Fault chain-based equipment structure intelligent fault locating and early warning method

Abstract

The application provides an equipment structure intelligent fault positioning and early warning method based on a fault chain, comprising the following steps: obtaining an equipment structure to be monitored, dividing the equipment structure to be monitored into N non-overlapping sub-regions; arranging a plurality of strain monitoring points in each sub-region; assuming that there is a crack in the current sub-region along a main expansion path, combining the strain measurement value, and obtaining a strain sample data set under different crack lengths through simulation; training a pre-constructed neural network model based on the strain sample data set, and outputting a crack estimation value; decoding the real-time strain data based on the regional fault chain to obtain a crack length time sequence of each sub-region; positioning a fault region according to the maximum value of the crack length time sequence of each sub-region; designing a double-criterion mechanism to early warn the fault region, and generating a corresponding early warning marker signal. The application provides a more stable and efficient technical path for health management of the equipment structure.

Description

Technical Field

[0001] This invention belongs to the field of fault chains, and particularly relates to a method for intelligent fault location and early warning of equipment structures based on fault chains. Background Technology

[0002] Complex equipment structures (such as airship skins, UAV wings, satellite solar panels, and wind turbine blades) operate for extended periods in complex stress environments. Under the combined effects of various loads, temperature and humidity changes, ultraviolet radiation, and wind erosion, they are highly susceptible to material performance degradation and structural damage evolution. As service life extends, these structural components often exhibit a series of latent damages, including cracks, fatigue failure, delamination, and debonding, which gradually evolve into functional failure or even catastrophic damage. Traditional structural health assessment methods rely heavily on manual inspections, offline non-destructive testing, or periodic maintenance. These methods have limited detection frequency and significant response delays, often failing to capture the initial stages of structural damage in real time, and particularly struggling to identify dynamic characteristics such as crack initiation and early propagation. Against this backdrop, structural health monitoring (SHM) technology has gradually become a key means of ensuring equipment safety. Through the real-time acquisition and analysis of physical quantities such as strain, stress, temperature, and acceleration, it enables damage identification and condition assessment. While existing research has attempted to use artificial intelligence methods such as neural networks to process monitoring data, identify structural anomalies from an image perspective, or achieve status determination through CAE embedded simulation, the former suffers from weak physical interpretability, insufficient early warning capabilities, and difficulties in edge deployment, while the latter is limited by bottlenecks such as poor real-time performance, high computational requirements, and poor model adaptability. Although these methods possess certain diagnostic capabilities in specific scenarios, they still fall short of meeting the engineering requirements for rapid, reliable, low-resource-consumption, and predictable handling of equipment structural faults. Therefore, the current key technological challenge lies in establishing a simplified criterion model that can integrate the laws of structural physical evolution while possessing real-time diagnostic and early warning capabilities, thereby balancing physical reliability, computational efficiency, and engineering adaptability. Summary of the Invention

[0003] The purpose of this invention is to propose an intelligent fault location and early warning method for equipment structures based on fault chains, which solves the problems of diagnostic lag, weak early warning capability, insufficient physical correlation and high deployment cost in existing structural health monitoring technologies.

[0004] To achieve the above objectives, this invention provides an intelligent fault location and early warning method for equipment structures based on fault chains, comprising the following steps:

[0005] Obtain the equipment structure to be monitored, and divide the equipment structure to be monitored into N non-overlapping sub-regions;

[0006] Multiple strain monitoring points are set up in each sub-region to capture local load response characteristics as strain measurements for each strain monitoring point.

[0007] For any of the sub-regions, assuming that there is a crack along the main propagation path in the current sub-region, and combining the strain measurements, a strain sample dataset under different crack lengths is obtained through simulation.

[0008] A pre-built neural network model is trained based on the strain sample dataset to output crack estimates;

[0009] When real-time strain data is available, the real-time strain data is decoded based on the regional fault chain to obtain the crack length time series of each sub-region.

[0010] The fault region is located based on the maximum value of the crack length time series of each of the sub-regions, and a continuous determination mechanism is introduced to verify the fault region.

[0011] Based on the crack length time series and the fault region, a dual-criteria mechanism is designed to provide early warning for the fault region and generate a corresponding early warning signal.

[0012] Furthermore, the method also includes:

[0013] When collecting the original strain measurement values ​​of any of the equipment structures to be monitored, outlier identification and correction are performed on the original strain measurement values, and a dual threshold mechanism is adopted in combination with global statistical criteria and local trend residual test.

[0014] Furthermore, for any of the sub-regions, assuming a crack exists along the main propagation path in the current sub-region, and combining the strain measurements, a strain sample dataset for different crack lengths is obtained through simulation, including:

[0015] Suppose that there is a crack along the path in the i-th sub-region, and select a discrete state of crack length to represent the propagation length of the crack along the path;

[0016] For each state along the crack path, strain values ​​at M deployment points are collected through numerical simulation or physical loading experiments to form a strain sample vector set.

[0017] Perform sample augmentation on the strain sample vector set to generate a data sample augmentation function;

[0018] The data sample augmentation function and the data sample augmentation function are combined to generate a strain sample dataset.

[0019] Furthermore, the process of performing sample augmentation on the strain sample vector set to generate a data sample augmentation function is as follows:

[0020]

[0021] Wherein, φ(D) ix ) is the data sample augmentation function, ω ij Indicates monitoring point P ij G in the principal direction of the crack i The response weights are set based on the displacement-strain coupling degree in the simulation; M is the total number of monitoring points j. λ represents the strain gradient residual of the monitoring point in region i on the orthogonal axis of the crack direction; λ is the structural response anomaly compensation factor, reflecting the disturbance effect of crack path offset on strain distribution.

[0022] Furthermore, the pre-built neural network model is a structurally constrained neural network model, wherein the model parameters are obtained by training by minimizing the constrained loss function;

[0023] Wherein, the constrained loss function is the objective function, expressed as:

[0024]

[0025] Where, θ i Here are the model parameters: x is the current crack, X is the total number of cracks, and f is the total number of cracks. i For a structurally constrained neural network model, φ(D) ix ) is the data sample augmentation function, L ix Ψ is the crack propagation length, β is the weight of the regularization term, and Ψ is the crack propagation length. i (θ i ) represents the regularization term for multi-factor structural constraints;

[0026] Wherein, α1 and α2 correspond to the weight coefficients, This is a penalty term for local response deviation;

[0027] The strain noise sensitivity suppression term automatically suppresses the model sensitivity of high-noise points in the input data by coupling the strain variance within the sample group with the higher-order partial derivatives of the response; M is the total number of monitoring points j; ε ij Let ρ be the strain measurement value of the j-th monitoring point in the sample input; ij It is the expected gradient response intensity calculated from the crack propagation direction and simulation results, used to guide the network weights to align with the main driving direction of the crack; ε ijx This represents the strain measurement value of the j-th monitoring point in the current crack x;

[0028] The structure-constrained neural network model adopts a dynamic batch iteration method and introduces an early stopping mechanism to prevent overfitting. At the same time, it monitors the stability of the mean square rate of change of the residuals to generate crack estimates.

[0029] Furthermore, when real-time strain data exists, decoding the real-time strain data based on the regional fault chain to obtain the crack length time series of each sub-region specifically includes:

[0030] Obtain the corresponding crack estimate;

[0031] A trend response enhancement mechanism based on a smoothing factor is introduced into the crack estimate to construct a time series of the crack estimate; wherein the smoothing factor is used to adjust the response ratio between the current estimate and the historical trend.

[0032] Furthermore, the fault region is located based on the maximum value of the crack length time series of each of the sub-regions, and a continuous determination mechanism is introduced to verify the fault region, specifically as follows:

[0033] Obtain the time series of crack lengths for all sub-regions;

[0034] At each moment, the estimated crack lengths of all sub-regions are compared, and the region number corresponding to the maximum value is extracted as the fault region.

[0035] The introduction of a continuous determination mechanism to verify the fault area specifically involves:

[0036] Set a minimum time confirmation length Z. Only when the fault location judgment results are consistent for Z consecutive time periods will the fault area corresponding to the current area number be determined as the true fault location.

[0037] Furthermore, based on the crack length time series and the fault region, a dual-criteria mechanism is designed to provide early warning for the fault region and generate a corresponding early warning flag signal, specifically as follows:

[0038] Set a crack alarm threshold L that matches the material and structure. r When a certain region S i Smooth crack estimation satisfy

[0039] ≥L r When the crack reaches the set limit crack threshold L, the area is declared to be in a warning state; if the crack continues to grow to the set limit crack threshold L, the warning state will be declared to be in a warning state. f If so, it is considered a critical failure of the structural function;

[0040] Simultaneously, a fitting curve based on sliding window polynomial fitting is constructed, and predictions are made for the most recent W sampling points to solve for the prediction time point under the current fitting curve. Then compare with the current time t s Compare to obtain the remaining time of the fault Where, when the current time point T fSmaller than a certain set response window T r At that time, the system issues an early warning signal, indicating that the structure is entering a state of imminent failure.

[0041] Furthermore, when collecting the original strain measurement values ​​of any of the monitored equipment structures, outlier identification and correction are performed on the original strain measurement values, and a dual threshold mechanism is adopted in combination with global statistical criteria and local trend residual test, specifically including:

[0042] A sliding statistical window is constructed on the observation sequence of each strain observation point to calculate its first quartile Q1 and third quartile Q3, and an outlier range control threshold τ is defined. Any observation value is substituted into the following expression:

[0043] ε ij (t s ) is considered abnormal if and only if: |ε ij (t s )-μ ij (s)|>τ·(Q3-Q1)

[0044] Where μ ij (s) represents the median value within the same sliding window, and τ is an adjustable weighting coefficient;

[0045] For abnormal data points, the crack estimate is introduced as a compensation reference, and backfilling or replacement is performed using a local weighted fitting function within the time window to ensure the sequence is stable and continuous.

[0046] Furthermore, the structure-constrained neural network model is deployed on an edge computing device for real-time processing of strain monitoring data from each sub-region.

[0047] The beneficial technical effects of the present invention are at least as follows:

[0048] This invention proposes a novel intelligent fault identification method that integrates the physical characteristics of structural evolution with real-time sensor data analysis. Based on the crack propagation behavior during actual structural service, this method constructs a quantitative mapping relationship between fault evolution and monitoring response. Through multi-point monitoring in typical areas, highly correlated real-time strain data is collected for damage degree assessment and location identification. Furthermore, an early warning mechanism oriented towards structural response characteristics is proposed, capable of predicting damage development trends based on continuous observation sequences, thereby providing early warning of potential fault risks. This method does not rely on large-scale simulation modeling or highly complex image coding and deep models, exhibiting low computational resource requirements and strong engineering feasibility, making it particularly suitable for edge device deployment scenarios. By introducing a physically constrained structural evolution response mechanism and combining it with actual monitoring signals to achieve fault location and trend prediction, the real-time performance, accuracy, and interpretability of structural fault identification are significantly improved, providing a more robust and efficient technical path for the health management of equipment structures. Attached Figure Description

[0049] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of the steps of the intelligent fault location and early warning method for equipment structure based on fault chain of the present invention.

[0051] Figure 2 This is a schematic diagram of a preferred embodiment of the present invention. Detailed Implementation

[0052] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0053] In one or more embodiments, such as Figure 1 As shown, a method for intelligent fault location and early warning of equipment structure based on fault chain is disclosed. The method includes the following steps:

[0054] Obtain the equipment structure to be monitored, and divide the equipment structure to be monitored into N non-overlapping sub-regions;

[0055] Multiple strain monitoring points are set up in each sub-region to capture local load response characteristics as strain measurements for each strain monitoring point.

[0056] For any of the sub-regions, assuming that there is a crack along the main propagation path in the current sub-region, and combining the strain measurements, a strain sample dataset under different crack lengths is obtained through simulation.

[0057] A pre-built neural network model is trained based on the strain sample dataset to output crack estimates;

[0058] When real-time strain data is available, the real-time strain data is decoded based on the regional fault chain to obtain the crack length time series of each sub-region.

[0059] The fault region is located based on the maximum value of the crack length time series of each of the sub-regions, and a continuous determination mechanism is introduced to verify the fault region.

[0060] Based on the crack length time series and the fault region, a dual-criteria mechanism is designed to provide early warning for the fault region and generate a corresponding early warning signal.

[0061] Furthermore, the method also includes:

[0062] When collecting the original strain measurement values ​​of any of the equipment structures to be monitored, outlier identification and correction are performed on the original strain measurement values, and a dual threshold mechanism is adopted in combination with global statistical criteria and local trend residual test.

[0063] The following is a further explanation of the plan:

[0064] Specifically, the equipment structure to be monitored is acquired, and the equipment structure to be monitored is divided into N non-overlapping sub-regions:

[0065] Before implementing fault monitoring and location, the target structure needs to be functionally divided into spatial areas to facilitate subsequent regional response analysis based on monitoring point deployment and data decoding strategies. The structural regional division described here is not a simple geometric segmentation, but rather a structural functional zoning that comprehensively considers material inhomogeneity, stress concentration distribution, typical load paths, and common crack initiation regions.

[0066] Let the entire structure of the monitored object be denoted as S, and this structure be divided into N non-overlapping sub-regions Si. i ,satisfy:

[0067]

[0068] Where S represents the overall structure; S i Let N be the i-th structural sub-region; N is the total number of structural sub-regions, and the suggested range is 3≤N≤10.

[0069] Each sub-region S iIt possesses independent strain response characteristics, and subsequent local damage modeling will be conducted in conjunction with its corresponding sensor deployment points. In practical engineering, the selection of N can be comprehensively determined based on the following three dimensions:

[0070] Overall dimensions and typical stress directions of the equipment structure;

[0071] Typical crack history distribution and spatial clustering of high-frequency damage areas;

[0072] The total number of deployable sensors and their distribution coverage capability.

[0073] Consider each region S i The centroid coordinates are C i =(x i ,y i Let its projected area be A. i The coverage consistency index of the structural partitioning is defined as:

[0074]

[0075] Wherein, η is the structural partitioning consistency index, used to evaluate the consistency of area uniformity and stress distribution; A i σ is the projected area of ​​the i-th region; i σ represents the average equivalent stress in the i-th region under the initial load condition; σ is the average stress of the N regions.

[0076] Various stress values ​​σ i The values ​​were obtained through initial finite element simulation of the structure, and were taken as the average steady-state equivalent stress of the nodes within the region. By adjusting the region division method, a reasonable balance can be achieved between area equilibrium and stress distribution stability, thereby improving the physical rationality and monitoring adaptability of the overall division.

[0077] Specifically, multiple strain monitoring points are deployed in each sub-region to capture local load response characteristics as strain measurements for each strain monitoring point:

[0078] After completing the structural region division, to ensure real-time perception of structural state changes, in each region S... i Multiple strain monitoring points are deployed internally to capture local load response characteristics. The layout of the strain monitoring points needs to be optimized based on typical crack propagation directions, stress concentration areas, and material anisotropy characteristics to improve the sensitivity of monitoring data to potential damage.

[0079] Suppose that M strain monitoring points are set up in the i-th sub-region, denoted as P. ijWhere j = 1, 2, ..., M. Each monitoring point acquires raw strain data of the structure during service through a high-sensitivity strain sensor. The sensor type can be a patch resistance strain gauge, fiber Bragg grating, or MEMS device, depending on the surface characteristics of the structure and the spatial layout conditions.

[0080] Each strain monitoring point P ij Its spatial position is represented by its two-dimensional coordinates in the local coordinate system of the structure as follows:

[0081] P ij =(x ij ,y ij (3)

[0082] Among them, (x ij ,y ij ) indicates that monitoring point j is in region S i The coordinates of the layout position within the structure are obtained by registering the three-dimensional model of the structure with the actual installation surface.

[0083] To enhance the response coverage capability of the monitoring array, a strain observation distribution index μ is introduced. i Used to evaluate the uniformity of deployment and the effectiveness of location, defined as follows:

[0084]

[0085] Among them, C i =(x i ,y i ) is region S i The centroid coordinates are given by ||·||, where ||·|| represents the Euclidean distance. μ i A larger value indicates a more dispersed distribution of monitoring points, which is beneficial for capturing multi-directional load response characteristics within the region; a value that is too small may lead to redundancy or information overlap.

[0086] The strain measurement value at each monitoring point is denoted as ε. ij (t) represents the instantaneous strain observation value at this point at time t. The observation value is transmitted to the edge processing module via the signal acquisition channel for subsequent fault chain decoding.

[0087] Specifically, for any of the aforementioned sub-regions, assuming a crack exists along the main propagation path within the current sub-region, and combining the strain measurements, a strain sample dataset for different crack lengths is obtained through simulation:

[0088] After completing the structural region division and monitoring point deployment, it is necessary to establish a quantitative mapping relationship between the structural load response and the degree of damage. This process relies on strain response samples collected under different crack propagation states. This step is located in the core intermediate layer of the entire method, which not only inherits the spatial information of the monitoring points, but also provides high-quality, structurally sensitive training samples for subsequent data-driven fault chain modeling. It is a key transition link between physical modeling and data modeling.

[0089] Let the i-th subregion S i A crack exists along the main propagation path G. i (The direction has been obtained from structural simulation or fatigue test analysis), select the discrete state of crack length L. ix Where x = 1, 2, ..., X, represents the crack along path G. i The extended length. Here, the crack state values ​​cover the entire process of the structure from healthy to ultimate failure. For each state, strain values ​​at M points are collected through numerical simulation or physical loading experiments to form a strain sample vector:

[0090] D ix =[ε i1x ,ε i2x ,...,ε iMx (5)

[0091] To improve the responsiveness of data samples to local crack propagation and the convergence of fault chain model construction, an innovative data sample enhancement function φ(D) is introduced. ix The original strain vector is reconstructed based on crack direction weights:

[0092]

[0093] Where, ω ij Indicates monitoring point P ij G in the principal direction of the crack i The response weights are set in conjunction with the displacement-strain coupling degree in the simulation; Let be the strain gradient residual of the monitoring point in the i-th region along the orthogonal axis of the crack direction; λ is the structural response anomaly compensation factor, reflecting the perturbation effect of crack path offset on the strain distribution. This construction can enhance the sensitivity of data in the boundary ambiguity stage and effectively improve the expressive power of nonlinear features in subsequent modeling.

[0094] like Figure 2 As shown, the preferred value for M is 8.

[0095] Meanwhile, to evaluate the distribution quality of sample data in the overall crack space, an improved sample coverage quality index γ is introduced. i This is used to measure the density and response variability of the collected data at critical crack stages:

[0096]

[0097] The higher the value of this index, the greater the differentiation of strain response between adjacent crack states, which is beneficial for clear convergence and uniform error distribution during model training.

[0098] In the numerical simulation construction phase, crack propagation simulation is based on XFEM and uses a region S... i The actual boundary conditions and load conditions are considered, and a fine-mesh high-fidelity crack front modeling method is introduced. In the experimental data acquisition scenario, an equally spaced crack propagation fatigue loading device can be used, and all strain channels can be recorded in parallel at a synchronous acquisition frequency to form a sample pair set {φ(D ix ),L ix}

[0099] The data foundation built in this step not only directly determines the mapping capability of the subsequent fault chain model, but also serves as a bridge connecting the physical process and the core mechanism of data reasoning in the overall patent solution.

[0100] Specifically, a pre-built neural network model is trained based on the strain sample dataset to output crack estimates:

[0101] The ability to estimate crack states directly determines whether the system can achieve early fault warning. Therefore, constructing a robust mapping model from monitoring data to crack length is a key step in the entire methodological chain. The fault chain modeling method proposed in this step, based on the crack-strain sample data obtained in the previous step, combines physical constraint information and the characteristic structure of the input data to design a lightweight and highly interpretable region-level estimation model, thereby achieving low-latency and high-precision prediction of structural states.

[0102] Suppose that in the i-th structural region, there is already a sample set. Where φ(D) ix ) is the monitoring data vector after strain direction weighted reconstruction, L ix This corresponds to the crack propagation length. To map this data relationship, a structurally constrained neural network model f is introduced. i (·;θ i ), where parameter θ i The model is obtained through training by minimizing the constrained loss function. The overall architecture retains a three-layer structure, but the key design lies in introducing a "trend-driven term" and a "strain stray suppression term" to the output response pattern, so that the network can better fit the inherent laws of structural physical evolution.

[0103] The complete objective function is as follows:

[0104]

[0105] The second item Ψ i (θ i The multi-factor structural constraint regularization term proposed for the patent scheme includes two core sub-terms:

[0106]

[0107] Where, θ i Here are the model parameters: x is the current crack, X is the total number of cracks, and f is the total number of cracks. i For a structurally constrained neural network model, φ(D) ix ) is the data sample augmentation function, L ix Ψ is the crack propagation length, β is the weight of the regularization term, and Ψ is the crack propagation length. i (θ i ) represents the regularization term for multi-factor structural constraints;

[0108] Wherein, α1 and α2 correspond to the weight coefficients, This is a penalty term for local response deviation;

[0109] The strain noise sensitivity suppression term automatically suppresses the model sensitivity of high-noise points in the input data by coupling the strain variance within the sample group with the higher-order partial derivatives of the response; M is the total number of monitoring points j; ε ij Let ρ be the strain measurement value of the j-th monitoring point in the sample input; ij It is the expected gradient response intensity calculated from the crack propagation direction and simulation results, used to guide the network weights to align with the main driving direction of the crack; ε ijx This represents the strain measurement value of the j-th monitoring point in the current crack x;

[0110] The model training process employs a dynamic batch iteration approach, incorporating an early stopping mechanism to prevent overfitting, while simultaneously monitoring the stability of the mean squared rate of change of the residuals. The final output model format is as follows:

[0111]

[0112] This fault chain model has the following technical features: First, it models each region independently, preserving local structural differences; second, it is lightweight and can be deployed on edge hardware; and third, it integrates physical driving direction and input disturbance suppression capabilities, achieving a steady-state mapping of the model to the actual damage evolution trend. This model will serve as a generator for crack length time series in subsequent steps, directly providing the input basis for fault location identification and early warning.

[0113] Specifically, when real-time strain data is available, the real-time strain data is decoded based on the regional fault chain to obtain the crack length time series of each sub-region:

[0114] The real-time estimation result of the crack length needs to be further transformed into a time-series structural evolution trajectory to support subsequent fault location and trend prediction. This step plays a crucial role in this invention, extending the static neural network decoding result into a time-series expression of the structural state, while simultaneously performing structured correction on the inference output to improve the stability and adaptability of the system under complex operating conditions.

[0115] Let D be the strain monitoring data for S consecutive sampling periods in the i-th region. is Where s = 1, 2, ..., S, each frame of data, after being input into the fault chain model, yields the corresponding crack estimate. This sequence reflects the damage evolution process of the structure within this region.

[0116] Considering that actual monitoring signals are often accompanied by random disturbances or instantaneous spikes, a trend response enhancement mechanism is introduced to improve the smoothness and reliability of the inference sequence and to construct the time series for crack estimation.

[0117]

[0118] Here, α is a smoothing factor used to adjust the response ratio between the current valuation and the historical trend. This processing mechanism effectively suppresses the amplification risk of short-term abnormal disturbances while ensuring sensitivity, and improves the reliability of the time series response at the microcrack stage.

[0119] The final smooth crack sequence This sequence will serve as the core basis for determining the location of structural damage and for early warning behavior. Because of its continuity, trend, and physical interpretability, subsequent comparative analysis between regions and threshold determination can be robustly performed based on it. The sequence is also compatible with the real-time inference requirements of edge hardware, making it suitable for high-frequency monitoring deployment scenarios under limited computing power.

[0120] Specifically, the fault region is located based on the maximum value of the crack length time series of each of the sub-regions, and a continuous determination mechanism is introduced to verify the fault region:

[0121] Fault location is the location discrimination module in a structural monitoring system that most directly reflects the spatial distribution of damage. This step utilizes the crack estimation time series obtained in the previous stage. The area with the most obvious crack propagation trend is identified from multiple monitoring regions and used as the location of the structural failure at the current moment. Simultaneously, a cross-time redundancy confirmation mechanism is introduced to ensure the robustness and continuity of the identified location, avoiding misjudgments due to occasional noise.

[0122] Suppose the overall structure has been divided into N regions, and for each sub-region S i There is a crack length time series. Where s represents the index of the discrete sampling point within the time window. At each time t s The crack length estimates for all regions are compared, and the region index corresponding to the maximum value is extracted to initially define the candidate fault location I at that moment. s for:

[0123]

[0124] Considering the potential for short-term fluctuations or measurement errors in the structural response, I is directly used. s Determining a fault location may lead to false tripping. Therefore, a time consistency determination mechanism is introduced, setting a minimum time confirmation length Z. Fault location determination is only performed when the fault location determination results are consistent over Z consecutive time intervals, i.e., I... s-Z+1 =I s-Z+2 =…=I s Only then did they manage to control the area. The location was determined to be the actual fault location. This mechanism effectively improves the localization process's resistance to occasional interference without significantly increasing latency.

[0125] The final output of this step is the fault location area number I of the current structure. s This information serves as a spatial basis for subsequent early warning mechanisms to determine whether a fault is becoming more severe and whether an external response is needed. Furthermore, this positioning mechanism possesses strong real-time performance and edge adaptability, enabling rapid operation on low-power devices through simple comparison and memory mechanisms.

[0126] Specifically, based on the crack length time series and the fault region, a dual-criteria mechanism is designed to provide early warning for the fault region and generate corresponding early warning flag signals:

[0127] After completing the temporal decoding of crack length and fault location, an early warning mechanism needs to be established during the structural operation process to issue a response signal before the fault evolution reaches a critical danger level. This step plays a "forward intervention" role in the system and is a key logical closed loop for realizing the transition from state perception to behavior-driven action.

[0128] This step employs a dual-criteria mechanism to achieve early warning: the first is a static judgment based on the crack threshold, used to identify fault states that have reached a dangerous level; the second is a dynamic prediction mechanism based on trend extrapolation, used to identify risk scenarios that are about to reach their limit.

[0129] First, set a crack alarm threshold L that matches the material and structure. r When a certain region S i Smooth crack estimation satisfy When the crack reaches the set limit crack threshold L, the area is declared to be in a warning state; if the crack continues to grow to the set limit crack threshold L, the warning state will be declared to be in a warning state. f If so, it is considered a critical failure of the structural function.

[0130] Considering that most structures do not fail instantaneously, but rather exhibit an accelerated crack evolution trend, it is necessary to analyze the sequence {L}. is Trend prediction is performed. A sliding window polynomial fitting method is introduced to model the prediction based on the most recent W sampling points. Taking a third-order fitting function as an example:

[0131]

[0132] Solve the fitted curve Predicted time point Then compare with the current time t s Compare to obtain the remaining time of the fault When T f Smaller than a certain set response window T r At that time, the system issues an early warning signal, indicating that the structure is entering a state of imminent failure.

[0133] This mechanism comprehensively considers both the absolute safety margin of the current structural state and the dynamic trend of the state change rate, resulting in higher recognition sensitivity and fault tolerance. Compared to traditional judgment methods that only set static thresholds, this method can significantly advance the maintenance window, preventing irreversible damage to structural functions. It is particularly suitable for high-risk scenarios and long-term unattended equipment platforms.

[0134] The output of this step is a warning signal E. i (s)∈{0,1} indicates whether the i-th region has entered the warning state, and is dynamically updated over time.

[0135] Specifically, the method further includes:

[0136] When collecting the original strain measurements of any of the equipment structures to be monitored, outlier identification and correction are performed on the original strain measurements, and a dual threshold mechanism is used in combination with global statistical criteria and local trend residual tests:

[0137] During the operation of a structural monitoring system, strain acquisition signals are highly susceptible to environmental interference, electromagnetic noise, equipment drift, or data packet loss, leading to outliers in the observation sequence. If these outliers are directly input into the fault chain model without processing, they may cause misjudgments, trend drift, or even warning failures. Therefore, it is necessary to introduce outlier identification and robust correction mechanisms before data flows into the model to improve the overall anti-interference capability of the system.

[0138] This step applies to the original monitoring sequence ε. ij (t s Preprocessing is performed, where i is the region number, j is the monitoring point number, and t is the monitoring point number. s The sampling time is specified. Considering that outliers may manifest as abrupt changes, slow drifts, or high-frequency disturbances, this step employs a dual-threshold identification mechanism: one is a global criterion based on statistical steady-state characteristics, and the other is a dynamic test based on local trend residuals. The combination of these two mechanisms effectively covers different anomaly patterns.

[0139] First, a sliding statistical window is constructed on the observation sequence of each monitoring point, and its first quartile Q1 and third quartile Q3 are calculated. An outlier range control threshold τ is defined, and any observation value is substituted into the following expression:

[0140] ε ij (t s ) is considered abnormal if and only if: |ε ij (t s )-μ ij (s)|>τ · (Q3-Q1) (14)

[0141] Where μ ij (s) represents the median value within the same sliding window, and τ is an adjustable weighting coefficient, with a recommended range of 1.5 to 3. The above judgment ensures that the system can automatically adapt to background drift caused by seasonality, load fluctuations, etc., avoiding the false negativeing ​​of valid abnormal responses.

[0142] For data points identified as anomalous, an estimate L based on the structural crack evolution trend is introduced. is As a compensation reference, backfilling or replacement is performed using a locally weighted fitting function within the time window to ensure the sequence is stable and continuous. This trend recovery mechanism effectively avoids the problem of broken sequence discontinuities caused by missing data, improving the integrity of the model input.

[0143] This module is lightweightly embedded between the acquisition link and the decoding model in the overall system architecture, with minimal impact on real-time performance. Its introduction significantly improves the system's adaptability to sudden disturbances, equipment noise, outliers, and drift errors, and is an essential guarantee for achieving long-term online operation and stability in edge deployment.

[0144] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0145] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0146] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0147] The various illustrative logic blocks, modules, and circuits described in conjunction with the embodiments disclosed herein can be implemented or performed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The general-purpose processor may be a microprocessor, but in alternatives, it may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0148] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0149] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0150] The prior description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent fault location and early warning of equipment structure based on fault chain, characterized in that, Includes the following steps: Obtain the equipment structure to be monitored, and divide the equipment structure to be monitored into N non-overlapping sub-regions; Multiple strain monitoring points are set up in each sub-region to capture local load response characteristics as strain measurements for each strain monitoring point. For any of the sub-regions, assuming that there is a crack along the main propagation path in the current sub-region, and combining the strain measurements, a strain sample dataset under different crack lengths is obtained through simulation. A pre-built neural network model is trained based on the strain sample dataset to output crack estimates; When real-time strain data is available, the real-time strain data is decoded based on the regional fault chain to obtain the crack length time series of each sub-region, specifically: A trend response enhancement mechanism is introduced to construct a time series of crack estimation; a trend response enhancement mechanism based on a smoothing factor is introduced to the crack estimation value to construct a time series of crack estimation; wherein, the smoothing factor is used to adjust the response ratio between the current estimation and the historical trend; The fault region is located based on the maximum value of the crack length time series of each of the sub-regions, and a continuous determination mechanism is introduced to verify the fault region, specifically as follows: Obtain the time series of crack lengths for all sub-regions; At each moment, the estimated crack lengths of all sub-regions are compared, and the region number corresponding to the maximum value is extracted as the fault region. The introduction of a continuous determination mechanism to verify the fault area specifically involves: Set a minimum time confirmation length Only when consecutive Only when the fault location determination results are consistent at each moment can the fault area corresponding to the current area number be determined as the true fault location. Based on the crack length time series and the fault region, a dual-criteria mechanism is designed to provide early warning for the fault region and generate a corresponding early warning signal.

2. The intelligent fault location and early warning method for equipment structure based on fault chain as described in claim 1, characterized in that, The method further includes: When collecting the original strain measurement values ​​of any of the equipment structures to be monitored, outlier identification and correction are performed on the original strain measurement values, and a dual threshold mechanism is adopted in combination with global statistical criteria and local trend residual test.

3. The intelligent fault location and early warning method for equipment structure based on fault chain as described in claim 2, characterized in that, For any of the aforementioned sub-regions, assuming a crack exists along the main propagation path within the current sub-region, and combining the strain measurements, a strain sample dataset for different crack lengths is obtained through simulation, including: Let the first There are cracks along the path in each sub-region. The crack length is selected as a discrete state to represent the propagation length of the crack along the path. For each state along the crack path, data is collected through numerical simulation or physical loading experiments. The strain values ​​at each deployment point constitute a strain sample vector set; Perform sample augmentation on the strain sample vector set to generate a data sample augmentation function; The data sample augmentation function and the data sample augmentation function are combined to generate a strain sample dataset.

4. The intelligent fault location and early warning method for equipment structure based on fault chain as described in claim 3, characterized in that, The strain sample vector set is augmented to generate a data sample augmentation function, specifically as follows: ； in, For data sample augmentation functions, Indicates monitoring point In the principal direction of the crack The response weights are set in conjunction with the displacement-strain coupling degree in the simulation; Let j be the total number of monitoring points; For the first The strain gradient residuals at monitoring points in the region along the orthogonal axes of the crack direction; It serves as a structural response anomaly compensation factor, reflecting the perturbation effect of crack path offset on strain distribution; For the first The strain measurement value of each monitoring point in the current crack x.

5. The intelligent fault location and early warning method for equipment structure based on fault chain as described in claim 4, characterized in that, The pre-built neural network model is a structurally constrained neural network model, wherein the model parameters are obtained by training by minimizing the constrained loss function; Wherein, the constrained loss function is the objective function, expressed as: ； ； in, Here are the model parameters, where x is the current crack and X is the total number of cracks. This is a structurally constrained neural network model. For data sample augmentation functions, The crack propagation length, As the weight of the regularization term, For multi-factor structural constraint regularization terms; in, and The corresponding weighting coefficients, This is a penalty term for local response deviation; The strain noise sensitivity suppression term automatically suppresses the model sensitivity of high-noise points in the input data by coupling the strain variance within the sample group with the higher-order response partial derivatives. Let j be the total number of monitoring points; For the first The strain measurement values ​​of each monitoring point in the sample input; It is the expected gradient response intensity calculated from the crack propagation direction and simulation results, used to guide the network weights to align with the main driving direction of the crack; For the first The strain measurement value of each monitoring point in the current crack x; The structure-constrained neural network model adopts a dynamic batch iteration method and introduces an early stopping mechanism to prevent overfitting. At the same time, it monitors the stability of the mean square rate of change of the residuals to generate crack estimates.

6. The intelligent fault location and early warning method for equipment structure based on fault chain as described in claim 1, characterized in that, When real-time strain data exists, the real-time strain data is decoded based on the regional fault chain to obtain the crack length time series of each sub-region, specifically including: Obtain the corresponding crack estimate; A trend response enhancement mechanism based on a smoothing factor is introduced into the crack estimate to construct a time series of the crack estimate; wherein the smoothing factor is used to adjust the response ratio between the current estimate and the historical trend.

7. The intelligent fault location and early warning method for equipment structure based on fault chain as described in claim 1, characterized in that, Based on the crack length time series and the fault region, a dual-criteria mechanism is designed to provide early warning for the fault region and generate a corresponding early warning signal, specifically as follows: Set a crack alarm threshold that matches the material and structure. When a certain area Smooth crack estimation satisfy When the crack reaches a certain threshold, the area is considered to be in a warning state; if it continues to grow to the set limit crack threshold... If so, it is considered a critical failure of the structural function; Simultaneously, a fitting curve based on sliding window polynomial fitting is constructed, and the most recent... Prediction is performed using each sampling point, and the prediction time point under the current fitted curve is calculated. And then compared with the current moment Compare to obtain the remaining time of the fault ; where, at the current time point Smaller than a certain set response window At that time, the system issues an early warning signal, indicating that the structure is entering a state of imminent failure.

8. The intelligent fault location and early warning method for equipment structure based on fault chain as described in claim 4, characterized in that, When collecting the original strain measurement values ​​of any of the monitored equipment structures, outlier identification and correction are performed on the original strain measurement values, and a dual threshold mechanism is adopted in combination with global statistical criteria and local trend residual test, specifically including: A sliding statistical window is constructed on the observation sequence of each strain observation point, and its first quartile is calculated. and the third and fourth quartiles And define an outlier range control threshold. As an adjustable weighting coefficient, substitute any strain measurement value into the following expression: , The current moment; in This represents the median value within the same sliding window; For abnormal data points, the crack estimate is introduced as a compensation reference, and backfilling or replacement is performed using a local weighted fitting function within the time window to ensure the sequence is stable and continuous.

9. The intelligent fault location and early warning method for equipment structure based on fault chain as described in claim 5, characterized in that, The structure-constrained neural network model is deployed on an edge computing device to process strain monitoring data from each sub-region in real time.

Images