Coal mining equipment state monitoring and diagnosing system based on digital twinning
By constructing a manifold topology and Riemannian geometric analysis, the problem of monitoring nonlinear coupling faults in coal mining equipment was solved, achieving highly sensitive fault diagnosis and real-time location, and improving the accuracy and reliability of equipment condition monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HENAN ZHENGLONG COAL IND CO LTD
- Filing Date
- 2025-12-12
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies are unable to effectively analyze the nonlinear coupling relationships in multi-source heterogeneous sensor data of coal mining equipment, resulting in fault characteristics being masked by high-dimensional noise, insufficient monitoring sensitivity, delayed early anomaly identification, and difficulty in accurately locating the root cause of the fault.
By employing manifold spatial mapping based on digital twins and Riemannian geometric analysis, a manifold topology is constructed. Through Riemannian metric tensor field and parallel movement prediction analysis, combined with the k-nearest neighbor algorithm and Euclidean distance norm, a deep perception of the health status of coal mining equipment and fault location are achieved.
It significantly improves the sensitivity of equipment condition monitoring and the immediacy of fault diagnosis, can accurately pinpoint the root cause of the fault, prevent early fault characteristics from being masked by high-dimensional noise, and achieve high signal-to-noise ratio anomaly detection and closed-loop diagnosis.
Smart Images
Figure CN122064908A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of coal mine machinery and equipment condition monitoring and fault diagnosis technology, specifically a coal mining equipment condition monitoring and diagnosis system based on digital twins. Background Technology
[0002] In modern coal mining operations, key equipment such as coal mining machines integrate multi-source heterogeneous sensors for vibration, current, attitude, and hydraulics. Their operation generates massive amounts of monitoring data with high-dimensional characteristics, and there are often complex nonlinear dynamic coupling relationships between various physical parameters. For the health status analysis of such equipment, existing technical solutions are generally based on traditional Euclidean geometric space theory, mainly relying on linear statistical analysis or simple threshold comparison for fault diagnosis. Since such methods are difficult to analyze the intrinsic geometric structure and manifold topological features within the data, they cannot effectively capture the subtle changes in the constraint relationship between variables when dealing with nonlinear coupled faults. This makes fault features easily masked by high-dimensional noise, resulting in insufficient monitoring sensitivity, delayed early anomaly identification, and difficulty in accurately locating the root cause of the fault. Therefore, how to break through the limitations of traditional Euclidean space analysis methods and deeply perceive the coupling anomalies between multiple variables from the geometric topology level, thereby improving the accuracy of equipment condition monitoring and the timeliness of fault diagnosis, has become an urgent technical problem to be solved. Summary of the Invention
[0003] To address the aforementioned technical problems, this invention provides a coal mining equipment condition monitoring and diagnostic system based on digital twins. Specifically, the technical solution of this invention includes: Equipment condition monitoring center, manifold space mapping unit, condition drift monitoring unit, curvature fault location unit, and closed-loop control decision unit; The equipment status monitoring center is used to retrieve sensor operating data of coal mining equipment and send the sensor operating data to the manifold space mapping unit; The manifold space mapping unit is used to perform manifold topology construction analysis on sensor operation data to obtain the original high-dimensional state vector, and to perform Riemannian metric tensor field construction analysis on the original high-dimensional state vector to obtain metric tensor components. The state drift monitoring unit is used to perform parallel movement prediction analysis based on the metric tensor components to obtain the health state prediction component. The health state prediction component is compared with the sensor operation data to obtain the state drift amount, and the state drift amount is compared with the preset confidence interval threshold. If the state drift is less than or equal to the preset information interval threshold, a normal operation signal is generated, and the closed-loop control decision unit maintains the current operation strategy. If the state drift exceeds the preset threshold, a drift anomaly signal is generated. In response to the drift anomaly signal, the curvature fault location unit is used to perform Riemann curvature singularity analysis based on the metric tensor components to obtain the cross-sectional curvature, and to perform fault contribution decoupling analysis on the cross-sectional curvature to obtain the fault contribution index. The closed-loop control decision unit is used to identify and process the fault contribution index and generate control commands.
[0004] Preferably, the manifold topology construction analysis process is as follows: The vibration data of the cutting section, the current data of the traction section, the tilt angle data of the rocker arm, and the pressure data of the hydraulic system of the coal mining equipment during the operation period are obtained, and the vibration data of the cutting section, the current data of the traction section, the tilt angle data of the rocker arm, and the pressure data of the hydraulic system are set as the original high-dimensional state vector; The Euclidean distance norm of each state point in the local tangent space is obtained in the original high-dimensional state vector, and the local adaptive kernel width of each state point is obtained through the k-nearest neighbor algorithm. An exponential decay operation is performed on the Euclidean distance norm and the local adaptive kernel width to obtain the adjacency weight matrix, which is used to characterize the geometric affinity between two state points on the manifold surface.
[0005] Preferably, the Riemannian metric tensor field construction and analysis process is as follows: Based on the adjacency weight matrix and the neighborhood point set of the original high-dimensional state vector, a local weighted covariance matrix centered on the data point is constructed. The precision matrix is obtained by inverting the local weighted covariance matrix. The precision matrix is set as a metric tensor component, which is used to characterize the allowable degrees of freedom of the physical quantities while keeping the manifold structure stable.
[0006] Preferably, the parallel movement prediction analysis process is as follows: The connection coefficient is calculated using the metric tensor components, and the connection coefficient is used to characterize the bending properties of the manifold. Obtain the health status component and the state displacement of the previous time step; Based on the connection coefficient and the state displacement of the previous time step, the health state component of the previous time step is approximated by a first-order calculation to obtain the health state prediction component of the current time step.
[0007] Preferably, the process for obtaining the state drift amount is as follows: The sensor operation data at the current moment is obtained, and the difference between the sensor operation data and the health status prediction component is calculated to obtain the state deviation vector; Obtain the metric tensor components; The state deviation vector is calculated by weighted square Euclidean distance based on the metric tensor components to obtain the geodesic deviation on the Riemann manifold, and the geodesic deviation is set as the state drift.
[0008] Preferably, the Riemann curvature singularity analysis process is as follows: Based on the metric tensor components and their second derivatives, a Riemann curvature tensor is constructed. Obtain the orthogonal basis vectors of the local tangent space; By performing tensor reduction and fractional operations on the Riemann curvature tensor and orthogonal basis vectors, the cross-sectional curvature on the plane spanned by the orthogonal basis vectors is obtained. The cross-sectional curvature is used to quantify the degree of geometric distortion of the manifold space.
[0009] Preferably, the fault contribution decoupling analysis process is as follows: Eigendecomposition of the Riemann curvature tensor yields the principal eigenvectors that result in the maximum cross-sectional curvature. These principal eigenvectors indicate the direction of the most severe tearing of the manifold space. Obtain the unit basis vectors representing each physical sensor in the original data space; Calculate the projection magnitude of the principal eigenvector onto the unit basis vectors of each physical sensor; Calculate the product of the projection modulus and the cross-sectional curvature, and set the product as the fault contribution index of the corresponding physical sensor.
[0010] Preferably, the operation process of the closed-loop control decision unit is as follows: Obtain the fault contribution index of each physical sensor; By comparing and analyzing the values of each fault contribution index, the largest fault contribution index with the highest value is selected. Identify the physical sensor corresponding to the maximum failure contribution index and determine that the mechanical component corresponding to the physical sensor has suffered structural damage; Trigger a preset control strategy for mechanical components that have suffered structural damage.
[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This system deeply senses the health status of coal mining equipment by constructing a manifold geometric space, mapping multi-source heterogeneous discrete physical sampling points into a continuous and smooth differential manifold, and constructing a Riemannian metric tensor field. This mechanism effectively solves the technical problem that traditional Euclidean space monitoring methods cannot handle high-dimensional nonlinear coupling faults. It can analyze the intrinsic geometric structure and manifold topological features inside the data, thereby keenly capturing the subtle changes in the coupling relationship between variables at the geometric topology level, and significantly improving the sensitivity of equipment status monitoring. 2. This system derives the health state prediction component using the parallel translation principle in Riemannian geometry, and calculates the geodesic deviation on the Riemann manifold as the state drift based on the inverse matrix of the metric tensor. This method uses the metric tensor as weights, which significantly amplifies the contribution of small deviations in the direction of strong system constraints to the drift, while suppressing fluctuations in the direction of degrees of freedom. This mechanism achieves high signal-to-noise ratio anomaly detection, effectively prevents early fault features from being masked by high-dimensional noise, and solves the problem of lagging early anomaly identification. 3. In the process of constructing the manifold topology, this system uses the k-nearest neighbor algorithm to obtain the local adaptive kernel width of each state point, and combines it with the Euclidean distance norm to calculate the adjacency weight matrix. By introducing the adaptive kernel width of point pair dependence, the system can adapt to the uneven distribution of monitoring data, effectively solve the problem of unbalanced mapping between sparse and dense regions, ensure the accuracy of the manifold topology, and provide a reliable mathematical basis for subsequent geometric analysis. 4. This system performs Riemann curvature singularity analysis and fault contribution degree decoupling analysis through the curvature fault location unit, which can map the abstract manifold geometric distortion back to the specific physical sensor dimension. The system quantifies the degree of geometric distortion by calculating the curvature of the cross section to identify the break in the physical coupling relationship, and calculates the fault contribution index by projecting the principal eigenvector onto the unit basis vector, thereby accurately locating the dominant physical sensor and corresponding mechanical component that causes the manifold tearing, realizing a closed-loop diagnosis from hidden fault perception to physical root cause location. Attached Figure Description
[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation
[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.
[0014] Example 1: Please see Figure 1 A digital twin-based coal mining equipment condition monitoring and diagnosis system includes an equipment condition monitoring center, a manifold space mapping unit, a condition drift monitoring unit, a curvature fault location unit, and a closed-loop control decision unit. The equipment status monitoring center is used to retrieve sensor operating data of coal mining equipment and send the sensor operating data to the manifold space mapping unit; The manifold space mapping unit is used to perform manifold topology construction analysis on sensor operation data to obtain the original high-dimensional state vector, and to perform Riemannian metric tensor field construction analysis on the original high-dimensional state vector to obtain metric tensor components. The state drift monitoring unit is used to perform parallel movement prediction analysis based on the metric tensor components to obtain the health state prediction component. The health state prediction component is compared with the sensor operation data to obtain the state drift amount, and the state drift amount is compared with the preset confidence interval threshold. If the state drift is less than or equal to the preset information interval threshold, a normal operation signal is generated, and the closed-loop control decision unit maintains the current operation strategy. If the state drift exceeds the preset threshold, a drift anomaly signal is generated. In response to the drift anomaly signal, the curvature fault location unit is used to perform Riemann curvature singularity analysis based on the metric tensor components to obtain the cross-sectional curvature, and to perform fault contribution decoupling analysis on the cross-sectional curvature to obtain the fault contribution index. The closed-loop control decision unit is used to identify and process the fault contribution index and generate control commands.
[0015] This embodiment discloses a coal mining equipment condition monitoring and diagnosis system based on digital twins. The system constructs a manifold geometric space to deeply perceive the health status of the coal mining equipment, solving the technical problem that traditional Euclidean space monitoring is difficult to handle high-dimensional nonlinear coupling faults. The manifold geometric space constructed by this system is a digital twin of the physical coal mining equipment in the mathematical dimension. The physical health status of the physical equipment is mapped in real time through the geometric evolution trajectory of the twin. The manifold topology corresponds to the physical constraint mechanism of the equipment, and the Riemann metric corresponds to the coupling strength between various physical parameters. The system mainly consists of an equipment status monitoring center, a manifold space mapping unit, a status drift monitoring unit, a curvature fault location unit, and a closed-loop control decision unit. The equipment status monitoring center is responsible for retrieving the sensor operation data of the coal mining equipment in real time. In this embodiment, the monitoring center is connected to the vibration accelerometer, current transformer, and attitude sensor on the coal mining machine body through an industrial Ethernet to collect multi-source heterogeneous data, and performs time synchronization and cleaning processing on these data, and sends the processed sensor operation data to the manifold space mapping unit. After receiving the data, the manifold space mapping unit performs manifold topology construction analysis on the sensor operation data, thereby mapping the discrete physical sampling points into a continuous and smooth differential manifold to obtain the original high-dimensional state vector. The unit performs Riemannian metric tensor field construction analysis on the original high-dimensional state vector to calculate the metric tensor components that characterize the geometric constraint relationship inside the data. The state drift monitoring unit performs parallel movement prediction analysis based on the calculated metric tensor components. This analysis utilizes the parallel movement principle in Riemannian geometry to derive the theoretical health state prediction component. The unit compares the health state prediction component with the actual observed sensor operating data to calculate the state drift. This state drift is essentially the geodesic deviation on the Riemannian manifold. The system compares this state drift with a preset confidence interval threshold. This confidence interval threshold is the 99.7th percentile of the drift probability distribution obtained based on the historical health operating data of the equipment. If the state drift is less than or equal to the preset confidence interval threshold, the system determines that the equipment is operating within a healthy manifold subspace and generates a normal operation signal. At this time, the closed-loop control decision unit maintains the current operating strategy unchanged. Conversely, if the state drift is greater than the preset confidence interval threshold, it indicates that the equipment state has deviated from the healthy geometric constraints, and the system generates a drift anomaly signal. In response to drift anomaly signals, the curvature fault location unit initiates Riemann curvature singularity analysis based on metric tensor components to calculate the cross-sectional curvature of the manifold space. This cross-sectional curvature quantifies the degree of geometric distortion of the data manifold. Furthermore, the unit performs fault contribution decoupling analysis on the cross-sectional curvature, mapping the abstract geometric distortion back to the physical sensor dimension and calculating the fault contribution index. The closed-loop control decision unit discriminates the fault contribution index, identifies the physical component causing the dominant fault, and generates corresponding control commands to adjust the equipment's operating state. Through these settings, the system can capture minute changes in the coupling relationships between variables at the geometric topology level, significantly improving the sensitivity of fault diagnosis.
[0016] Example 2: The manifold topology construction analysis process is as follows: The vibration data of the cutting section, the current data of the traction section, the tilt angle data of the rocker arm, and the pressure data of the hydraulic system of the coal mining equipment during the operation period are obtained, and the vibration data of the cutting section, the current data of the traction section, the tilt angle data of the rocker arm, and the pressure data of the hydraulic system are set as the original high-dimensional state vector; The Euclidean distance norm of each state point in the local tangent space is obtained in the original high-dimensional state vector, and the local adaptive kernel width of each state point is obtained through the k-nearest neighbor algorithm. An exponential decay operation is performed on the Euclidean distance norm and the local adaptive kernel width to obtain the adjacency weight matrix, which is used to characterize the geometric affinity between two state points on the manifold surface.
[0017] This embodiment further illustrates the specific process of manifold topology construction analysis performed by the manifold space mapping unit; The system obtains information on the operating time of the coal mining equipment. Key physical parameters within the system include vibration data of the cutting section, current data of the traction section, rocker arm tilt angle data, and hydraulic system pressure data. These data are normalized using the Z-Score standardization method to eliminate the influence of different dimensional physical quantities on the weighting of the Euclidean distance calculation. The calculation formula is as follows: ,in, For the first The raw observations from the 3D sensor, This is the statistical mean of this dimension in the pre-stored historical health dataset. The statistical standard deviation for this dimension; the processed data is set to... The original high-dimensional state vector at time step ; To define the geometric connectivity between state points on the manifold surface, the system needs to calculate the Euclidean distance norm of the state points in the local tangent space. Simultaneously, to adapt to the non-uniformity of data distribution, the system employs the k-nearest neighbor algorithm to obtain the local adaptive kernel width for each state point. Specifically, for each state point… The system searches its historical database for its first... Find the nearest neighbor sample points and set the Euclidean distance between them as the local adaptive kernel width. ; here The value of is determined based on the square root of the total sample size, for example, the range is 10 to 20; The system performs exponential decay operations on the Euclidean distance norm and the local adaptive kernel width to obtain the adjacency weight matrix. This matrix is used to characterize the geometric affinity between two state points on the manifold surface. The system uses a multi-scale adaptive scaling algorithm to calculate the weights, and the calculation formula is as follows: ; in, State point The Euclidean distance to its k-th nearest neighbor. Representing state points The Euclidean distance to its k-th nearest neighbor is calculated. This step effectively solves the problem of unbalanced mapping between sparse and dense regions by introducing an adaptive kernel width based on point pair dependencies, thus ensuring the accuracy of the manifold topology.
[0018] Example 3: The process of constructing and analyzing the Riemannian metric tensor field is as follows: Based on the adjacency weight matrix and the neighborhood point set of the original high-dimensional state vector, a local weighted covariance matrix centered on the data point is constructed. The precision matrix is obtained by inverting the local weighted covariance matrix. The precision matrix is set as a metric tensor component, which is used to characterize the allowable degrees of freedom of the physical quantities while keeping the manifold structure stable.
[0019] This embodiment further illustrates the specific process of performing Riemannian metric tensor field construction analysis by the manifold space mapping unit; Based on the adjacency weight matrix obtained from the preceding steps and the neighborhood point set of the original high-dimensional state vector, the system constructs a system with the current data point... The local weighted covariance matrix centered at the tangent space reflects the degree of dispersion of the data distribution within the local tangent space; the specific calculation is as follows: ; in, Represented by data points The local weighted covariance matrix centered at the center. Representing data points The neighborhood set of points is determined by the k-nearest neighbor algorithm; The system performs an inversion operation on the local weighted covariance matrix to obtain the precision matrix; this precision matrix is directly set as the metric tensor component in Riemannian geometry. ,Right now: ; in, Represents the metric tensor components. and The index subscript is the dimension of the state vector; this metric tensor component is used to characterize the allowable degrees of freedom of the changes in each physical quantity while keeping the manifold structure stable; in a physical sense, if there are strong coupling constraints between some physical variables, their covariance is small and the inverted metric tensor value is large, which means that a small change that violates the constraint will be amplified into a huge geometric distance in Riemann space, thereby achieving sensitive capture of abnormal patterns.
[0020] Example 4: The parallel movement prediction analysis process is as follows: The connection coefficient is calculated using the metric tensor components, and the connection coefficient is used to characterize the bending properties of the manifold. Obtain the health status component and the state displacement of the previous time step; Based on the connection coefficient and the state displacement of the previous time step, the health state component of the previous time step is approximated by a first-order calculation to obtain the health state prediction component of the current time step.
[0021] This embodiment further illustrates the specific process by which the state drift monitoring unit performs parallel movement prediction analysis; The system calculates the connection coefficient using metric tensor components; prior to this, partial derivatives were calculated on discrete state point clouds. The system employs the local moving least squares method to construct the differential operator; specifically, for each state point... For each component of the metric tensor, select its Construct a locally quadratic fitting surface from its nearest neighbors. The fitted surface uses a polynomial model in the form of a second-order Taylor expansion, and its expression is: ; in, For the dimension of the state vector, These are the undetermined fitting coefficients obtained using the weighted least squares method; the weight function used in the weighted least squares method. Let be the Gaussian kernel function, and its expression is: ,in, From nearest neighbor to central state point Euclidean distance, The smoothing parameter takes the value of a locally adaptive kernel width. 1.5 times; of which, , The state vector is represented at the th... , The coordinate components of the dimension; the fitted surface Specifically used to measure tensor component values within an approximate local neighborhood. The distribution of the system; the partial derivatives are directly obtained based on this analytical expression, i.e. The analytical derivative of the fitted surface is calculated and used as an approximate partial derivative value at discrete points. The connection coefficient, also known as the Christopher symbol, is used to characterize the bending properties of the manifold, that is, to describe the rotation rule when a vector moves in the bending space; its calculation formula is as follows: ; in, This represents the second type of Christopher symbol. The components of the metric tensor inverse matrix are represented. This represents partial derivative operations; The system obtains the health status components from the previous moment. and the state displacement of the previous moment Based on the connection coefficient and the state shift of the previous time step, the system performs a first-order approximation calculation on the health state component of the previous time step to obtain the predicted health state component of the current time step. This process simulates the parallel movement of a vector along a geodesic line of a manifold, and the calculation formula is as follows: ; By introducing a connection coefficient correction, the prediction model is no longer a simple linear extrapolation based on Euclidean space, but strictly follows the dynamic evolution prediction of the manifold geometry.
[0022] Example 5: The process of obtaining the state drift is as follows: The sensor operation data at the current moment is obtained, and the difference between the sensor operation data and the health status prediction component is calculated to obtain the state deviation vector; Obtain the metric tensor components; The state deviation vector is calculated by weighted square Euclidean distance based on the metric tensor components to obtain the geodesic deviation on the Riemann manifold, and the geodesic deviation is set as the state drift.
[0023] This embodiment further illustrates the process of obtaining the state drift amount; The system acquires the sensor's operational data at the current moment and normalizes the data according to the Z-Score normalization parameters in Example 2 to obtain a normalized observation vector. The normalized observation vector is then compared with the health status prediction component obtained in the preceding steps. The difference is calculated to obtain the state deviation vector; the system obtains the metric tensor components. ; To accurately quantify the degree of deviation, the system performs a weighted squared Euclidean distance calculation on the state deviation vector based on the metric tensor components to obtain the geodesic deviation on the Riemannian manifold; this geodesic deviation is set as the state drift. The calculation formula is as follows: ; in, This index represents the state drift. It uses the metric tensor as weights, which significantly amplifies the contribution of small deviations in the direction of strong system constraints to the drift, while suppressing fluctuations in the direction of degrees of freedom, thereby achieving high signal-to-noise ratio anomaly detection.
[0024] Example 6: The analysis process for the Riemann curvature singularity is as follows: Based on the metric tensor components and their second derivatives, a Riemann curvature tensor is constructed. Obtain the orthogonal basis vectors of the local tangent space; By performing tensor reduction and fractional operations on the Riemann curvature tensor and orthogonal basis vectors, the cross-sectional curvature on the plane spanned by the orthogonal basis vectors is obtained. The cross-sectional curvature is used to quantify the degree of geometric distortion of the manifold space.
[0025] This embodiment further illustrates the specific process by which the curvature fault location unit performs Riemann curvature singularity analysis; When an anomaly in state drift is detected, the system uses the metric tensor components and their second derivatives, where the second derivative... The value is directly taken from the locally quadratic fitted surface in Example 4. Second-order coefficients in the analytical expression ,Right now No additional difference operations are required; the fourth-order Riemann curvature tensor is constructed using the second-type Christopher notation and its derivative, and its calculation formula is as follows: ; Then, by performing a descent operation on the metric tensor index, we obtain... ; Constructing a fourth-order Riemann curvature tensor Meanwhile, in order to quantify the specific impact of the fault on the current trajectory, the system constructs a set of characteristic orthogonal basis vectors in the local tangent space. and ; where vector The normalized velocity vector selected as the system state at the current moment, i.e. Vectors represent the direction of system evolution. The selected state deviation vector is the one that is normalized. This represents the escape direction of the system deviating from the healthy manifold; The system utilizes the Riemann curvature tensor and orthogonal basis vectors to perform tensor reduction and fractional operations, calculating the plane spanned by the orthogonal basis vectors. Cross-sectional curvature The curvature of this section is used to quantify the degree of geometric distortion in the manifold space. The calculation model is as follows: ; in, It represents the curvature of the cross section; physically, a sudden change in the curvature of the cross section corresponds to the breakage or reorganization of the physical coupling relationship inside the system. By calculating this index, the system can identify hidden structural faults from the geometric topology level.
[0026] Example 7: The decoupling analysis process for fault contribution is as follows: Eigendecomposition of the Riemann curvature tensor yields the principal eigenvectors that result in the maximum cross-sectional curvature. These principal eigenvectors indicate the direction of the most severe tearing of the manifold space. Obtain the unit basis vectors representing each physical sensor in the original data space; Calculate the projection magnitude of the principal eigenvector onto the unit basis vectors of each physical sensor; Calculate the product of the projection modulus and the cross-sectional curvature, and set the product as the fault contribution index of the corresponding physical sensor.
[0027] This embodiment further illustrates the specific process of fault contribution degree decoupling analysis; For the fourth-order Riemann curvature tensor Perform tensor shrinking operations to obtain the second-order Ricci curvature tensor. The calculation formula is: Perform eigenvalue decomposition on the second-order Ricci curvature tensor and obtain the eigenvector corresponding to the largest eigenvalue as the principal eigenvector. The principal eigenvector geometrically indicates the direction of the most severe tearing in the manifold space; the system obtains the unit basis vectors representing each physical sensor in the original data space. For example, for the first The number of sensors, whose unit basis vectors are in the th... One component is 1, and the rest are 0; The system calculates the projection modulus of the principal eigenvector onto the unit basis vectors of each physical sensor; the system calculates the product of this projection modulus and the cross-sectional curvature, and sets this product as the fault contribution index of the corresponding physical sensor. The calculation formula is as follows: ; in, Represents the principal eigenvector In the Component values in each physical sensor dimension; due to unit basis vectors As an orthonormal basis, its projection operation is simplified to the absolute values of the corresponding components of the principal eigenvectors; Indicates the first The failure contribution index of a sensor; this index maps abstract geometric curvature back to specific physical dimensions. The larger the value, the greater the contribution of the physical parameter corresponding to the sensor to the manifold distortion, that is, the parameter is the main factor causing the failure.
[0028] Example 8: The operation process of the closed-loop control decision unit is as follows: Obtain the fault contribution index of each physical sensor; By comparing and analyzing the values of each fault contribution index, the largest fault contribution index with the highest value is selected. Identify the physical sensor corresponding to the maximum failure contribution index and determine that the mechanical component corresponding to the physical sensor has suffered structural damage; Trigger a preset control strategy for mechanical components that have suffered structural damage.
[0029] This embodiment further illustrates the operation process of the closed-loop control decision unit; The closed-loop control decision unit obtains the fault contribution index of each physical sensor; the unit performs numerical comparison analysis of each fault contribution index and selects the maximum fault contribution index with the largest value; the system locks the physical sensor corresponding to the maximum fault contribution index and determines that the mechanical component corresponding to the physical sensor has suffered structural damage. Based on the judgment results, the system triggers a preset control strategy for the mechanical components that have suffered structural damage. For example, if the system determines that the fault contribution index of the traction current sensor is the largest, it identifies it as a traction overload fault. The system will automatically generate a control command to reduce the traction speed by 20% and increase the coolant flow to prevent equipment damage. This process realizes fully automatic closed-loop control from fault perception, location to handling.
[0030] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A condition monitoring and diagnostic system for coal mining equipment based on digital twins, characterized in that, It includes an equipment condition monitoring center, a manifold space mapping unit, a condition drift monitoring unit, a curvature fault location unit, and a closed-loop control decision unit; The equipment status monitoring center is used to retrieve sensor operating data of coal mining equipment and send the sensor operating data to the manifold space mapping unit; The manifold space mapping unit is used to perform manifold topology construction analysis on sensor operation data to obtain the original high-dimensional state vector, and to perform Riemannian metric tensor field construction analysis on the original high-dimensional state vector to obtain metric tensor components. The state drift monitoring unit is used to perform parallel movement prediction analysis based on the metric tensor components to obtain the health state prediction component. The health state prediction component is compared with the sensor operation data to obtain the state drift amount, and the state drift amount is compared with the preset confidence interval threshold. If the state drift is less than or equal to the preset information interval threshold, a normal operation signal is generated, and the closed-loop control decision unit maintains the current operation strategy. If the state drift exceeds the preset threshold, a drift anomaly signal is generated. In response to the drift anomaly signal, the curvature fault location unit is used to perform Riemann curvature singularity analysis based on the metric tensor components to obtain the cross-sectional curvature, and to perform fault contribution decoupling analysis on the cross-sectional curvature to obtain the fault contribution index. The closed-loop control decision unit is used to identify and process the fault contribution index and generate control commands.
2. The coal mining equipment condition monitoring and diagnosis system based on digital twin as described in claim 1, characterized in that, The manifold topology construction and analysis process is as follows: The vibration data of the cutting section, the current data of the traction section, the tilt angle data of the rocker arm, and the pressure data of the hydraulic system of the coal mining equipment during the operation period are obtained, and the vibration data of the cutting section, the current data of the traction section, the tilt angle data of the rocker arm, and the pressure data of the hydraulic system are set as the original high-dimensional state vector; The Euclidean distance norm of each state point in the local tangent space is obtained in the original high-dimensional state vector, and the local adaptive kernel width of each state point is obtained through the k-nearest neighbor algorithm. An exponential decay operation is performed on the Euclidean distance norm and the local adaptive kernel width to obtain the adjacency weight matrix, which is used to characterize the geometric affinity between two state points on the manifold surface.
3. The coal mining equipment condition monitoring and diagnosis system based on digital twin according to claim 2, characterized in that, The Riemannian metric tensor field construction and analysis process is as follows: Based on the adjacency weight matrix and the neighborhood point set of the original high-dimensional state vector, a local weighted covariance matrix centered on the data point is constructed. The precision matrix is obtained by inverting the local weighted covariance matrix. The precision matrix is set as a metric tensor component, which is used to characterize the allowable degrees of freedom of the physical quantities while keeping the manifold structure stable.
4. The coal mining equipment condition monitoring and diagnosis system based on digital twin according to claim 1, characterized in that, The parallel movement prediction and analysis process is as follows: The connection coefficient is calculated using the metric tensor components, and the connection coefficient is used to characterize the bending properties of the manifold. Obtain the health status component and the state displacement of the previous time step; Based on the connection coefficient and the state displacement of the previous time step, the health state component of the previous time step is approximated by a first-order calculation to obtain the health state prediction component of the current time step.
5. The coal mining equipment condition monitoring and diagnosis system based on digital twin according to claim 4, characterized in that, The process for obtaining the state drift amount is as follows: The sensor operation data at the current moment is obtained, and the difference between the sensor operation data and the health status prediction component is calculated to obtain the state deviation vector; Obtain the metric tensor components; The state deviation vector is calculated by weighted square Euclidean distance based on the metric tensor components to obtain the geodesic deviation on the Riemann manifold, and the geodesic deviation is set as the state drift.
6. The coal mining equipment condition monitoring and diagnosis system based on digital twin according to claim 1, characterized in that, The Riemann curvature singularity analysis process is as follows: Based on the metric tensor components and their second derivatives, a Riemann curvature tensor is constructed. Obtain the orthogonal basis vectors of the local tangent space; By performing tensor reduction and fractional operations on the Riemann curvature tensor and orthogonal basis vectors, the cross-sectional curvature on the plane spanned by the orthogonal basis vectors is obtained. The cross-sectional curvature is used to quantify the degree of geometric distortion of the manifold space.
7. The coal mining equipment condition monitoring and diagnosis system based on digital twin according to claim 6, characterized in that, The decoupling analysis process for the fault contribution is as follows: Eigendecomposition of the Riemann curvature tensor yields the principal eigenvectors that result in the maximum cross-sectional curvature. These principal eigenvectors indicate the direction of the most severe tearing of the manifold space. Obtain the unit basis vectors representing each physical sensor in the original data space; Calculate the projection magnitude of the principal eigenvector onto the unit basis vectors of each physical sensor; Calculate the product of the projection modulus and the cross-sectional curvature, and set the product as the fault contribution index of the corresponding physical sensor.
8. The coal mining equipment condition monitoring and diagnosis system based on digital twin according to claim 7, characterized in that, The operation process of the closed-loop control decision unit is as follows: Obtain the fault contribution index of each physical sensor; By comparing and analyzing the values of each fault contribution index, the largest fault contribution index with the highest value is selected. Identify the physical sensor corresponding to the maximum failure contribution index and determine that the mechanical component corresponding to the physical sensor has suffered structural damage; Trigger a preset control strategy for mechanical components that have suffered structural damage.