Multi-physical quantity measurement method and device of passive multi-parameter optical fiber sensing unit

Abstract

The application relates to the technical field of physical quantity measurement, and discloses a multi-physical quantity measurement method and device of a passive multi-quantity optical fiber sensing unit, which synchronously collects temperature, humidity, air pressure and tilt angle four-quantity optical signals, and quantifies the quantity coupling strength and the causal conduction time delay by real-time calculation of a timing cross correlation function; an initial weighted causal graph is constructed based on the timing causal index, a chain enhancement path is automatically detected, an initial abnormal node is located and a subsequent fault probability is iteratively predicted; then, a dynamic arrow chain is drawn along the enhancement path in a three-dimensional digital twin model starting from the initial abnormal node, and is rendered according to the fault probability; the application converts discrete multi-physical quantity abnormal signals into a topological and visual causal chain, realizes full-link transparent mapping of an abnormal source-propagation path-potential fault node, and significantly improves the prediction accuracy of the evolution trend of a complex structure state and the timeliness of operation and maintenance decisions.

Description

Technical Field

[0001] This invention relates to the field of physical quantity measurement technology, and more specifically, to a method and apparatus for measuring multiple physical quantities using a passive multi-parameter fiber optic sensing unit. Background Technology

[0002] In the field of physical quantity monitoring of complex structures, accurately capturing the dynamic correlations between multiple physical quantities and achieving early warning of faults is of paramount importance for ensuring the safe operation of systems. As the complexity of monitored objects increases, monitoring a single physical quantity is no longer sufficient to meet the needs of a comprehensive assessment of structural condition; therefore, multi-physical quantity collaborative monitoring technology has become a key means of ensuring structural safety. However, existing technologies still have shortcomings in quantifying and tracking the causal relationships between multiple physical quantities, leading to inaccurate fault location and delayed prediction, making it difficult to meet the requirements of high-precision, real-time monitoring.

[0003] Chinese patent application CN107402027A proposes a physical quantity measurement method based on an intensity-modulated fiber optic sensor. By setting up a dual-optical-path structure of a sensing unit and a reference unit, it eliminates system intensity noise interference caused by optical pulse source fluctuations and optical path device disturbances without introducing new disturbance sources, significantly improving the stability and minimum resolution of the fiber optic intensity-modulated sensor. However, this technical solution focuses on noise suppression and accuracy optimization in the measurement of a single physical quantity, without addressing the quantitative analysis of the dynamic correlation strength between parameters in multi-physical-quantity collaborative monitoring scenarios, and without constructing a causal propagation delay model between multiple parameters. When multiple parameters of the monitored object exhibit anomalies, this method cannot identify the propagation path between anomalies, nor can it determine the location of the initial anomaly node.

[0004] Chinese patent application CN119354244A proposes a multi-parameter dynamic monitoring system and method based on fiber Bragg gratings. It achieves synchronous measurement of multiple parameters through a dense fiber Bragg grating array, eliminating the need for multiple independent sensors and effectively reducing system complexity and cost. Simultaneously, it improves the accuracy of monitoring data by leveraging the synchronous acquisition characteristics of reflectance spectra. However, this technical solution only addresses the issue of "synchronous measurement" of multiple parameters, failing to conduct in-depth analysis of the causal relationships between multiple physical quantities. It neither quantifies the intensity of mutual influence between different parameters nor establishes a time-series model for the propagation of parameter anomalies. Consequently, it can only achieve parallel acquisition and presentation of multi-parameter data, unable to track the dynamic process of anomalies propagating from the initial node to subsequent nodes, and even less able to predict potential fault nodes in advance, making it difficult to meet the forward-looking fault warning requirements of complex structures.

[0005] In the monitoring of the main girder structure of large bridges, when the temperature rises abnormally, if the causal transmission delay between temperature and tilt angle is not quantified, it will be impossible to determine whether the temperature rise is the cause of the tilt angle change (temperature rise leads to thermal expansion and contraction of the structure, causing tilt change) or the result (structural tilt leads to abnormal local temperature distribution). Such misjudgment of causal relationship will lead to incorrect fault tracing. If a dynamic causal graph is not constructed and chain reinforcement paths are not detected, it is impossible to predict subsequent possible fault nodes in advance (such as abnormal air pressure caused by the aggravation of tilt angle abnormality), making it impossible for maintenance personnel to take preventive measures before the fault chain reaction occurs. In the existing technology, due to the lack of accurate calculation of the temporal cross-correlation between multiple physical quantities and quantitative analysis of causal transmission delay, the tracking of abnormal causal chains is inaccurate, fault prediction is lagging, and it is impossible to automatically generate clear and accurate abnormal causal chains in the three-dimensional digital twin system. As a result, it is impossible to achieve an intuitive display of "who is abnormal first, along which path, and where the next fault point is", resulting in an excessively short maintenance decision window, a significant increase in the risk of chain failures, and even the potential for partial failure of the bridge structure. Summary of the Invention

[0006] This invention is applicable to health monitoring scenarios for large structures such as main beams and supports of large bridges, core tubes of high-rise buildings, and wind turbine towers, as well as multi-physical quantity status monitoring scenarios for key load-bearing components of precision industrial equipment (such as machine tool spindle support structures). In these scenarios, it can simultaneously achieve accurate measurement of temperature, humidity, air pressure, and tilt angle, and dynamic tracking of abnormal causal chains, meeting the core requirements of these scenarios for accurate physical quantity measurement, timely anomaly tracing, and forward-looking fault prediction.

[0007] To overcome the aforementioned deficiencies of existing technologies, this invention provides a method and apparatus for measuring multiple physical quantities in a passive multi-parameter fiber optic sensing unit. Through a closed-loop process of "temporal cross-correlation → causal graph → digital twin arrow chain," the abnormal signals of multiple physical quantities are transformed into directly observable causal propagation trajectories. This immediately reveals the topological characteristics of the initial abnormal node in the graph, where the in-degree is zero or the weighted in-degree is minimal. The failure probability of subsequent nodes is quantified in advance through dynamic iteration of coupling strength. In a three-dimensional digital twin scenario, a dynamic arrow chain with varying thickness and color that evolves with risk is formed. This allows maintenance personnel to intuitively grasp "who is abnormal first, along which path, and where the next failure point is" without having to interpret lengthy data. This enables source handling to be completed with higher certainty and a shorter response window, significantly reducing the risk of cascading failures caused by inaccurate causal localization or delayed prediction.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] Methods for measuring multiple physical quantities using passive multi-parameter fiber optic sensing units include:

[0010] The system collects and preprocesses the raw optical signals of multiple physical parameters of the monitored object, calculates the temporal cross-correlation function between each physical parameter in real time, and quantifies the parameter coupling strength and causal propagation delay that characterize the correlation of multiple physical parameters based on the temporal cross-correlation function between each physical parameter. The multiple physical parameters include four physical parameters: temperature, humidity, air pressure, and tilt angle.

[0011] An initial weighted causal graph is constructed based on parametric coupling strength and causal propagation delay. Chain-like reinforcement paths in the initial weighted causal graph are detected and marked. The initial anomalous nodes of the chain-like reinforcement paths are located and the probability of subsequent failure nodes of all nodes other than the initial anomalous nodes is predicted.

[0012] Starting from the initial abnormal node, a dynamic arrow chain is drawn in the three-dimensional digital twin model of the monitored object along the chain-like enhancement path, and the dynamic arrow chain is rendered according to the probability of subsequent fault nodes.

[0013] The method for real-time calculation of the temporal cross-correlation function between various physical parameters includes:

[0014] Apply a sliding time window to the multi-parameter data set with timestamps to perform outlier repair and normalization, and obtain the normalized multi-parameter data sequence.

[0015] From the normalized multi-parameter data sequence, extract the time series of any pair of physical parameters i and j, and define them as the time series pair to be correlated; where i and j are the indices of the physical parameters, i≠j;

[0016] Based on the time series pairs to be correlated, calculate the temporal cross-correlation function R of physical parameters i and j. ij (τ).

[0017] The method for calculating the temporal cross-correlation function of physical parameters i and j based on the time series pairs to be correlated includes:

[0018] Set a time delay τ, and based on the time delay τ, perform a time shift on one of the time series in the time series pair to be correlated to generate a time delay sequence;

[0019] The covariance of the other time series in the time series pair to be correlated is calculated one by one with the time delay series. The covariance is then divided by the product of the standard deviations of the time series of physical parameter i and the time series of physical parameter j to obtain the time series cross-correlation function R of physical parameters i and j. ij (τ).

[0020] The methods for quantifying the parametric coupling strength and causal propagation delay include:

[0021] In the time-series cross-correlation function Rij Peak detection is performed on the curve (τ), and the global maximum value is searched and determined. The vertical axis value of the global maximum value is defined as the maximum correlation coefficient value, and the horizontal axis value is defined as the peak delay.

[0022] The parametric coupling strength and causal propagation delay are determined based on the maximum correlation coefficient value and peak delay.

[0023] The method for determining parametric coupling strength and causal propagation delay based on the maximum correlation coefficient value and peak delay includes:

[0024] Based on the maximum correlation coefficient value, it is determined whether there is a significant coupling relationship between physical parameters i and j. If there is a significant coupling relationship between physical parameters i and j, the maximum correlation coefficient value is recorded as the parameter coupling strength, the peak delay is recorded as the causal propagation delay, and physical parameters i and j with a significant coupling relationship are defined as a significant coupled parameter pair.

[0025] The method for constructing the initial weighted causal graph includes:

[0026] Create a basic graph containing four nodes: temperature, humidity, air pressure, and tilt angle.

[0027] Iterate through all parameter coupling strengths and causal propagation delays, and establish directed edges between corresponding nodes for each significant coupled parameter pair;

[0028] The direction of the directed edge is determined by the positive or negative value of the causal propagation delay, and the parameter coupling strength is assigned as the weight of the corresponding directed edge to form an initial weighted causal graph describing the static correlation of multiple physical parameters.

[0029] The method for detecting the chain-like enhancement path includes:

[0030] The new parametric coupling strength is continuously calculated, and the new parametric coupling strength is compared with the weight of the corresponding edge in the initial weighted causal graph to calculate the weight increase; if the weight increase is greater than the dynamic threshold, the edge is marked as an enhanced edge.

[0031] In the initial weighted causal graph containing augmenting edge labels, a depth-first search algorithm is used to identify paths consisting of at least three consecutive augmenting edges as chain augmenting paths.

[0032] The method for locating the initial abnormal node includes:

[0033] The weights of the initial weighted causal graph containing the reinforcing edge labels are updated to new parametric coupling strengths, while the labels of the chain-reinforcing paths are preserved, resulting in a dynamically updated weighted causal graph.

[0034] Calculate the in-degree and weighted in-degree of each node in the labeled chain-reinforcement path in the dynamically updated weighted causal graph, locate and output the initial anomalous node.

[0035] The method for locating the initial abnormal node also includes:

[0036] Extract all nodes on the marked chain-like augmented path to form a path node set;

[0037] In the dynamically updated weighted causal graph, calculate the in-degree of each node in the set of path nodes, and check if there is a node with an in-degree of 0.

[0038] If there is a node with an in-degree of 0, then the node with an in-degree of 0 is directly identified as the initial abnormal node; if there is no node, then the weighted in-degree of each node in the path node set is calculated, and the node with the smallest weighted in-degree is selected as the initial abnormal node.

[0039] A multi-physical quantity measurement device for a passive multi-parameter fiber optic sensing unit, used to implement the aforementioned multi-physical quantity measurement method for a passive multi-parameter fiber optic sensing unit, the device comprising:

[0040] The parameter correlation calculation module is used to collect and preprocess the raw optical signals of multiple physical parameters of the monitored object, calculate the temporal cross-correlation function between each physical parameter in real time, and quantify the parameter coupling strength and causal propagation delay that characterize the correlation of multiple physical parameters based on the temporal cross-correlation function between each physical parameter; the multiple physical parameters include four physical parameters: temperature, humidity, air pressure and tilt angle.

[0041] Fault prediction module: It is used to construct an initial weighted causal graph based on parametric coupling strength and causal propagation delay, detect and mark chain reinforcement paths in the initial weighted causal graph, locate the initial abnormal nodes of the chain reinforcement paths, and predict the probability of subsequent fault nodes of all nodes other than the initial abnormal nodes.

[0042] Visualization module: Used to draw a dynamic arrow chain in the 3D digital twin model of the monitored object, starting from the initial abnormal node and following the chain-like enhancement path, and to render the dynamic arrow chain according to the probability of subsequent fault nodes.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] This invention transforms traditional multi-physical quantity monitoring data, which can only trigger isolated alarms, into directly observable causal propagation trajectories through a closed-loop process of "temporal cross-correlation → causal graph → digital twin arrow chain." This immediately reveals the topological characteristics of initial abnormal nodes in the graph, where the in-degree is zero or the weighted in-degree is minimal. The failure probability of subsequent nodes is quantified in advance through dynamic iteration of coupling strength. This forms a dynamic arrow chain in the 3D digital twin scenario, with the thickness and color evolving with risk. Maintenance personnel can intuitively grasp "who is abnormal first, along which path, and where the next failure point is" without having to interpret lengthy data. This allows for more certainty and a shorter response window to complete source handling, significantly reducing the risk of cascading failures caused by inaccurate causal localization or delayed prediction. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of a method for measuring multiple physical quantities using a passive multi-parameter fiber optic sensing unit, provided in an embodiment of the present invention.

[0047] Figure 2 This is a schematic diagram of a data acquisition and timestamp synchronization process provided in an embodiment of the present invention;

[0048] Figure 3 A flowchart of a method for constructing an initial weighted causal graph provided in an embodiment of the present invention;

[0049] Figure 4 A schematic diagram illustrating the principle of multi-level information integration and display provided in an embodiment of the present invention;

[0050] Figure 5 This is a functional block diagram of a multi-physical quantity measurement device for a passive multi-parameter fiber optic sensing unit provided in an embodiment of the present invention. Detailed Implementation

[0051] 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 only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1

[0053] Please see Figure 1 As shown, this embodiment provides a method for measuring multiple physical quantities using a passive multi-parameter fiber optic sensing unit, including:

[0054] Step S10: Collect and preprocess the original optical signals of multiple physical parameters of the monitored object, calculate the temporal cross-correlation function between each physical parameter in real time, and quantify the parameter coupling strength and causal propagation delay that characterize the correlation of multiple physical parameters based on the temporal cross-correlation function between each physical parameter.

[0055] Further, step S10 includes:

[0056] Step S11: Collect raw optical signals of four physical parameters—temperature, humidity, air pressure, and tilt angle—from multiple passive multi-parameter fiber optic sensing units on the monitored object, and use a unified time reference for timestamp synchronization to generate a multi-parameter data set with timestamps.

[0057] The deployment of passive multi-parameter fiber optic sensing units needs to be determined based on the structural characteristics and key stress areas of the monitored object, such as the mid-span and supports of the main beam structure of large bridges, the corners and middle of the core tube of high-rise buildings, and the stress concentration points of key load-bearing components of precision industrial equipment, to ensure that the sensing unit can capture the real physical state changes of the monitored object. This sensing unit integrates four functional sensing sections, each based on different optical principles to sense specific physical parameters: the temperature sensing section uses a Bragg grating structure, utilizing the characteristic of reflected wavelength shift caused by temperature changes in the grating period to convert temperature changes into detectable light signal changes; the humidity sensing section uses a microcavity structure, leveraging changes in ambient humidity to induce changes in the refractive index of the microcavity wall material, leading to a shift in the microcavity resonant wavelength and achieving optical humidity sensing; the air pressure sensing section uses a Fabry-Perot cavity structure, changing the cavity length through the pressure difference between the inside and outside of the cavity, causing the transmitted light intensity to change with the cavity length, converting air pressure changes into light intensity signal changes; the tilt angle sensing section utilizes the fiber bending loss effect, when the monitored object is tilted, the fiber in the sensing unit bends to varying degrees, resulting in differences in transmitted light intensity loss, and the tilt angle is inverted through the change in light intensity.

[0058] During data acquisition, the sensing unit periodically samples four physical parameters: temperature (T), humidity (H), air pressure (P), and tilt angle (θ). Each sampling forms a complete sampling cycle containing the original optical signals of the four parameters. To address the timing discrepancies caused by inconsistent sampling time bases for multiple parameters in existing technologies, a unified time synchronization mechanism needs to be introduced: such as... Figure 2As shown, a GPS timing module is used as a high-precision time reference source. This module generates a time signal with microsecond-level precision by receiving satellite signals and broadcasts it to all passive multi-parameter fiber optic sensing units. Each sensing unit's end-side processor receives this time signal in real time. After completing one full sampling cycle—that is, simultaneously acquiring the raw optical signals corresponding to T, H, P, and θ—it immediately associates and binds the raw optical signal data of these four parameters with the currently received microsecond-level timestamp through a data binding unit, generating a structured data unit. This data unit contains the raw optical signal values ​​of the four physical parameters and a uniquely corresponding timestamp, defined as a timestamped multi-parameter data group.

[0059] During this process, timestamp binding must be completed within microseconds after the sampling cycle ends to avoid timestamp mismatch due to processing delays. This ensures that the original optical signal of each parameter accurately corresponds to its acquisition time. For example, in wind turbine tower monitoring, temperature abrupt changes at the top of the tower and tilt changes at the bottom may have a transmission delay of several minutes. If the sampling time bases for the two are inconsistent, subsequent analysis will misjudge the time deviation as physical transmission delay, resulting in false correlations. However, by binding timestamps at the microsecond level, the time records of temperature abrupt changes and tilt changes can be ensured to be absolutely accurate, laying the foundation for subsequent capture of the true physical delay.

[0060] Step S11 addresses the technical issue of inconsistent sampling time bases for multiple parameters in heterogeneous and complex monitoring environments. By achieving high-precision time synchronization and binding with real-time data, it eliminates spurious correlations introduced by sampling time deviations, providing a consistent data source for subsequent steps. Step S11 ensures that the time series of any two physical parameters are strictly aligned on the time axis, allowing the correlation between parameters in subsequent analysis to reflect true physical coupling rather than spurious correlations caused by time deviations. Microsecond-level timestamps provide a precise time measurement benchmark for subsequent calculations of causal propagation delays. Insufficient timestamp precision would significantly increase the calculation error of causal propagation delays, affecting the accuracy of causal relationship judgments. The multi-parameter data set with timestamps forms a structured data foundation, enabling subsequent step S12 to perform sliding window processing based on a unified time dimension. Without this structured data, the data within the window in S12 would be unable to be effectively cleaned and standardized due to time discrepancies, resulting in unreliable input for subsequent cross-correlation calculations.

[0061] Step S12: Apply a sliding time window to the multi-parameter data group with timestamps to perform outlier repair and normalization processing to obtain the normalized multi-parameter data sequence.

[0062] The length W of the sliding time window needs to be determined based on the dynamic response characteristics of the monitored object. This determination is based on factors including the vibration period of the monitored object, the rate of change of physical parameters, and the required time resolution. For monitored objects with slow dynamic responses, such as dam structures, whose physical parameters have long change periods, the window length W needs to be set to the number of sampling points that can cover a complete change period to ensure that the window contains sufficient change information. For monitored objects with fast dynamic responses, such as precision rotating equipment, whose physical parameters change frequently, the window length W needs to be set to a smaller number of sampling points to capture rapidly changing details. For example, in dam structure monitoring, W can be set to the duration corresponding to 600 sampling points, and in precision rotating equipment monitoring, W can be set to the duration corresponding to 100 sampling points.

[0063] Outlier removal employs a method based on the absolute deviation of the median (MAD). The specific implementation process is as follows: First, all data points for a specific physical parameter are extracted within a sliding time window, forming a data sequence within that window. The median M of this data sequence is calculated. The absolute deviation of each data point within the window from the median M is calculated, and the median of all absolute deviations is denoted as MAD. MAD is multiplied by a commonly used statistical coefficient k to obtain the outlier threshold Th. k is determined based on the data distribution characteristics; for approximately normally distributed data, k is set to 1.4826, ensuring that approximately 95% of the normal data falls within the outlier threshold range. If the absolute deviation of a data point within the window is greater than Th, the data point is determined to be an outlier. Outlier repair uses linear interpolation, where two adjacent normal data points are used to calculate a replacement value through linear fitting, replacing the original outlier. If the outlier is located at either end of the sequence, the values ​​of adjacent normal data points are used for replacement, ultimately obtaining the repaired data sequence within the window.

[0064] Normalization is performed independently on the repaired data sequence within the window for each physical parameter, using a linear normalization method. The calculation formula is: x norm =(x current -x min ) / (x max -x min ), where x norm For the normalized value of the parameter, x current For the parameter data points within the window to be normalized, x min x is the minimum value of the data sequence after parameter repair within this window. max This represents the maximum value of the parameter-corrected data sequence within this window. During the calculation, it is necessary to first traverse all corrected data points within the window to determine x. min With x max Then for each x currentPerform normalization calculations one by one to ensure that all normalized values ​​fall within the [0,1] interval.

[0065] Step S12 addresses the technical problems of low signal-to-noise ratio and inconsistent dimensions of multiple parameters in the original sensor data. In existing technologies, the original data is susceptible to electromagnetic interference, instantaneous vibrations, and other factors, resulting in glitch data. Furthermore, the dimensions of temperature, humidity, air pressure, and tilt angle differ significantly, making direct correlation comparison impossible. By using a sliding time window to remove outliers, glitch data can be effectively eliminated, improving the data signal-to-noise ratio and preventing outliers from interfering with subsequent correlation analysis. Through normalization, the influence of dimensions is eliminated, making the variation amplitudes of different physical parameters comparable within the same numerical range. Outlier removal employs the MAD (Mean-Standard Deviation) method, which is more robust to extreme outliers than the traditional mean-standard deviation method. The mean-standard deviation method is susceptible to extreme values, leading to threshold deviations, while MAD, based on the median, is less affected by extreme values ​​and can more accurately identify outlier data, ensuring the authenticity of the repaired data. The dynamic processing method using sliding time windows can adapt to the dynamic changes in the physical parameters of the monitored object. The window length is adjusted according to the characteristics of the monitored object, ensuring that the data within each window reflects the parameter change characteristics of the current period. This provides a segmented data foundation in the time dimension for subsequent dynamic updates of the causal map. If a fixed window or no window processing method is used, it is impossible to capture the dynamic changes of parameters in a timely manner, causing subsequent correlation analysis to lag behind the actual physical state changes.

[0066] Step S13: Using the normalized multi-parameter data sequence, calculate the time-series cross-correlation function between each physical parameter in real time;

[0067] Further, step S13 includes:

[0068] Step S131: Extract the time series of any pair of physical parameters i and j from the normalized multi-parameter data sequence, and define them as the time series pair to be correlated; where i and j are the indices of the physical parameters, i≠j;

[0069] Step S132: Set a time delay τ, and based on the time delay τ, perform a time shift on one of the time series in the time series pair to be correlated to generate a time delay sequence;

[0070] Step S133: Calculate the covariance of the other time series in the time series pair to be correlated with the time delay series one by one. Divide the covariance by the product of the standard deviations of the time series of physical parameter i and the time series of physical parameter j to obtain the time series cross-correlation function R of physical parameters i and j. ij (τ).

[0071] Specifically, the core of step S13 is to reveal the correlation between different physical parameters with time delay through quantitative calculation based on the normalized multi-parameter data sequence, providing a quantitative basis for subsequent causal relationship determination. Step S131 requires first clarifying the composition of the normalized multi-parameter data sequence. The normalized multi-parameter data sequence includes the normalized time series of four physical parameters: temperature T, humidity H, air pressure P, and tilt angle θ. Each sequence is plotted with time on the horizontal axis and normalized value on the vertical axis, and the time nodes of each sequence correspond one-to-one with the timestamps of step S11. The extraction of physical parameter pairs needs to cover all non-repeating parameter combinations. This involves selecting any two different time series of physical parameters from the four physical parameters to form a time series pair for correlation calculation. As shown in Table 1, there are a total of 6 physical parameter pairs: (T,H), (T,P), (T,θ), (H,P), (H,θ), and (P,θ), corresponding to the correlation analysis of temperature and humidity, temperature and air pressure, temperature and tilt angle, humidity and air pressure, humidity and tilt angle, and air pressure and tilt angle, respectively. Indices i and j are used to distinguish different physical parameters. For example, if i=1 represents temperature, i=2 represents humidity, i=3 represents air pressure, and i=4 represents tilt angle, then the physical parameter pair (T,H) corresponds to i=1 and j=2, the physical parameter pair (H,P) corresponds to i=2 and j=3, and so on. This ensures that each physical parameter pair has a unique and non-repeating identifier, providing a clear index basis for the subsequent differentiation and recording of cross-correlation functions. The extraction process must ensure that the time series length of each physical parameter pair is consistent, equal to the length W of the sliding time window in step S12, and that the time nodes completely correspond. That is, the k1th data point of the time series of physical parameter i and the k1th data point of the time series of physical parameter j both correspond to the sampling data at the same timestamp in step S11, avoiding calculation errors caused by mismatches in sequence length or time nodes. If the sequence lengths of the parameter pairs are inconsistent, subsequent time shifts and related calculations will not correspond, and an accurate cross-correlation function cannot be obtained.

[0072] The time delay τ needs to be set in conjunction with the physical response characteristics of the monitored object, and its value range is [-W / 2, W / 2], where W is the length of the sliding time window, and the unit is consistent with the sampling interval (e.g., seconds, milliseconds). The sign of τ represents the lead-lag relationship between parameters. The time shift operation is performed on one of the time series in the time series pair to be correlated. The specific time series to be shifted must maintain the consistency of the calculation logic. For example, to uniformly shift the time series of physical parameter j: let the time series of physical parameter i be i(t), t be the time variable, t=1,2,...,W, corresponding to W time nodes in the window, and the original time series of physical parameter j be j(t). When the time delay is set to τ, j(t) is shifted τ units along the time axis to generate the time delay sequence j(t-τ). During the translation process, if τ is positive, the first τ data points of j(t-τ) have no corresponding original data and are filled with data from the beginning of the sequence; the last τ data points exceed the length of the original sequence and are filled with data from the end of the sequence, ensuring that the length of the time delay sequence j(t-τ) is still W, consistent with the length of the time sequence i(t) of the physical parameter i, and meeting the sequence length requirements for subsequent covariance calculation.

[0073] For example, in oil pipeline monitoring, if we analyze the correlation between temperature T (i=1) and air pressure P (j=3), and set the value range of τ to [-50, 50] (assuming W=100, sampling interval 1 second), when τ=30, the air pressure sequence j(t) is shifted backward by 30 seconds to generate j(t-30). At this time, the first 30 data points of j(t-30) are filled with j(1), and the last 30 data points are filled with j(100) to ensure that the length is consistent with the temperature sequence i(t) (t=1-100), so that subsequent related calculations can be performed. If time shift is not performed, and only the correlation of parameters at the same time node is calculated, the time delay correlation between parameters cannot be captured. For example, the physical phenomenon of air pressure change after the temperature rises by 30 seconds will be ignored because the delay is not considered, resulting in incomplete judgment of the correlation.

[0074] Temporal cross-correlation function R ij The calculation of (τ) is based on the ratio of the sample covariance to the standard deviation, and the formula is: Where Cov[i(t),j(t-τ)] is the sample covariance of the time series i(t) of physical parameter i and the time delay series j(t-τ) of physical parameter j, and σ i Let σ be the standard deviation of the time series i(t) of physical parameter i. j Let be the standard deviation of the time series j(t) of the physical parameter j. Each time delay τ corresponds to an R0. ij By iterating through all values ​​of τ∈[-W / 2,W / 2], the time-series cross-correlation function curve of the physical parameter pair (i,j) can be obtained. This curve is plotted with τ as the horizontal axis and R as the horizontal axis.ij (τ) is the vertical axis, R ij The value of (τ) ranges from [-1, 1]. The closer the absolute value is to 1, the stronger the correlation between physical parameters i and j under that τ. ij When (τ) is positive, it indicates that the two are positively correlated, that is, when one parameter increases, the other parameter also increases; R ij When (τ) is negative, it indicates that the two are negatively correlated, that is, when one parameter increases, the other parameter decreases. Table 1 is a table showing the correspondence between physical parameter pairs and time-series cross-correlation functions.

[0075] Table 1. Correspondence between physical parameter pairs and temporal cross-correlation functions

[0076]

[0077] Step S13 addresses the technical problem in existing technologies that cannot quantitatively describe the lead-lag relationship and dynamic correlation strength among multiple physical parameters. Existing technologies often rely on intuitive observation of the trends in parameter time series to determine correlation, failing to quantify the degree of correlation under different time delays or accurately determine the lead-lag duration between parameters. Step S13, through the calculation of the time-series cross-correlation function, transforms the dynamic correlation between parameters into a quantitative function curve, achieving a quantitative representation of the correlation relationship. The time-series cross-correlation function R... ij (τ) can capture the correlation characteristics between physical parameters as time delay changes. It can not only determine whether a correlation exists, but also determine the time delay when the correlation is strongest, providing a direct basis for determining the causal transmission delay in the subsequent step S14. If the time-series cross-correlation function R is missing... ij (τ) makes it impossible to accurately pinpoint the time delay between physical parameters, leading to errors in determining the time sequence of causal relationships. By calculating the ratio of covariance to standard deviation, the influence of the dispersion of physical parameters on the strength of the correlation is eliminated. For example, if two physical parameters with different dispersions are correlated solely based on covariance, the larger dispersion might result in a larger covariance value, but the actual correlation strength might be weak. R... ij (τ) Standard deviation normalization makes the quantification results of correlation strength more objective; the cross-correlation functions of the six physical parameter pairs comprehensively cover the correlation relationships between all physical parameters, providing a complete correlation data foundation for constructing the initial weighted causal graph in the subsequent step S21. If the function of a certain physical parameter pair is missing, the causal graph will have missing correlation information and cannot fully reflect the causal network between multiple parameters. Step S13 works closely with the preceding and following steps: the normalized multi-parameter data sequence generated in step S12 provides data with uniform dimensions and high signal-to-noise ratio for cross-correlation calculation. If the data is not normalized, the dimensional differences between parameters will lead to excessive differences in covariance and standard deviation values, R ij(τ) cannot accurately reflect the correlation strength; the time-series cross-correlation function generated in step S13 is the only input for peak detection and threshold judgment in step S14. Without this function, step S14 cannot extract the parametric coupling strength and causal transmission delay, and the subsequent causal graph construction loses the core data support, and the entire abnormal causal chain tracking process cannot proceed.

[0078] Step S14: Peak detection and threshold judgment are performed on the temporal cross-correlation function between each physical parameter to quantify the parameter coupling strength and causal propagation delay.

[0079] Further, step S14 includes:

[0080] Step S141, in the time-series cross-correlation function R ij Peak detection is performed on the curve (τ), and the global maximum value is searched and determined. The vertical axis value of the global maximum value is defined as the maximum correlation coefficient value, and the horizontal axis value is defined as the peak delay.

[0081] Step S142: Based on the maximum correlation coefficient value, determine whether there is a significant coupling relationship between physical parameters i and j. If there is a significant coupling relationship between physical parameters i and j, record the maximum correlation coefficient value as the parameter coupling strength, record the peak delay as the causal propagation delay, and define the physical parameters i and j with a significant coupling relationship as a significant coupled parameter pair.

[0082] Specifically, step S14 transforms the abstract functional relationship of the temporal cross-correlation function into a quantified causal feature index, providing structured input for subsequent causal graph construction.

[0083] Peak detection requires targeting the time-series cross-correlation function R. ij The specific implementation process of peak detection for (τ) curve expansion is as follows: First, traverse all data points within the range of τ values ​​to identify local peaks – which must satisfy the R value corresponding to a certain τ. ij The value of (τ) is simultaneously greater than the R corresponding to its left adjacent τ-Δτ and its right adjacent τ+Δτ. ij The (τ) value is used to exclude spurious peaks when τ is close to ±W / 2 due to data edge padding. Δτ is the minimum step size of τ, consistent with the sampling interval, to ensure coverage of all possible delay nodes. Then, the point with the largest value is selected from all identified local peaks and defined as the global maximum point. The vertical axis of the global maximum point is the maximum correlation coefficient value, reflecting the strongest correlation between physical parameters i and j under all time delays. The horizontal axis is the peak delay, reflecting the time difference when physical parameters i and j reach the strongest correlation. A positive value indicates that the change of physical parameter i leads physical parameter j, and a negative value indicates that the change of physical parameter j leads physical parameter i.

[0084] Threshold determination requires first identifying the relevant threshold. The setting of the relevant threshold needs to consider the structural characteristics of the monitored object and the intensity of environmental interference. The specific method is as follows: First, collect multiple sets of historical data of the monitored object under normal operating conditions. The amount of historical data needs to cover at least one complete environmental change cycle, such as one day or one week. Repeat step S10 to calculate the maximum correlation coefficient value of each physical parameter pair, and statistically analyze the distribution characteristics of these values, such as mean, standard deviation, and quantiles. Take the 95th quantile of the maximum correlation coefficient value distribution under normal conditions as the basic threshold, and then adjust it according to the structural stiffness of the monitored object. The higher the structural stiffness, the more important it is to adjust accordingly. For structures like concrete dams and steel structure factories, the smaller the impact of environmental interference on parameter correlation, the higher the basic threshold can be (α) to obtain the final correlation threshold. The value of α ranges from 0.05 to 0.15. For example, in the monitoring of concrete dams, α is set to 0.1 to avoid misjudging weak environmental interference correlations as valid correlations. Conversely, for structures with lower structural stiffness, such as flexible pipes and cable net structures, the greater the impact of environmental interference, the lower the basic threshold can be (β) to obtain the final correlation threshold. The value of β ranges from 0.03 to 0.1. For example, in the monitoring of flexible pipes, β is set to 0.05 to avoid missing weak correlations caused by structural flexibility. If the maximum correlation coefficient is greater than the preset correlation threshold, it is determined that there is a significant coupling relationship between physical parameters i and j. The maximum correlation coefficient is recorded as the parameter coupling strength. The larger the value, the tighter the correlation between the parameters, reflecting the degree of mutual influence in a physical sense. The peak delay is recorded as the causal propagation delay, providing a quantitative basis for determining the causal direction in the future. This pair of physical parameters is marked as a significantly coupled parameter pair. If the maximum correlation coefficient is less than or equal to the correlation threshold, it is determined that there is no significant coupling relationship between the two. In this case, the maximum correlation coefficient and peak delay of the pair of physical parameters need to be recorded as historical reference data and stored in the local database. They are not used in the subsequent edge construction of the initial weighted causal graph to avoid meaningless correlation information increasing the topological complexity of the graph, and at the same time, they provide data support for the subsequent system feedback optimization closed loop.

[0085] Step S14 addresses the technical problem in existing technologies that cannot accurately extract effective causal association information from time-series cross-correlation functions. Existing technologies often rely solely on observing the trend of the function curve to determine whether parameters are correlated, failing to quantify the strength and time delay of the association. Furthermore, they are prone to misjudging weak associations caused by noise as valid associations, leading to distortion in subsequent causal analysis. Peak detection technology can accurately locate the strongest association point of parameter pairs across all time delays, ensuring the representativeness of the extracted association information. By combining dynamic threshold judgments based on the characteristics of the monitored objects, genuine significant coupling relationships can be screened out, eliminating false weak associations. This invention, for the first time, quantifies the dynamic correlation between physical parameters into a binary index of "parameter coupling strength - causal transmission delay," forming a structured causal information unit. This provides a clear basis for edge weights and directions in step S21, which is crucial for constructing the initial weighted causal graph. Without step S14, step S21 would be unable to determine the core attributes of the graph edges, leading to a lack of foundation for causal graph construction. The dynamic setting mechanism for relevant thresholds allows the system to adapt to different monitoring objects, such as rigid dams and flexible pipelines, avoiding missed or false detections caused by uniform thresholds and improving the method's versatility. It also addresses the handling of parameter pairs without significant coupling relationships. This approach not only simplifies the topology of the graph and reduces interference in path identification, but also enables the dynamic capture of correlations through historical data storage. When the correlation strength of a certain parameter pair continues to rise, it can be quickly incorporated into the analysis without recalculating the entire process data. This effect goes beyond the scope of basic data processing and provides support for the system to dynamically adapt to changes in the state of the monitored object. The accurate extraction of peak delay provides a quantitative basis for judging the causal direction. Combined with physical logic, such as temperature changes usually precede tilting changes caused by thermal expansion and contraction of the structure, the causal roles between physical parameters can be clearly defined, avoiding the subjectivity of causal direction judgment in traditional methods.

[0086] Step S20: Construct an initial weighted causal graph based on parametric coupling strength and causal propagation delay, detect and mark chain-like reinforcement paths in the initial weighted causal graph, locate the initial anomalous nodes of the chain-like reinforcement paths, and predict the probability of subsequent fault nodes of all nodes other than the initial anomalous nodes.

[0087] Further, step S20 includes:

[0088] Step S21: Integrate all parameter coupling strengths and causal propagation delays to construct an initial weighted causal graph;

[0089] Please see Figure 3 As shown, step S21 further includes:

[0090] Step S211: Create a basic graph containing four nodes: temperature node, humidity node, air pressure node, and tilt angle node.

[0091] Step S212: Traverse all parameter coupling strengths and causal propagation delays, and establish directed edges between corresponding nodes for each significant coupled parameter pair;

[0092] Step S213: Determine the direction of the directed edge based on the positive or negative value of the causal propagation delay, assign the parameter coupling strength to the weight of the corresponding directed edge, and form an initial weighted causal graph describing the static correlation of multiple physical parameters.

[0093] Specifically, the core of step S21 is to transform the discrete causal information output from step S14—namely, parameter coupling strength, causal propagation delay, and significantly coupled parameter pairs—into a structured network topology. This provides a visualized and computable causal carrier for subsequent dynamic updates and anomaly tracing. Step S211 requires first creating the node structure of the basic graph. The node types strictly correspond to the physical parameters monitored in step S10, including four core nodes: temperature node T, humidity node H, air pressure node P, and tilt angle node θ. Each node needs to integrate two types of attributes: first, data attributes, including a real-time monitoring value interface for the corresponding physical parameter and a historical causal information storage unit. The real-time monitoring value interface is used to receive normalized data updated subsequently in step S10, and the historical causal information storage unit records the coupling strength and historical delay changes between the node and other nodes; second, topological attributes, including the node's neighbor node list and the node's spatial coordinate mapping relationship in the 3D digital twin model, providing a location basis for visualization in step S30, and the node's neighbor node list is used for subsequent path search. The creation of nodes must ensure uniqueness and integrity. Each node identifier must be consistent with the parameter index in step S13 (i=1 corresponds to T, i=2 corresponds to H, i=3 corresponds to P, and i=4 corresponds to θ) to avoid index confusion during subsequent association.

[0094] Directed edges are established solely based on significant coupled parameter pairs. The traversal logic must cover all non-repeating significant coupled parameter pairs: First, extract all combinations marked as "significant coupled parameter pairs" from the output of step S14, and traverse them in ascending order of parameter index, such as traversing i=1 and j=2, i=1 and j=3, and then traversing i=2 and j=3, etc., to avoid building duplicate edges; For each significant coupled parameter pair (i,j), establish a physical connection between the corresponding node i and node j in the basic graph, which is defined as a directed edge. Directed edges are chosen instead of undirected edges because causal relationships have a clear temporal order. Undirected edges cannot distinguish causal roles, which will lead to the inability to determine the direction of fault propagation during subsequent path tracing. This design solves the defect of traditional association graphs that "only reflect association, not causation".

[0095] Step S213 completes the assignment of direction and weight for directed edges: the direction determination is based on the causal propagation delay τ. peak If τ peakA positive τ indicates that the change in physical parameter i precedes that of physical parameter j, i.e., i is the cause and j is the effect. Therefore, the direction of the directed edge points from node i to node j. peak A negative value indicates that the change in physical parameter j precedes that of physical parameter i, meaning j is the cause and i is the effect. Therefore, the direction of the directed edge points from node j to node i. Weight assignment directly uses the parameter coupling strength, as this quantifies the tightness of the causal influence between physical parameters. A higher weight value indicates a stronger influence of the cause node on the effect node. This setting provides a comparable quantitative benchmark for detecting enhanced edges in the subsequent step S22. After assigning directions and weights, the basic graph is transformed into an initial weighted causal graph, which contains nodes and directed edges, and can fully describe the static causal network relationships between multiple physical parameters.

[0096] For example, in a tunnel monitoring scenario, step S14 outputs three significantly coupled parameter pairs: (H, P) (parameter coupling strength 0.72, τ peak =-5 seconds), (H,θ) (parametric coupling strength 0.68, τ peak =-8 seconds), (T,H) (parametric coupling strength 0.65, τ peak =+10 seconds). Step S211: Create four nodes: T, H, P, and θ. Each node integrates the real-time monitoring interface and spatial coordinates of the corresponding parameters within the tunnel. For example, node H corresponds to the coordinates of the sensing unit in the middle of the tunnel sidewall. Step S212: Traverse the three significantly coupled parameter pairs and establish directed edges between H and P, H and θ, and T and H respectively. Step S213: Determine the direction: τ of (H, P) peak =-5 seconds (P leads H), the side direction is P→H; τ of (H,θ) peak =-8 seconds (θ leads H), the side direction is θ→H; τ of (T,H) peak =+10 seconds (T leads H), the edge direction is T→H; the weights are assigned to 0.72, 0.68 and 0.65 respectively, and the final initial weighted causal graph presents a "multiple causes and one effect" topological structure of "T→H←P, T→H←θ", which intuitively shows that node H is the common object of influence of nodes T, P and θ.

[0097] Step S21 addresses the limitation in existing technologies where discrete physical parameters cannot form a system-level causal network for causal information, resulting in an inability to intuitively present the global causal relationships between multiple physical parameters and a lack of structured support for subsequent dynamic updates and anomaly tracing. By creating nodes, establishing directed edges, and assigning directions and weights, discrete coupling strength and time delay information are integrated into a topological graph, transforming previously isolated causal relationships into a visualized and computable network, thus overcoming the deficiency of traditional discrete data in terms of global correlation. This invention achieves a global topological representation of causal relationships with multiple physical parameters, clearly identifying complex structures such as "one cause with multiple effects," "multiple causes with one effect," and "chain causality." Maintenance personnel can grasp the entire causal network of the system without analyzing discrete data one by one, significantly improving the understandability of causal relationships. The direction setting of directed edges strictly follows the temporal order of causal propagation delay, ensuring that the causal relationships in the graph are consistent with physical reality, avoiding the source tracing errors caused by the inability of traditional undirected graphs to distinguish causal roles. The structured design of edge weights and node attributes provides crucial support for subsequent steps. Step S22 requires detecting enhanced edges based on changes in edge weights; step S23 requires tracing initial abnormal nodes based on the in-degree of nodes and edge directions; and step S30 requires achieving 3D visualization based on the spatial coordinates of nodes. Without this structured design, subsequent steps would lack a unified data interface and computational benchmark, hindering collaborative progress.

[0098] Step S22: Continuously calculate the new parametric coupling strength, and based on the new parametric coupling strength and the initial weighted causal graph, detect and mark chain-like enhancement paths to form a dynamically updated weighted causal graph;

[0099] Further, step S22 includes:

[0100] Step S221: Compare the new parametric coupling strength with the weight of the corresponding edge in the initial weighted causal graph and calculate the weight increase; if the weight increase is greater than the dynamic threshold, mark the edge as an enhanced edge.

[0101] Step S222: In the initial weighted causal graph containing the enhanced edge labels, a depth-first search algorithm is used to identify paths consisting of at least three consecutive enhanced edges as chain-like enhanced paths and to label them.

[0102] Step S223: Update the weights of the initial weighted causal graph containing the enhanced edge labels to the new parametric coupling strength, while retaining the labels of the chain-like enhanced paths, to obtain a dynamically updated weighted causal graph.

[0103] Specifically, step S22 identifies the chain path of anomaly propagation by continuously monitoring the changing trend of causal relationships, thus addressing the deficiency that the initial static map cannot capture real-time anomalies.

[0104] In step S221, a new parameter coupling strength is first obtained. This new parameter coupling strength is generated by repeatedly executing step S10 (collection-preprocessing-cross-correlation-peak detection) at a preset cycle. After the new parameter coupling strength is generated, it needs to be compared with the current weight of the corresponding edge in the initial weighted causal graph to calculate the weight increase ΔW. A dynamic threshold is set, and the determination of the dynamic threshold needs to be based on the weight fluctuation characteristics of the monitored object under normal operating conditions: first, weight change data for at least 10 update cycles under normal operating conditions of the monitored object is collected, and the maximum value of normal fluctuation ΔW is calculated. norma,max ; The dynamic threshold Th dyn Set as ΔW norma,max The value of k2 is k2 times the value of Th, with k2 ranging from 1.2 to 1.8. This value is adjusted based on the stability requirements of the monitored object; the higher the stability requirement, the larger k2. If ΔW > Th dyn This indicates that the causal relationship strength of this edge is significantly enhanced compared to the normal state, and it is marked as an enhanced edge; if ΔW≤Th dyn If the edge remains unchanged, it will not be marked. A dynamic threshold is chosen over a fixed threshold because the normal weight fluctuation range differs for different monitored objects. For example, the weight fluctuation of a flexible pipeline is greater than that of a rigid dam. A fixed threshold may misjudge normal fluctuations as enhanced edges or miss small-amplitude enhanced signals that exceed the normal range for that object. A dynamic threshold improves the accuracy of enhanced edge detection by adapting to the normal fluctuation characteristics of different objects.

[0105] Chain-based augmentation path identification employs a depth-first search algorithm. The implementation process of the depth-first search algorithm is as follows: First, start from each node of the initial weighted causal graph containing augmentation edge labels as the starting node for path search; Starting from the starting node, follow the direction of the directed edges to ensure that the search is along the causal propagation direction, recursively traversing all outgoing edges, and determining whether each outgoing edge is marked as an augmentation edge; If the current edge is an augmentation edge, continue traversing the outgoing edges of the next node pointed to by the edge, and record the number of consecutive augmentation edges; When the number of consecutive augmentation edges reaches or exceeds three, stop the recursion of the current branch, and mark the path (starting node → intermediate node 1 → intermediate node 2 → ... → ending node) as a chain-based augmentation path; After traversing all starting nodes and branches, summarize all chain-based augmentation paths that meet the conditions, and exclude duplicate paths, such as the same path found by different starting nodes. The criterion of "at least three consecutive reinforcing edges" is set because a single or two reinforcing edges may be caused by accidental environmental interference, such as instantaneous electromagnetic pulses or short-term temperature fluctuations, and do not have the continuity of fault propagation. Three or more consecutive reinforcing edges indicate that the reinforcement of causal relationship presents a chain-like transmission characteristic, which conforms to the physical law of faults gradually spreading from the initial node to subsequent nodes, and can effectively avoid misjudging accidental interference as an abnormal causal chain.

[0106] The weights of all edges in the initial weighted causal graph containing the enhanced edge markers are replaced with the new parameter coupling strength calculated in step S221 to ensure that the graph weights are consistent with the current causal association state. The chain-like enhanced paths marked in step S222 are retained, and information such as the node sequence, number of consecutive enhanced edges, and identification time of the path are stored as independent fields in the graph to provide path basis for locating the initial abnormal nodes in the subsequent step S23. The updated graph is defined as a dynamically updated weighted causal graph. The dynamically updated weighted causal graph contains both the current causal association strength and records the chain-like paths of abnormal propagation, which can simultaneously reflect the real-time state of the system and the trajectory of abnormal evolution.

[0107] Step S22 addresses the problem that the static characteristics of the initial weighted causal graph cannot reflect the dynamic changes in causal relationships. Existing technologies cannot identify the chain-like reinforcement trend of fault propagation, causing the tracking of abnormal causal chains to lag behind the actual fault evolution. By periodically updating coupling strength, using dynamic thresholds to detect reinforcing edges, and employing depth-first search to identify chain-like paths, real-time monitoring of causal relationship changes and capture of abnormal paths are achieved, overcoming the limitation of static graphs in adapting to changes in system state. The dynamic update mechanism ensures that the graph always reflects the current causal relationship state of the system. Real-time weight updates can capture the reinforcement trend of causal relationships in the early stages of anomalies, such as the gradual strengthening of the influence of faulty nodes on adjacent nodes, providing a time window for fault prediction compared to the traditional method of waiting for fault symptoms to appear. The detection of reinforcing edges focuses on causal relationships with "significantly enhanced correlation strength," a core signal of fault propagation. When a fault occurs, the initial abnormal node gradually influences subsequent nodes through the causal chain, manifested as a continuous increase in the coupling strength of the corresponding edge. Dynamic threshold filtering can accurately extract this signal, eliminating interference from normal fluctuations. The identification of chain-like reinforcement paths... Visualizing the abnormal propagation trajectory allows operations and maintenance personnel to directly grasp the specific path of the fault spreading from the initial node to subsequent nodes, such as T→H→P→θ, solving the problem that traditional discrete data cannot locate the fault propagation trajectory. The retention of path markings and the synchronous updating of weights provide key inputs for subsequent steps. Step S23 needs to locate the initial abnormal node based on the marked chain path, step S24 needs to calculate the transition probability based on the updated weights, and step S30 needs to render a dynamic arrow chain based on the path and weights. If step S22 is missing, subsequent tracing, prediction, and visualization will lose accurate data on the abnormal chain, resulting in an overly broad analysis scope or incorrect direction.

[0108] Step S23: Calculate the in-degree and weighted in-degree of each node in the marked chain-like enhancement path in the dynamically updated weighted causal graph, locate and output the initial anomalous node;

[0109] Further, step S23 includes:

[0110] Step S231: Extract all nodes on the marked chain-like enhanced path to form a path node set;

[0111] Step S232: Calculate the in-degree of each node in the set of path nodes in the dynamically updated weighted causal graph, and check if there is a node with an in-degree of 0.

[0112] Step S233: If there is a node with an in-degree of 0, then the node with an in-degree of 0 is directly identified as the initial abnormal node; if not, then the weighted in-degree of each node in the path node set is further calculated, and the node with the smallest weighted in-degree is selected as the initial abnormal node.

[0113] Step S23 uses graph theory topology analysis to locate the initial abnormal node of the chain-like reinforcement path, solving the problem that traditional methods cannot accurately identify the source of the fault from complex causal chains. Step S231 extracts the node set of the chain-like reinforcement path. The extraction process requires traversing all nodes in the path, removing duplicate nodes, and forming a path node set. The method for removing duplicate nodes is to retain the first occurrence of the node if the path has branches or loops, avoiding repeated calculations. The node extraction must strictly match the node sequence of the path in step S22 to ensure that the set contains all nodes on the path that participate in causal transmission, without omitting intermediate nodes. If intermediate nodes are omitted, such as only the start and end points are extracted, subsequent in-degree calculations will lose the causal relationship information of key nodes, leading to an incorrect tracing direction.

[0114] In-degree calculation must be performed within the dynamically updated weighted causal graph, not just within the set of path nodes. In-degree is defined as the number of directed edges pointing to a given node in the dynamically updated weighted causal graph. The calculation logic is as follows: for each node in the set of path nodes, traverse the endpoints of all directed edges in the dynamically updated weighted causal graph, and count the total number of edges ending at that node; this is the in-degree of that node. The physical meaning of in-degree is the "number of causal input sources" for a node. An in-degree of 0 means that no other node influences the node through causal relationships, which aligns with the physical characteristic of an initial anomalous node: "no preceding causal input, and the anomalous node itself occurs first." During the calculation, it is necessary to distinguish between "inner edges" and "outer edges". Inner edges are edges in chain-reinforced paths (such as T→H), while outer edges are edges in dynamically updated weighted causal graphs that point to path nodes but are not in chain-reinforced paths, such as other edges that are not marked as reinforcement edges. Both need to be included in the in-degree statistics because outer edges may reflect the potential causal input of nodes. If only inner edges are counted, the causal influence of non-reinforced edges will be ignored, resulting in an underestimation of the in-degree and misjudging the initial abnormal nodes.

[0115] The initial abnormal node determination follows the logic of "in-degree priority, weighted in-degree supplementation": If there is a node with an in-degree of 0 in the path node set, this node is directly determined as an initial abnormal node, because an in-degree of 0 indicates that no node drives it through causal relationships, and its abnormality can only originate from itself, such as initial damage to the structure itself or anomalies directly detected by the sensing unit, which meets the "source attribute" of the initial abnormal node; If the in-degree of all nodes is greater than 0, such as when the causal chain forms a closed loop or there are multiple source inputs, then the weighted in-degree needs to be calculated. The weighted in-degree is defined as the sum of the weights of all directed edges pointing to the node in the dynamically updated weighted causal graph. The weight is the parametric coupling strength in step S14. The calculation process is as follows: For each node, traverse all directed edges pointing to it, and accumulate the weight of each edge to obtain the weighted in-degree. The physical meaning of weighted in-degree is the "total strength of causal influence received by a node". The node with the smallest weighted in-degree indicates that it is least affected by the causal influence of other nodes and is more likely to be the first node to experience an anomaly in the causal chain. Since the anomaly of the initial abnormal node is generated by itself, it is far less affected by other nodes than the subsequently driven nodes, so its weighted in-degree is naturally the smallest.

[0116] For example, in a building settlement monitoring scenario, the chain-like enhancement path marked in step S22 is θ (tilt angle) → P (air pressure) → H (humidity) → T (temperature), and step S231 extracts the path node set as {θ, P, H, T}; step S232 calculates the in-degree in the dynamically updated weighted causal graph: the in-degree of θ is 0 (no other node points to it), the in-degree of P is 1 (only θ → P), the in-degree of H is 1 (only P → H), and the in-degree of T is 1 (only H → T); step S233 determines θ as the initial abnormal node because its in-degree is 0, which meets the source characteristic of "no preceding causal input". If the in-degree of all path nodes in the scenario is 1, such as the path θ→P→H, and there is a non-enhancing edge P→θ, making the in-degree of θ 1, then the weighted in-degree is calculated: the weighted in-degree of θ is the weight of the edge P→θ (0.3), the weighted in-degree of P is the weight of the edge θ→P (0.8), and the weighted in-degree of H is the weight of the edge P→H (0.6). At this time, θ, which has the smallest weighted in-degree, is determined to be the initial abnormal node because it is least affected by the causal influence of other nodes.

[0117] Step S23 addresses the problem in existing technologies that cannot accurately locate the initial abnormal node from a chain of causal relationships. Traditional methods often rely on the anomaly threshold of a single parameter for judgment; if a parameter exceeds the safe range, it is determined to be the source, ignoring the causal transmission order between parameters. This easily leads to misjudging subsequent driven nodes as the source, such as misjudging the anomaly of H as the source when it is actually caused by the causal transmission from θ→P→H. By using techniques such as path node extraction, in-degree calculation, and weighted in-degree supplementation, combined with the topological characteristics of the causal chain, accurate source location is achieved. Extracting the path node set narrows the tracing scope, analyzing only the nodes involved in anomaly propagation and avoiding calculations on irrelevant nodes in the entire graph, significantly improving tracing efficiency. Directly calculating the in-degree of the entire graph introduces interference from irrelevant nodes, leading to a dispersed tracing direction. The hierarchical determination of in-degree and weighted in-degree ensures tracing accuracy. The determination of 0 in-degree directly corresponds to the physical essence of "no prior causality." Weighted in-degree supplements and solves the tracing problem in scenarios with the same in-degree, avoiding subjective judgment errors of traditional methods. Step S23 works in conjunction with the dynamically updated weighted causal graph, incorporating the influence of the path outside the in-degree calculation to ensure that the tracing results conform to the global causal relationship of the system, rather than being limited to local paths. Ignoring the path outside the in-degree may miss key influences on the causal input of nodes, leading to tracing errors. The calculation of weighted in-degree can indirectly reflect the "causal vulnerability" of nodes. Nodes with consistently low weighted in-degree are more likely to have anomalies originating from themselves rather than external drivers, and can be a key focus for daily maintenance, expanding the system's preventive maintenance function. The chain-like enhancement path marked in step S22 provides a clear node extraction range for step S23. Without this path, step S23 would lose its source object and would have to blindly calculate in the entire graph, resulting in extremely low efficiency. The initial abnormal node located in step S23 is the only starting point of the Markov chain model in step S24. Without this starting point, step S24 cannot initialize the abnormal probability vector, and fault prediction would lose its foundation. The parametric coupling strength in step S14, as the edge weight of the dynamically updated weighted causal graph, directly affects the calculation result of the weighted in-degree. The accuracy of the weights ensures that the weighted in-degree can truly reflect the causal influence strength of the node. If the weights are distorted, the ranking of the weighted in-degree will lose its meaning.

[0118] Step S24: Starting from the initial abnormal node, apply the Markov chain model to the dynamically updated weighted causal graph to calculate the probability of subsequent fault nodes for all nodes other than the initial abnormal node.

[0119] Further, step S24 includes:

[0120] Step S241: Based on the edge weights of the dynamically updated weighted causal graph, calculate the state transition probability from each node in the graph to all its neighboring nodes, and construct the Markov transition probability matrix.

[0121] Step S242: Set the anomalous probability corresponding to the initial anomalous node to 1, and set the anomalous probability of all other nodes to 0, to form an initialized anomalous probability vector.

[0122] Step S243: The abnormal probability vector is iteratively calculated using the Markov transition probability matrix. The abnormal probability vector is updated at each time step, and finally, the subsequent fault node probabilities that evolve over time are generated for all nodes except the initial abnormal node.

[0123] Step S24 quantifies the probability of subsequent faulty nodes using a Markov chain model, addressing the problem that traditional methods cannot predict the trend of fault propagation from causal chains. The construction of the Markov transition probability matrix in step S241 is based on the edge weights of a dynamically updated weighted causal graph. The matrix dimension is N×N, where N is the total number of nodes in the dynamically updated weighted causal graph. For example, with 4 nodes, it would be a 4×4 matrix. The matrix element P(i,j) represents the state transition probability from node i to node j, physically representing the probability that node j will become abnormal after node i becomes abnormal. The calculation logic for the transition probability is as follows: For each node i, traverse all its outgoing edges (outgoing edges are directed edges from node i to other nodes). Divide the weight of each outgoing edge by the sum of the weights of all outgoing edges of node i to obtain the corresponding transition probability P(i,j). If node i has no outgoing edges, the transition probability of its corresponding row is 0; if node i has only one outgoing edge (i→j), then P(i,j)=1. The calculation must satisfy the "probability normalization" constraint of Markov chains, that is, the sum of the transition probabilities of each node's corresponding row is 1, to ensure compliance with the probability axiom. If normalization is not performed, the sum of the transition probabilities will be greater than or less than 1, which will cause the probability values ​​calculated in subsequent iterations to exceed the range of [0,1] and lose physical meaning.

[0124] The initialization of the anomaly probability vector must be based on the located initial anomaly node. The vector dimension is consistent with the transition probability matrix, being N-dimensional. The vector elements of the anomaly probability vector V represent the anomaly probability of each node at the initial moment. The vector elements corresponding to the initial anomaly node are set to 1, indicating that the node has been determined to be anomaly at the initial moment; the vector elements corresponding to other nodes are set to 0, indicating that there are no signs of anomaly at the initial moment. This initialization conforms to the physical law that "faults propagate from the source." At the initial moment, only the source has anomalies, and the anomalies of subsequent nodes need to be generated gradually through causal transmission. If the initial probability vector is set incorrectly, such as setting multiple nodes to 1, it will cause the probability values ​​calculated in subsequent iterations to be superimposed and distorted, failing to reflect the true fault propagation order.

[0125] The iterative calculation in step S243 needs to proceed step by step. The anomaly probability vector V(t') at each time step t' is obtained by multiplying the vector V(t'-1) from the previous time step t'-1 by the transition probability matrix P. The calculation formula is V(t') = V(t'-1) × P. The specific calculation logic is: the anomaly probability of node g at time step t' That is, the anomalous probability of node g is equal to the sum of the products of the anomalous probabilities of all nodes i that can drive it at time step t'-1, and the transition probability of i→g, where i traverses all nodes, and N is the total number of nodes. This represents the probability of node i being an anomaly at time step t'-1; This represents the state transition probability from node i to node g. Iteration must continue until a stopping condition is met: the maximum difference between the anomaly probability vectors of two adjacent time steps is less than a preset convergence threshold. The convergence threshold is set according to the monitoring accuracy requirements, with an exemplary range of 0.001-0.01. At this point, the probability values ​​tend to stabilize, indicating that the fault propagation trend has reached a steady state, and further iterations will not result in significant changes. The difference is calculated using absolute differences. After iteration stops, the anomaly probability vector for each time step is output. The elements in the vector, excluding the initial anomaly node, represent the subsequent fault node probability for each node. The higher the probability value, the higher the likelihood that the node will experience an anomaly at the corresponding time step.

[0126] The technical problem addressed in step S24 is that existing technologies cannot quantify the probability of failure of subsequent nodes in a causal chain. Traditional methods can only qualitatively determine that "a certain node may fail," without providing probabilistic results, making it difficult for maintenance personnel to assess risk priorities. By constructing a transition probability matrix, initializing anomaly probability vectors, and iteratively calculating, combined with the "no aftereffect" property of Markov chains—that is, the future state of a node is only related to its current state and not to its historical state—probabilistic prediction of fault propagation can be achieved. The probability output of time evolution reflects the dynamic process of fault propagation. Operation and maintenance personnel can grasp the risk distribution at different time steps, which is convenient for formulating phased intervention strategies and avoiding the lag of traditional qualitative judgment. The transition probability is calculated based on the parametric coupling strength, ensuring that the probability value can reflect the tightness of causal influence. The higher the coupling strength, the greater the transition probability and the faster the node failure probability rises, which is consistent with the physical reality of "strong causal association driving rapid fault propagation". Step S24 works in conjunction with the dynamically updated weighted causal graph. The transition probability is adjusted in real time with the update of the graph weight, ensuring that the prediction result is consistent with the current causal state of the system and avoiding the defects of static models that cannot adapt to state changes. The cumulative effect of probability during the iteration process can capture the fault scenario of "multi-source driven". If a node is driven by two predecessor nodes, its failure probability is the sum of the products of the two predecessor probabilities and the corresponding transition probability, which is more in line with the complex situation of multi-path fault propagation in reality. Traditional methods cannot realize the probability calculation of such multi-source superposition. The initial abnormal node in step S23 provides a unique probability initialization benchmark for step S242. Without this node, the initialization vector has no clear starting point, and the prediction loses its direction. The dynamically updated weighted causal graph edge weights in step S22 are the sole basis for calculating the transition probability in step S241. The dynamic update of the weights ensures that the transition probability can reflect the latest causal relationship strength. If the weights are not updated, the transition probability will be calculated based on outdated causal relationships, and the prediction result will be distorted. The parametric coupling strength in step S14 directly determines the edge weight size. The quantification accuracy of the coupling strength ensures that the transition probability can truly reflect the degree of causal influence. If the coupling strength is calculated incorrectly, the order of transition probabilities will be reversed, causing high-risk nodes to be misjudged as low-risk.

[0127] Step S30: Starting from the initial abnormal node, draw a dynamic arrow chain in the three-dimensional digital twin model of the monitored object along the chain-like enhancement path, and render the dynamic arrow chain according to the probability of subsequent fault nodes.

[0128] The core of step S30 is to transform the abstract causal chains and probabilistic data generated in steps S22, S23, and S24 into intuitive and actionable decision-making basis through the visualization of the 3D digital twin model and the standardized output of intelligent early warning reports. This solves the problems of difficult understanding and slow decision-making for maintenance personnel under traditional pure data presentation methods. Its implementation process requires close integration with the preliminary analysis results and unfolds in three key stages: 3D rendering of dynamic arrow chains, multi-level information integration and display, and automatic generation of intelligent early warning reports.

[0129] The 3D rendering of the dynamic arrow chain requires first completing the "mapping between abstract path and physical space": First, extract the actual installation coordinates of the passive multi-parameter fiber optic sensing units corresponding to each node from the initial weighted causal graph constructed in step S21. These coordinates must be consistent with the coordinate system of the 3D digital twin model of the monitored object. Coordinate acquisition must be based on the survey data during sensing unit deployment. For example, in bridge monitoring, the coordinates of node T correspond to the 3D position of the sensing unit at the mid-span of the main beam, and node H corresponds to the 3D position of the sensing unit on the side wall of the pier, ensuring that the spatial position of the node in the twin model completely matches the physical world. Then, starting from the initial abnormal node, draw continuous directional arrows along the marked chain-like enhancement path in the 3D digital twin model of the monitored object, defining this as a dynamic arrow chain. The rendering rules for arrows need to consider both "causal influence strength" and "fault risk level": The dynamic arrow chain is rendered based on the parameter coupling strength and the probability of subsequent fault nodes. Specifically, arrow thickness is positively correlated with parameter coupling strength; the higher the coupling strength, the thicker the arrow, intuitively reflecting the closer the causal influence of the cause node on the effect node. Arrow color is positively correlated with the probability of subsequent fault nodes; the higher the probability, the closer the color is to a high-risk color scheme, such as dark red; the lower the probability, the closer the color is to a low-risk color scheme, such as light yellow. This rendering rule was chosen because, in pure data format, it is difficult for operations and maintenance personnel to quickly distinguish the differences in influence strength and risk level. The dual dimensions of thickness and color enable parallel information transmission, significantly reducing the cost of understanding.

[0130] Multi-level information integration and display requires overlaying four types of key information on top of the dynamic arrow chain to form a complete information view from macro to micro and from real-time to historical: such as Figure 4As shown, the first layer consists of floating labels for real-time parameter values. Each node displays the real-time normalized value of its corresponding physical parameter. The data originates from the latest normalized data sequence output in step S12. The labels are dynamically bound to the node's location, allowing maintenance personnel to view the real-time value simply by clicking on the node, thus resolving the issue of "only seeing the path, not the current status." The second layer is an interactive historical trend curve. Clicking on any node will display the normalized value change curve for that parameter over the past N' update cycles. N' is set according to monitoring needs, such as the past 24 cycles. The curve data originates from the historical stored data of each cycle in step S10. The trend curve allows tracing the evolution of the parameter from normal to abnormal, aiding in determining the time point when the anomaly occurred. The third layer... The first layer is a semi-transparent, dynamically updated weighted causal graph. The dynamically updated weighted causal graph generated in step S22 is superimposed on the twin model in a semi-transparent form. The nodes in the graph correspond one-to-one with the nodes in the twin model, and the weights and directions of the edges remain unchanged. This design allows maintenance personnel to grasp the global causal relationship of the system while viewing the physical space path, avoiding the disconnect between local paths and global causality. The fourth layer is an early warning information panel. The panel is fixed to one side of the twin model interface and sorts the subsequent fault nodes generated in step S24 from high to low probability. It displays the node name and current probability value of each node. The panel data is updated in real time as the probability iterates, providing maintenance personnel with a clear risk priority ranking.

[0131] The automatic generation of intelligent early warning reports must meet the requirements of "trigger-based output and standardized content": First, the report triggering conditions are set. When the probability of any node in the subsequent fault node probabilities output in step S24 exceeds the warning threshold, the system automatically triggers the report generation mechanism. The determination of the warning threshold needs to be combined with the safety operation standards of the monitored object. The specific method is as follows: collect the "critical value of the node probability before the fault occurred" from the historical fault cases of the monitored object, and calculate the 90th percentile of the critical value as the warning threshold. If there is no historical fault data for the monitored object, the safety threshold of similar monitored objects is referenced. An exemplary warning threshold range is 0.6-0.8. The report strictly adheres to a fixed structure to ensure standardization: The first part is the anomaly source analysis, detailing the name of the initial anomaly node located in step S23, the type of physical parameters corresponding to that node, the node's spatial location in the twin model, and the criteria for determining it as the initial anomaly node, such as an in-degree of 0 or the minimum weighted in-degree. The second part is the causal propagation path, describing the chain-like reinforcement path marked in step S22 in textual form, and explaining the parametric coupling strength of each edge, reflecting the differences in causal impact along the path. The third part is the risk assessment, listing the names of all nodes except the initial anomaly node, and subsequent risks in tabular form. The system first identifies the probability of a faulty node and the corresponding time step to clarify the urgency of risk for different nodes. The fourth part provides handling recommendations. The system matches handling strategies from a pre-set solution library that correspond to the "initial abnormal node type + chain-like enhanced path characteristics." This pre-set solution library is built upon fault handling cases of similar monitored objects. Each case includes information such as the abnormality type, propagation path, handling measures, and effect verification. For example, "When the initial abnormal node is an angle of inclination and the path is θ→P→H, it is recommended to prioritize checking the stability of the surrounding structure of the angle of inclination sensor unit and simultaneously monitor the air pressure change trend," ensuring the recommendations are targeted and actionable. After the report is generated, it is automatically stored in the historical report library and can be pushed to maintenance personnel via email or system notification, achieving "early warnings for abnormalities and immediate solutions for early warnings."

[0132] Step S30 addresses two core problems in existing technologies: First, complex causal relationships and probability data are presented in pure numerical or tabular form, requiring maintenance personnel to spend a significant amount of time interpreting the data logic, making it difficult to quickly grasp the full picture of anomalies and leading to delayed decision-making. Second, anomaly-related information (such as real-time status, historical trends, global causality, and risk priority) is scattered across different modules, requiring maintenance personnel to query and integrate across modules, resulting in low information retrieval efficiency and the potential for missing key information. By using dynamic arrow chain 3D rendering technology, abstract chain-like enhanced paths are transformed into visual arrows in physical space. Combined with dual encoding of thickness and color, this directly maps the intensity of causal impact and the level of fault risk, solving the problem of "pure data being difficult to understand." Through multi-level information integration, real-time data, historical trends, global causality, and risk ranking are centralized on the same interface, eliminating the need for cross-module queries and solving the problem of "information dispersion." Through the standardized generation of intelligent early warning reports, the anomaly source, propagation path, risk assessment, and handling suggestions are output in a fixed structure, avoiding the subjectivity of human interpretation and information omissions, thus solving the problem of "unsystematic decision-making basis." The 3D rendering of dynamic arrow chains allows maintenance personnel to directly "see" the fault propagation path in a realistic physical space scene, eliminating the need to interpret complex data sequences. For example, in bridge monitoring, maintenance personnel can intuitively identify the thickest and darkest red arrow in the path "pier tilt angle node (θ) → bridge deck air pressure node (P)" using a twin model, immediately determining that this path has the strongest causal impact and the highest risk, rather than comparing parameter coupling strength and probability values ​​one by one. This visualization reduces anomaly understanding time to less than one-third of traditional pure data methods, significantly improving decision-making efficiency. This effect stems from the technical means of "physical space mapping + multi-dimensional encoding," which transforms abstract data into visual information that conforms to human spatial cognitive habits, reducing cognitive load. Multi-level information integration allows maintenance personnel to drill down from the macroscopic dynamic arrow chain to the microscopic historical trend curve, and then connect it to the globally dynamically updated weighted causal graph, ultimately clarifying risk priorities through the early warning panel. This "overall view-detail-global" information connection avoids the decision-making bias of "seeing only the part and not the whole" caused by traditional single information modules. The fixed structure of intelligent early warning reports is: source of anomaly - propagation path - risk assessment - handling suggestions. This ensures that the output content of each early warning is consistent and avoids the subjectivity and omissions of human recording. For example, when different maintenance personnel handle the same type of anomaly, they all operate according to the "initial anomaly node judgment basis" and "matching handling suggestions" in the report, rather than relying on personal experience.

[0133] Example 2

[0134] This embodiment, based on Embodiment 1, provides a multi-physical quantity measurement device using a passive multi-parameter fiber optic sensing unit, such as... Figure 5 As shown, it includes:

[0135] The parameter correlation calculation module is used to collect and preprocess the raw optical signals of multiple physical parameters of the monitored object, calculate the temporal cross-correlation function between each physical parameter in real time, and quantify the parameter coupling strength and causal propagation delay that characterize the correlation of multiple physical parameters based on the temporal cross-correlation function between each physical parameter; the multiple physical parameters include four physical parameters: temperature, humidity, air pressure and tilt angle.

[0136] Fault prediction module: It is used to construct an initial weighted causal graph based on parametric coupling strength and causal propagation delay, detect and mark chain reinforcement paths in the initial weighted causal graph, locate the initial abnormal nodes of the chain reinforcement paths, and predict the probability of subsequent fault nodes of all nodes other than the initial abnormal nodes.

[0137] Visualization module: Used to draw a dynamic arrow chain in the 3D digital twin model of the monitored object, starting from the initial abnormal node and following the chain-like enhancement path, and to render the dynamic arrow chain according to the probability of subsequent fault nodes.

[0138] Furthermore, in the parameter correlation calculation module, the method for real-time calculation of the temporal cross-correlation function between various physical parameters includes:

[0139] Step S131: Extract the time series of any pair of physical parameters i and j from the normalized multi-parameter data sequence, and define them as the time series pair to be correlated; where i and j are the indices of the physical parameters, i≠j;

[0140] Step S132: Set a time delay τ, and based on the time delay τ, perform a time shift on one of the time series in the time series pair to be correlated to generate a time delay sequence;

[0141] Step S133: Calculate the covariance of the other time series in the time series pair to be correlated with the time delay series one by one. Divide the covariance by the product of the standard deviations of the time series of physical parameter i and the time series of physical parameter j to obtain the time series cross-correlation function R of physical parameters i and j. ij (τ).

[0142] Furthermore, in the fault prediction module, the method for constructing the initial weighted causal graph includes:

[0143] Step S211: Create a basic graph containing four nodes: temperature node, humidity node, air pressure node, and tilt angle node.

[0144] Step S212: Traverse all parameter coupling strengths and causal propagation delays, and establish directed edges between corresponding nodes for each significant coupled parameter pair;

[0145] Step S213: Determine the direction of the directed edge based on the positive or negative value of the causal propagation delay, assign the parameter coupling strength to the weight of the corresponding directed edge, and form an initial weighted causal graph describing the static correlation of multiple physical parameters.

[0146] Furthermore, in the fault prediction module, the method for locating the initial abnormal node of the chain-like enhancement path includes:

[0147] Step S231: Extract all nodes on the marked chain-like enhanced path to form a path node set;

[0148] Step S232: Calculate the in-degree of each node in the set of path nodes in the dynamically updated weighted causal graph, and check if there is a node with an in-degree of 0.

[0149] Step S233: If there is a node with an in-degree of 0, then the node with an in-degree of 0 is directly identified as the initial abnormal node; if not, then the weighted in-degree of each node in the path node set is further calculated, and the node with the smallest weighted in-degree is selected as the initial abnormal node.

[0150] The methods and apparatus of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated.

[0151] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0152] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for measuring multiple physical quantities using a passive multi-parameter fiber optic sensing unit, characterized in that, The method includes: The system collects and preprocesses the raw optical signals of multiple physical parameters of the monitored object, calculates the temporal cross-correlation function between each physical parameter in real time, and quantifies the parameter coupling strength and causal propagation delay that characterize the correlation of multiple physical parameters based on the temporal cross-correlation function between each physical parameter. The multiple physical parameters include four physical parameters: temperature, humidity, air pressure, and tilt angle. An initial weighted causal graph is constructed based on parametric coupling strength and causal propagation delay. Chain-like reinforcement paths in the initial weighted causal graph are detected and marked. The initial anomalous nodes of the chain-like reinforcement paths are located and the probability of subsequent failure nodes of all nodes other than the initial anomalous nodes is predicted. Starting from the initial abnormal node, a dynamic arrow chain is drawn in the three-dimensional digital twin model of the monitored object along the chain-like enhancement path, and the dynamic arrow chain is rendered according to the probability of subsequent fault nodes. The method for real-time calculation of the temporal cross-correlation function between various physical parameters includes: A sliding time window is applied to the timestamped multi-parameter data set for outlier repair and normalization, resulting in a normalized multi-parameter data sequence. From this normalized sequence, the time series of any pair of physical parameters i and j are extracted and defined as the time series pair to be correlated. Here, i and j are indices of the physical parameters, i ≠ j. Based on the time series pair, the temporal cross-correlation function R of physical parameters i and j is calculated. ij (τ); The method for calculating the temporal cross-correlation function of physical parameters i and j based on the time series pair to be correlated includes: setting a time delay τ; shifting one time series in the time series pair to be correlated based on the time delay τ to generate a time-delayed sequence; calculating the covariance of the other time series in the time series pair to be correlated with the time-delayed sequence one by one; and dividing the covariance by the product of the standard deviation of the time series of physical parameter i and the standard deviation of the time series of physical parameter j to obtain the temporal cross-correlation function R of physical parameters i and j. ij (τ); The quantification method for the parametric coupling strength and causal propagation delay includes: in the time-series cross-correlation function R ij Peak detection is performed on the curve (τ) to search for and determine the global maximum point. The vertical axis value of the global maximum point is defined as the maximum correlation coefficient value, and the horizontal axis value is defined as the peak delay. The parametric coupling strength and causal propagation delay are determined based on the maximum correlation coefficient value and the peak delay. The method for determining the parameter coupling strength and causal propagation delay based on the maximum correlation coefficient value and the peak delay includes: determining whether there is a significant coupling relationship between physical parameters i and j based on the maximum correlation coefficient value; if there is a significant coupling relationship between physical parameters i and j, then the maximum correlation coefficient value is recorded as the parameter coupling strength, the peak delay is recorded as the causal propagation delay, and physical parameters i and j with a significant coupling relationship are defined as a significant coupled parameter pair. The method for constructing the initial weighted causal graph includes: creating a basic graph containing four nodes: temperature node, humidity node, air pressure node, and tilt angle node; traversing all parameter coupling strengths and causal propagation delays, establishing directed edges between corresponding nodes for each significant coupled parameter pair; determining the direction of the directed edges based on the positive or negative value of the causal propagation delay, assigning the parameter coupling strength as the weight of the corresponding directed edge, and forming an initial weighted causal graph describing the static correlation of multiple physical parameters. The method for detecting chain-like enhancement paths includes: continuously calculating new parametric coupling strengths, comparing the new parametric coupling strengths with the weights of the corresponding edges in the initial weighted causal graph, and calculating the weight increase; if the weight increase is greater than a dynamic threshold, then marking the edge as an enhancement edge; in the initial weighted causal graph containing enhancement edge markings, using a depth-first search algorithm to identify paths consisting of at least three consecutive enhancement edges as chain-like enhancement paths.

2. The method for measuring multiple physical quantities using a passive multi-parameter fiber optic sensing unit according to claim 1, characterized in that, The method for locating the initial abnormal node includes: The weights of the initial weighted causal graph containing the reinforcing edge labels are updated to new parametric coupling strengths, while the labels of the chain-reinforcing paths are preserved, resulting in a dynamically updated weighted causal graph. Calculate the in-degree and weighted in-degree of each node in the labeled chain-reinforcement path in the dynamically updated weighted causal graph, locate and output the initial anomalous node.

3. The method for measuring multiple physical quantities of a passive multi-parameter fiber optic sensing unit according to claim 2, characterized in that, The method for locating the initial abnormal node also includes: Extract all nodes on the marked chain-like augmented path to form a path node set; In the dynamically updated weighted causal graph, calculate the in-degree of each node in the set of path nodes, and check if there is a node with an in-degree of 0. If there is a node with an in-degree of 0, then the node with an in-degree of 0 is directly identified as the initial abnormal node; if there is no node, then the weighted in-degree of each node in the path node set is calculated, and the node with the smallest weighted in-degree is selected as the initial abnormal node.

4. A multi-physical quantity measurement device for a passive multi-parameter fiber optic sensing unit, used to implement the multi-physical quantity measurement method of the passive multi-parameter fiber optic sensing unit according to any one of claims 1-3, characterized in that, The device includes: The parameter correlation calculation module is used to collect and preprocess the raw optical signals of multiple physical parameters of the monitored object, calculate the temporal cross-correlation function between each physical parameter in real time, and quantify the parameter coupling strength and causal propagation delay that characterize the correlation of multiple physical parameters based on the temporal cross-correlation function between each physical parameter; the multiple physical parameters include four physical parameters: temperature, humidity, air pressure and tilt angle. Fault prediction module: It is used to construct an initial weighted causal graph based on parametric coupling strength and causal propagation delay, detect and mark chain reinforcement paths in the initial weighted causal graph, locate the initial abnormal nodes of the chain reinforcement paths, and predict the probability of subsequent fault nodes of all nodes other than the initial abnormal nodes. Visualization module: Used to draw a dynamic arrow chain in the 3D digital twin model of the monitored object, starting from the initial abnormal node and following the chain-like enhancement path, and to render the dynamic arrow chain according to the probability of subsequent fault nodes.

Images