A fiber-optic gyroscope fault monitoring method and system

Abstract

The application relates to the technical field of optical fiber sensing and fault diagnosis, in particular to a fiber-optic gyroscope fault monitoring method and system; the method comprises the following steps: extracting four-dimensional features of phase difference, light intensity fluctuation, polarization state drift and temperature drift from a fiber-optic gyroscope output signal, and constructing a high-dimensional feature matrix after unifying a time reference; a dynamic prediction model is established based on historical data to generate a residual matrix, the residual matrix is orthogonally separated into an internal structure abnormal signal and an environmental disturbance signal through covariance feature decomposition; a structure residual sequence is mapped into a topology graph, an evolution index is calculated in real time, and a fault gradual change trend is captured by combining sliding window gradient analysis; a multi-dimensional vector is constructed by fusing topology features, an adaptive classifier of an online Gaussian mixture model is used to identify a fault mode and quantify a health index, and the health measurement result is fed back to the prediction model and topology analysis parameters to realize closed-loop optimization. The application realizes full-process adaptive monitoring from abnormal detection to health measurement.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensing and fault diagnosis technology, specifically to a method and system for monitoring faults in a fiber optic gyroscope. Background Technology

[0002] As the core sensing unit of a high-precision inertial navigation system, the long-term operational reliability of fiber optic gyroscopes is directly related to the navigation accuracy and stability under complex working conditions. Suppression of environmental interference under multi-physics coupling, effective extraction of early minor fault characteristics, and dynamic quantification of the health status throughout the entire life cycle have become key requirements for ensuring system reliability.

[0003] Chinese invention patent CN112595340B discloses an online fault detection method for fiber optic gyroscopes. Since light intensity information is more sensitive to damage to fiber optic fibers, fiber optic fiber breaks can be more easily detected through light intensity information. Therefore, this invention demodulates an overmodulated signal (#imgabs0#) applied to the Y-waveguide, and outputs the demodulated signal after summing and smoothing, thus obtaining light intensity information without increasing hardware costs. Real-time light intensity detection is then used to determine if there are any hidden dangers such as scratches on the fiber optic fibers during the production process, and whether the fiber optic fiber splicing is normal. Simultaneously, during the overall temperature cycling process, changes in light intensity can be monitored in real time. By detecting even slight changes in light intensity, it is possible to determine if there are any quality issues with the fiber optic gyroscope. In other words, this invention can determine the quality status of each gyroscope in real time by detecting the light intensity signal of each gyroscope online, providing reference information for expert systems to select the appropriate gyroscope.

[0004] However, existing monitoring schemes rely primarily on phase difference analysis, while also paying attention to auxiliary indicators such as light intensity fluctuations. While these schemes are valuable in improving system maintainability, as application scenarios expand to complex electromagnetic environments, wide temperature range conditions, and long-term unattended scenarios, higher demands are placed on the ability to refine the analysis of the coupling effects of multiple physical quantities, the ability to detect minor gradual faults in their early stages, and the ability to continuously optimize diagnostic models through long-term online self-adaptation. There is an urgent need to develop a comprehensive monitoring mechanism that integrates multi-dimensional feature collaborative analysis, dynamic signal separation, and topology evolution modeling to support the construction of an intelligent operation and maintenance system for next-generation fiber optic gyroscopes. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a method and system for monitoring faults in fiber optic gyroscopes.

[0006] The technical solution of this invention: a method for fault monitoring of a fiber optic gyroscope, comprising the following specific implementation steps:

[0007] S1. Extract four-dimensional features from the output signal of the fiber optic gyroscope, including: phase difference features, light intensity fluctuation features, polarization state drift features and temperature drift features. Construct a high-dimensional feature matrix after unifying the four-dimensional features with a time reference.

[0008] S2. Based on historical normal operation data, a dynamic prediction model is established for the high-dimensional feature matrix to obtain the prediction feature matrix. The residual matrix between the actual observation matrix and the prediction feature matrix is ​​calculated and normalized. Then, the normalized residual matrix is ​​orthogonally separated into a structural residual matrix reflecting the abnormal signal of the internal device and an environmental residual matrix reflecting the disturbance signal of the external environment through covariance eigendecomposition.

[0009] S3. Map the structural residual matrix to a topological graph. The nodes of the topological graph correspond to four-dimensional features, and the edge weights reflect the residual correlation between features. Extract the node degree, clustering coefficient, and global topological change of the topological graph to form a topological evolution feature vector, and perform sliding window gradient analysis on the vector to capture the gradual trend of fault change.

[0010] S4. A multi-dimensional feature vector is constructed by fusing the topological evolution feature vector and its gradient. An adaptive classifier using an online Gaussian mixture model is used to identify the fault mode and calculate the health index. The health measurement results are fed back to the dynamic prediction model and topological analysis parameters for closed-loop optimization.

[0011] Preferably, the construction of the high-dimensional feature matrix in step S1 specifically includes:

[0012] Define phase difference : ;

[0013] Define the characteristic of light intensity fluctuation I(t): ;

[0014] Define the polarization state drift characteristic P(t): ;

[0015] Define the temperature drift characteristic T(t): ;

[0016] Where L represents the fiber loop length; λ represents the wavelength of the light source; c represents the speed of light; Ω(t) represents the gyroscope rotational angular velocity; I0 represents the initial intensity of the light source; K represents the interference visibility factor; S1(t), S2(t), and S3(t) represent the Stokes parameters; T sensor (t) represents the real-time temperature of the sensor; T ref (t) represents the design reference temperature;

[0017] Align the above four-dimensional features over time to construct a high-dimensional feature matrix M(t):

[0018] ;

[0019] Where n is the total number of time sampling points.

[0020] Preferably, the orthogonal separation of the normalized residual matrix in step S2 specifically includes:

[0021] The normalized residual matrix is ​​then centered using column mean.

[0022] Calculate the covariance matrix of the centered residual matrix and perform eigenvalue decomposition, then select the principal eigenvectors to construct the orthogonal projection matrix;

[0023] By projecting the centered residual onto an orthogonal subspace composed of principal eigenvectors, structural residuals reflecting internal device anomalies are obtained.

[0024] The centered residual is projected onto the complement of the orthogonal subspace to obtain the environmental residual that reflects the external disturbance.

[0025] Preferably, in step S3, mapping the structural residual matrix to a topological graph specifically involves:

[0026] Define a topology graph, where nodes correspond to four feature dimensions: phase difference, light intensity fluctuation, polarization state drift, and temperature drift.

[0027] The edge weights are obtained by normalizing the covariance of the structural residual sequences corresponding to the two feature dimensions, and are used to represent the residual correlation between nodes.

[0028] Preferably, the topology evolution feature vector in step S3 includes node degree, node clustering coefficient, and global topology change.

[0029] Node degree is the number of edges whose weights exceed a set threshold in the edges connecting each node.

[0030] The node clustering coefficient is the ratio of the actual number of triangles formed between a node and its neighboring nodes to the maximum number of triangles that can be formed.

[0031] The global topological change is the sum of the squares of the differences between the weights of all edges in the topological graph at the current time and the corresponding edge weights at the previous time.

[0032] Preferably, in step S3, performing sliding window gradient analysis on the vector to capture the gradual trend of fault change specifically involves:

[0033] The node degree, node clustering coefficient, and global topology change are combined to form a topology feature vector;

[0034] Calculate the gradient of the topological feature vector within the sliding time window;

[0035] Determine whether the gradient exceeds a set threshold, and use this as an abnormal indication of the fault gradual change trend.

[0036] Preferably, the adaptive classifier using an online Gaussian mixture model in step S4 is specifically as follows:

[0037] The classifier contains several Gaussian components, and the probability distribution of each Gaussian component is defined by the mean vector and the covariance matrix, which respectively represent a fault mode.

[0038] For the input multidimensional feature vector, calculate the probability of it belonging to each Gaussian component, and take the fault mode represented by the component with the highest probability as the recognition result.

[0039] The classifier uses an online learning mechanism to recursively update the mean vector, covariance matrix, and prior weights of each Gaussian component as new data arrives.

[0040] The preferred health index is:

[0041] ;

[0042] Where m represents the total number of failure modes; H(t) represents the health index at time t; Indicates fault category C i Severity weights; P(C i |X(t)) represents the conditional probability, that is, given the input feature vector X(t), the fiber optic gyroscope belongs to the i-th fault mode C. i The probability of.

[0043] Preferably, the closed-loop optimization in step S4 specifically includes:

[0044] When the health index falls below the threshold and continues to change, the parameter update mechanism is triggered.

[0045] The update mechanism includes updating the weights and historical window length of the dynamic prediction model, adjusting the edge weight threshold and gradient judgment threshold in the topology graph analysis, and updating the mean vector, covariance matrix, and prior weights of the Gaussian components in the online Gaussian mixture model classifier.

[0046] The technical solution of the present invention: a fiber optic gyroscope fault monitoring system, used in the aforementioned fiber optic gyroscope fault monitoring method, comprising:

[0047] Memory;

[0048] processor;

[0049] A computer program stored in the memory and capable of running on the processor;

[0050] The processor executes the aforementioned fiber optic gyroscope fault monitoring method.

[0051] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects:

[0052] This invention designs a fault monitoring method and system for fiber optic gyroscopes. It constructs a comprehensive monitoring system by fusing multi-dimensional features such as phase difference, light intensity fluctuation, polarization drift, and temperature drift. The system utilizes orthogonal decomposition technology of dynamic prediction residuals to accurately separate inherent abnormal signals of the device from environmental disturbances. The structural residual sequence is mapped into a dynamic topology graph. By analyzing the gradient characteristics of node degree, clustering coefficient, and global topology change, it effectively captures minute, gradual fault defects that are difficult to identify using traditional methods. Furthermore, based on an adaptive classification mechanism using an online Gaussian mixture model, it achieves dynamic learning of fault mode probabilities and recursive parameter updates. Finally, it generates a quantitative health index and drives a closed-loop feedback mechanism to optimize the prediction model weights, topology analysis thresholds, and classifier parameters in real time. This significantly improves the anti-interference capability and long-term monitoring stability of fiber optic gyroscopes in high-noise environments, effectively addressing common industry challenges such as high false alarm rates caused by environmental coupling, delayed identification of gradual faults, and system parameter drift. It forms a closed-loop adaptive optimization process from anomaly detection to health assessment. Attached Figure Description

[0053] Figure 1 This is a flowchart of a fiber optic gyroscope fault monitoring method proposed in this invention. Detailed Implementation

[0054] Example 1, as Figure 1 As shown, the present invention proposes a method for fault monitoring of a fiber optic gyroscope, which includes the following specific implementation steps:

[0055] S1. Extract multi-dimensional features from the output signal of the fiber optic gyroscope and map different types of signals to the same time reference to form a high-dimensional feature matrix, providing a stable data foundation for subsequent fault detection. The specific implementation process is as follows:

[0056] S11. Introduce multi-source feature acquisition, including:

[0057] (1) Phase difference : Obtained through optical interferometry, it represents the change in phase difference of light waves in the fiber optic loop and directly reflects the rotational angular velocity. ;

[0058] Where L represents the fiber optic loop length, which is determined by the gyroscope design; λ represents the wavelength of the light source; c represents the speed of light; and Ω(t) represents the angular velocity of the gyroscope.

[0059] (2) Light intensity fluctuation characteristic I(t): Measured by a photodetector, it reflects the amplitude change of the interference fringes and is used to detect light source power attenuation or loop coupling anomalies. ;

[0060] Where I0 represents the initial intensity of the light source; K represents the interference visibility factor, i.e. the clarity of the interference fringes;

[0061] (3) Polarization state drift characteristic P(t): reflects the change in polarization state within the fiber loop and is used to identify polarization coupling anomalies. ;

[0062] Where S1(t), S2(t), and S3(t) represent Stokes parameters, which characterize the projection of the polarization state onto spherical coordinates;

[0063] (4) Temperature drift characteristic T(t): reflects the influence of ambient temperature on the performance of the gyroscope.

[0064] ;

[0065] Among them, T sensor (t) represents the real-time temperature of the sensor; T ref (t) represents the design reference temperature;

[0066] S12. After unifying the time alignment of the multi-source features, construct a high-dimensional feature matrix M(t):

[0067] ;

[0068] Where n is the total number of time sampling points; each row represents the multidimensional features at the same time point; and each column represents the time series of the same dimension.

[0069] S2. Based on the high-dimensional feature matrix M(t) generated in step S1, a dynamic prediction model is constructed and the residuals are calculated to accurately separate the abnormal signals of real devices from environmental disturbances, providing high-quality input for topology evolution analysis and health measurement. The specific implementation process is as follows:

[0070] S21. Based on the high-dimensional feature matrix M(t) generated in step S1, and using historical normal operation data, establish a multi-step prediction model. Predicting future characteristics:

[0071] ;

[0072] Where p is the length of the historical time window; This represents the predicted feature matrix, i.e., the expected normal state.

[0073] It should be noted that the multi-step prediction model The model consists of an input layer, a feature encoding layer, a temporal memory layer, and an output layer. The input layer receives a multi-dimensional feature matrix, including phase difference, light intensity fluctuation, polarization state drift, and temperature drift information. The feature encoding layer performs nonlinear mapping and fusion of features in each dimension through a fully connected network to extract the intrinsic correlation of multi-source features. The temporal memory layer adopts a gated recurrent unit structure, which can capture the dynamic dependence and changing trend of features within a historical time window, and realize the memory and update of multi-step time series prediction. The output layer maps the state of the temporal memory layer to the predicted multi-dimensional feature values, realizing the joint prediction of multiple future time points. The entire model has a compact and reasonable structure, and can adaptively adjust parameters to adapt to the operating characteristics of different fiber optic gyroscopes, while ensuring the stability and high sensitivity of the prediction.

[0074] S22. Perform a difference operation between the actual observation matrix and the prediction matrix: ;

[0075] The residual matrix R(t) is then normalized. ;

[0076] Where R(t) represents the residual matrix, and each column represents the residual time series of the corresponding feature dimension; σ M This represents the historical standard deviation of each feature dimension; ε represents a small positive number to avoid division by zero. This represents the residual matrix after normalization.

[0077] S23. Calculate the normalized residual matrix. It is decomposed into internal structural anomaly signals and external environmental disturbance signals, specifically:

[0078] A1. Remove the mean from each feature column so that the data is distributed around the zero mean: ;

[0079] in, This represents the matrix after the residual matrix has been centered by column mean. The normalized residual matrix represents the time mean of each feature column. Calculate the average value;

[0080] A2. By calculating the residual covariance matrix and performing eigenvalue decomposition, the principal eigenvectors are selected to construct an orthogonal projection matrix, forming an orthogonal subspace Q used to capture internal device anomalies, specifically:

[0081] Calculate the covariance matrix C (representing the residual covariance among features): ;

[0082] Perform eigenvalue decomposition: ;

[0083] Where V represents the eigenvector matrix, with each column representing the principal direction of the covariance; Λ represents the eigenvalue diagonal matrix, i.e., the magnitude of the variance;

[0084] Select the principal eigenvectors to form the orthogonal projection matrix Q: ;

[0085] Among them, V s This represents a submatrix composed of eigenvectors corresponding to the k largest eigenvalues; k is adaptively selected based on the proportion of internal faults to the feature variance.

[0086] A3. The centered residual is projected onto an orthogonal subspace to obtain the structural residual, and the environmental residual is obtained by filling in the space. A sliding window smoothing process is then applied to both types of residuals to reduce the impact of transient noise. Specifically:

[0087] Calculate structural residuals (internal anomaly components): ;

[0088] Calculate the environmental residuals (external disturbance components): ;

[0089] Among them, R s (t) represents the structural residual matrix, reflecting abnormal signals of internal devices; R e (t) represents the environmental residual matrix, reflecting the external disturbance signal; E represents the identity matrix.

[0090] S3. Map the structural residual sequence to a topological graph and analyze the evolution characteristics of the topology over time to capture the gradual trend of minute internal faults. The specific implementation process is as follows:

[0091] S31. Based on the structural residual matrix R output in step S2 s (t), define the topological graph G(t) = (V, E(t));

[0092] Node V={v1,v2,v3,v4} corresponds to each feature dimension (phase, light intensity, polarization state, temperature drift);

[0093] The edge set E(t) represents the residual correlation between nodes, and the edge weight w ij (t) reflects the correlation of residuals between nodes: ;

[0094] Where v1, v2, v3, and v4 represent the phase difference node, light intensity node, polarization state node, and temperature drift node, respectively; Let represent the covariance between the i-th and j-th feature residual sequences; and These represent the standard deviations of the corresponding residual sequences, used for normalization to avoid the influence of dimensional differences on the edge weights; To represent small positive numbers and prevent division by zero errors; Let R represent the structural residual sequence of the i-th feature. s The i-th column of (t);

[0095] S32. Extract key indicators of topology evolution over time to reflect the gradual characteristics of internal faults, specifically:

[0096] Analyze the node degree k i (t): ;

[0097] Where d is the total number of nodes, which is 4 in this embodiment; θ represents the edge weight threshold, which is derived from historical normal data statistics and used to determine strong correlation; This indicates an indicator function that takes the value 1 if the condition is met, and 0 otherwise.

[0098] Analyze the node clustering coefficient C i (t): ;

[0099] Among them, T i (t) represents node v i The number of triangles is the number of closed loops formed when a node is connected to two of its neighbors simultaneously.

[0100] Calculate global topology changes : ;

[0101] Among them, w ij (t-1) represents the edge weight at the previous time step;

[0102] S33, Topological feature vectors Perform sliding window gradient analysis:

[0103] ;

[0104] Perform anomaly detection: ;

[0105] Where F(t) represents the topological feature vector, which consists of node degree, clustering coefficient and global change; δt represents the sliding window length, used to calculate the local gradient; γ represents the gradient of the topological feature, that is, the rate at which the topological feature changes over time; γ represents a set threshold used to determine whether the gradient is abnormal. This represents an exception indicator variable, where 1 indicates an exception and 0 indicates normal operation.

[0106] S4. The topology evolution trend information output in step S3 is converted into specific fault modes, and the overall health status of the fiber optic gyroscope is quantitatively scored. Through an adaptive classification mechanism and health measurement model, an early fault warning and long-term monitoring closed loop is achieved. The specific implementation process is as follows:

[0107] S41. Using the node degree, clustering coefficient, global topology change, and time gradient parameters obtained in step S3, integrate them into a multi-dimensional feature vector, that is, construct an input vector X(t) that can comprehensively reflect the system's operating characteristics:

[0108] ;

[0109] Where X(t) represents the multidimensional feature vector constructed at time t, which is used to characterize the comprehensive information of the current state of the fiber optic gyroscope;

[0110] S42. An adaptive classification mechanism is introduced, which can adjust the classification boundary in real time according to the dynamic changes of the operating data, thereby adapting to the nonlinear drift and complex fault characteristics exhibited by the fiber optic gyroscope during long-term operation. Specifically:

[0111] An adaptive classifier based on an online Gaussian mixture model (Online GMM) is used:

[0112] ;

[0113] ;

[0114] Where K1 represents the number of Gaussian components in the current classifier, i.e., the number of known or adaptively learned fault modes; C i This represents the i-th fault mode category, including but not limited to optical path breakage, polarization perturbation, light source attenuation, and detector noise degradation; π i μ represents the prior weight of the i-th category, reflecting the relative frequency or importance of this failure mode during long-term operation; i Σ represents the mean vector of the i-th fault mode, i.e., the center position of this category in the feature space. It is obtained from historical training data or expert knowledge at initialization and is continuously updated during subsequent operation; i N(X(t)|μ) represents the covariance matrix of the i-th failure mode, i.e., the dispersion and correlation of the mode across different feature dimensions; i ,Σ i ) represents the Gaussian distribution function, which measures the similarity of the input feature vector X(t) to the i-th pattern; d is the dimension of the feature vector; |Σ i | is the determinant of the covariance matrix; P(C i |X(t)) represents the conditional probability, that is, given the input feature vector X(t), the fiber optic gyroscope belongs to the i-th fault mode C. i The probability of;

[0115] An online learning mechanism is introduced to recursively update the parameters each time new data arrives:

[0116] ;

[0117] ;

[0118] in, This represents the mean vector of the i-th type of pattern at time t, which is the system's estimate of the center position of the pattern at this time. This represents the mean vector updated at time t+1, i.e., the mode center corrected by the current observation value X(t); This represents the learning rate (update rate), with a value range of (0,1), and is used to control the degree of influence of new data on parameter updates; Let represent the covariance matrix of the i-th type of mode at time t, which is an estimate of the distribution range of the mode. This represents the updated covariance matrix, which incorporates the influence of the current observation data, making the model more closely reflect the new operating state.

[0119] It should be noted that the number of Gaussian components K1 in the current classifier is not a fixed constant, but is dynamically adjusted with the data flow through online incremental learning: when the probability of a certain type of residual pattern is consistently low, the system will reduce its weight π. i Until it is eliminated; when a completely new feature distribution appears, the system will introduce a new Gaussian component to expand the summation range;

[0120] S43. Quantify the health status of the fiber optic gyroscope based on the failure mode probability and topology evolution intensity, convert the results into a quantitative health status evaluation, and calculate the health index H(t):

[0121] ;

[0122] Where m represents the total number of failure modes; H(t) represents the health index at time t; Indicates fault category C i The severity weights are used to measure the degree of impact of different failure modes on the overall health status;

[0123] S44. The health measurement results are fed back with historical data to optimize the prediction model and topology analysis parameters in steps S2 and S3, namely:

[0124] When H(t) <H threshold Furthermore, when the parameters continue to change, a model parameter update mechanism is triggered:

[0125] Update prediction model The weights and historical window p;

[0126] Adjust the topology graph threshold θ and the gradient threshold (anomaly decoupling parameter) γ;

[0127] Update the classification model mean μ i Covariance Σ i and category weight π i ;

[0128] Accordingly, this mechanism allows monitoring results to be fed back into preceding steps, achieving a closed-loop adaptive process from anomaly detection to pattern recognition and then to health measurement.

[0129] Example 2: The present invention proposes a fiber optic gyroscope fault monitoring system, which is used to execute a fiber optic gyroscope fault monitoring method proposed in Example 1, comprising:

[0130] Memory;

[0131] processor;

[0132] A computer program stored in the memory and capable of running on the processor;

[0133] The processor executes a computer program to implement a fiber optic gyroscope fault monitoring method as described in Embodiment 1 above.

[0134] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method of fiber optic gyroscope fault monitoring, the method comprising: The method comprises the following specific implementation steps: S1, extracting four-dimensional features from the fiber-optic gyroscope output signal, including phase difference features, light intensity fluctuation features, polarization state drift features, and temperature drift features, and constructing a high-dimensional feature matrix after unifying the time reference of the four-dimensional features; The step S1 of constructing the high-dimensional feature matrix specifically comprises: Defining phase difference : ; Defining the light intensity fluctuation characteristic I(t): ; Defining the polarization state drift characteristic P(t): ; Define the temperature drift characteristic T(t) as: ; where L represents the fiber loop length; λ represents the light source wavelength; c represents the speed of light; Ω(t) represents the gyroscope rotational angular velocity; I0represents the light source initial intensity; K represents the interference visibility factor; S1(t), S2(t), and S3(t) represent the Stokes parameters; T sensor (t) represents the sensor real-time temperature; T ref (t) represents the design reference temperature; aligning the above four-dimensional features in time to construct a high-dimensional feature matrix M(t): ； wherein n is the total number of time sampling points; S2, establishing a dynamic prediction model based on historical normal operation data to obtain a predicted feature matrix, calculating a residual matrix of an actual observation matrix and the predicted feature matrix and performing normalization processing, and then orthogonally separating the normalized residual matrix into a structural residual matrix reflecting internal device abnormal signals and an environmental residual matrix reflecting external environmental disturbance signals through covariance feature decomposition; S3, mapping the structural residual matrix into a topology graph, wherein the nodes of the topology graph correspond to the four-dimensional features, and the edge weight values reflect the residual correlation between the features, extracting the node degree, clustering coefficient, and global topology change amount of the topology graph to form a topology evolution feature vector, and performing sliding window gradient analysis on the vector to capture the fault gradual change trend; S4, fusing the topology evolution feature vector and its gradient to construct a multi-dimensional feature vector, using an adaptive classifier of an online Gaussian mixture model to identify the fault mode and calculate the health index, and feeding back the health measurement result to the dynamic prediction model and the topology analysis parameters for closed-loop optimization.

2. The fiber-optic gyroscope fault monitoring method of claim 1, wherein, The step S2 of orthogonally separating the normalized residual matrix specifically comprises: performing column mean centering processing on the normalized residual matrix; calculating the covariance matrix of the centered residual matrix and performing feature decomposition, and selecting the principal eigenvectors to construct an orthogonal projection matrix; projecting the centered residual to the orthogonal subspace composed of the principal eigenvectors to obtain the structural residual reflecting the internal device abnormalities; projecting the centered residual to the complement space of the orthogonal subspace to obtain the environmental residual reflecting the external disturbance.

3. A fiber optic gyroscope fault monitoring method according to claim 2, wherein, The step S3 of mapping the structural residual matrix into a topology graph specifically comprises: defining the topology graph, wherein the nodes correspond to the four feature dimensions of phase difference, light intensity fluctuation, polarization state drift, and temperature drift; the edge weight values are obtained by normalizing the covariance of the structural residual sequences corresponding to two feature dimensions, and are used to represent the residual correlation between the nodes.

4. A fiber optic gyroscope fault monitoring method according to claim 3, wherein, The topology evolution feature vector in step S3 comprises node degree, node clustering coefficient, and global topology change amount; the node degree is the number of edges whose edge weight exceeds a certain threshold in the edges connected to each node; the node clustering coefficient is the ratio of the number of triangles actually formed between the node and its neighbor nodes to the maximum number of triangles that can be formed; the global topology change amount is the sum of the squares of the differences between all edge weights of the current topology graph and the corresponding edge weights at the previous moment.

5. A fiber optic gyroscope fault monitoring method according to claim 4, wherein, The step S3 of performing sliding window gradient analysis on the vector to capture the fault gradual change trend specifically comprises: combining the node degree, node clustering coefficient, and global topology change amount to form a topology feature vector; calculating the gradient of the topology feature vector within a sliding time window; judging whether the gradient exceeds a certain threshold, which is used as an abnormal indication of the fault gradual change trend.

6. A fiber optic gyroscope fault monitoring method according to claim 5, wherein, The step S4 of using an adaptive classifier of an online Gaussian mixture model specifically comprises: The classifier contains several Gaussian components, and the probability distribution of each Gaussian component is defined by a mean vector and a covariance matrix, representing a failure mode respectively; For the input multi-dimensional feature vector, the probability of belonging to each Gaussian component is calculated, and the failure mode represented by the component with the maximum probability is taken as the recognition result; The classifier updates the mean vector, covariance matrix and prior weight of each Gaussian component recursively through an online learning mechanism as new data arrives.

7. A fiber optic gyroscope fault monitoring method according to claim 6, wherein, The health index is: ； where m represents the total number of failure modes; H(t) represents the health index at time t; represents the severity weight of the failure category C i ; P(C i | X(t)) represents the conditional probability, i.e. the probability that the fiber-optic gyroscope belongs to the i-th failure mode C i under the condition that the input feature vector is X(t).

8. A fiber optic gyroscope fault monitoring method according to claim 7, wherein, The closed-loop optimization in step S4 specifically includes: When the health index is below the threshold and continuously changes, a parameter updating mechanism is triggered; The updating mechanism includes updating the weight and history window length of the dynamic prediction model, adjusting the edge weight threshold and gradient judgment threshold in the topology graph analysis, and updating the mean vector, covariance matrix and prior weight of the Gaussian component in the online Gaussian mixture model classifier.

9. A fiber-optic gyroscope failure monitoring system for performing the fiber-optic gyroscope failure monitoring method of any one of claims 1-8, wherein, Comprise: a memory; a processor; a computer program stored in the memory and executable on the processor; The processor executes a fiber-optic gyroscope fault monitoring method according to any one of claims 1-8.

Images