minimum performance factor and an original data acquisition frequency, and output an adaptive control instruction.

[0032] Compared with the prior art, the present application has the following beneficial effects:

[0033] 1. The present application innovatively combines a physical failure model and a data-driven model to quantitatively evaluate the degradation state of the capacitor and the flash memory, respectively, and obtains a comprehensive health degree that takes into account chronic aging and acute impact through two-layer fusion processing, so that the evaluation result is more comprehensive and accurate.

[0034] 2. The present application uses an anomaly detection model to quantitatively evaluate the acute impact of external environments such as vibration, humidity and voltage fluctuation to form an environment penalty factor, which is fused with the basic health degree through a multiplication penalty model, so that the real state of the gateway under harsh working conditions can be accurately reflected.

[0035] 3. The application estimates the instantaneous degradation rate using a state space model instead of directly extrapolating the health score, which can extract a more realistic degradation trend from noisy observations, making the remaining useful life prediction result smoother, more robust and more sensitive;

[0036] 4. The application sets double thresholds for health and remaining life, automatically triggers predictive maintenance alarms and work orders, and changes the traditional passive response to active predictive maintenance, effectively avoiding line downtime and data interruption caused by unexpected failure. BRIEF DESCRIPTION OF DRAWINGS

[0037] The application will be further explained in conjunction with the accompanying drawings and examples:

[0038] Figure 1 is a flowchart of the method of the application;

[0039] Figure 2 is a structural diagram of the system of the application. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in conjunction with specific examples.

[0041] Example 1:

[0042] Please refer to Figure 1 , the data acquisition method of the edge gateway, which comprises:

[0043] S1, collecting multi-dimensional state data of the edge gateway, including working condition and environmental data, and key component degradation data;

[0044] S2, based on the working condition and environmental data and the preset physical failure model, calculating the comprehensive degradation rate of the capacitor and determining the capacitor health; and based on the key component degradation data and the preset data-driven model, determining the flash memory health;

[0045] S3, combining the capacitor health and the flash memory health, and calculating the basic health according to the preset component importance weight; and based on the working condition and environmental data and the pre-trained anomaly detection model, determining the environmental penalty factor;

[0046] S4, fusing the basic health and the environmental penalty factor through the preset multiplication penalty model to determine the gateway comprehensive health score; and based on the time series of the gateway comprehensive health score, estimating the instantaneous degradation rate through the preset state space model to determine the remaining useful life;

[0047] S5, based on the real-time on-board temperature in the working condition and environment data, the preset safety temperature threshold, the preset load-temperature control coefficient, the preset minimum performance factor and the original data acquisition frequency, a corrected data acquisition frequency is calculated through load adaptive closed-loop control logic, and an adaptive control instruction is output.

[0048] The embodiment of the application provides a data acquisition method of an edge gateway, which is the core of the technical scheme of the application, and aims to realize accurate health degree evaluation, residual life prediction and adaptive operation regulation and control of the edge gateway through multi-dimensional data fusion, hybrid driving modeling and closed-loop control.

[0049] In a specific embodiment, the method comprises:

[0050] Collecting multi-dimensional state data of the edge gateway; the purpose of this step is to provide comprehensive, multi-dimensional real-time data input for subsequent health degree evaluation and degradation modeling;

[0051] The multi-dimensional state data refers to a data set which is divided into two key subsets, working condition and environment data and key component degradation data, in the specific technical environment of the application.

[0052] The working condition and environment data are used to represent the real-time running load of the gateway and the external environmental conditions; in this embodiment, it can specifically include but is not limited to the CPU occupancy rate, network throughput, on-board temperature, environmental humidity, three-axis vibration acceleration and power input voltage fluctuation collected through integrated sensors;

[0053] The key component degradation data are used to represent the physical wear and state deterioration of the internal core components of the gateway such as flash memory and power module, and are direct indicators reflecting the physical degradation; in this embodiment, it can specifically include but is not limited to the cumulative erase count of NAND flash, ECC error correction count, bad block growth number and output voltage ripple of the power module;

[0054] Based on the working condition and environment data and the preset physical failure model, the comprehensive degradation rate of the capacitor is calculated, and the capacitor health degree is determined; and based on the key component degradation data and the preset data-driven model, the flash memory health degree is determined; the purpose of this step is to quantitatively evaluate the health status of the two most critical vulnerable components, i.e. the capacitor and the flash memory, through the hybrid driving modeling method.

[0055] For the capacitor, its degradation is influenced by temperature, load and electric stress; the preset physical failure model PoF model is used for calculation in this embodiment, and the model contains two levels.

[0056] In order to quantitatively evaluate the basic acceleration effect of temperature on capacitor aging, the basic thermal aging rate is calculated based on the Arrhenius model :

[0057]

[0058] wherein: is the base thermal aging rate, unit is calculated from the formula; is the real-time on-board temperature, unit is provided by the working condition and environmental data collected in the previous steps; is the pre-set material and geometric proportion factor, unit is ; is the material constant related to the activation energy, unit is ; and are calibrated based on the specification of the capacitor model or through accelerated life test ALT data; to clarify the calibration process, those skilled in the art can obtain a set of calibration data points, each of which contains a specific constant ambient temperature and its corresponding average failure time , by performing, for example, least squares regression analysis on , the parameters and can be determined;

[0059] To simulate the nonlinear accelerated aging effect when high temperature, high load and high voltage ripple are coupled, working condition load and electrical stress are introduced as acceleration factors to calculate the comprehensive degradation rate of the capacitor :

[0060]

[0061] wherein: is the comprehensive degradation rate of the capacitor, unit is calculated from the formula; is the base thermal aging rate, calculated from the previous step; is the CPU occupancy rate, ranging from 0 to 1, provided by the working condition and environmental data collected in the previous steps; is the real-time voltage ripple, provided by the key component degradation data collected in the previous steps; is the pre-set rated reference ripple; and are pre-set load and voltage stress sensitivity coefficients, , and are pre-set parameters based on hardware design such as heat dissipation capacity, power supply margin and experimental data calibration, used to adjust the influence weight of different working condition stress on aging;

[0062] To illustrate the calibration process, those skilled in the art can obtain a set of calibration data through controlled variable experiments, for example, at a constant temperature. and constant ripple Below, measure different CPU utilization rates. corresponding Change, and then constancy and constant Below, measure different corresponding The changes were finally determined by fitting the data using multiple regression analysis. and ;

[0063] Through the The total damage is obtained by integrating over time. And finally determine the health of the capacitor. :

[0064]

[0065] in: It is a preset capacitor failure damage threshold, which can be determined based on historical failure data statistics or accelerated aging experiments.

[0066] For flash memory, its health is determined by both inherent wear and real-time error correction status; this embodiment uses a preset data-driven model to determine the flash memory health. The technical motivation behind this model is to transform multiple discrete SMART attributes, i.e., key component degradation data, into a single, quantified flash health score.

[0067]

[0068] in: for The flash memory health status at any given time, with a value ranging from 0 to 1, is calculated using this formula; The cumulative number of erase / write cycles is provided by degradation data of key components collected in previous steps; The preset maximum number of erase / write cycles threshold; The number of real-time ECC error corrections is provided by the degradation data of key components collected in the preceding steps; The preset failure threshold for the number of ECC cycles; The real-time bad block growth number is provided by the degradation data of key components collected in the preceding steps; The preset failure threshold for the number of bad blocks; and These are the weighting coefficients; , and are all preset threshold values according to the specification book of NAND Flash particles such as TLC / QLC type; weight with satisfies , according to the historical failure data analysis FMEA calibration, for balancing the different effects of correctable errors and permanent bad blocks on health;

[0069] Combine the capacitor health and the flash health, and calculate the basic health according to the preset component importance weight; and determine the environmental penalty factor based on the working condition and environmental data and the pre-trained anomaly detection model; the purpose of this step is to fuse the component-level health obtained in the previous steps and the instantaneous impact risk from the environment to evaluate the overall state of the gateway; for the fusion of the chronic aging of internal components, calculate the basic health :

[0070]

[0071] wherein: is the basic health, the value range is 0-1, calculated by the formula; and are the capacitor health and the flash health calculated in the previous steps, respectively; and are the preset component importance weights, satisfying , calibrated according to the influence of component failure on the fundamental value of the system such as production continuity;

[0072] is to quantify the acute impact of the external environment, based on the working condition and environmental data collected in the previous steps, specifically , , , the vibration real-time anomaly score , the humidity real-time anomaly score and the power supply real-time anomaly score are calculated by the pre-trained anomaly detection model such as LSTM model; the pre-trained anomaly detection model uses the data of the gateway under normal service conditions as the baseline training set for training, so that it can identify abnormal patterns deviating from the normal; the range of these scores is from 0 indicating normal to 1 indicating serious anomaly; determine the environmental penalty factor :

[0073]

[0074] wherein: is the environmental penalty factor, the value range is 0-1, calculated by the formula; , , are the above real-time anomaly scores; , , is the environmental stress penalty weight, calibrated according to the sensitivity of the gateway deployment scenario to specific environmental stress; The processing ensures that the penalty factor always remains in the 0-1 interval;

[0075] The basic health degree and the environmental penalty factor are fused by a preset multiplication penalty model to determine the gateway comprehensive health degree score; and based on the time series of the gateway comprehensive health degree score, an instantaneous degradation rate is estimated by a preset state space model to determine the remaining useful life; this step is the core output of the embodiment, providing the final health state evaluation and future life prediction;

[0076] To obtain a final score that can reflect chronic aging and acute impact, a preset multiplication penalty model is used to calculate the gateway comprehensive health degree score ; The technical motivation of using the multiplication model is to ensure always remains in the 0-1 interval, and to repair the defect that the subtraction model may cause negative values;

[0077]

[0078] Wherein: is the gateway comprehensive health degree score, with a value range of 0-1, calculated by the formula; is the basic health degree calculated in the previous step; is the environmental penalty factor calculated in the previous step; when the environment is harsh increases, will be punished to decrease on the basis of ;

[0079] To predict the future life, the embodiment takes the historical time series of as the observation value, applies a preset state space model such as the particle filter or Kalman filter framework, to estimate a key hidden state, i.e. the instantaneous degradation rate ;

[0080] is the instantaneous degradation rate, with a unit of , estimated by the state space model, which reflects the current deterioration speed; the estimation process of the state space model depends on the preset process noise and observation noise covariance matrices, which are tuned according to the historical data set; based on the rate, the remaining useful life is calculated by linear extrapolation :

[0081]

[0082] Wherein: for The remaining useful life predicted at any time, in units of It is calculated using this formula; This represents the current overall health score of the gateway. The estimated instantaneous degradation rate; To ensure robustness of the calculation, a preset failure threshold is used as the denominator. Form, in which It is a preset, extremely small positive number, for example This measure can effectively prevent issues caused by unstable equipment status. It avoids division-by-zero errors caused by data approaching zero and can prevent errors caused by data noise. When a negative value occurs, a negative lifetime with no physical meaning is calculated, such as 0.15. This threshold is a failure judgment standard preset based on historical data and operational requirements. This formula maintains consistency in dimensions; the difference in dimensionless health scores is divided by... The rate, to obtain The remaining lifetime of a dimension;

[0083] Based on the real-time onboard temperature, preset safe temperature threshold, preset load-temperature control coefficient, preset minimum performance factor, and original data acquisition frequency in the operating condition and environmental data, the corrected data acquisition frequency is calculated through load adaptive closed-loop control logic, and adaptive control commands are output. The purpose of this step is to build a closed-loop control system that actively adjusts the gateway operating conditions to ensure its long-term survival and the fundamental value of the system when harsh environments such as high temperatures are detected.

[0084] In this embodiment, the load adaptive closed-loop control logic achieves graceful performance degradation through the following formula:

[0085]

[0086] in: The corrected data acquisition frequency, calculated using this formula, is used as the output of the adaptive control command. This is the original data acquisition frequency, which is a preset value; Real-time onboard temperature is provided by operating condition and environmental data collected in previous steps; For example, a preset safe temperature threshold. The maximum rated operating temperature of the gateway hardware, especially the capacitors, is preset. This is the preset load-temperature control coefficient, in units of... or ,make sure It is a dimensionless value; This is the preset minimum performance factor, with a value ranging from 0 to 1, for example, 0.3; and The calibration is based on the tolerance of the industrial scenario to the real-time performance of the data, ensuring that even after the frequency is reduced, the collected data can still meet the basic needs of the upper-level system.

[0087] when hour, Will be less than 1, It will be reduced but will not be lower than This reduces CPU load. This reduces Slow down the rate of capacitor degradation ;

[0088] The method described in this embodiment collects comprehensive multidimensional state data and innovatively combines the Physical Failure Model (PoF) and a data-driven model to accurately model the degradation of capacitors and flash memory, respectively. This method employs a two-layer fusion approach—inter-component fusion and environmental penalty fusion—and utilizes a multiplicative penalty model to obtain a comprehensive gateway health score that simultaneously reflects both chronic aging and acute impact. To further improve prediction performance, this scheme utilizes a state-space model from... Extracting instantaneous degradation rate from sequence This results in a more robust remaining service life. Prediction; This solution also constructs a load-adaptive closed-loop control logic, enabling the gateway to proactively cope with harsh environments and achieve graceful performance degradation; In summary, this invention provides a complete technical solution from data acquisition, hybrid modeling, fusion evaluation, lifetime prediction to adaptive closed-loop control, significantly improving the accuracy, robustness, and intelligence level of edge gateway health management, transforming traditional passive maintenance into proactive predictive maintenance, and maximizing the continuity of industrial processes and asset security.

[0089] Example 2:

[0090] Operating and environmental data include: CPU utilization; network throughput; onboard temperature; ambient humidity; triaxial vibration acceleration; and power input voltage fluctuation.

[0091] This embodiment, based on the method described in Embodiment 1, specifically defines the working conditions and environmental data in the preceding steps;

[0092] Operating and environmental data include: CPU utilization; network throughput; onboard temperature; ambient humidity; triaxial vibration acceleration; and power input voltage fluctuation.

[0093] CPU utilization and onboard temperature It is to calculate the overall degradation rate of the capacitor. Three-axis vibration acceleration , ambient humidity , and power supply input voltage fluctuation are the core inputs of the environmental penalty factor , because they are the direct basis for the pre-trained anomaly detection model to evaluate the acute impact of the external environment; network throughput is another important indicator reflecting the working condition of the gateway, which can be used for more detailed load modeling or anomaly detection;

[0094] This embodiment ensures the completeness of the key information required for subsequent hybrid drive modeling and fusion evaluation by explicitly defining the specific composition of the working condition and environmental data; this specific data combination has a clear technical direction: and support the physical failure model, while , and support the anomaly detection model; this enables the present application to capture both chronic degradation caused by internal working conditions and acute impact caused by external environments, thereby significantly improving the accuracy and comprehensiveness of the gateway comprehensive health score .

[0095] Embodiment 3:

[0096] The key component degradation data includes: the cumulative erase count of the NANDFlash; the ECC correction count; the number of bad block growth; and the output voltage ripple of the power module.

[0097] This embodiment specifically defines the key component degradation data in the pre-step based on the method described in Embodiment 1;

[0098] The key component degradation data includes: the cumulative erase count of the NANDFlash; the ECC correction count; the number of bad block growth; and the output voltage ripple of the power module.

[0099] The cumulative erase count of the NANDFlash is a key parameter in the flash data drive model for measuring long-term wear-out, i.e., the first failure mode; the ECC correction count and the number of bad block growth are key parameters for measuring instantaneous state deterioration, i.e., the second failure mode; the output voltage ripple of the power module is a key parameter in the capacitive physical failure model for quantifying the electrical stress acceleration effect;

[0100] By explicitly defining the specific composition of the key component degradation data, the embodiment ensures that the two core branches of the hybrid drive model, namely the capacitance PoF model and the flash data-driven model, obtain the most direct and most critical physical state observations; 、 、 The combination of the above makes the flash health assessment be able to take into account both long-term wear and instantaneous state, and The introduction of the above makes the capacitance degradation rate be able to quantify the influence of electrical stress; therefore, the embodiment greatly enhances the physical authenticity and accuracy of the component-level health solution.

[0101] Embodiment 4:

[0102] The capacitance comprehensive degradation rate is calculated by the physical failure model, including:

[0103] Based on the Arrhenius model and the on-board temperature, the basic thermal aging rate is calculated;

[0104] The capacitance comprehensive degradation rate is calculated by combining the basic thermal aging rate, CPU occupancy and output voltage ripple, and introducing preset load and voltage stress sensitivity coefficients.

[0105] The embodiment further defines the process of calculating the capacitance comprehensive degradation rate by the physical failure model in the preceding step on the basis of the method described in Embodiment 1;

[0106] The process specifically includes:

[0107] Based on the Arrhenius model and the on-board temperature, the basic thermal aging rate is calculated;

[0108] The embodiment uses the Arrhenius equation to quantify the basic effect of temperature on aging: Wherein is the on-board temperature; this As a basic rate, it reflects the aging speed under thermal stress only;

[0109] The capacitance comprehensive degradation rate is calculated by combining the basic thermal aging rate, CPU occupancy and output voltage ripple, and introducing preset load and voltage stress sensitivity coefficients.

[0110] As described in the implementation manner of Embodiment 1, the embodiment further couples the influences of the working condition load and the electrical stress on the basis of , and calculates the capacitance comprehensive degradation rate by introducing preset load and voltage stress sensitivity coefficients 、 :

[0111]

[0112] This design simulates the nonlinear accelerated aging effect of the coupling of high temperature, high CPU load, and high voltage ripple on the capacitor life under real working conditions. Those skilled in the art can understand that the linear relationship is an effective approximation and preferred simplified implementation of the accelerated effect. In a scenario with higher accuracy requirements, the linear term can be replaced by other nonlinear functions, such as a power-law function or an exponential function to more accurately depict the dose-effect relationship between stress and aging rate.

[0113] The embodiment provides a more accurate capacitor degradation modeling method; compared with the traditional Arrhenius model which only depends on temperature, the scheme has a significant gain effect: it establishes a physically reasonable degradation baseline, and innovatively introduces CPU occupancy and output voltage ripple, two key non-thermal stress sources, through multiplication factors and and The coupling model calculated by the coupling model more accurately reflects the actual aging speed of the gateway under real, harsh, and multi-stress coupling conditions, thereby greatly improving the accuracy of capacitor health evaluation.

[0114] Embodiment 5:

[0115] Determine the environmental penalty factor, including:

[0116] The three-axis vibration acceleration, the environmental humidity, and the power input voltage fluctuation are taken as inputs of the anomaly detection model, and vibration real-time anomaly scores, humidity real-time anomaly scores, and power real-time anomaly scores are calculated respectively;

[0117] The vibration real-time anomaly scores, the humidity real-time anomaly scores, the power real-time anomaly scores, and the preset environmental stress penalty weight are combined, and the environmental penalty factor is determined through minimum clipping processing.

[0118] The embodiment specifically limits the process of determining the environmental penalty factor in the preceding step on the basis of the method described in Embodiment 1;

[0119] The process specifically includes:

[0120] The three-axis vibration acceleration, the environmental humidity, and the power input voltage fluctuation are taken as inputs of the anomaly detection model, and vibration real-time anomaly scores, humidity real-time anomaly scores, and power real-time anomaly scores are calculated respectively;

[0121] ​The pre-step collected in the embodiment i.e. three-axis vibration acceleration, i.e. ambient humidity, i.e. power input voltage fluctuation, as the input of the pre-trained anomaly detection model such as LSTM, which is trained by normal working condition data and can output normalized vibration real-time anomaly score , humidity real-time anomaly score and power real-time anomaly score ;

[0122] Combine the vibration real-time anomaly score, humidity real-time anomaly score, power real-time anomaly score, and the preset environmental pressure penalty weight, and determine the environmental penalty factor through minimum clipping processing;

[0123] The embodiment calculates the environmental penalty factor by weighted fusion of the above three anomaly scores , and especially adopts minimum clipping processing, i.e. :

[0124]

[0125] Among them , , is the preset environmental pressure penalty weight;

[0126] The embodiment provides a robust and interpretable environmental risk quantification method; its gain effect is that it converts the original, physical quantity dimensionally different sensor data , , into unified, 0-1 interval real-time anomaly scores , , , solves the problem of multi-source heterogeneous data fusion; and determines the environmental penalty factor through minimum clipping processing, ensures that even in the case of multiple environmental factors being simultaneously extremely abnormal, causing the weighted sum to exceed 1, will be limited to 1; this avoids the occurrence of negative values when calculating in the subsequent steps, guarantees the mathematical stability and physical meaning of the gateway comprehensive health degree score i.e. always in the 0-1 interval, makes the penalty model more robust and reliable.

[0127] Embodiment 6:

[0128] Determine the remaining useful life, including:

[0129] The time series of the gateway comprehensive health score is taken as an observation value, and a state space model in a particle filter or Kalman filter framework is applied to estimate the instantaneous degradation rate;

[0130] Based on the instantaneous degradation rate, the current gateway comprehensive health score, and a preset failure threshold, the remaining useful life is calculated through linear extrapolation.

[0131] This embodiment specifically limits the process of determining the remaining useful life in the preceding step based on the method described in Embodiment 1;

[0132] The process specifically includes:

[0133] The time series of the gateway comprehensive health score is taken as an observation value, and a state space model in a particle filter or Kalman filter framework is applied to estimate the instantaneous degradation rate;

[0134] This embodiment does not directly use the time series of the gateway comprehensive health score for extrapolation, but takes it as an observation value of a state space model; a particle filter or Kalman filter framework familiar to those skilled in the art is used for this purpose; the technical motivation of this model is that is a noisy observation value, and the real instantaneous degradation rate of the system is a hidden state; through a filtering algorithm, the real instantaneous degradation rate can be more accurately estimated from the noisy ; ;

[0135] Based on the instantaneous degradation rate, the current gateway comprehensive health score, and a preset failure threshold, the remaining useful life is calculated through linear extrapolation.

[0136] After obtaining a more reliable estimation value, this embodiment uses a linear extrapolation method to calculate the remaining useful life based on the current gateway comprehensive health score , the instantaneous degradation rate , and a preset failure threshold :

[0137]

[0138] This embodiment provides a more robust and sensitive RUL prediction method; compared to directly fitting and extrapolating the time series, the gain effect of this scheme lies in the estimation-extrapolation two-step method: it uses a state space model such as particle filtering or Kalman filtering to extract a smoother and more real degradation trend instantaneous degradation rate from the noisy observation value ; then this more reliable is used to estimate the remaining useful life. ​​linear extrapolation is performed; this decoupling of state estimation and extrapolation calculation makes the RUL prediction more sensitive to changes in the degradation rate, while avoiding the instantaneous noise disturbance , thus obtaining more accurate and stable remaining useful life prediction results.

[0139] Embodiment 7:

[0140] The method further comprises:

[0141] In response to the gateway comprehensive health score being less than a preset health threshold, or the remaining useful life being less than a preset time threshold, triggering a predictive maintenance alarm and automatically generating a maintenance work order.

[0142] The embodiment adds a subsequent decision and action step to the method of Embodiment 1;

[0143] In response to the gateway comprehensive health score being less than a preset health threshold, or the remaining useful life being less than a preset time threshold, triggering a predictive maintenance alarm and automatically generating a maintenance work order.

[0144] The embodiment sets two parallel triggering conditions, i.e., or logic:

[0145] The gateway comprehensive health score is calculated by the previous step to be less than a preset health threshold , for example, 0.3; this condition is used to capture the case where the current state is already at a dangerous level;

[0146] The remaining useful life is calculated by the previous step to be less than a preset time threshold , for example, 20 days; this condition is used to capture the case where the current state is still acceptable, but the degradation rate is too fast, and the device will soon fail;

[0147] and The setting is based on the maintenance window, i.e., to ensure that the warning time must be greater than the average period required for spare parts procurement, logistics, and engineer dispatch, to ensure that replacement is completed before unplanned downtime occurs;

[0148] Once either condition is met, the system will automatically trigger a predictive maintenance alarm and automatically generate a maintenance work order, which can further indicate the primary risk point, e.g., the power module health Hc is below the threshold;

[0149] The embodiment abstracts the data and ​The transformation into specific, executable maintenance actions constitutes a closed loop from prediction to maintenance; by adopting a double threshold trigger logic of health or RUL, the scheme ensures the timeliness and completeness of the early warning, which can capture both the state difference caused by chronic degradation and the speed caused by acute impact; the function of automatically generating maintenance work orders automates the predictive maintenance process, ensuring that maintenance decisions can be executed before unplanned downtime occurs, significantly improving operational efficiency and system reliability.

[0150] Embodiment 8:

[0151] Referring to Figure 2 , the predictive maintenance system of the edge gateway comprises:

[0152] The data acquisition module is configured to acquire multi-dimensional state data of the edge gateway, the multi-dimensional state data comprising: working condition and environment data, and key component degradation data.

[0153] The predictive maintenance module comprises:

[0154] The health degree calculation unit is configured to calculate a comprehensive degradation rate of the capacitor based on the working condition and environment data and a preset physical failure model, and determine a capacitor health degree; and determine a flash memory health degree based on the key component degradation data and a preset data-driven model.

[0155] The fusion evaluation unit is configured to combine the capacitor health degree and the flash memory health degree, and calculate a basic health degree according to a preset component importance weight; and determine an environment penalty factor based on the working condition and environment data and a pre-trained anomaly detection model.

[0156] The comprehensive processing unit is configured to fuse and process the basic health degree and the environment penalty factor through a preset multiplication penalty model to determine a gateway comprehensive health score.

[0157] The life prediction unit is configured to estimate an instantaneous degradation rate based on a time series of the gateway comprehensive health score through a preset state space model to determine a remaining useful life.

[0158] The closed-loop control unit is configured to calculate a corrected data acquisition frequency through a load adaptive closed-loop control logic based on a real-time on-board temperature in the working condition and environment data, a preset safety temperature threshold, a preset load-temperature control coefficient, a preset minimum performance factor, and an original data acquisition frequency, and output an adaptive control instruction.

[0159] The embodiment provides a predictive maintenance system of an edge gateway, which is configured to execute the method of any one of the preceding embodiments 1-7; in a preferred embodiment, the system comprises:

[0160] The data collection module is configured to collect multi-dimensional state data of the edge gateway, including working condition and environment data, key component degradation data, and so on. 、 、 、 、 、 、 、 ;

[0161] The predictive maintenance module is a core processing unit of the system and is configured to perform steps S2 to S5. The predictive maintenance module can be further divided into multiple functional units.

[0162] The health degree calculation unit is configured to perform step S2. The health degree calculation unit is configured to calculate the comprehensive degradation rate of the capacitor and determine the capacitor health degree based on the working condition and environment data and a preset physical failure model, such as The health degree calculation unit is configured to determine the flash memory health degree based on the key component degradation data and a preset data-driven model, such as

[0163] The fusion evaluation unit is configured to perform step S3. The fusion evaluation unit is configured to combine the capacitor health degree and the flash memory health degree , and calculate the basic health degree based on a preset component importance weight 、 such as based on the working condition and environment data 、 、 and a pre-trained anomaly detection model, and determine the environment penalty factor such as

[0164] The comprehensive processing unit is configured to perform the first part of step S4. The comprehensive processing unit is configured to combine the basic health degree and the environment penalty factor by a preset multiplication penalty model, such as to determine the gateway comprehensive health score ;

[0165] The life prediction unit is configured to perform the second part of step S4. The life prediction unit is configured to estimate the instantaneous degradation rate based on the time series of the gateway comprehensive health score by a preset state space model, such as a particle filter ​​​​​determining the remaining useful life As formula

[0166] The closed-loop control unit aims to execute the S5 step; the unit is configured to calculate the modified data acquisition frequency , the preset safety temperature threshold , the preset load-temperature control coefficient , the preset minimum performance factor and the original data acquisition frequency , through the load adaptive closed-loop control logic such as formula, and output adaptive control instructions ;

[0167] The embodiment provides a complete, modular edge gateway predictive maintenance system architecture; the data acquisition module provides a data basis; the predictive maintenance module realizes complete functional mapping of the methods described in embodiments 1-7 at the system level through its internal five logic units, namely health degree calculation, fusion evaluation, comprehensive processing, life prediction and closed-loop control; this clear modularization and unitization division decouples the complex multi-dimensional data analysis, hybrid modeling, fusion evaluation, life prediction and adaptive control processes into clear functional entities; this not only makes the system logic clear, easy to implement and maintain, but also creates an end-to-end intelligent system from data input, i.e., the data acquisition module, to intelligent analysis and prediction, i.e., the first four units of the predictive maintenance module, to active closed-loop response, i.e., the closed-loop control unit, with extremely high scalability and robustness, which can effectively realize the predictive maintenance and autonomous health management of the edge gateway.

[0168] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A data collection method of an edge gateway, characterized by, Comprising: S1, collecting multi-dimensional state data of the edge gateway, the multi-dimensional state data comprising: working condition and environment data, and key component degradation data; S2, based on the working condition and environment data and a preset physical failure model, calculating a comprehensive degradation rate of the capacitor and determining a capacitor health degree; and based on the key component degradation data and a preset data-driven model, determining a flash memory health degree; S3, combining the capacitor health degree and the flash memory health degree, and calculating a basic health degree according to a preset component importance weight; and based on the working condition and environment data and a pre-trained anomaly detection model, determining an environment penalty factor; S4, fusing and processing the basic health degree and the environment penalty factor through a preset multiplication penalty model to determine a gateway comprehensive health degree score; and based on a time sequence of the gateway comprehensive health degree score, estimating an instantaneous degradation rate through a preset state space model to determine a remaining useful life; S5, based on real-time on-board temperature in the working condition and environment data, a preset safety temperature threshold, a preset load-temperature control coefficient, a preset minimum performance factor and an original data acquisition frequency, calculating a corrected data acquisition frequency through a load adaptive closed-loop control logic, and outputting an adaptive control instruction.

2. The data collection method of the edge gateway according to claim 1, wherein, The working condition and environment data include: CPU occupancy rate; network throughput; on-board temperature; environmental humidity; three-axis vibration acceleration; and power input voltage fluctuation.

3. The data collection method of the edge gateway according to claim 1, wherein, The key component degradation data include: cumulative erase count of NANDFlash; ECC error correction count; bad block growth number; and output voltage ripple of the power module.

4. The data collection method of the edge gateway according to claim 1, wherein, The comprehensive degradation rate of the capacitor is calculated through the physical failure model, comprising: calculating a basic thermal aging rate based on the Arrhenius model and the on-board temperature; combining the basic thermal aging rate, the CPU occupancy rate and the output voltage ripple, and introducing preset load and voltage stress sensitivity coefficients to calculate the comprehensive degradation rate of the capacitor.

5. The data collection method of the edge gateway according to claim 1, wherein, The environment penalty factor is determined, comprising: taking the three-axis vibration acceleration, the environmental humidity and the power input voltage fluctuation as inputs of the anomaly detection model to calculate vibration real-time abnormal score, humidity real-time abnormal score and power real-time abnormal score respectively; combining the vibration real-time abnormal score, the humidity real-time abnormal score, the power real-time abnormal score and a preset environmental stress penalty weight to determine the environment penalty factor through minimum amplitude limiting processing.

6. The data collection method of the edge gateway according to claim 1, wherein, The remaining useful life is determined, comprising: taking the time sequence of the gateway comprehensive health degree score as an observation value, and applying a state space model of a particle filter or a Kalman filter framework to estimate the instantaneous degradation rate; based on the instantaneous degradation rate, the current gateway comprehensive health degree score and a preset failure threshold, calculating the remaining useful life through linear extrapolation.

7. The data collection method of the edge gateway according to claim 1, wherein, Further comprising: in response to the gateway comprehensive health degree score being less than a preset health degree threshold or the remaining useful life being less than a preset time threshold, triggering a predictive maintenance alarm and automatically generating a maintenance work order.

8. A predictive maintenance system for an edge gateway based on the data collection method for an edge gateway according to any one of claims 1 to 7, characterized in that, Comprising: a data acquisition module for collecting multi-dimensional state data of the edge gateway, the multi-dimensional state data comprising: working condition and environment data, and key component degradation data; a predictive maintenance module, comprising: The health degree calculation unit is configured to calculate a comprehensive degradation rate of the capacitor based on the working condition and environment data and a preset physical failure model, and determine a capacitor health degree; and determine a flash memory health degree based on key component degradation data and a preset data-driven model; The fusion evaluation unit is configured to combine the capacitor health degree and the flash memory health degree, calculate a basic health degree according to a preset component importance weight, and determine an environment penalty factor based on the working condition and environment data and a pre-trained anomaly detection model; The comprehensive processing unit is configured to fuse and process the basic health degree and the environment penalty factor through a preset multiplication penalty model to determine a gateway comprehensive health degree score; The life prediction unit is configured to estimate an instantaneous degradation rate based on a time sequence of the gateway comprehensive health degree score through a preset state space model to determine a remaining useful life; The closed-loop control unit is configured to calculate a corrected data acquisition frequency through load adaptive closed-loop control logic based on a real-time on-board temperature in the working condition and environment data, a preset safety temperature threshold, a preset load-temperature control coefficient, a preset minimum performance factor and an original data acquisition frequency, and output an adaptive control instruction.

Images