Grain screening fault diagnosis method and system based on knowledge graph

Abstract

The invention relates to the technical field of industrial equipment pre-diagnosis, in particular to a knowledge graph-based grain screening fault diagnosis method and system. The method comprises the following steps: synchronously acquiring multi-channel vibration and power data through a precise clock protocol; extracting complex domain rheological impedance representing energy consumption and energy storage of the material; constructing a dynamic physical knowledge graph for mapping the state of the equipment, and quantifying the amplitude attenuation and phase lag weight of a graph edge by using a viscous real part and an elastic imaginary part of impedance; the low-frequency component responding to the material and the high-frequency detail component responding to the fault are decoupled by applying atlas wavelet transform and a multi-scale filter bank, and diagnosis is realized through high-frequency energy characteristics. According to the method, fusion of a physical mechanism and data driving is realized, nonlinear interference of materials can be effectively stripped, and weak faults can be accurately identified.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment pre-diagnosis technology, and in particular to a method and system for diagnosing grain screening faults based on knowledge graphs. Background Technology

[0002] Large vibrating screens are key equipment in grain processing and storage. Under long-term high-intensity alternating load conditions, components such as the screen box side plates, excitation beams, and support springs are prone to mechanical failures such as fatigue cracks, loose bolts, or breakage.

[0003] Existing fault diagnosis technologies for vibrating screens mainly include time-domain monitoring methods and frequency-domain diagnostic methods. Time-domain monitoring methods typically use root mean square (RMS) values ​​or peak values ​​to determine vibration intensity. In actual production, grain, as a bulk particle group, has rheological properties. The thickness of the grain accumulation on the screen surface and its moisture content will change the equivalent mass and damping coefficient of the screening system. The signal fluctuations generated by this coupling effect of variable mass and variable damping can easily mask the impact signals caused by weak mechanical faults. When the feed rate increases, the additional damping effect of the material will suppress the screen body amplitude. Even if there is a fault in the screen body that reduces structural stiffness, the vibration amplitude may still appear to be a normal value, leading to missed detections.

[0004] Frequency domain diagnostic methods monitor the drift of natural frequencies through spectral analysis. Most existing technologies simplify the screening system as a linear system, neglecting the viscoelastic characteristics of the material. Grain particles not only provide damping energy dissipation but also additional stiffness energy storage. Changes in material properties can alter the phase hysteresis effect of vibration waves, manifesting as broadening or shifting of resonance peaks in the spectrum. Standard spectral analysis methods struggle to distinguish between frequency changes caused by material rheological properties and those caused by structural faults.

[0005] Some existing technologies attempt to use graph algorithms to process multi-sensor data. Conventional graph algorithms typically use scalar weights to represent the connection strength between nodes. This representation only considers amplitude correlation and cannot describe the dual effects of amplitude attenuation and phase lag when vibration waves propagate in viscoelastic media. When the operating conditions of the screening machine fluctuate drastically, graph algorithms based solely on scalar weights struggle to effectively separate global signal changes caused by material and local signal abrupt changes caused by faults during the feature extraction stage, leading to a decrease in diagnostic accuracy under heavy or variable load conditions.

[0006] Therefore, establishing a method that can integrate material rheology mechanism and graph signal processing to decouple nonlinear interference caused by materials from vibration signals, thereby identifying mechanical faults covered by material signals, is a technical problem that needs to be solved. Summary of the Invention

[0007] To address the issues that the variable mass and damping characteristics of grain particles can easily mask early mechanical faults in the screen body, and that existing graph algorithms cannot effectively separate operational fluctuations from structural faults due to a lack of phase dimension information, this invention provides a grain screening fault diagnosis method and system based on knowledge graphs.

[0008] In a first aspect, the present invention provides a grain screening fault diagnosis method based on knowledge graphs, employing the following technical solution: A knowledge graph-based method for diagnosing grain screening faults includes the following steps: Based on a precision clock synchronization protocol, the triaxial acceleration vibration signals of each node of the grain screening machine and the instantaneous power parameters of the motor are collected synchronously. Based on the triaxial acceleration vibration signal and the instantaneous power parameters, the complex domain rheological impedance tensor is calculated; it includes the real part of viscous damping that characterizes the energy absorption of the material and the imaginary part of elastic reactance that characterizes the energy storage. A dynamic physical knowledge graph is constructed, and the complex domain rheological impedance tensor is mapped to complex correlation weights between nodes; the complex correlation weights use the real part of the viscous damping to characterize the amplitude attenuation and the imaginary part of the elastic reactance to characterize the phase hysteresis. A graph wavelet transform is performed on the dynamic physical knowledge graph loaded with the triaxial acceleration vibration signal, and a multi-scale filter bank is used to decouple the low-frequency smooth component of the response material load and the high-frequency detail component of the response mechanical structure. The energy characteristics of the high-frequency detail components are extracted, and mechanical faults are determined when they exceed a dynamic threshold.

[0009] This invention introduces complex domain rheological impedance as a priori physical knowledge to construct a dynamic physical knowledge graph containing phase information. By utilizing the multi-scale characteristics of wavelet transform in the graph, it achieves orthogonal decoupling of material load signals and mechanical fault signals in the graph frequency domain, thus solving the problem of difficulty in extracting weak faults under mixed signals.

[0010] Preferably, the step of synchronously acquiring the triaxial acceleration vibration signals of each node of the grain screening machine based on a precision clock synchronization protocol specifically includes: The clock of the sensor array deployed at the four corners of the screen box and the excitation beam is locked using a protocol to ensure that the phase synchronization error of each channel is less than a preset microsecond threshold. By aggregating the sampling data from various measuring points at the same time, a third-order vibration state tensor with node, spatial, and temporal dimensions is constructed.

[0011] This invention ensures strict alignment of multi-channel signals in time and space through precise clock synchronization, eliminating transmission delay errors between channels and providing an accurate data basis for subsequent calculation of phase lag in complex impedance.

[0012] Preferably, the components of the complex domain rheological impedance tensor satisfy the following relationship:

[0013] In the formula, For the first The sampling time of the first sampling moment The values ​​of complex impedance components in each spatial direction. The imaginary unit; For the first The sampling time of the first sampling moment The real part of the viscous damping in each spatial direction. For the first The sampling time of the first sampling moment The imaginary part of the elastic reactance in each spatial direction; The real part of the viscous damping satisfies the following relationship:

[0014] In the formula, For the first At the sampling time, the motor is in the... The active power component values ​​output in each spatial direction. For the first At the sampling time, the sieve body was at the... Vibration velocity modulus in a spatial direction.

[0015] This invention extracts the ratio of active power to the square of velocity as the real part of viscous damping. This index can independently characterize the energy dissipation characteristics of materials due to friction and viscosity, thereby quantifying the working condition characteristic of materials becoming stickier or thicker, and avoiding misjudging it as a fault.

[0016] Preferably, the imaginary part of the elastic reactance satisfies the following relationship:

[0017] In the formula, For the first At the sampling time, the motor is in the... The reactive power component values ​​output in each spatial direction.

[0018] This invention extracts the ratio of reactive power to the square of velocity as the imaginary part of the elastic reactance. This index can independently characterize the energy storage characteristics and natural frequency drift of the system. Since mechanical faults mainly affect the energy storage structure, this index is highly sensitive to structural damage.

[0019] Preferably, the complex association weights satisfy the following relation:

[0020] In the formula, For the first Each sampling time node With nodes The complex correlation weight values ​​between them The baseline weight values ​​are based on geometric location. For nodes With nodes The physical Euclidean distance between them. The preset amplitude attenuation constant, The preset wave velocity phase constant, The imaginary unit, For the first The sampling time of the first sampling moment The imaginary part of the elastic reactance in each spatial direction. For the first The sampling time of the first sampling moment The real part of the viscous damping in each spatial direction. It is the imaginary unit.

[0021] This invention achieves dynamic mapping of physical impedance to spectrum weights through this relationship. The real part exponent term corrects the amplitude attenuation of the signal, and the imaginary part exponent term corrects the phase lag of the signal. This enables the constructed spectrum to adjust the topology strength in real time according to the material state, eliminating spectrum mismatch caused by fluctuations in operating conditions.

[0022] Preferably, the spectral wavelet transform includes: Based on the complex correlation weights, a complex Laplacian matrix is ​​constructed and eigenvalue decomposition is performed to obtain the eigenvector matrix. ;use Transform the spectral signal to the spectral domain and then compare it with the low-pass kernel function. and high-pass kernel function The low-frequency smooth component and the high-frequency detail component are obtained by dot product and inverse transform decoupling. The specific calculation method is as follows:

[0023] In the formula, For component type index, retrieve value or These correspond to the low-frequency smoothing component and the high-frequency detail component, respectively. This is the diagonal matrix of the filter kernel function for the corresponding frequency band; The first obtained by decoupling the filter kernel function for the corresponding frequency band The signal matrix at each sampling time; This is the conjugate transpose of the eigenvector matrix; For the first The original signal tensor at each sampling time.

[0024] This invention utilizes graphical Fourier transform to project mixed signals from the spatial domain onto the graphical frequency domain. By designing specific filter kernel functions, it forcibly separates material signals, which exhibit spatial smoothness, from fault signals, which exhibit spatial abrupt changes, thus achieving signal-to-noise separation.

[0025] Preferably, the low-pass kernel function covers the slowly varying frequency range of material flow, and the high-pass kernel function covers the abrupt frequency range of structural fracture.

[0026] Preferably, the extraction of the energy features of the high-frequency detail components includes:

[0027] In the formula, For the first The energy characteristic values ​​of high-frequency detail components at each sampling time. This represents the total number of nodes in the knowledge graph. For the first The sampling time of the first sampling moment The high-frequency detail component values ​​of each node. This is the L2 norm symbol.

[0028] Preferably, the dynamic threshold is calculated as follows:

[0029] In the formula, For the first The dynamic threshold at each sampling time. The preset baseline threshold, This is the sensitivity adjustment coefficient. For the first The average real part of the viscous damping in all spatial directions at each sampling time.

[0030] Secondly, the present invention provides a grain screening fault diagnosis system based on knowledge graphs, which adopts the following technical solution: A knowledge graph-based grain screening fault diagnosis system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned knowledge graph-based grain screening fault diagnosis method is implemented.

[0031] By adopting the above technical solution, a computer program for the knowledge graph-based grain screening fault diagnosis method is generated and stored in a memory for loading and execution by a processor. This allows for the creation of a terminal device based on the memory and processor, facilitating its use.

[0032] The present invention has the following technical effects: This invention does not simply regard the influence of materials as noise. Instead, it analyzes the complex domain rheological impedance by decomposing and processing it. By utilizing the two orthogonal dimensions of the real part of viscous damping and the imaginary part of elastic reactance, it thoroughly distinguishes the two different signal sources of material viscosity change and energy consumption and structural damage and deformation from a physical perspective.

[0033] Furthermore, this invention constructs a dynamic physical knowledge graph containing complex weights. By introducing imaginary part weight correction, it solves the defect that traditional scalar graphs cannot characterize the phase lag of vibration waves, enabling the graph to adaptively adjust its topology to follow load changes and ensuring the robustness of feature extraction.

[0034] Furthermore, this invention utilizes the multi-scale characteristics of graph wavelet transform in the graph frequency domain, combined with an adaptive threshold mechanism, to achieve accurate discovery of weak fault signals masked by high-damping materials, effectively improving the accuracy of fault detection. Attached Figure Description

[0035] Figure 1 This is a flowchart of a method in a knowledge graph-based grain screening fault diagnosis method provided in an embodiment of the present invention; Figure 2 This is a timing diagram of viscous damping provided in an embodiment of the present invention; Figure 3 Vibration velocity timing diagram provided for embodiments of the present invention; Figure 4 A comparison chart of the dynamic threshold determination effect provided in the embodiments of the present invention. Detailed Implementation

[0036] This invention discloses a method for diagnosing grain screening faults based on knowledge graphs, referring to... Figure 1 This includes steps S1-S5: S1: Based on a precision clock synchronization protocol, the triaxial acceleration vibration signals of each node of the grain screening machine and the instantaneous power parameters of the motor are collected synchronously.

[0037] It should be noted that during the operation of a grain screening machine, the vibration wave transmitted from the excitation source to each monitoring point will generate a phase lag at the microsecond level. This phase difference contains information about the stiffness distribution of the screen structure. If a conventional independent clock acquisition method is used, the sampling time deviation between each sensor often reaches the millisecond level. This asynchronous error will confuse the true physical phase lag, causing distortion of the imaginary part of the subsequent complex impedance calculation, which can easily lead to misjudgment of changes in structural stiffness. Therefore, in order to accurately obtain phase information, a precision clock synchronization mechanism needs to be introduced.

[0038] Preferably, as an example, based on a precision clock synchronization protocol, the triaxial acceleration vibration signals of each node of the grain screening machine and the instantaneous power parameters of the motor are synchronously acquired, including: A distributed acquisition architecture based on the IEEE 1588V2 precise time protocol is adopted, utilizing data acquisition points deployed at the four corners of the screen box, the center of the excitation beam, and the spring supports. A high-frequency MEMS triaxial accelerometer and a smart meter at the motor drive end serve as the data acquisition devices for each node. The global sampling time is locked by hardware stamping, and the triaxial acceleration vibration signals of each node and the instantaneous power parameters of the motor are collected synchronously at the same time. The instantaneous power parameters include active power and reactive power.

[0039] The specific implementation method for locking the global sampling time using hardware stamping is as follows:

[0040] In the formula, For the first The synchronization error value of each node relative to the master clock; For the first The sampling time values ​​of each node; The reference time value for the master clock; For example, a preset microsecond-level error threshold is used. Pick Microseconds.

[0041] It is understandable that in the relational expression This value characterizes the system's time synchronization accuracy; the smaller the value, the higher the alignment of the sensors on the time axis. Less than This allows the error to be controlled at the microsecond level, ensuring that the phase difference of the vibration wave between different measurement points is determined solely by the physical transmission path, thus eliminating interference from asynchronous hardware sampling.

[0042] This directly verifies that the acquisition scheme accurately locks the phase reference through the hardware protocol, effectively eliminating the hidden danger of physical phase lag due to timing errors, which makes accurate identification impossible.

[0043] S2: Based on the triaxial acceleration vibration signal and the instantaneous power parameters, calculate the complex domain rheological impedance tensor; it includes the real part of viscous damping that characterizes the energy absorption of the material and the imaginary part of elastic reactance that characterizes the energy storage.

[0044] It should be noted that the grain particles on the screen surface are not rigid bodies, but rather a complex viscoelastic medium. This medium exhibits both fluid-like viscosity and spring-like elasticity during vibration. Analyzing solely using vibration amplitude would neglect the modulation effect of material elasticity on the system's natural frequency, making it impossible to distinguish between frequency drift caused by material accumulation and frequency drift caused by a decrease in structural stiffness. Therefore, to accurately describe the coupling state between the material and the screen body, a complex-domain rheological impedance must be introduced.

[0045] Preferably, as an example, based on the triaxial acceleration vibration signal and the instantaneous power parameter, a complex domain rheological impedance tensor is calculated; it includes a viscous damping real part characterizing the energy absorbed by the material and an imaginary elastic reactance characterizing the stored energy, including: Using the vibration velocity signal obtained by integrating the triaxial acceleration vibration signal and the instantaneous power parameter as inputs, the complex domain rheological impedance tensor is calculated hourly using the power flow conservation principle.

[0046] For the The sampling time of the first sampling moment The calculation methods for each component of the complex domain rheological impedance tensor in each spatial direction are as follows:

[0047]

[0048]

[0049] In the formula, For the first The sampling time of the first sampling moment The values ​​of complex impedance components in each spatial direction. The imaginary unit; For the first The sampling time of the first sampling moment The real part of the viscous damping in each spatial direction. For the first The sampling time of the first sampling moment The imaginary part of the elastic reactance in each spatial direction; For the first At the sampling time, the motor is in the... The active power component values ​​output in each spatial direction. For the first At the sampling time, the sieve body was at the... Vibration velocity modulus in one spatial direction; For the first At the sampling time, the motor is in the... The reactive power component values ​​output in each spatial direction.

[0050] It is understandable that the numerator of the relational formula contains This characterizes the energy rate at which the driving system does work against resistance; the larger this value, the more irreversible energy the system consumes; the denominator of the relational expression contains... It characterizes the kinetic energy level of the screen body; the larger the value, the more intense the vibration of the screen body. This describes the energy dissipation rate per unit kinetic energy. When A higher value indicates that, at the same vibration speed, the motor needs to output more active power to maintain the motion. This means that the viscous frictional resistance between the material and the screen surface has increased significantly, indicating that the state of the material has changed. For example, excessive moisture content may cause adhesion, or excessive feed rate may lead to increased layer thickness. This change is a normal fluctuation in operating conditions, rather than a mechanical failure.

[0051] In the numerator of the relational formula This characterizes the rate at which kinetic and potential energy are exchanged within a system. A larger absolute value indicates a more vigorous energy exchange within the system. The denominator of the relational expression contains... This value represents the kinetic energy level of the sieve body; the larger the value, the more intense the vibration of the sieve body. It describes the degree of kinetic-potential energy exchange mismatch, when A sudden change in the value indicates that, under the same vibration conditions, the reactive power required to maintain the vibration has changed. Since changes in material mass are usually gradual, this sudden change signifies a disruption in the system's energy storage structure, thus indicating a drift in the device's natural frequency. Specifically, if An abnormal increase often corresponds to a sudden decrease in spring stiffness or an abnormal loss of vibrating mass, such as component detachment. This change is a mechanical structural failure.

[0052] This intuitively verifies that by deconstructing a single vibration signal into a real part of energy consumption and an imaginary part of energy storage, orthogonal decoupling of material rheological properties and structural mechanical properties is achieved, thereby effectively solving the problem of misjudgment of faults caused by the elastic modulation of materials.

[0053] Figure 2 This is a time-series graph of viscous damping; the graph shows how the real part of the average viscous damping changes over time, reflecting the thickness and viscosity of the material on the screen surface, i.e., the load. The curve fluctuates little in the first half, but after the 10th second, the feed rate suddenly increases, and the curve shows a significant jump.

[0054] Figure 3 This is a vibration velocity time series diagram, showing the change of vibration velocity modulus over time. The gray background area in the diagram marks the time period of heavy-load, high-damping conditions, corresponding to the production stage with large material feed rate and high viscosity.

[0055] A comparison of the two figures above clearly shows that as the viscous damping value in the upper figure increases, the vibration velocity exhibits a significant downward trend, thus proving that under heavy load conditions, the decrease in vibration amplitude is a normal physical phenomenon, not a equipment malfunction. If judgment is based solely on vibration amplitude, traditional methods are prone to misinterpreting this load-induced vibration attenuation as a fault, or failing to detect weak fault signals when the overall vibration is suppressed. This provides a solid physical basis for introducing complex-domain rheological impedance for decoupling in this invention.

[0056] S3: Construct a dynamic physical knowledge graph and map the complex domain rheological impedance tensor to complex correlation weights between nodes; the complex correlation weights use the real part of the viscous damping to quantize amplitude attenuation and the imaginary part of the elastic reactance to quantize phase lag.

[0057] It should be noted that traditional fault diagnosis maps typically construct a static topology based solely on the sensor's geometric location or a data topology based solely on signal statistical correlation, neglecting the physical attenuation characteristics of vibration waves propagating through the medium. For example, when the load changes, both the attenuation rate and wave velocity of the vibration wave will change. If the map weights are not adjusted accordingly, the algorithm may incorrectly classify signal attenuation caused by material thickening as sensor failure or structural fracture. Therefore, to make the map possess temporal dynamics, a dynamic physical knowledge graph needs to be constructed.

[0058] Preferably, as an example, a dynamic physical knowledge graph is constructed, mapping the complex domain rheological impedance tensor to complex correlation weights between nodes; the complex correlation weights quantize amplitude attenuation using the real part of the viscous damping and quantize phase hysteresis using the imaginary part of the elastic reactance, including: Based on the mechanical structure of the screening machine and the connection relationships of the mechanical structure, the entity nodes and edge connections of the graph are defined. The connection strength between nodes is dynamically calculated using the complex domain rheological impedance tensor as a correction factor, thereby generating complex correlation weights.

[0059] The specific calculation method for the complex correlation weight is as follows:

[0060] In the formula, For the first Each sampling time node With nodes The complex correlation weight values ​​between them The baseline weight values ​​are based on geometric location. For nodes With nodes The physical Euclidean distance between them. The preset amplitude attenuation constant, The preset wave velocity phase constant, The imaginary unit, For the first The sampling time of the first sampling moment The imaginary part of the elastic reactance in each spatial direction. For the first The sampling time of the first sampling moment The real part of the viscous damping in each spatial direction. It is the imaginary unit.

[0061] Understandable This value characterizes the energy consumption characteristics of the material at the current moment. The larger the value, the more viscous the material or the more intense the internal friction. This characterizes the propagation path length of the vibration wave. A larger value indicates a greater distance between nodes and a stronger cumulative effect of signal attenuation. The larger the value, The smaller the value, the smaller the calculated complex correlation weight. When the screening machine is fully loaded with wet grain, the more viscous the material or the more intense the internal friction, the greater the signal strength between nodes will be attenuated. This can easily lead to misjudgment of equipment failure. Therefore, this invention actively lowers the expected signal strength by reducing the connection weight between nodes, thereby effectively reducing misjudgment of equipment failure.

[0062] It characterizes the energy storage properties of the system, when When the value changes, for example, an increase in the feed rate leads to an increase in the equivalent mass, the phase angle in the exponential term changes. The change indicates that the signal originates from the node. Transmitted to node The arrival time has shifted. When drastic fluctuations in operating conditions cause changes in wave velocity, traditional graph algorithms cannot extract features due to phase misalignment. Therefore, this invention utilizes the imaginary part... The real-time rotating weight vector maintains phase synchronization between the spectrum and physical fluctuations, ensuring accurate capture of true structural fault signals under any load.

[0063] S4: Perform a graph wavelet transform on the dynamic physical knowledge graph of the loaded triaxial acceleration vibration signal, and use a multi-scale filter bank to decouple the low-frequency smooth component of the response material load and the high-frequency detail component of the response mechanical structure.

[0064] It should be noted that signal fluctuations caused by material flow are typically spatially distributed, low-frequency, and gradually changing, while signal distortions caused by mechanical faults are typically localized, high-frequency, and abrupt. Conventional time-frequency analysis struggles to separate these two in mixed signals, easily leading to material signals overshadowing fault characteristics. Therefore, to achieve accurate fault feature extraction, spectral wavelet transform technology needs to be introduced.

[0065] Preferably, as an example, a graph wavelet transform is performed on the dynamic physical knowledge graph of the loaded triaxial acceleration vibration signal, and a multi-scale filter bank is used to decouple the low-frequency smooth component of the response to the material load and the high-frequency detail component of the response to the mechanical structure, including: Based on the complex correlation weights, a complex Laplacian matrix is ​​constructed and eigenvalue decomposition is performed to obtain the eigenvector matrix. ; Using low-pass and high-pass kernel functions, the low-frequency smoothing component and the high-frequency detail component are obtained through inverse transform decoupling. The specific calculation method is as follows:

[0066] In the formula, For component type index, retrieve value or These correspond to the low-frequency smoothing component and the high-frequency detail component, respectively. This is the diagonal matrix of the filter kernel function for the corresponding frequency band; The first obtained by decoupling the filter kernel function for the corresponding frequency band The signal matrix at each sampling time; This is the conjugate transpose of the eigenvector matrix; For the first The original signal tensor at each sampling time.

[0067] Understandable The graphical Fourier transform converts spatial vibration signals into spectral coefficients at graphical frequencies. For wavelet scale selection, when At this time, the kernel function assigns high weights to small eigenvalues, physically preserving the spatial smoothing fluctuations caused by materials. At that time, the kernel function assigns high weights to large eigenvalues, which physically preserves the spatial abrupt fluctuations caused by the fracture of the equipment structure; To perform filtering on the paving coefficients, The inverse Fourier transform of the graph restores the filtered spectral coefficients back to the original sensor signal space.

[0068] This intuitively verifies that the separation operation in the frequency domain accurately decouples the signal characteristics of different physical sources, effectively eliminating the hidden danger of material background noise masking weak fault characteristics.

[0069] S5: Extract the energy characteristics of the high-frequency detail components, and determine mechanical faults when they exceed the dynamic threshold.

[0070] It should be noted that after wavelet decoupling, the high-frequency components mainly contain vibrational energy caused by structural discontinuities, but a small amount of high-frequency noise caused by high-viscosity materials may still remain. If a fixed threshold is used, the actual fault signal will be significantly suppressed when the material is extremely viscous, resulting in a signal strength lower than the fixed threshold and missed detection. Therefore, to ensure the robustness of the diagnosis, energy features need to be extracted and used in conjunction with a dynamic threshold for judgment.

[0071] Preferably, as an example, extracting the energy characteristics of the high-frequency detail components and determining a mechanical fault when they exceed a dynamic threshold includes: Using the high-frequency detail components as input, the energy characteristics of each node at each time moment are calculated and compared with a dynamic threshold. If the energy characteristics are greater than the dynamic threshold, it is determined to be a mechanical fault.

[0072] The energy characteristics are calculated as follows:

[0073] In the formula, For the first The energy characteristic values ​​of high-frequency detail components at each sampling time. This represents the total number of nodes in the knowledge graph. For the first The sampling time of the first sampling moment The high-frequency detail component values ​​of each node. This is the L2 norm symbol.

[0074] The dynamic threshold is calculated as follows:

[0075] In the formula, For the first The dynamic threshold at each sampling time; This is a preset baseline threshold. This is the sensitivity adjustment coefficient; For the first The average real part of the viscous damping in all spatial directions at each sampling time.

[0076] Understandably, in the formula It represents the pure fault impact energy after removing the material background; the larger the value, the more severe the structural damage.

[0077] In relational expressions This characterizes the judgment threshold when the average damping of the material is detected. As the threshold increases, the denominator increases, and the threshold value increases. Automatically lower.

[0078] This intuitively verifies that the algorithm, by establishing a negative correlation mechanism between threshold and damping, accurately matches the fault sensitivity under different operating conditions, effectively eliminating the hidden danger of missing weak faults under heavy load and high damping conditions.

[0079] Figure 4 This is a comparison chart of the effects of dynamic threshold determination. The solid green line represents the energy characteristics of high-frequency detail components, the dashed red line represents the dynamic threshold, and the gray dotted line represents the preset baseline threshold, which represents the fixed alarm line set in traditional monitoring methods. The black arrows indicate the changes in the threshold when damping increases, the time of fault occurrence, and the missed detection situation of traditional methods.

[0080] By observing the red dashed line, it can be seen that after entering the heavy load condition at the 10th second, the dynamic threshold automatically and significantly decreases as the damping increases. When the fault occurs at the 15th second, due to the masking effect of high damping, the absolute value of the fault characteristic is actually not high, and is significantly lower than the traditional fixed threshold in gray. However, since the red dynamic threshold of the present invention has been reduced in advance, the green curve successfully breaks through the red dashed line, thereby triggering the alarm. This intuitively proves that the present invention has extremely high detection sensitivity and robustness under complex working conditions.

[0081] This invention also discloses a knowledge graph-based grain screening fault diagnosis system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a knowledge graph-based grain screening fault diagnosis method according to this invention.

[0082] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0083] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (DRAM), high-bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

Claims

1. A method for diagnosing grain screening faults based on knowledge graphs, characterized in that, Including the following steps: Based on a precision clock synchronization protocol, the triaxial acceleration vibration signals of each node of the grain screening machine and the instantaneous power parameters of the motor are collected synchronously. Based on the triaxial acceleration vibration signal and the instantaneous power parameters, the complex domain rheological impedance tensor is calculated; It includes the real part of the viscous damping that characterizes the energy absorption of the material and the imaginary part of the elastic reactance that characterizes the energy storage. A dynamic physical knowledge graph is constructed, and the complex domain rheological impedance tensor is mapped to complex correlation weights between nodes; the complex correlation weights use the real part of the viscous damping to characterize the amplitude attenuation and the imaginary part of the elastic reactance to characterize the phase hysteresis. A graph wavelet transform is performed on the dynamic physical knowledge graph loaded with the triaxial acceleration vibration signal, and a multi-scale filter bank is used to decouple the low-frequency smooth component of the response material load and the high-frequency detail component of the response mechanical structure. The energy characteristics of the high-frequency detail components are extracted, and mechanical faults are determined when they exceed a dynamic threshold.

2. The method for diagnosing grain screening faults based on knowledge graphs according to claim 1, characterized in that, The method of synchronously acquiring triaxial acceleration vibration signals of each node of the grain screening machine based on a precision clock synchronization protocol specifically includes: The clock of the sensor array deployed at the four corners of the screen box and the excitation beam is locked using a protocol to ensure that the phase synchronization error of each channel is less than a preset microsecond threshold. By aggregating the sampling data from various measuring points at the same time, a third-order vibration state tensor with node, spatial, and temporal dimensions is constructed.

3. The method for diagnosing grain screening faults based on knowledge graphs according to claim 1, characterized in that, The components of the complex domain rheological impedance tensor satisfy the following relationship: In the formula, For the first The sampling time of the first sampling moment The values ​​of complex impedance components in each spatial direction. The imaginary unit; For the first The sampling time of the first sampling moment The real part of the viscous damping in each spatial direction. For the first The sampling time of the first sampling moment The imaginary part of the elastic reactance in each spatial direction; The real part of the viscous damping satisfies the following relationship: In the formula, For the first At the sampling time, the motor is in the... The active power component values ​​output in each spatial direction. For the first At the sampling time, the sieve body was at the... Vibration velocity modulus in a spatial direction.

4. The method for diagnosing grain screening faults based on knowledge graphs according to claim 3, characterized in that, The imaginary part of the elastic reactance satisfies the following relationship: In the formula, For the first At the sampling time, the motor is in the... The reactive power component values ​​output in each spatial direction.

5. The method for diagnosing grain screening faults based on knowledge graphs according to claim 1, characterized in that, The complex correlation weights satisfy the following relation: In the formula, For the first Each sampling time node With nodes The complex correlation weight values ​​between them The baseline weight values ​​are based on geometric location. For nodes With nodes The physical Euclidean distance between them. The preset amplitude attenuation constant, The preset wave velocity phase constant, The imaginary unit, For the first The sampling time of the first sampling moment The imaginary part of the elastic reactance in each spatial direction. For the first The sampling time of the first sampling moment The real part of the viscous damping in each spatial direction. It is the imaginary unit.

6. The method for diagnosing grain screening faults based on knowledge graphs according to claim 1, characterized in that, The graph wavelet transform includes: Based on the complex correlation weights, a complex Laplacian matrix is ​​constructed and eigenvalue decomposition is performed to obtain the eigenvector matrix. ;use Transform the spectral signal to the spectral domain and then compare it with the low-pass kernel function. and high-pass kernel function The low-frequency smooth component and the high-frequency detail component are obtained by dot product and inverse transform decoupling. The specific calculation method is as follows: In the formula, For component type index, retrieve value or These correspond to the low-frequency smoothing component and the high-frequency detail component, respectively. This is the diagonal matrix of the filter kernel function for the corresponding frequency band; The first obtained by decoupling the filter kernel function for the corresponding frequency band The signal matrix at each sampling time; This is the conjugate transpose of the eigenvector matrix; For the first The original signal tensor at each sampling time.

7. The method for diagnosing grain screening faults based on knowledge graphs according to claim 1, characterized in that, The low-pass kernel function covers the slowly varying frequency range of material flow, while the high-pass kernel function covers the abrupt frequency range of structural fracture.

8. The method for diagnosing grain screening faults based on knowledge graphs according to claim 1, characterized in that, The extraction of energy features from the high-frequency detail components includes: In the formula, For the first The energy characteristic values ​​of high-frequency detail components at each sampling time. This represents the total number of nodes in the knowledge graph. For the first The sampling time of the first sampling moment The high-frequency detail component values ​​of each node. This is the L2 norm symbol.

9. The method for diagnosing grain screening faults based on knowledge graphs according to claim 1, characterized in that, The dynamic threshold is calculated as follows: In the formula, For the first The dynamic threshold at each sampling time. The preset baseline threshold, This is the sensitivity adjustment coefficient. For the first The average real part of the viscous damping in all spatial directions at each sampling time.

10. A grain screening fault diagnosis system based on knowledge graph, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a knowledge graph-based grain screening fault diagnosis method according to any one of claims 1-9.

Images