A collaborative fatigue traceability method and system based on virtual working condition simulation and a medium
By using a collaborative fatigue tracing method based on virtual working condition simulation, the false alarm problem of structural health monitoring system in closed coal yards was solved, and accurate damage identification and fatigue life prediction were achieved under strong noise background, improving the system's identification accuracy and the scientific nature of fatigue assessment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUADIAN ZHENGZHOU MECHANICAL DESIGN INST
- Filing Date
- 2026-04-30
- Publication Date
- 2026-08-04
AI Technical Summary
Existing structural health monitoring systems struggle to distinguish between normal equipment impacts and abnormal structural damage in enclosed coal yards, leading to frequent false alarms. Furthermore, traditional methods struggle to handle nonlinear load components in strong noise environments, resulting in overly conservative fatigue assessments or the presence of safety blind spots.
A collaborative fatigue tracing method based on virtual working condition simulation is adopted. By acquiring multi-source data in real time, a multi-source spatiotemporal correlation sequence of equipment working conditions and structural response is established. The mechanical transmission relationship between load and response is defined by a physical-guided causal graph model. Dynamic loads that cannot be directly measured are calculated, and unobserved confounding factors are eliminated. The intrinsic damage of the structure is identified by combining the causal residual statistical distribution benchmark and high-frequency energy envelope. A nonlinear cumulative damage correction factor is introduced to calculate the remaining fatigue life.
It enables accurate identification of intrinsic structural damage without interrupting operations, reduces false alarm rates, improves the scientific rigor and transparency of fatigue life prediction, ensures the stability of identification accuracy under complex production cycles, and provides nanosecond-level preventative maintenance data support.
Smart Images
Figure CN122508980A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial structural health monitoring technology, specifically to a collaborative fatigue tracing method, system, and medium based on virtual working condition simulation. Background Technology
[0002] Within a closed coal yard, the rotation, travel, and cantilever pitch of the stacker-reclaimer generate significant dynamic excitations; existing structural health monitoring systems often confuse equipment-induced vibrations with structural abnormal damage, leading to frequent false alarms; existing filtering algorithms struggle to handle nonlinear load components that are deeply coupled with operating conditions.
[0003] Existing technology and its limitations: 1. Traditional methods rely on spectral filtering or modal analysis, which can easily mask minor structural degradations, i.e. weak signals, under the background of severe vibration of the stacker-reclaimer, i.e. strong noise.
[0004] 2. Traditional structural health monitoring systems often cause false alarms because they cannot distinguish between normal equipment impacts and abnormal structural damage.
[0005] 3. Existing fatigue assessments are often based on total stress amplitude without considering the causal contribution of load components, resulting in overly conservative assessment results or the existence of safety blind spots.
[0006] 4. The inertial braking during equipment start-up and shutdown is an extremely non-stationary process, and conventional algorithms fail during this stage. Summary of the Invention
[0007] This invention proposes a collaborative fatigue tracing method, system, and medium based on virtual working condition simulation to solve the problems mentioned in the background.
[0008] To achieve the above objectives, the present invention adopts the following technical solution: a collaborative fatigue tracing method based on virtual working condition simulation, comprising the following steps: S1. Real-time acquisition of multi-source data to establish a multi-source spatiotemporal correlation sequence of equipment operating conditions and structural response; S2. Input a multi-source spatiotemporal correlation sequence into a physically guided causal graph model to define the mechanical transfer relationship between load and response; S3. The mechanical transfer relationship between load and response uses the multiple hysteresis terms of motor current as tool variables to estimate dynamic loads that cannot be directly measured, eliminate unobserved confounding factors, and establish a baseline for the statistical distribution of causal residuals. S4. The causal residual statistical distribution benchmark supports Do-calculus counterfactual calculation, performs virtual working condition intervention, removes the dynamic vibration component of equipment operation, retains the static load effect, and calculates the causal residual between the measured value and the virtual predicted value. S5. The causal residual is determined by the dual conditions of deviation probability and high-frequency energy envelope to identify structural endogenous damage abnormal disturbances. Combined with spatial distribution characteristics, the damage nodes of the grid structure are located and the residual sequence is converted into standard stress amplitude. S6. The standard stress amplitude introduces a nonlinear cumulative damage correction factor to correct the Miner linear law, calculate the remaining fatigue life of key nodes of the grid structure, and trigger graded damage early warning and operation and maintenance.
[0009] Preferably, the physical-guided causal graph model in S2 is a linear superposition model, and its parameterization form is as follows:
[0010] in, For the total response variables of the key nodes of the grid structure, The dynamic excitation load variable generated by equipment operation. For the static mass load variable of the equipment, For temperature field distribution variables, , For load-response path coefficients, , For regression coefficients, It is a set of environmental confounding factors. It is independent Gaussian noise.
[0011] Preferably, the multiple lag terms of the motor current constitute a redundant information estimation system:
[0012] in, The dynamic load variable of the device at time t is . This is the intercept term for the first-stage regression. , , These are the regression coefficients of the motor current on the dynamic load at the current moment and the two moments before, respectively. , , These are the motor current values at the current moment and the two moments before. This is the motor current lag term. This represents the regression residuals for the first stage.
[0013] Preferably, the reliability of the linear superposition model is guaranteed by the following engineering constraints: Physical constraints: The structural stiffness matrix is known, the force transmission path is determined by the grid topology, and the cause-effect graph is sparsely connected; Timing constraints: The timestamp of the device state variable must strictly precede the timestamp of the structural response variable; Redundant information constraints: Multiple motor current hysteresis terms are jointly calculated to ensure the robustness and identifiability of the load-response coefficient.
[0014] Preferably, the validity of the calculated dynamic load that cannot be directly measured is verified through the following steps: First stage calculation: ; Second-stage calculation: ; Current independence verification: verifying residuals With each current signal correlation coefficient ; in, This is the intercept term for the second-stage regression. These are the predicted dynamic load values obtained from the first-stage regression. The causal effect coefficient of dynamic load. The path factor for static loads. This refers to the residuals from the second-stage regression. Redundant information consistency verification: Calculate the Sargan-Hansen J statistic to verify the consistency of the inference results; Confidence interval of the calculated result: When the current independence condition deviates slightly, the confidence interval of the load-response coefficient is estimated by the Bootstrap method.
[0015] Preferably, the causal residual is used to identify structural endogenous damage anomalous perturbations through a dual-condition determination of deviation probability and high-frequency energy envelope, including:
[0016] in, The statistical distribution of residuals in the baseline period. , These are the mean and standard deviation of the residuals in the base period. The standard normal cumulative distribution function is... For causal residuals, This indicates the probability that the residual deviation from the baseline distribution is no less than the current observation value. Two-condition fusion logic: residual deviation probability Used to quantify the degree of statistical anomaly in causal residuals; High-frequency energy envelope Used to detect high-frequency impact signals generated by localized structural damage; Fusion rules: When or An alarm is triggered at any time; Threshold setting method: Anomaly detection threshold Based on the significance level, safety-critical components are selected. General components are taken ; The high-frequency energy envelope threshold; The high-frequency energy envelope threshold is determined by the following adaptive method: Data collected during normal operating conditions A high-frequency energy envelope sample; Calculate the sample mean and standard deviation ; Set the threshold to ,in Determined based on the measured distribution of samples under normal operating conditions.
[0017] Preferably, the nonlinear cumulative damage correction factor employs a linear acceleration correction model related to the damage degree to modify the Miner linear cumulative damage rule:
[0018]
[0019] in, The damage acceleration coefficient, For the process Cumulative damage after one stress cycle This indicates that the structure has completely failed; For the first Damage increment caused by secondary stress cycles For the first The actual number of cycles at each stress level. For the first Fatigue life under stress level 1 For the first Nonlinear correction index for secondary stress cycles For the first Cumulative damage level before the next cycle This represents the total number of stress cycles.
[0020] Preferably, the remaining fatigue life assessment is based on the following conditions: Assumption: The rate of damage evolution remains constant during the prediction period; Extrapolation confidence interval: The prediction results should be accompanied by a confidence interval with a confidence level of not less than 90%; Application Notes: Recalibration should be performed when the damage enters the accelerated propagation phase. parameter.
[0021] A collaborative fatigue tracing system based on virtual working condition simulation includes: Sensor networks are used to acquire multi-source data in real time and establish a multi-source spatiotemporal correlation sequence between equipment operating conditions and structural responses. Edge computing units are used to input multi-source spatiotemporal correlation sequences into a physically guided causal graph model, defining the load-response mechanical transfer relationship. This relationship uses multiple hysteresis terms of motor current as instrumental variables to estimate dynamic loads that cannot be directly measured, eliminating unobserved confounding factors and establishing a causal residual statistical distribution benchmark. This benchmark supports Do-calculus counterfactual calculations, performs virtual operating condition intervention, removes dynamic vibration components from equipment operation while retaining static load effects, and calculates the causal residuals between measured values and virtual predicted values. The load-response estimation processor is used to identify structural endogenous damage abnormal disturbances by determining the causal residuals through a dual-condition judgment of deviation probability and high-frequency energy envelope. Combined with spatial distribution characteristics, it locates the damaged nodes of the space frame and transforms the residual sequence into a standard stress amplitude. The standard stress amplitude introduces a nonlinear cumulative damage correction factor to correct the Miner linear law, calculates the remaining fatigue life of key nodes of the space frame, and triggers graded damage early warning and operation and maintenance.
[0022] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.
[0023] As can be seen from the above technical solution, the present invention provides a collaborative fatigue tracing method based on virtual working condition simulation. Compared with the prior art, the present invention has the following advantages: 1. By implementing virtual working condition intervention, this invention can simulate shutdown at the computational level, accurately isolate the intrinsic damage components of the structure, realize online pure damage identification without interrupting operations, and improve the detection effect.
[0024] 2. This invention establishes an unbiased mapping between load and response through a physics-guided load-response model, which significantly reduces the false alarm rate and operation and maintenance costs of structural early warning.
[0025] 3. By stripping away the pure damage component and combining it with a nonlinear cumulative damage correction factor, this invention can more realistically reflect the fatigue evolution trend of structures under actual complex service environments. It provides nanosecond-level high-precision data support for preventative maintenance, improving the scientific rigor and transparency of fatigue life prediction.
[0026] 4. By introducing residual dynamic load attenuation correction, this invention ensures that the system can maintain stable recognition accuracy under complex production cycles such as frequent speed adjustments and reversals of equipment, eliminates false damage judgments caused by transient fluctuations, and enhances the system robustness of unsteady transition processes. Attached Figure Description
[0027] Figure 1 This is a flowchart illustrating a collaborative fatigue tracing method based on virtual working condition simulation according to the present invention.
[0028] Figure 2 This is a convergence curve of the state determination accuracy of the present invention.
[0029] Figure 3 This is a comparison chart of the average state determination delay under the method of the present invention and the traditional method.
[0030] Figure 4 This is a graph showing the changes in critical node availability and bandwidth utilization in this invention. Detailed Implementation
[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.
[0033] like Figure 1 As shown in the figure, a collaborative fatigue tracing method based on virtual working condition simulation in this embodiment includes the following steps: S1. Real-time acquisition of multi-source data to establish a multi-source spatiotemporal correlation sequence of equipment operating conditions and structural response; S2. Input a multi-source spatiotemporal correlation sequence into a physically guided causal graph model to define the mechanical transfer relationship between load and response; S3. The mechanical transfer relationship between load and response uses the multiple hysteresis terms of motor current as tool variables to estimate dynamic loads that cannot be directly measured, eliminate unobserved confounding factors, and establish a baseline for the statistical distribution of causal residuals. S4. The causal residual statistical distribution benchmark supports Do-calculus counterfactual calculation, performs virtual working condition intervention, removes the dynamic vibration component of equipment operation, retains the static load effect, and calculates the causal residual between the measured value and the virtual predicted value. S5. Causal residuals are determined by the dual conditions of deviation probability and high-frequency energy envelope to identify structural endogenous damage and abnormal disturbances. Combined with spatial distribution characteristics, the damage nodes of the grid structure are located and the residual sequence is transformed into standard stress amplitude. S6 introduces a nonlinear cumulative damage correction factor into the standard stress amplitude, corrects the Miner linear law, calculates the remaining fatigue life of key nodes of the grid structure, and triggers graded damage early warning and operation and maintenance handling.
[0034] Specifically, the physics-guided causal graph model in S2 is a linear superposition model, and its parameterization form is as follows:
[0035] in, The total response variable of the key nodes of the grid structure, with dimensions of (Stress) or (Acceleration) The dynamic excitation load variable generated by equipment operation. For the static mass load variable of the equipment, For temperature field distribution variables, , The load-response path coefficient has the following dimensions: , A positive sign indicates that an increase in dynamic load leads to an increase in structural response. A positive sign indicates that an increase in static load leads to an increase in structural response. , For regression coefficients, Dimensions are , Dimensional view And so on, It is a collection of environmental confounding factors (wind load, coal pile side pressure, etc.). It is independent Gaussian noise.
[0036] Specifically, the multiple lag terms of motor current constitute a redundant information estimation system:
[0037] in, Let be the dynamic load variable of the device at time , with dimensions . The dynamic vibration force generated by the rotation, travel, and pitching movements of the stacker-reclaimer The first-stage regression intercept term represents the reference dynamic load when there is no current input. , , These are the regression coefficients of the motor current on the dynamic load at the current moment and the two moments before, respectively, reflecting the immediate and delayed effects of current changes on the load. , , These are the motor current values at the current moment and the two previous moments, respectively, with dimensions of... , The first-stage regression residual represents the dynamic load component that cannot be explained by the current.
[0038] Specifically, the reliability of the linear superposition model is guaranteed by the following engineering constraints: Physical constraints: The structural stiffness matrix is known, the force transmission path is determined by the grid topology, and the cause-effect graph is sparsely connected; Timing constraints: The timestamp of the device state variable must strictly precede the timestamp of the structural response variable; Redundant information constraints: Multiple motor current hysteresis terms are jointly calculated to ensure the robustness and identifiability of the load-response coefficient.
[0039] Specifically, dynamic loads Methods for obtaining engineering projects: Due to the dynamic load of the stacker-reclaimer Since direct measurement is not possible (the device lacks a load sensor), this invention employs the following indirect acquisition method to construct the training dataset for the first-stage regression: 1. Kinematic parameter estimation method: using rotation angles acquired in real time by the PLC system. Pitch angle Walking position Using kinematic parameters and the equipment's kinematic model, the coordinates of the bucket wheel center are calculated. Then, the equivalent dynamic nodal loads at each grid node are calculated using the finite element influence surface function (established by MIDASGen). This is then used as... The proxy value; 2. Calibration Condition Measurement Method: During the equipment installation and commissioning phase, dynamic load data under typical operating conditions are collected using a temporarily installed force sensor (model: Kistler9077B, range 0-50kN, accuracy ±0.5%). This data is then regressed and fitted with synchronously collected motor current data to determine the first-stage regression coefficients. The initial value; 3. Online Update Strategy: After the equipment is put into operation, kinematic parameter estimation methods are used to continuously provide updates. The proxy value is used to recalibrate the regression coefficients quarterly using the most recent month's synchronized data to compensate for parameter drift caused by equipment wear and tear.
[0040] Specifically, the effectiveness of estimating dynamic loads that cannot be directly measured is verified through the following steps: First stage calculation: ; Second-stage calculation: ; Current independence verification: verifying residuals With each current signal correlation coefficient ; in, This is the intercept term for the second-stage regression. These are the predicted dynamic load values obtained from the first-stage regression. The causal effect coefficient of dynamic load. The path factor for static loads. This refers to the residuals from the second-stage regression. Redundant information consistency verification: Calculate the Sargan-Hansen J statistic to verify the consistency of the inference results; Confidence interval of the calculated result: When the current independence condition deviates slightly, the confidence interval of the load-response coefficient is estimated by the Bootstrap method.
[0041] Causal residuals are used to identify structural endogenous damage anomalous perturbations through a dual-condition determination of deviation probability and high-frequency energy envelope, including:
[0042] in, This refers to the statistical distribution of residuals during the baseline period, i.e., the statistical distribution of residuals collected during the period when the equipment is operating normally but without any signs of structural damage. , The mean and standard deviation of the residuals for the base period, with dimensions equal to... same, The standard normal cumulative distribution function is... For causal residuals, the dimension is or , This represents the probability that the residual deviation from the baseline distribution is no less than the current observation value. It is dimensionless and has a range of values. , Let be a random variable in the residual distribution of the base period, with dimensions and . same; Specifically, the two-condition fusion logic: residual deviation probability Used to quantify the degree of statistical anomaly in causal residuals; High-frequency energy envelope Used to detect high-frequency impact signals generated by localized structural damage; Fusion rules: When or An alarm is triggered at any time; Threshold setting method: Anomaly detection threshold Based on the significance level, safety-critical components are selected. General components are taken ; The high-frequency energy envelope threshold; Specifically, the high-frequency energy envelope threshold is determined using the following adaptive method: Data collected during normal operating conditions A high-frequency energy envelope sample; Calculate the sample mean and standard deviation :
[0043]
[0044] in, For the first One high-frequency energy envelope sample value; Set the threshold to ,in Determined based on the measured distribution of samples under normal operating conditions: If the distribution is approximately normal, then ; If a heavy-tailed feature exists, an adaptive formula is used: Among them, kurtosis .
[0045] Specifically, the selection criteria for the high-frequency energy envelope frequency band are as follows: Frequency band selection instructions: The experimental observation characteristic frequency range of bolt loosening impact is: The energy is mainly concentrated in ; This invention selects As the analysis frequency band, it also avoids interference from 50Hz industrial noise and its harmonics; When the sampling rate is 2kHz, according to the Nyquist criterion, the upper limit of the effective analysis frequency band is 1kHz, and the upper limit of 500Hz is reasonable.
[0046] Exception detection logic:
[0047] in, This represents the tail probability of the causal residual; For significance level, safety-critical components are taken as... General components are taken ; The high-frequency energy envelope value at the current moment, with dimensions of ; The high-frequency energy envelope threshold, with dimensions of... same; The logical OR operator indicates that an exception is triggered if either of the two conditions is met. This is a logical implication operator, indicating that an exception flag operation is executed when the condition is true.
[0048] The nonlinear cumulative damage correction factor employs a linear acceleration correction model related to the damage severity to modify Miner's linear cumulative damage law.
[0049]
[0050] in, The damage acceleration factor was obtained by fitting fatigue tests on specimens with similar structures. The range of empirical values is , For the process The cumulative damage after one stress cycle, dimensionless, with a range of values. , This indicates that the structure has completely failed; For the first The damage increment generated by the stress cycle is dimensionless and is calculated recursively. For the first The actual number of cycles at each stress level is dimensionless and obtained statistically by the rainflow counting method. For the first Fatigue life under stress level 1, dimensionless, determined by SN curve, according to standard GB50017 or measured SN curve; For the first Nonlinear correction exponent during secondary stress cycles, dimensionless, initial value. ; For the first The cumulative damage before the next cycle is dimensionless. This is the initial state; The damage acceleration factor is dimensionless and obtained by fitting fatigue tests on specimens with similar structures. Its empirical value range is [insert range here]. ; The total number of stress cycles is dimensionless.
[0051] Specifically, the remaining fatigue life assessment is based on the following conditions: Assumption: The rate of damage evolution remains constant during the prediction period; Extrapolation confidence interval: The prediction results should be accompanied by a confidence interval with a confidence level of not less than 90%; Application Notes: Recalibration should be performed when the damage enters the accelerated propagation phase. parameter.
[0052] Specifically, lifetime prediction and uncertainty:
[0053] in, This is the predicted value of remaining fatigue life, measured in years or hours. The current cumulative damage level is dimensionless and has a range of values. ; The damage evolution rate is expressed in units of . It is determined by least squares estimation based on the most recent observation data; To predict uncertainty, dimensions and The same method is used to calculate the confidence interval. For degrees of freedom of Distribution critical value, The significance level is dimensionless. The standard error of the damage evolution rate estimate, with dimensions of same; For estimation The number of samples, dimensionless. ; like Figure 4 As shown, a collaborative fatigue tracing system based on virtual working condition simulation includes: Sensor networks are used to acquire multi-source data in real time and establish a multi-source spatiotemporal correlation sequence between equipment operating conditions and structural responses. Edge computing units are used to input multi-source spatiotemporal correlation sequences into a physical-guided causal graph model, defining the load-response mechanical transfer relationship. This relationship uses multiple hysteresis terms of motor current as instrumental variables to estimate dynamic loads that cannot be directly measured, eliminating unobserved confounding factors and establishing a causal residual statistical distribution benchmark. This benchmark supports Do-calculus counterfactual calculations, performs virtual operating condition interventions, removes dynamic vibration components from equipment operation while retaining static load effects, and calculates the causal residuals between measured and virtual predicted values. The load-response estimation processor is used to identify structural endogenous damage anomalies by determining causal residuals through a dual-condition judgment of deviation probability and high-frequency energy envelope. Combined with spatial distribution characteristics, it locates the damaged nodes of the space frame and transforms the residual sequence into a standard stress amplitude. The standard stress amplitude introduces a nonlinear cumulative damage correction factor to correct the Miner linear law, calculates the remaining fatigue life of key nodes of the space frame, and triggers graded damage early warning and operation and maintenance.
[0054] In practical applications, three methods were used to conduct parallel monitoring for a period of 6 months in the same 310m span coal yard. During the experiment, all methods operated under the same working conditions and shared the raw data collected by the same sensor network.
[0055] Experimental conditions: Experimental location: A 310m-span enclosed coal yard; Experimental period: 6 months; During the experiment, the stacker-reclaimer accumulated approximately 4,320 hours of operation. Sensor network configuration: 4 accelerometers (2kHz sampling), 120 fiber optic strain sensors (100Hz sampling), and several displacement sensors; All methods were tested in parallel using the same raw data to ensure the fairness of the control experiment; Experimental data acquisition frequencies: accelerometer 2kHz, strain gauge 100Hz, PLC 50Hz; Manual verification cycle: On-site inspections are conducted every two weeks to confirm whether the alarm is a genuine damage; Experimental data description: Total number of observed events: (Sampling every 30 minutes, totaling approximately 8640 hours over 6 months); Alarm triggering conditions: When or An alarm is triggered at any time; False alarm determination: Alarms confirmed by manual verification to be caused by non-structural damage are considered false alarms; Method for identifying real damage events: During the experiment, real damage events were confirmed twice through regular manual inspections (once every two weeks) combined with ultrasonic non-destructive testing. Both events were successfully detected by this method, with zero missed detections. Confidence interval calculation method: The 90% confidence interval for lifetime prediction is calculated using the Bootstrap quantile method (number of resampling cycles). ) calculation, i.e., taking and The quantiles are used as the upper and lower confidence bounds.
[0056] like Figure 2 As shown, key experimental data: 1. Comparison of false alarm rates: During the 6-month operation period, this method triggered 23 valid alarms, of which 2 were confirmed as false alarms after manual verification. The overall false alarm rate was 2 / 8640 = 0.023%, which is significantly lower than the 0.31% of the traditional spectrum filtering method and the 1.24% of the simple threshold method. This method reduces the false alarm rate by approximately 92.6% compared to traditional spectrum filtering methods, approximately 94.1% compared to machine learning methods, and approximately 96.6% compared to statistical regression methods.
[0057] 2. Weak signal detection capability: At a signal-to-noise ratio (SNR) of Under simulated operating conditions (simulated signals constructed based on measured system parameters), this method successfully detected... Structural strain damage components of the order of magnitude; Traditional methods in SNR Effective detection only begins at that time.
[0058] 3. Ablation experiment: Experimental design: Run the full method and ablation variants of each module independently 10 times on the same dataset to calculate statistical significance; Sample size description: Each independent experiment used monitoring data for 7 consecutive days (approximately 336 sample points, sampled once every 30 minutes), with a total of 10 experiments covering different seasonal working conditions, 2-3 times each in spring, summer, autumn and winter, to ensure the generalizability of the results; Ablation Dimension 1 Instrumental Variables (IVs) Module: The mean false alarm rate of the complete method was 0.023%, while the mean false alarm rate of the IV removal method was 0.112%. Paired t-test , The differences are highly statistically significant. Conclusion: Removal of the instrumental variable module led to a significant increase in the false alarm rate, confirming the key role of module IV in eliminating confounding factor bias; Ablation Dimension 2: PTP Nanosecond-Level Synchronization Module The average false alarm rate of the complete method was 0.023%, while the average false alarm rate of the PTP removal method (using NTP millisecond-level synchronization instead of PTP nanosecond-level synchronization) was 0.087%. Paired t-test , The differences are statistically significant. Conclusion: Removal of PTP nanosecond-level synchronization leads to a decrease in current-load time alignment accuracy (synchronization error deteriorates from <100ns to 1-5ms), and the first-stage regression... The false alarm rate decreased from 0.73 to 0.58, while the false alarm rate increased by about 3.8 times, confirming the key role of high-precision spatiotemporal alignment in the reliability of 2SLS estimation. Ablation Dimension 3: Transitional State Correction Module The mean false alarm rate of the complete method is 0.023%, while the mean false alarm rate of the transition state correction method is 0.065%. Paired t-test , The differences are statistically significant. Conclusion: The removal of the transient state correction led to false alarms of braking residual vibration during equipment start-up and shutdown, increasing the false alarm rate by approximately 2.8 times, confirming the crucial role of residual dynamic load attenuation correction in eliminating false alarms during start-up and shutdown. Ablation Dimension 4: Nonlinear Damage Correction Module The mean prediction error of the complete method for remaining fatigue life is 18.5%, while the mean prediction error of the nonlinear correction method is 34.2%. Paired t-test , The differences are statistically significant. Conclusion: Nonlinear damage correction factor The removal of [something] led to a decrease in the accuracy of fatigue life prediction, confirming that the acceleration effect of nonlinear correction in the high damage stage cannot be ignored.
[0059] like Figure 3 As shown, the nonlinear damage correction factor is illustrated. The evolution curve. When D=0 =1, degenerating into Miner's linear law; as damage accumulates... The value gradually increases, reflecting the accelerated crack propagation effect. The figure indicates the applicable range D < 0.5. In the early to middle stages and when D ≥ 0.5, it is necessary to switch to the Paris law crack propagation model.
[0060] 4. Statistical significance test: This invention vs. traditional spectral filtering: Chi-square test , The difference is extremely significant; This invention vs. machine learning methods: Chi-square test , The difference is extremely significant; This invention vs. statistical regression methods: Chi-square test , The difference is extremely significant; Conclusion: This invention significantly outperforms all comparative methods in terms of false alarm rate, and the statistical test results support the significance of the technical effect.
[0061] 5. Comparison of computational efficiency: Table 1. Computational efficiency of the present invention and existing methods
[0062] Conclusion: The computational efficiency of this invention is slightly lower than that of traditional methods, but it still meets the requirements of real-time monitoring and has a significant advantage in terms of false alarm rate.
[0063] Experimental conclusion: The comparative experimental results show that the present invention reduces the false alarm rate by more than an order of magnitude while maintaining zero missed detections, which is significantly better than traditional filtering methods and simple thresholding methods, verifying the technical advantages of causal tracing over association identification.
[0064] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk), etc.
[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.
[0066] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0067] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A collaborative fatigue tracing method based on virtual working condition simulation, characterized in that, Includes the following steps: S1. Real-time acquisition of multi-source data to establish a multi-source spatiotemporal correlation sequence of equipment operating conditions and structural response; S2. Input a multi-source spatiotemporal correlation sequence into a physically guided causal graph model to define the mechanical transfer relationship between load and response; S3. The mechanical transfer relationship between load and response uses the multiple hysteresis terms of motor current as tool variables to estimate dynamic loads that cannot be directly measured, eliminate unobserved confounding factors, and establish a baseline for the statistical distribution of causal residuals. S4. The causal residual statistical distribution benchmark supports counterfactual calculation, performs virtual working condition intervention, and calculates the causal residual between the measured value and the virtual predicted value. S5. The causal residual is determined by the dual conditions of deviation probability and high-frequency energy envelope to identify structural endogenous damage abnormal disturbances. Combined with spatial distribution characteristics, the damage nodes of the grid structure are located and the residual sequence is converted into standard stress amplitude. S6. The standard stress amplitude introduces a nonlinear cumulative damage correction factor to calculate the remaining fatigue life of key nodes of the space frame and trigger graded damage early warning and operation and maintenance.
2. The collaborative fatigue tracing method based on virtual working condition simulation according to claim 1, characterized in that: The physical-guided causal graph model in S2 is a linear superposition model, and its parameterization form is as follows: in, For the total response variables of the key nodes of the grid structure, The dynamic excitation load variable generated by equipment operation. For the static mass load variable of the equipment, For temperature field distribution variables, , For load-response path coefficients, , For regression coefficients, It is a set of environmental confounding factors. It is independent Gaussian noise.
3. The collaborative fatigue tracing method based on virtual working condition simulation according to claim 1, characterized in that: The multiple hysteresis terms of the motor current constitute a redundant information estimation system: in, The dynamic load variable of the device at time t is . This is the intercept term for the first-stage regression. , , These are the regression coefficients of the motor current on the dynamic load at the current moment and the two moments before, respectively. , , These are the motor current values at the current moment and the two moments before. This is the motor current lag term. This represents the regression residuals for the first stage.
4. The collaborative fatigue tracing method based on virtual working condition simulation according to claim 3, characterized in that: The reliability of the linear superposition model is guaranteed by the following engineering constraints: Physical constraints: The structural stiffness matrix is known, the force transmission path is determined by the grid topology, and the cause-effect graph is sparsely connected; Timing constraints: The timestamp of the device state variable must strictly precede the timestamp of the structural response variable; Redundant information constraints: Multiple motor current hysteresis terms are jointly calculated to ensure the robustness and identifiability of the load-response coefficient.
5. The collaborative fatigue tracing method based on virtual working condition simulation according to claim 4, characterized in that: The validity of the calculated dynamic loads that cannot be directly measured is verified through the following steps: First stage calculation: ; Second-stage calculation: ; Current independence verification: verifying residuals With each current signal correlation coefficient ; in, This is the intercept term for the second-stage regression. These are the predicted dynamic load values obtained from the first-stage regression. The causal effect coefficient of dynamic load. The path factor for static loads. This refers to the residuals from the second-stage regression. Redundant information consistency verification: Calculate the Sargan-Hansen J statistic to verify the consistency of the inference results; Confidence interval of the calculated result: When the current independence condition deviates slightly, the confidence interval of the load-response coefficient is estimated by the Bootstrap method.
6. The collaborative fatigue tracing method based on virtual working condition simulation according to claim 5, characterized in that: The causal residuals are determined by a dual-condition assessment of deviation probability and high-frequency energy envelope to identify structural endogenous damage anomalous perturbations, including: in, The statistical distribution of residuals in the baseline period. , These are the mean and standard deviation of the residuals in the base period. The standard normal cumulative distribution function is... For causal residuals, This indicates the probability that the residual deviation from the baseline distribution is no less than the current observation value. Two-condition fusion logic: residual deviation probability Used to quantify the degree of statistical anomaly in causal residuals; High-frequency energy envelope Used to detect high-frequency impact signals generated by localized structural damage; Fusion rules: When or An alarm is triggered at any time; Threshold setting method: Anomaly detection threshold Based on the significance level, safety-critical components are selected. General components are taken ; The high-frequency energy envelope threshold; The high-frequency energy envelope threshold is determined by the following adaptive method: Data collected during normal operating conditions A high-frequency energy envelope sample; Calculate the sample mean and standard deviation ; Set the threshold to ,in Determined based on the measured distribution of samples under normal operating conditions.
7. The collaborative fatigue tracing method based on virtual working condition simulation according to claim 6, characterized in that: The nonlinear cumulative damage correction factor adopts a linear acceleration correction model related to the damage degree to modify the Miner linear cumulative damage law: in, The damage acceleration coefficient, For the process Cumulative damage after one stress cycle This indicates that the structure has completely failed; For the first Damage increment caused by secondary stress cycles For the first The actual number of cycles at each stress level. For the first Fatigue life under stress level 1 For the first Nonlinear correction index for secondary stress cycles For the first Cumulative damage level before the next cycle This represents the total number of stress cycles.
8. The collaborative fatigue tracing method based on virtual working condition simulation according to claim 7, characterized in that: The remaining fatigue life assessment is based on the following conditions: Assumption: The rate of damage evolution remains constant during the prediction period; Extrapolation confidence interval: The prediction results should be accompanied by a confidence interval with a confidence level of not less than 90%; Application Notes: Recalibration should be performed when the damage enters the accelerated propagation phase. parameter.
9. A collaborative fatigue tracing system based on virtual working condition simulation, employing the collaborative fatigue tracing method based on virtual working condition simulation as described in any one of claims 1-8, characterized in that, include: Sensor networks are used to acquire multi-source data in real time and establish a multi-source spatiotemporal correlation sequence between equipment operating conditions and structural responses. Edge computing units are used to input multi-source spatiotemporal correlation sequences into a physically guided causal graph model, defining the mechanical transfer relationship between load and response. The load-response mechanical transfer relationship uses the motor current multi-hysteresis term as the instrument variable to estimate the dynamic load that cannot be directly measured, eliminate the bias of unobserved confounding factors, and establish a causal residual statistical distribution benchmark. The causal residual statistical distribution benchmark supports Do-calculus counterfactual calculation, performs virtual working condition intervention, removes the dynamic vibration component of equipment operation, retains the static load effect, and calculates the causal residual between the measured value and the virtual predicted value. The load-response estimation processor is used to identify structural endogenous damage abnormal disturbances by determining the causal residuals through a dual-condition judgment of deviation probability and high-frequency energy envelope. Combined with spatial distribution characteristics, it locates the damaged nodes of the space frame and transforms the residual sequence into a standard stress amplitude. The standard stress amplitude introduces a nonlinear cumulative damage correction factor to correct the Miner linear law, calculates the remaining fatigue life of key nodes of the space frame, and triggers graded damage early warning and operation and maintenance.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 8.