A device anomaly detection and self-diagnosis method based on component function model

Abstract

The application relates to the field of simulation modeling technology and discloses a device abnormality detection and self-diagnosis method based on a component function model, which comprises the following steps: step S1, constructing a measuring point mapping table; step S2, extracting a phase synchronization sequence; step S3, obtaining a disturbance response sequence; step S4, generating a diagnosis residual sequence; step S5, calculating a node mismatch amount; step S6, mapping to obtain an abnormality classification; and step S7, outputting an abnormality three-dimensional model and a diagnosis text. The application realizes the unified binding of the geometric structure and the function logic of a device by establishing a three-dimensional component function model and constructing a function directed graph and a function partition set; realizes the physical driving type abnormality feature extraction which is independent of real fault samples by obtaining the response sensitivity of each function node through virtual disturbance simulation; and realizes the intuitive visual display of a diagnosis result by calculating the phase migration function mismatch feature and combining the topological tracing of the function directed graph.

Description

Technical Field

[0001] This invention relates to the field of simulation modeling technology, and more specifically, to a method for equipment anomaly detection and self-diagnosis based on component functional models. Background Technology

[0002] Rotary pumps, as general-purpose fluid transport equipment, are widely used in various fields of industrial production. Currently, equipment anomaly detection and diagnosis technologies are mainly divided into three categories: signal analysis-based, machine learning-based, and digital twin simulation-based. Signal analysis-based methods typically extract time-domain and frequency-domain features from external sensor signals such as vibration and pressure for fault identification; machine learning-based methods train classification models using a large number of labeled samples to achieve automatic identification of multiple types of faults; and digital twin-based methods construct three-dimensional models of the equipment and conduct anomaly analysis by comparing simulation results with measured data. These technologies can achieve certain diagnostic effects under steady-state operating conditions, meeting the needs of some routine operating scenarios.

[0003] Under operating conditions involving variable speed, variable flow rate, and strongly unsteady flow fields, existing technologies have several limitations. Pressure contour maps, velocity contour maps, or external vibration signals are often processed as isolated inputs, making it difficult to reflect the functional relationships between various components within the equipment. Fixed-time-window feature extraction methods exhibit phase drift when the rotational speed fluctuates, leading to decreased feature stability. Signals generated by anomalies in upstream components are amplified or delayed in downstream components, causing different anomalies to exhibit similar external responses under the same operating conditions, while the response characteristics of the same anomaly differ under different operating conditions. Furthermore, the number of real-world fault samples is usually limited, making it difficult for training models relying on fault labels to cover the combined boundary conditions of multiple anomalies.

[0004] With the increasing demands for equipment reliability and operational efficiency in industrial production, there is an urgent need for a method for detecting and diagnosing equipment anomalies that can adapt to complex and variable operating conditions. This method should effectively distinguish the abnormal characteristics of different components, accurately identify the location and type of anomalies, reduce misjudgments and missed diagnoses caused by similar responses, and provide reliable technical support for equipment maintenance. Summary of the Invention

[0005] This invention provides a method for equipment anomaly detection and self-diagnosis based on a component functional model, which solves the technical problems mentioned in the background.

[0006] This invention provides a method for equipment anomaly detection and self-diagnosis based on a component functional model, comprising the following steps: Step S1: Introduce the functional topology model into the 3D model corresponding to the physical equipment, perform mesh generation, establish a functional directed graph and functional partition set, and construct a measurement point mapping table in combination with the measurement point installation location; Step S2: Collect the running time series based on the measurement point mapping table and determine the time boundary. Perform resampling processing with the rotor rotation phase as the independent variable to extract the phase synchronization sequence. Step S3: Use the time boundary and phase synchronization sequence to perform three-dimensional flow field solution, generate reference response sequence, apply functional perturbation to the functional partition set, and obtain perturbation response sequence; Step S4: Compare the phase synchronization sequence and the reference response sequence to extract the difference sequence, perform normalization mapping calculation, and generate the diagnostic residual sequence; Step S5: Extract functional sensitivity using the perturbation response sequence and the reference response sequence, search for the coherent extrema of the functional sensitivity corresponding to the diagnostic residual sequence, extract the phase shift, integrate the residual energy sequence, and calculate the node mismatch. Step S6: Input the node mismatch into the functional directed graph, compare it with the health envelope boundary to determine the nodes that exceed the limit, extract the abnormal source nodes according to the topology level, and map to obtain the abnormal classification. Step S7: Combining the anomaly source node, anomaly classification, diagnostic residual sequence, and phase offset, mark the corresponding functional partition in the functional partition set, and output the anomaly 3D model and diagnostic text.

[0007] The beneficial effects of this invention are as follows: By establishing a three-dimensional component functional model and constructing a directed functional graph and functional partition set, this invention achieves a unified binding between the equipment's geometric structure and functional logic, enabling anomaly diagnosis results to be mapped to specific component functional units; by converting the variable speed time series into a unified phase sequence through phase synchronization resampling, it achieves comparable analysis of signals under different operating conditions, reducing the interference of speed fluctuations on residual calculation; by obtaining the response sensitivity of each functional node through virtual disturbance simulation, it achieves physical-driven anomaly feature extraction that does not rely on real fault samples; by calculating phase migration functional mismatch features and combining them with topological sourcing of the directed functional graph, it assists in anomaly source identification and reduces the risk of misclassifying downstream components affected by propagation as anomaly sources; by generating a three-dimensional anomaly localization model with component functional semantics, it achieves an intuitive visualization of diagnostic results. This helps improve the accuracy and engineering practicality of rotating pump anomaly diagnosis. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a device anomaly detection and self-diagnosis method based on a component functional model according to the present invention. Figure 2 This is a schematic diagram of the node mismatch calculation process of the present invention. Detailed Implementation

[0009] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0010] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" indicate that the element or object preceding the term encompasses the elements or objects listed following the term and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0011] like Figure 1 As shown, a method for equipment anomaly detection and self-diagnosis based on component functional models includes the following steps: Step S1: Introduce the functional topology model into the 3D model corresponding to the physical equipment, perform mesh generation, establish a functional directed graph and functional partition set, and construct a measurement point mapping table in combination with the measurement point installation location; Step S2: Collect the running time series based on the measurement point mapping table and determine the time boundary. Perform resampling processing with the rotor rotation phase as the independent variable to extract the phase synchronization sequence. Step S3: Use the time boundary and phase synchronization sequence to perform three-dimensional flow field solution, generate reference response sequence, apply functional perturbation to the functional partition set, and obtain perturbation response sequence; Step S4: Compare the phase synchronization sequence and the reference response sequence to extract the difference sequence, perform normalization mapping calculation, and generate the diagnostic residual sequence; Step S5: Extract functional sensitivity using the perturbation response sequence and the reference response sequence, search for the coherent extrema of the functional sensitivity corresponding to the diagnostic residual sequence, extract the phase shift, integrate the residual energy sequence, and calculate the node mismatch. Step S6: Input the node mismatch into the functional directed graph, compare it with the health envelope boundary to determine the nodes that exceed the limit, extract the abnormal source nodes according to the topology level, and map to obtain the abnormal classification. Step S7: Combining the anomaly source node, anomaly classification, diagnostic residual sequence, and phase offset, mark the corresponding functional partition in the functional partition set, and output the anomaly 3D model and diagnostic text.

[0012] In one embodiment of the present invention, step S1 specifically includes the following steps: Step S101: Import the 3D model corresponding to the physical equipment, and perform mesh generation based on the inlet flow channel, impeller, volute or guide vane, shaft system, coupling, bearing housing, sealing gap and outlet section. The 3D model comes from the equipment as-built CAD model, the calibrated 3D point cloud inverse model, or a fusion model of the two. The inlet flow channel, impeller, volute or guide vane, shaft system, coupling, bearing housing, sealing gap and outlet section have unique 3D partition identifiers. The three-dimensional component functional model is a computer model that simultaneously includes the three-dimensional geometric partitions of the equipment, the simulation calculation domain, and the causal chain of the component's functions. It is used to carry the geometric and functional mapping relationship for all subsequent diagnostic calculations.

[0013] The sources of the 3D model include the as-built CAD model of the equipment, the calibrated 3D point cloud inverse model, or a fusion model of the two. The specific choice is determined based on the available data on site.

[0014] Step S102: Bind the functional topology model to the mesh generation result and establish a functional directed graph that includes inlet flow guidance function, impeller energy conversion function, volute pressure recovery function, shaft torque transmission function and bearing support function. A functional directed graph is a graph structure that uses directional edges to describe the causal propagation order of component functions. The number of nodes is the same as the number of functional nodes, and the direction of the edges is the same as the propagation direction of the abnormal influence.

[0015] Functional nodes are functional units in a directed functional graph, including five basic types: inlet flow guiding function, impeller energy conversion function, volute pressure recovery function, shaft torque transmission function, and bearing support function.

[0016] Step S103: Bind each functional node in the functional directed graph to a unique three-dimensional mesh partition or solid partition to form a functional partition set corresponding to the functional directed graph, wherein the functional node and the three-dimensional mesh partition or solid partition are bound one-to-one. A functional partition set is a set of 3D mesh or solid partitions that are bound one-to-one with each functional node in a functional directed graph, with each partition corresponding to an independent functional node.

[0017] Step S104: Based on the installation location of the measuring points, map the inlet pressure measuring point, outlet pressure measuring point, flow measuring point, speed measuring point, and bearing housing radial vibration measuring point to the cross section or support point in the three-dimensional model to generate a measuring point mapping table. Among them, the inlet pressure measuring point, outlet pressure measuring point, flow measuring point, speed measuring point, and bearing housing radial vibration measuring point have a unified clock stamp.

[0018] The measuring point mapping table is an association table that maps the installation position of each physical measuring point to the corresponding section or support point of the three-dimensional model, and is used to establish the spatial correspondence between the measured data and the three-dimensional model.

[0019] The inlet pressure is the fluid pressure at the inlet section of the rotating pump, which can be obtained by installing a pressure transmitter at the inlet flange.

[0020] The outlet pressure is the fluid pressure at the outlet section of the rotating pump, which can be obtained by installing a pressure transmitter at the outlet flange.

[0021] Flow rate is the volume of fluid delivered by a rotating pump per unit time, and can be obtained by installing an electromagnetic flow meter or an ultrasonic flow meter on the inlet or outlet pipe.

[0022] Rotational speed is the number of revolutions the rotor makes per unit time, which can be obtained by installing a key phase sensor or photoelectric encoder on the shaft.

[0023] The radial vibration of the bearing housing is the vibration acceleration of the bearing housing in the radial direction, which can be obtained by installing acceleration sensors in the horizontal and vertical directions of the bearing housing respectively.

[0024] A unified clock stamp serves as a shared time reference for all measurement points and can be achieved through synchronization via a synchronization acquisition card or a network time protocol.

[0025] Furthermore, the functional topology model is constructed according to the sequence of energy transfer and functional execution in the rotating pump: First, the main functional units of the equipment are identified, including the inlet guiding function, impeller energy conversion function, volute pressure recovery function, shaft torque transmission function, and bearing support function. Then, unidirectional causal connections between functional nodes are established according to the direction of energy flow. Energy enters the equipment from the inlet, reaches the impeller after rectification by the inlet guiding function, and is converted into fluid pressure energy by the impeller energy conversion function. Subsequently, it enters the volute or guide vanes, where the volute pressure recovery function converts some kinetic energy into pressure energy, and finally exits through the outlet. Simultaneously, the shaft torque transmission function transmits the torque from the drive end to the impeller, and the bearing support function provides support constraints for the entire rotor system. The directional edge represents the dominant anomaly influence direction confirmed by simulation or experiment; the functional directed graph only retains the dominant direction used for anomaly tracing, and feedback influence is recorded as an additional coupling relationship and does not participate in the topology hierarchy division. For multi-stage centrifugal pumps or other rotating pumps with more complex structures, corresponding functional nodes can be added according to the actual structure, such as interstage guiding function, balance disc function, etc.

[0026] Furthermore, the mesh is generated using three independent computational domains: the fluid domain, the rotor domain, and the support domain. The fluid domain includes all areas through which the fluid flows, such as the inlet channel, impeller channel, volute or guide vane channel, and outlet channel. The rotor domain includes all solid components that rotate with the rotor, such as the shaft, impeller, and coupling. The support domain includes all fixed support components, such as bearing housings, casing, and foundation. Each computational domain is further subdivided according to component boundaries to ensure that each functional node corresponds to an independent mesh group. During mesh generation, it is necessary to ensure a one-to-one correspondence between mesh nodes at interfaces, or to use an interface interpolation algorithm to ensure the continuity of data transmission.

[0027] Furthermore, the binding of functional nodes to 3D partitions is established through mesh group identifiers or entity attribute tags: after meshing is completed, a unique identifier number is assigned to each mesh group, and a unique functional number is assigned to each functional node, establishing a mapping relationship between mesh group identifiers and functional numbers. Each functional node can only be bound to one 3D partition, and each 3D partition can only be bound to one functional node; many-to-many binding relationships are not allowed.

[0028] It should be noted that this embodiment binds the functional directed graph to the three-dimensional mesh partitions one by one, so that each functional node corresponds to a specific geometric region in the three-dimensional model; by establishing a functional partition set and a measurement point mapping table, a unified mapping of geometric structure, functional logic and measured data is achieved; thus, the diagnostic results of any functional level can be directly traced back to the specific location in the three-dimensional model, providing a basis for the visualization and localization of anomaly sources.

[0029] In one embodiment of the present invention, step S2 specifically includes the following steps: Step S201: Collect the running time series of inlet pressure, outlet pressure, flow rate, rotational speed and bearing housing radial vibration based on the measuring point mapping table, and perform time alignment according to the unified clock stamp of each measuring point. Inlet pressure, outlet pressure, flow rate and bearing housing radial vibration are converted to SI units respectively, and rotational speed is unified to r / min. Step S202: Determine the time boundary based on the diagnostic window covering the complete rotation cycle, and perform time interpolation on the missing moments using adjacent valid points of the same acquisition channel, wherein the sampling frequency satisfies the Nyquist condition of the highest diagnostic order relative to the current maximum shaft frequency. The diagnostic window is a continuous data range selected for a single diagnosis, preferably a duration covering 10 to 20 complete rotor rotation cycles.

[0030] The time boundary is the start and end time corresponding to the diagnostic window, representing the time range covered by this diagnosis.

[0031] Step S203: Using the rotor rotation phase as the independent variable, perform phase binning resampling processing on the running time series; Specifically, the formula for calculating the rotor rotation phase is as follows: in, Indicates the first The rotor rotation phase at each sampling time, in radians, is processed as a pure number. Indicates the first Rotational speed at each sampling time, in units of , Indicates the first The interval between adjacent sampling times, in units of , This represents the conversion constant for rotational speed units. This indicates that the phase is converted to one rotation period; Phase dividers are several continuous phase intervals into which a complete rotor rotation cycle is divided, with each divider corresponding to a fixed phase range.

[0032] The rotor rotation phase is a physical quantity that characterizes the current rotational orientation of the rotor. It has a cycle period of one rotation period and its value ranges from 0 to the angle corresponding to one complete circle.

[0033] Step S204: Extract the phase synchronization sequence of inlet pressure, outlet pressure, flow rate and bearing housing radial vibration according to the preset number of phase sub-boxes, and output the time boundary; Specifically, the formula for calculating the phase synchronization value in the phase synchronization sequence is as follows: in, Indicates the first The measured channel is at the first Measured values ​​at each sampling time Indicates the first The measured channel is at the first Phase synchronization value of each phase sub-bin This represents median operations. Indicates the first One phase interval, This represents the preset number of phase bins, specifically a fixed positive integer determined before diagnosis based on the number of blades and the highest diagnostic order. This indicates the range of values ​​for the phase bin index.

[0034] The phase synchronization sequence is a sequence of operating data that has been resampled with the rotor rotation phase as the independent variable, and it characterizes the law of change of each physical quantity with the rotor orientation.

[0035] The preset number of phase bins is the total number of phase intervals pre-set before diagnosis, preferably 360, with a range of 120 to 720.

[0036] Furthermore, for missing sampling moments, linear interpolation results from two adjacent valid data points within the same channel are used to fill in the gaps. The linear interpolation is calculated as follows: the value of the missing moment equals the value of the previous valid point plus the difference between the values ​​of the next valid point and the previous valid point, multiplied by the time difference between the missing moment and the previous valid point, and then divided by the time difference between the next valid point and the previous valid point. Data from other channels is not used for substitution because the physical quantities in different channels have different variation patterns and time characteristics, and cross-channel substitution would introduce additional errors. Linear interpolation is only performed on a single missing sampling point; that is, the interval between valid sampling moments on both sides of the missing point does not exceed two sampling intervals. When more than two consecutive sampling points are missing, the corresponding data segment is marked as invalid.

[0037] Furthermore, all measured data points falling within the same phase bin are sorted by numerical value, and the value at the middle position is taken as the phase synchronization value for that bin. If the number of data points is odd, the value at the exact middle after sorting is taken; if the number of data points is even, the average of the two middle values ​​is taken. Median calculation has stronger anti-pulse interference capability than mean calculation and can effectively suppress the influence of random impact noise and outliers. In actual industrial sites, vibration and pressure signals are often affected by various pulse interferences, such as water hammer in pipelines, mechanical impacts, and electromagnetic interference. These pulse interferences can cause abnormal jumps in the values ​​of individual sampling points. If mean calculation is used, these outliers will have a significant impact on the final result; while median calculation is not sensitive to outliers and can effectively eliminate these interferences, improving the reliability of the phase synchronization value.

[0038] Furthermore, the highest diagnostic order is typically set to 2 to 3 times the blade passage order. The blade passage order equals the number of blades multiplied by the shaft frequency order, i.e., the number of times the blade passes through the volute diaphragm or guide vane per revolution. Setting the highest diagnostic order to 2 to 3 times the blade passage order covers the main harmonic components such as the fundamental frequency, second harmonic, and third harmonic, meeting the needs of most anomaly diagnoses. If the equipment has other important excitation sources, the highest diagnostic order can be appropriately increased, but it should generally not exceed 5 times the blade passage order; otherwise, it will lead to too many phase bins, too few data points in each bin, and reduced statistical stability of the phase synchronization value.

[0039] It should be noted that traditional signal processing uses time as the independent variable. Under variable speed operation conditions, the rotor rotates at different angles within the same time interval, leading to signal phase misalignment. This embodiment employs phase synchronization resampling technology, using the rotor rotation phase as a unified reference, converting the time axis into a phase axis, so that each sampling point corresponds to a specific rotor orientation. Phase bin resampling through median calculation can effectively suppress pulse interference in industrial environments. This processing method eliminates the influence of speed fluctuations on signal phase, making signals under different operating conditions comparable and providing a unified phase reference for subsequent residual calculations.

[0040] In one embodiment of the present invention, step S3 specifically includes the following steps: Step S301: Input the time boundary, phase synchronization sequence, three-dimensional model mesh and functional partition set into the three-dimensional flow field solution. The three-dimensional flow field solution uses the same three-dimensional model mesh, the same fluid domain boundary naming and the same phase output point. The medium density and dynamic viscosity are taken as medium parameters within the time boundary and kept constant within the time boundary. Step S302: Generate a reference response sequence based on the flow rate boundary, rotation speed boundary, and outlet pressure boundary. The reference response sequence includes the outlet pressure head, outlet pressure pulsation, and radial load in the bearing support direction. The flow rate boundary is taken as the average value of the flow rate phase or the average value of the window. The rotation speed boundary is taken from the rotation speed data within the time boundary. The outlet pressure boundary is taken as the average value of the equipment process back pressure or the outlet pressure. The baseline response sequence is a three-dimensional simulation output sequence of the equipment under current operating conditions without any abnormalities. It includes the outlet pressure head, outlet pressure pulsation, and radial load in the bearing support direction, characterizing the theoretical response level of the equipment in a healthy state.

[0041] Step S303: Apply functional perturbation to a single functional partition in the functional partition set, while keeping the remaining boundaries and solution settings unchanged. In this step, only one functional partition is applied at a time. Local flow loss perturbation is applied to flow channel functional partitions, and support functional perturbation is applied to shaft system or support functional partitions in the corresponding rotor dynamics sub-model. Functional perturbation is a dimensionless relative perturbation applied to the model parameters of a certain functional node. The preferred relative perturbation ratio is 10%, and the value range is 5% to 20%.

[0042] Local flow loss disturbances are disturbances applied to functional zones of flow channels to simulate the effects of anomalies such as flow channel blockage and flow separation on flow characteristics.

[0043] Support function disturbance is a disturbance applied to the shaft system or support-type functional zones to simulate the impact of anomalies such as support loosening and stiffness degradation on the structural response.

[0044] Step S304: Obtain the disturbance response sequence corresponding to each functional partition one by one.

[0045] The disturbance response sequence is a three-dimensional simulation output sequence after applying a virtual disturbance to a single functional node, used to calculate the sensitivity of that functional node to the measurable boundary response.

[0046] Furthermore, local flow loss disturbances are achieved by increasing the viscous drag coefficient within the corresponding three-dimensional partition: an additional drag source term is added to the momentum equation in the fluid calculation. This drag source term is proportional to the local flow velocity, and the proportionality coefficient is the viscous drag coefficient. By increasing the viscous drag coefficient within the target functional partition, the increased flow loss in that region can be simulated, corresponding to anomalies such as channel blockage, scaling, and flow separation in actual operation. The disturbance ratio is typically taken as 5% to 20%, meaning the viscous drag coefficient is increased by 5% to 20% from its original value. The disturbance ratio should not be too small, otherwise the disturbance response will be overwhelmed by numerical noise; nor should it be too large, otherwise it will lead to a fundamental change in the flow field structure.

[0047] Furthermore, the support function disturbance is achieved by altering the support stiffness coefficient in the corresponding rotor dynamics sub-model. In rotor dynamics calculations, bearing supports are typically simplified to spring-damped elements with radial and axial stiffness and damping coefficients. By adjusting the support stiffness coefficient corresponding to the target functional node, anomalies such as support loosening, wear, or stiffness degradation can be simulated. The disturbance ratio is typically taken as 10% to 30%, meaning the support stiffness coefficient is reduced by 10% to 30% from its original value. The reduction in support stiffness corresponds to the actual fault modes of support loosening or stiffness degradation, leading to an increase in the vibration response of the rotor system.

[0048] Furthermore, the convergence criteria for solving the three-dimensional flow field include pressure residuals less than 10⁻⁵, continuity residuals less than 10⁻⁶, velocity component residuals less than 10⁻⁵, and turbulence parameter residuals less than 10⁻⁴. In addition to the residual criteria, it is also necessary to monitor changes in key physical quantities, including outlet head, outlet flow rate, and radial load in the bearing support direction. All monitored physical quantities must remain stable for at least 20 time steps, with fluctuations not exceeding 1%. Strict adherence to the convergence criteria is a prerequisite for ensuring the comparability of nominal and perturbation simulation results.

[0049] It should be noted that traditional data-driven anomaly diagnosis methods rely on training with a large number of real fault samples, but fault samples are scarce in engineering. This embodiment uses virtual disturbance simulation to apply local flow loss disturbances or support function disturbances to each functional node in a 3D model, obtaining the sensitivity sequence of each functional node to the measurable boundary response. This method does not rely on real fault samples; it generates component-level anomaly feature templates through physical simulation, providing a physically driven basis for subsequent anomaly source matching, and solving the problem of poor generalization ability of diagnostic models caused by insufficient fault samples.

[0050] In one embodiment of the present invention, step S4 specifically includes the following steps: Step S401: Perform differential pressure conversion on the inlet pressure phase synchronization sequence and the outlet pressure phase synchronization sequence to generate the measured pressure head sequence; Specifically, the formula for calculating the measured pressure head in the measured pressure head sequence is as follows: in, Indicates the first The measured pressure head of each phase sub-box, in units of , Indicates the first The phase synchronization value of the outlet pressure of each phase compartment, in units of , Indicates the first The inlet pressure phase synchronization value of each phase sub-box, in units of , This represents the density of the medium, in units of... , Represents gravitational acceleration, with units of . ; The measured head sequence is a head sequence calculated from the pressure difference between the inlet and outlet, which characterizes the change of the impeller's energy conversion capability with phase.

[0051] Step S402: Perform order bandpass processing on the radial vibration phase synchronization sequence of the bearing housing, and generate a radial force proxy sequence by combining it with the equivalent mass; Specifically, the formula for calculating the radial force surrogate value in the radial force surrogate sequence is as follows: in, Indicates the first The radial force proxy value of each phase sub-box, in units of , Represents the equivalent mass of the rotor and bearings, in units of . , Indicates the first Phase synchronization value of radial vibration acceleration of bearing housing in each phase sub-box, in units of , This represents the order bandpass operator determined by the shaft frequency order and the blade pass order. This represents the set of orders of the bandpass operators, where the equivalent mass comes from the three-dimensional solid mass of the rotor and bearing and the support model; The radial force proxy sequence is a force sequence calculated from the radial vibration acceleration and equivalent mass of the bearing housing, used for dimensionless comparison with the radial load in the three-dimensional simulation.

[0052] The order bandpass operator is a bandpass filter set around the shaft frequency order and the blade pass order. It is used to extract the vibration components related to the rotational excitation and filter out background noise unrelated to rotation.

[0053] The equivalent mass is the equivalent total mass of the rotor and bearing system, which is determined by calculation based on the three-dimensional solid mass and support model of the specific equipment.

[0054] Step S403: The measured pressure head sequence, outlet pressure pulsation sequence, and radial force proxy sequence are used as the measured observation sequence. The outlet pressure pulsation sequence is obtained by subtracting the phase mean of the same diagnostic window from the outlet pressure phase synchronization sequence. The measured observation sequence is a multi-channel observation sequence composed of the measured pressure head sequence, the outlet pressure pulsation sequence, and the radial force proxy sequence, which characterizes the actual operating status of the equipment in multiple observation dimensions.

[0055] Step S404: Compare the measured observation sequence with the benchmark response sequence using the same observation type, and extract the difference sequence; Step S405: Perform normalization mapping calculation on the difference sequence to generate a diagnostic residual sequence; Specifically, the formula for calculating the diagnostic residual value in the diagnostic residual sequence is as follows: in, Indicates the first Observation type in the first Diagnostic residual values ​​for each phase sub-bin Indicates the first Observation type in the first The measured values ​​of each phase sub-bin were taken sequentially, including the measured pressure head, outlet pressure pulsation, and radial force surrogate value. Indicates the first Observation type in the first The reference response value of each phase bin. Indicates the first Amplitude reference values ​​for observation types, Indicates the first Numerical stability terms for observation types.

[0056] The diagnostic residual sequence is a dimensionless sequence obtained by normalizing the difference between the measured observation sequence and the baseline response sequence, and it characterizes the degree of deviation between the measured response and the health baseline.

[0057] The amplitude reference is the amplitude reference value of the corresponding observation in the current nominal simulation or health baseline, which characterizes the typical amplitude level of the observation and is used for the normalization of the residuals.

[0058] The numerical stability term is a small quantity used to avoid the denominator being zero during normalization calculations. It is preferably set to 10 to the power of -5 of the corresponding amplitude reference value, and its value range is from 10 to the power of -6 to 10 to the power of -4 of the corresponding amplitude reference value.

[0059] Furthermore, the order bandpass operator is implemented using an order tracking method based on Fast Fourier Transform (FFT): First, the phase-synchronized vibration signal undergoes a FFT to convert the signal from the phase domain to the order domain. Then, bandpass filters are set around the shaft frequency order and the blade pass order, retaining signal components near these orders and filtering out components of other orders. Each bandpass filter is centered on the target order, with a passband half-width of 0.5, meaning 0.5 orders above and below the center order, for a total bandwidth of 1 order. Through order bandpass processing, background noise and random interference unrelated to rotation can be effectively filtered out, improving the signal-to-noise ratio of the vibration signal. In practical applications, the bandwidth of the bandpass filter can be adjusted according to the characteristics of the specific equipment and diagnostic requirements. For equipment with large speed fluctuations, the bandwidth can be appropriately widened; for equipment with stable speeds, the bandwidth can be appropriately narrowed to obtain better filtering effects.

[0060] Furthermore, the amplitude reference (peak-to-peak value reference) is taken as the maximum amplitude of the corresponding observation in the current nominal simulation within one rotation cycle. The maximum amplitude refers to the difference between the maximum and minimum values ​​of the observation within a complete rotation cycle. Using the maximum amplitude of the nominal simulation as a reference ensures that the normalized benchmark of the residuals is consistent with the simulation results, avoiding abnormally large or small residuals due to improper selection of the reference value. When the equipment is in a healthy state, the measured response should be close to the nominal simulation response, and the amplitude of the normalized residuals should be much less than 1. Normalization processing enables the comparison of observed signals with different physical dimensions on the same numerical scale. The residuals of pressure signals and vibration signals are converted into dimensionless relative deviations, eliminating the influence of differences in dimensions and numerical ranges on subsequent analysis.

[0061] Furthermore, the numerical stability term is typically taken to be between 10^-6 and 10^-4, corresponding to the magnitude reference value. The purpose of the numerical stability term is to prevent the denominator from approaching zero in the normalization calculation when the magnitude reference value is very small, which would lead to an abnormally amplified residual value. The numerical stability term should not be too large, otherwise the residual will be excessively compressed; nor should it be too small, otherwise numerical instability problems cannot be effectively avoided.

[0062] It should be noted that diagnosing pump malfunctions requires the simultaneous use of multiple types of measurement signals, such as pressure and vibration. These signals have different physical dimensions and numerical ranges. Directly using the original signals will result in the dominant signal having larger dimensions and amplitudes. This embodiment constructs a diagnostic residual sequence with the same dimensions, phase, and observation type, converting measured signals with different physical dimensions into uniform dimensionless residuals. Normalization is performed using amplitude reference quantities and numerical stability terms to reduce the impact of differences in dimensions and amplitude ranges on the functional projection results, enabling comparison of different types of signals on a unified numerical scale.

[0063] In one embodiment of the present invention, such as Figure 2 As shown, step S5 specifically includes the following steps: Step S501: Extract the functional sensitivity of each functional node to each observation type using the perturbation response sequence and the baseline response sequence; Specifically, the formula for calculating functional sensitivity is as follows: in, Represents functional nodes For the first Observation type in the first Functional sensitivity of each phase sub-bin Represents functional nodes The first after being disturbed by function Observation type in the first The disturbance response value of each phase sub-bin. Indicates the first Observation type in the first The reference response value of each phase bin. Indicates that it is applied to the functional node. The proportion of functional perturbations on the surface. Indicates the first Amplitude reference values ​​for observation types, Indicates the first Numerical stability terms for observation types; Functional sensitivity is the degree of influence of a unit disturbance of a functional node on the measurable boundary response, characterizing the strength and phase characteristics of the effect of the functional node on the observed response when the node is abnormal.

[0064] Step S502: Search for the coherent extreme values ​​of the diagnostic residual sequence and functional sensitivity within one rotation cycle, and extract the phase offset of each functional node. Specifically, the formula for calculating the phase offset is as follows: in, Represents functional nodes The phase offset, in units of the number of phase bins and processed by index. This represents the set of cyclic phase offsets within one rotation cycle. This represents the candidate offset index in the cyclic phase offset set. Indicates the first The phase index, after subtracting the candidate offset index from each phase bin, is cyclically mapped to the first through the second... Each phase box, Indicates the number of observation types. Indicates the first Observation type in the first Diagnostic residual values ​​for each phase sub-bin Represents functional nodes For the first Functional sensitivity of observation type in cyclic mapping phase binning; The coherence extremum is the maximum correlation between the diagnostic residual sequence and the functional sensitivity sequence after different phase shifts, representing the best match between the residual and the sensitivity of that functional node.

[0065] Phase offset is the phase delay caused when an anomaly propagates from a functional node to a measurement point. It is expressed in units of the number of phase bins and characterizes the degree of phase lag in anomaly propagation.

[0066] The cyclic phase offset set is a set of all possible phase offsets within one rotation cycle, including all integer offsets from zero to the preset number of phase bins minus one.

[0067] Step S503: Perform cyclic alignment of functional sensitivity based on phase offset and calculate direction coefficient; Specifically, the formula for calculating the direction coefficient is as follows: in, Represents functional nodes directional coefficient, The numerically stable term representing the direction coefficient. This indicates the functional sensitivity that has been cyclically aligned according to the phase offset of the functional nodes. Indicates the first Observation type in the first Diagnostic residual values ​​for each phase sub-bin Indicates the preset number of phase bins. Indicates the number of observation types; The direction factor is the normalized coherence between the measured residual vector and the phase-shifted functional sensitivity vector, and it characterizes the similarity in shape between the residual waveform and the functional sensitivity waveform.

[0068] Step S504: Perform a weighted allocation on the squared values ​​of the diagnostic residual sequence based on the functional sensitivity after cyclic alignment to obtain the residual energy sequence; Specifically, the formula for calculating the projected value of the residual energy in the residual energy sequence is as follows: in, Represents functional nodes The residual energy projection value, The numerical stability term representing the projected residual energy value. Indicates the first Observation type in the first Diagnostic residual values ​​for each phase sub-bin Indicates that it has been sorted by functional node The functional sensitivity of phase offset cyclic alignment; The residual energy sequence is a node-level energy sequence obtained by weighting the global residual energy according to the sensitivity of each functional node, representing the amount of residual energy allocated to each functional node.

[0069] Step S505: Integrate the residual energy sequence and direction coefficients to calculate the node mismatch of each functional node; Specifically, the formula for calculating node mismatch is as follows: in, Represents functional nodes The amount of node mismatch, Represents functional nodes The residual energy projection value, Represents functional nodes Directional coefficient.

[0070] The node mismatch is an index that characterizes the degree of abnormality of a functional node, obtained by combining residual energy and directional consistency, and represents the degree of deviation of the functional node from its healthy state.

[0071] Furthermore, the cyclic mapping is implemented as follows to ensure that the phase index is always within the valid range: when the phase index minus the phase offset is less than 1, the total number of phase bins is added to bring it back to the valid range; when the phase index minus the phase offset is greater than the total number of phase bins, the total number of phase bins is subtracted to bring it back to the valid range. The physical basis of the cyclic mapping is the periodicity of rotor rotation. The phase distribution within one rotation cycle has the characteristic of cyclic repetition; after a phase offset exceeds one cycle, it will repeat the previous state. Through cyclic mapping, any phase offset can be equivalent to one rotation cycle, simplifying the calculation of phase transition. The cyclic mapping is essentially a modulo operation on the phase index, ensuring that the phase index is always within the valid range and avoiding array out-of-bounds or invalid index problems.

[0072] Furthermore, the search for coherent extrema employs an traversal method to find the optimal matching position between the residual and the sensitivity: it traverses all possible offsets in the cyclic phase offset set, calculates the dot product of the residual sequence and the functional sensitivity sequence for each offset, and takes the offset corresponding to the maximum dot product as the phase offset. The dot product is calculated by multiplying the residual value by the corresponding sensitivity value after the phase offset for each phase bin and each observation type, and then summing all products. The larger the dot product, the more similar the residual and the sensitivity at that offset are, and the higher the degree of matching. This traversal search method ensures that the globally optimal phase offset is found, avoiding getting trapped in local optima. In practical calculations, to improve search efficiency, a strategy of coarse search followed by fine search can also be adopted. A coarse search with a larger step size is performed first to find the approximate optimal position, and then a fine search is performed near that position with a smaller step size. This reduces computational load while maintaining accuracy. The dot product operation essentially measures the similarity between two vectors; the dot product reaches its maximum value when the waveform shape and phase of the two vectors are completely identical.

[0073] Furthermore, the search range for phase transfer is strictly limited to one complete rotor rotation cycle, and searches beyond this range are not allowed. Limiting the search range to one rotation cycle covers all possible physical propagation delays while avoiding meaningless searches, thus improving computational efficiency.

[0074] It should be noted that anomalies in different parts of the rotating pump may produce similar residual waveforms at external measuring points, making it difficult to distinguish the source of the anomaly based solely on the residual amplitude and waveform. This embodiment integrates residual energy and directional consistency information by calculating phase-transfer functional mismatch characteristics; it performs phase matching between the sensitivity template of each functional node and the residual sequence, and distributes the global residual energy to each functional node according to the degree of matching, obtaining the node-level mismatch. In other words, this embodiment utilizes the phase characteristics of anomaly propagation to reduce the confusion when different anomalies produce similar residuals, providing a basis for anomaly source localization.

[0075] In one embodiment of the present invention, step S6 specifically includes the following steps: Step S601: Input the node mismatch amount into the functional directed graph and compare it with the health envelope boundary stored by functional node. The health envelope boundary is established offline based on multi-condition health data after the equipment is put into operation or maintenance and confirmed to be healthy, and stored by functional node. The health envelope boundary represents the upper limit of node mismatch for each functional node under the device's health status. It is established offline using multi-condition health data and is used to determine whether a node is abnormal.

[0076] Step S602: Functional nodes whose node mismatch exceeds the corresponding health envelope boundary are judged as over-limit nodes; Specifically, the formula for calculating the candidate node with the largest node mismatch is as follows: in, Indicates the candidate anomaly source node. This represents the set of functional nodes in a directed graph. Represents functional nodes The amount of node mismatch; Specifically, the anomaly detection formula is as follows: in, Indicates candidate anomaly source nodes The amount of node mismatch, Indicates candidate anomaly source nodes When there are multiple out-of-limit nodes in the health envelope boundary, the out-of-limit nodes located at the upstream level are extracted as the source nodes of the anomaly according to the topological hierarchy of the functional directed graph. The out-of-limit nodes at the same topological hierarchy are determined as the source nodes of the anomaly according to the sorting result of the node mismatch (from largest to smallest). An out-of-limit node is a functional node whose mismatch exceeds the boundary of the corresponding health envelope, indicating that the abnormality of the node exceeds the normal fluctuation range.

[0077] The source node of the anomaly is the initial functional node that causes the equipment anomaly. It is the most upstream node among all the out-of-limit nodes and represents the true source of the equipment anomaly.

[0078] Step S603: Obtain the anomaly classification based on the functional node type of the anomaly source node. Among them, the inlet guiding function is mapped to inlet blockage or cavitation anomalies; the impeller energy conversion function is mapped to impeller local damage, erosion or imbalance anomalies; the volute pressure recovery function is mapped to volute or guide vane flow separation or local blockage anomalies; the shaft torque transmission function is mapped to coupling eccentricity or shaft misalignment anomalies; and the bearing support function is mapped to bearing support loosening, wear or stiffness degradation anomalies.

[0079] Anomalies are classified into specific anomaly types based on the functional type mapping of the anomaly source node, representing the specific nature and failure mode of the anomaly.

[0080] Furthermore, the functional directed graph topology hierarchy is divided according to the causal order of anomaly propagation. An upstream functional node is a unit whose anomaly affects the function of downstream nodes. The hierarchy begins at the upstream inlet and extends downstream level by level. The inlet guiding function is at the upstream level because anomalies in the inlet flow channel affect the inlet conditions of all subsequent components. The impeller energy conversion function is at the second level; it is directly affected by the inlet guiding function, and its anomaly also affects the downstream volute pressure recovery function and bearing support function. The volute pressure recovery function is at the third level. The shaft torque transmission function is at the second level, the same level as the impeller energy conversion function. The bearing support function is at the downstream level.

[0081] Furthermore, after the equipment is put into operation or confirmed to be healthy during maintenance, multi-condition data covering the entire normal operating speed and flow range are collected to establish the health envelope boundary. The speed range should cover the entire interval from the lowest to the highest operating speed, and the flow range should cover the entire interval from low flow rate to high flow rate. At least three sets of data are collected for each operating condition, and each set of data contains signals of at least 10 complete rotation cycles. Then, the node mismatch of each functional node is calculated. For each functional node, node mismatch data under all healthy operating conditions are collected, and the 95th percentile is taken as the health envelope boundary of that node. The health envelope boundary is stored separately for each functional node and is not updated in the current diagnostic window. In practical applications, the establishment of the health envelope boundary needs to ensure that the equipment is indeed in a healthy state. Data collection is usually carried out immediately after maintenance or in the early stages of new equipment commissioning to ensure the reliability of the baseline data. After the health envelope boundary is established, it also needs to be reviewed and updated regularly, generally every six months or year, to adapt to the natural degradation of equipment performance.

[0082] Furthermore, when multiple nodes exceed limits simultaneously, the upstreammost node in the directed functional graph is selected first. The upstreammost node is the one upstream of which no other node exceeds limits in the directed functional graph; it is the starting point of the anomaly propagation. If multiple nodes exceed limits at the same topological level, the node with the largest mismatch is selected as the source node of the anomaly. The severity of the anomaly can be assessed based on the proportion by which the node mismatch exceeds the healthy envelope boundary; the larger the proportion, the more severe the anomaly. Typically, the severity of the anomaly can be divided into three levels: mild, moderate, and severe, each corresponding to a different range of exceedance proportions.

[0083] It should be noted that when an anomaly occurs in a certain part of the equipment, the impact of the anomaly will propagate downstream along the functional chain, causing multiple downstream functional nodes to exhibit abnormal characteristics. Simply determining whether a node exceeds its limits is insufficient to identify the true source of the anomaly. This embodiment uses the topology of a directed functional graph to determine the anomaly source, leveraging the unidirectional propagation of anomaly effects from upstream functional nodes to downstream components to distinguish between the true source component and downstream components only affected by the propagation of the anomaly. By using a top-priority rule and a mismatch comparison rule between components at the same level, multiple nodes simultaneously exceeding their limits are ranked as candidate anomaly sources.

[0084] In one embodiment of the present invention, step S7 specifically includes the following steps: Step S701: Extract the corresponding functional partition from the functional partition set in combination with the abnormal source node, wherein the corresponding functional partition is the bound partition of the abnormal source node in the functional partition set. Step S702: Determine the abnormal phase position of the corresponding functional partition by combining the diagnostic residual sequence and the phase offset. The abnormal phase position is determined based on the phase bin that contributes the most to the node mismatch in the diagnostic residual sequence, and in combination with the phase offset. The abnormal phase position is the rotor phase position that contributes the most to the measurable boundary response to the abnormality, and it represents the rotor orientation where the abnormality has the most significant impact.

[0085] Step S703: Combine the exception classification with the name of the exception component and the exception classification in the corresponding functional partition; Step S704: Mark the corresponding functional partition in the functional partition set and output the abnormal 3D model and diagnostic text, without introducing new diagnostic weights, manually correcting labels or closing loop updates. An abnormal 3D model is a 3D model that marks the location of abnormal components and the type of abnormality, used to visually demonstrate the location and nature of the abnormality.

[0086] The diagnostic text is a text report containing the name of the abnormal component, the type of abnormality, and information on the three-dimensional partition of the abnormality, used to record and transmit diagnostic conclusions.

[0087] When no node exceeds the limit, all functional partitions in the functional partition set are marked as normal, and a normal 3D model and diagnostic text are output.

[0088] A normal 3D model is a 3D model in which all functional nodes are marked as normal, used to indicate that the device is in a healthy state.

[0089] Furthermore, the location of the anomalous phase is determined using the following method to accurately pinpoint the phase location with the most significant anomaly impact: For each phase bin, calculate the sum of squared diagnostic residuals for all observation types within that bin. Take the phase bin corresponding to the maximum sum of squares, and then adjust it in conjunction with the phase offset of the anomaly source node to obtain the final anomalous phase location. The sum of squared residuals reflects the total degree of anomaly across all observation dimensions at that phase location; a larger sum of squares indicates a more significant anomaly at that phase location. After finding the phase bin with the largest sum of squared residuals, further adjustment is needed in conjunction with the phase offset of the anomaly source node, because the residuals are observed at the measurement point location, while the anomaly source is located at the functional node location, resulting in a phase difference between the two. The adjustment method involves subtracting the phase offset from the index of the phase bin with the largest sum of squared residuals, and then adjusting it to the effective range through cyclic mapping. This yields the anomalous phase location corresponding to the anomaly source location.

[0090] Furthermore, functional partition marking is achieved by changing the color of the abnormal 3D partition to red, increasing its transparency to 50%, and adding a highlighted border. Simultaneously, text labels are added to the corresponding locations on the 3D model, displaying the name of the abnormal component and the type of abnormality. For complex faults, multiple abnormal partitions can be marked simultaneously, distinguishing between primary and secondary abnormal sources using different shades of red or different border styles. Users can interact with the 3D model using a mouse or touchscreen, including rotation, zoom, and translation, to observe the position and shape of abnormal components from different angles. Users can also show or hide specific components, view their internal structure, or switch between different physical quantity cloud map displays by clicking or selecting. These interactive operations are only for viewing and display and do not change the diagnostic results themselves. Diagnostic text is generated according to a fixed format template to ensure consistency and standardization of the output. Diagnostic text typically includes several parts: basic equipment information, operating condition parameters, diagnostic conclusions, anomaly severity assessment, and recommended handling measures. Each part has fixed field names and content formats for easy subsequent data management and system integration.

[0091] It should be noted that this embodiment transforms abstract numerical diagnostic results into an intuitive 3D visualization model, directly marking the location and type of abnormal components on the 3D model of the equipment. Through red semi-transparent highlighting and text labels, the location and type of abnormal components are clearly presented, thereby improving the interpretability and usability of the diagnostic results and facilitating quick understanding and handling by on-site maintenance personnel.

[0092] This invention can be deployed in equipment condition monitoring systems in industrial settings as a diagnostic module for predictive maintenance of rotating pumps. In actual deployment, the first step is to import the equipment's 3D model and construct its functional topology. This is typically done before equipment commissioning or during the initial deployment of the diagnostic system, and is a one-time configuration task. The 3D model can be derived from the as-built CAD model provided by the equipment manufacturer, or it can be modeled by reverse engineering point cloud data obtained through on-site 3D scanning. After the model is built, corresponding measuring sensors need to be installed on the equipment, including pressure transmitters, flow meters, speed sensors, and vibration acceleration sensors. All sensors are connected to a synchronous acquisition system to ensure a unified time reference.

[0093] The final output of this invention includes two parts: an abnormal 3D model and diagnostic text. The abnormal 3D model is a visual model generated based on the original 3D model of the equipment, with specially marked functional areas identified as abnormal. Specifically, when the diagnostic result determines that the equipment is abnormal, the system marks the functional area corresponding to the abnormal source node in the 3D model as red semi-transparent and adds a highlighted border. Simultaneously, it displays text labels near this area, indicating the name of the abnormal component and the type of abnormality. For example, when the diagnosis determines that the impeller energy conversion function is abnormal, the impeller area in the 3D model will be marked as red semi-transparent, with specific abnormality type text such as impeller local damage or impeller erosion displayed next to it.

[0094] Diagnostic reports typically include the equipment number, diagnosis time, operating parameters, name of the malfunctioning component, malfunction type, malfunction severity, and recommended corrective actions. For example, a diagnostic report for a centrifugal pump might read: "Single-stage centrifugal pump, equipment number P-101, diagnosis time: [date and time missing], operating conditions: speed 2980 rpm, flow rate 320 cubic meters per hour, diagnosis conclusion: impeller energy conversion function malfunction, malfunction type: localized impeller damage, malfunction severity: moderate, recommendation: shutdown for inspection and impeller repair or replacement as needed." Diagnostic reports can be directly pushed to the maintenance personnel's terminal devices.

[0095] The content of this embodiment has been described above, but this embodiment is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this embodiment, all of which are within the protection scope of this embodiment.

Claims

1. A method for equipment anomaly detection and self-diagnosis based on component functional models, characterized in that, Includes the following steps: Step S1: Introduce the functional topology model into the 3D model corresponding to the physical equipment, perform mesh generation, establish a functional directed graph and functional partition set, and construct a measurement point mapping table in combination with the measurement point installation location; Step S2: Collect the running time series based on the measurement point mapping table and determine the time boundary. Perform resampling processing with the rotor rotation phase as the independent variable to extract the phase synchronization sequence. Step S3: Use the time boundary and phase synchronization sequence to perform three-dimensional flow field solution, generate reference response sequence, apply functional perturbation to the functional partition set, and obtain perturbation response sequence; Step S4: Compare the phase synchronization sequence and the reference response sequence to extract the difference sequence, perform normalization mapping calculation, and generate the diagnostic residual sequence; Step S5: Extract functional sensitivity using the perturbation response sequence and the reference response sequence, search for the coherent extrema of the functional sensitivity corresponding to the diagnostic residual sequence, extract the phase shift, integrate the residual energy sequence, and calculate the node mismatch. Step S6: Input the node mismatch into the functional directed graph, compare it with the health envelope boundary to determine the nodes that exceed the limit, extract the abnormal source nodes according to the topology level, and map to obtain the abnormal classification. Step S7: Combining the anomaly source node, anomaly classification, diagnostic residual sequence, and phase offset, mark the corresponding functional partition in the functional partition set, and output the anomaly 3D model and diagnostic text.

2. The method for equipment anomaly detection and self-diagnosis based on component functional models according to claim 1, characterized in that, Import the 3D model corresponding to the physical equipment, and perform mesh generation based on the inlet flow channel, impeller, volute or guide vane, shaft system, coupling, bearing housing, sealing gap and outlet section; bind the functional topology model to the mesh generation result, and establish a functional directed graph that includes inlet flow guidance function, impeller energy conversion function, volute pressure recovery function, shaft system torque transmission function and bearing support function; Each functional node in the directed graph is bound to a unique 3D mesh partition or solid partition to form a set of functional partitions corresponding to the directed graph. Based on the installation location of the measuring points, the inlet pressure measuring point, outlet pressure measuring point, flow measuring point, speed measuring point, and bearing housing radial vibration measuring point are mapped to the cross section or support point in the three-dimensional model to generate a measuring point mapping table.

3. The equipment anomaly detection and self-diagnosis method based on component functional model according to claim 1, characterized in that, The operating time series of inlet pressure, outlet pressure, flow rate, speed and bearing housing radial vibration are collected based on the measuring point mapping table, and the execution time is aligned according to the unified clock stamp of each measuring point. The time boundary is determined based on the diagnostic window covering the complete rotation cycle, and time interpolation is performed on the missing moments using adjacent valid points of the same acquisition channel; Phase binning resampling is performed on the running time series with the rotor rotation phase as the independent variable; Based on the preset number of phase bins, the phase synchronization sequence of inlet pressure, outlet pressure, flow rate, and radial vibration of bearing housing is extracted, and the time boundary is output.

4. The equipment anomaly detection and self-diagnosis method based on component functional model according to claim 1, characterized in that, Input the time boundary, phase synchronization sequence, three-dimensional model mesh, and functional partition set into the three-dimensional flow field for solution; A reference response sequence is generated based on the flow rate boundary, speed boundary, and outlet pressure boundary. The reference response sequence includes the outlet head, outlet pressure pulsation, and radial load in the bearing support direction. Apply a functional perturbation to a single functional partition in the functional partition set, while keeping the remaining boundary and solution settings unchanged; Obtain the disturbance response sequence corresponding to each functional partition one by one.

5. The equipment anomaly detection and self-diagnosis method based on a component functional model according to claim 1, characterized in that, Perform differential pressure conversion on the inlet pressure phase synchronization sequence and the outlet pressure phase synchronization sequence to generate a measured head sequence; The radial vibration phase synchronization sequence of the bearing housing is subjected to order bandpass processing, and a radial force proxy sequence is generated by combining it with the equivalent mass.

6. The equipment anomaly detection and self-diagnosis method based on a component functional model according to claim 5, characterized in that, The measured head sequence, the outlet pressure pulsation sequence obtained by subtracting the phase mean of the same diagnostic window from the outlet pressure phase synchronization sequence, and the radial force proxy sequence are used as the measured observation sequences. The measured observation sequence is compared with the baseline response sequence using the same observation type, and the difference sequence is extracted. Perform normalization mapping calculation on the difference sequence to generate a diagnostic residual sequence.

7. The equipment anomaly detection and self-diagnosis method based on component functional model according to claim 1, characterized in that, The functional sensitivity of each functional node to each observation type is extracted using the perturbation response sequence and the baseline response sequence; Search for the coherent extreme values ​​of the diagnostic residual sequence and functional sensitivity within one rotation cycle, and extract the phase offset of each functional node; The functional sensitivity is cyclically aligned based on the phase offset, and the direction coefficient is calculated.

8. The equipment anomaly detection and self-diagnosis method based on a component functional model according to claim 7, characterized in that, The squared values ​​of the diagnostic residual sequence are weighted according to the functional sensitivity after cyclic alignment to obtain the residual energy sequence; By integrating the residual energy sequence and direction coefficients, the node mismatch of each functional node is calculated.

9. The equipment anomaly detection and self-diagnosis method based on component functional model according to claim 1, characterized in that, Input the node mismatch into the functional directed graph and compare it with the health envelope boundary stored by functional node; Functional nodes whose mismatch exceeds the corresponding health envelope boundary are classified as over-limit nodes. When there are multiple out-of-limit nodes, the out-of-limit nodes located at the upstream level are extracted as the source nodes of the anomaly according to the topological hierarchy of the functional directed graph. Out-of-limit nodes at the same topological hierarchy are determined as the source nodes of the anomaly according to the sorting result of the node mismatch. Anomalies are classified based on the mapping of the functional node type to which the anomaly source node belongs.

10. The method for equipment anomaly detection and self-diagnosis based on a component functional model according to claim 1, characterized in that, Extract the corresponding functional partition from the functional partition set by combining the anomaly source node; The abnormal phase location of the corresponding functional zone is determined by combining the diagnostic residual sequence and the phase offset. Write the abnormal component name and abnormal category in the corresponding functional partition based on the abnormality classification; Mark the corresponding functional partition in the functional partition set and output the abnormal 3D model and diagnostic text; When no node exceeds the limit, all functional partitions in the functional partition set are marked as normal, and a normal 3D model and diagnostic text are output.

Images