Intelligent diagnosis method, device and storage medium for couplings based on spectral imaging

Abstract

The application discloses a coupling intelligent diagnosis method and device based on spectral imaging and a storage medium, and relates to the technical field of data processing.The steps comprise the following steps: a phase-locked cursor acquisition environment is established, and a quasi-static spectrum image is obtained by eliminating ambiguity in cooperation with a crossed polarized light path; a manifold mapping sparse spectrum projection operator is used to extract transient health characteristics through single-step linear operation; a driving adaptive inertial memory recursive state model is constructed, and dynamic benchmarks and nonlinear residuals are calculated to determine coupling transient fracture and gradual misalignment faults.The application effectively solves the problems of visual standing wave blind area, high-speed motion blur and spectral aliasing in traditional frequency flash sampling, and realizes real-time and accurate diagnosis under the condition of low computing power at the edge.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method, device, and storage medium for intelligent diagnosis of couplings based on spectral imaging. Background Technology

[0002] In the field of intelligent operation and maintenance of rotating machinery, couplings, as the critical components of power transmission, directly affect the safety of the entire transmission chain. Currently, the industry mainly relies on contact vibration sensors for fault diagnosis. However, this traditional method based on mechanical response faces insurmountable physical barriers in practical applications: First, the invasiveness of "contact measurement" means that deploying sensors directly on rotating components for high-speed couplings requires expensive conductive slip rings or wireless telemetry systems; second, the physical field hysteresis of fault evolution means that vibration signals are often the macroscopic consequences of structural damage accumulating to a certain extent, rather than the microscopic causes, resulting in a significant delay in the warning window; and finally, the lack of spatial resolution due to single-point integration means that the output of a single-point sensor is the time-domain integral of the total field energy, which cannot accurately locate the specific topological position of the fault, and for high-damping non-metallic materials (such as rubber), their strong attenuation characteristics easily mask weak early fault signals.

[0003] To overcome the limitations of contact-based measurements, non-contact monitoring technologies based on optical field sensing are increasingly being introduced into coupling diagnostics. Existing technologies typically utilize infrared thermal imaging or multispectral cameras to attempt to infer the health status by capturing the temperature field distribution or spectral reflectance changes on the coupling surface. Compared to vibration sensors, spectral imaging technology has a natural WYSIWYG advantage: it can achieve full-field monitoring in a non-contact manner. Theoretically, it can detect abnormal temperature rises caused by friction (earlier than vibration) or changes in the chemical composition of the material surface (such as the spectral fingerprints of rubber aging and metal corrosion), enabling early warning at the nascent stage of a fault, thus compensating for the shortcomings of vibration analysis in terms of sensitivity and spatial location.

[0004] Despite the immense potential of spectral imaging, its direct application to the diagnosis of couplings on high-speed, high-brightness metal surfaces still presents certain technical challenges. First, the exposure and readout mechanisms of existing industrial cameras are ill-suited for high-speed rotating targets, causing micron-level crack textures to be smoothed out within the integration time, resulting in severe spectral aliasing. Second, the low emissivity and high reflectivity of bright metal surfaces easily reflect ambient thermal radiation as false hotspots, masking the true diffuse reflection damage texture. Finally, hyperspectral data cubes face the curse of dimensionality; existing embedded edge computing devices struggle to complete dimensionality reduction and feature extraction of high-dimensional data within milliseconds, leading to fatal delays in the system's response to transient faults (such as brittle fracture of diaphragms).

[0005] Therefore, the present invention provides a method, device and storage medium for intelligent diagnosis of couplings based on spectral imaging. Summary of the Invention

[0006] The purpose of this invention is to provide a method, device, and storage medium for intelligent diagnosis of couplings based on spectral imaging, so as to solve the existing problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a coupling intelligent diagnostic method based on spectral imaging, comprising the following steps: Step S1: Establish a phase-locked vernier acquisition environment, monitor the rotational speed and frequency of the coupling under test in real time, and construct a phase-locked trigger signal to drive a multispectral stroboscopic light source and cooperate with an orthogonal polarization imaging optical path to perform non-continuous discrete-time sampling on the high-speed rotating coupling, thereby obtaining a quasi-static spectral image sequence. Step S2: Pre-set a linear transformation matrix based on offline big data computation in the edge computing device. The matrix is ​​used to directly map high-dimensional raw spectral data to a fault-sensitive low-dimensional feature subspace to obtain a manifold mapping sparse spectral projection operator. Step S3: Input the single-frame quasi-static spectral image data into the manifold mapping sparse spectral projection operator, and output the transient health state feature vector containing texture entropy components and heat distribution components through single-step linear operation; Step S4: Construct a driving adaptive inertial memory recursive state model, configured to input the transient health state feature vector into a recursive model with a time memory decay mechanism, and calculate the dynamic reference vector at the current moment and the nonlinear residual vector relative to the reference. Step S5: Based on the amplitude mutation characteristics and time-series accumulation characteristics of the nonlinear residual vector, determine the transient fracture fault and the gradual misalignment fault of the coupling.

[0008] A further improvement of the present invention is that step S1 specifically includes the following steps: S11. Obtain the real-time rotational frequency of the coupling; and set the number of phase slices, wherein the number of phase slices is selected as a positive integer in the Fibonacci sequence to construct a non-harmonic overlapping sampling interval; S12. The vernier strobe frequency is obtained by dividing the number of phase slices by the sum of that number and one, and then multiplying by the real-time rotation frequency. S13. The acquisition is triggered by the stroboscopic frequency of the vernier, so that the acquired quasi-static spectral image sequence is presented in the time domain as a super-resolution reconstructed image that rotates slowly according to a preset slip rate.

[0009] A further improvement of this invention lies in the process of obtaining the manifold mapping sparse spectral projection operator: In the offline phase, full-band spectral data containing several types of faults are acquired and fed into the spectral pre-screening strategy to obtain a set of high-confidence candidate bands. The sparse regression algorithm is applied to the high-confidence candidate band set for topological feature selection, and the sparse band combination that is most sensitive to the microcrack texture and abnormal temperature rise on the metal surface is selected. Construct the optimal projection relationship from the full-band space to the low-dimensional manifold space defined by the sparse band combination, obtain the optimal weight matrix by the least squares method, and solidify the weight matrix into the manifold mapping sparse spectral projection operator.

[0010] A further improvement of the present invention is that the spectral pre-screening strategy includes calculating the spectral information entropy of each band in the entire band, eliminating invalid bands whose information entropy is lower than a preset noise threshold; calculating the spectral cross-correlation matrix of the remaining bands, identifying and eliminating collinear bands whose correlation coefficient is greater than a preset redundancy threshold, thereby generating a set of high-confidence candidate bands.

[0011] A further improvement of this invention is that the process of obtaining the transient health state feature vector includes: performing matrix multiplication on the collected image pixel gray value vector and the manifold mapping sparse spectral projection operator; and outputting the result vector, wherein the first dimension is defined as the spectral texture entropy index, used to characterize the physical integrity of the material surface; and the second dimension is defined as the thermal field coupling strength index, used to characterize the degree of abnormal aggregation of the local temperature field.

[0012] A further improvement of this invention is that the construction process of the driving adaptive inertial memory recursive state model includes: Step S41: Perform cold reference clamp initialization, construct a two-dimensional state space containing spectral texture entropy components and thermal field coupling strength components, collect multiple frames of transient health state feature vectors within the preset stable window at the initial stage of coupling startup and calculate their arithmetic mean, and solidify the arithmetic mean as the initial dynamic reference vector at time zero. Step S42: Construct a fault evolution memory decay matrix and establish a two-dimensional diagonal matrix as the system inertia operator. The first diagonal element corresponding to the spectral texture entropy component is set as the texture inertia factor, and its value is preset to a minimum value based on the signal-to-noise ratio of the imaging system to suppress high-frequency noise. The second diagonal element corresponding to the thermal field coupling intensity component is set as the thermal inertia factor, and its value is preset based on the specific heat capacity and thermal conductivity of the coupling metal material to simulate the hysteresis effect of physical heat conduction. Step S43: Update the dynamic reference vector in real time according to the discrete time step. The dynamic reference vector at the current time is composed of the transient health state feature vector at the previous time multiplied by the fault evolution memory decay matrix and the difference matrix between the dynamic reference vector at the previous time multiplied by the identity matrix and the fault evolution memory decay matrix. Step S44: Calculate the difference vector between the transient health state feature vector at the current moment and the updated dynamic reference vector at the current moment, and take the absolute value of each component of the difference vector to obtain the nonlinear residual vector used for final fault determination.

[0013] A further improvement of the present invention is that step S5 specifically includes the following steps: Step S51: Extract the spectral texture entropy exponential residual component from the nonlinear residual vector; Step S52: If the amplitude of the residual component of the spectral texture entropy index in a single frame is greater than the preset brittle fracture threshold, it is determined that the coupling has experienced diaphragm breakage or surface peeling, and an instantaneous stop command is generated. Step S53: Extract the thermal field coupling intensity exponential residual component from the nonlinear residual vector; Step S54: Perform time-domain integration on the residual component of the thermal field coupling strength index. If the integral value is greater than the preset thermal accumulation failure threshold, it is determined that the coupling has misalignment or lubrication failure, and a maintenance warning signal is generated.

[0014] A further improvement of the present invention is that the method further includes an anti-interference step based on decoupling of virtual and real features: When the increase in the thermal coupling strength index within a preset time window exceeds a preset increase threshold, the spectral texture entropy index under the same spatial coordinates is invoked. If the spectral texture entropy index falls entirely within the spectral texture entropy index benchmark range within the preset time window, then the current thermal anomaly is determined to be artifact noise caused by environmental thermal radiation reflection, and the process returns to step S4 to suppress the weight of the region in the adaptive inertial memory recursive state model.

[0015] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described intelligent diagnostic method for couplings based on spectral imaging.

[0016] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned intelligent diagnostic method for couplings based on spectral imaging.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention first constructs a non-harmonic overlapping sampling interval by using vernier phase-locked acquisition, which solves the problems of visual standing wave blind zone and severe motion blur and spectral aliasing caused by high-speed rotation caused by traditional stroboscopic sampling at specific rotation speeds. It realizes the super-resolution quasi-static reconstruction of the entire circumferential surface of the coupling without the need for expensive high-speed cameras, which significantly improves the spatial imaging clarity of microcracks.

[0018] 2. By using the manifold mapping sparse spectral projection operator, the complex nonlinear feature extraction process based on big data is pre-fixed into the optimal linear transformation matrix for offline solution. This solves the contradiction between the curse of dimensionality of hyperspectral data and the limited computing power of embedded edge computing devices, ensuring that the system can capture transient faults such as diaphragm rupture.

[0019] 3. By combining orthogonal polarization optical paths with virtual and real feature decoupling and anti-interference steps, and utilizing the physical differences between specular reflection and diffuse reflection polarization on metal surfaces, as well as the multi-mode interlocking logic that thermal anomalies must be accompanied by texture anomalies, the problem of photothermal coupling deception caused by the low emissivity and high reflectivity characteristics of bright metal coupling surfaces (such as false alarms of reflected heat from bypass steam pipes) is solved. This achieves accurate locking of real physical damage with zero false alarms in complex industrial photothermal environments.

[0020] 4. By driving an adaptive inertial memory recursive state model, and setting a physical inertial factor based on the thermal properties of the coupling material and the signal-to-noise ratio of the imaging system, a deterministic recursive system with dynamic benchmark update capability is constructed, which solves the problems that traditional static thresholds cannot adapt to natural temperature drift under working conditions and lack the ability to distinguish between gradual and sudden faults. Attached Figure Description

[0021] Figure 1 This is a flowchart of a coupling intelligent diagnostic method based on spectral imaging according to the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0023] The term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0024] Example 1 Figure 1This embodiment presents a flowchart of a coupling intelligent diagnostic method based on spectral imaging, the steps of which are as follows: Step S1: Establish a phase-locked vernier acquisition environment; monitor the rotational speed and frequency of the coupling under test in real time, and construct a phase-locked trigger signal based on the vernier effect principle; According to the Fresnel reflection principle, specular reflection from a metal surface maintains its original polarization state of 0° and is completely blocked by a 90° analyzer; however, diffuse reflection from a damaged area will depolarize and pass smoothly through the lens. Therefore, this trigger signal is used to drive a multispectral stroboscopic light source and, in conjunction with an orthogonal polarization imaging optical path, a polarizer with a polarization direction of 0° is installed at the multispectral light source, and an analyzer with a polarization direction of 90° is installed in front of the camera lens. Discontinuous discrete-time sampling is performed on the high-speed rotating coupling to obtain a quasi-static spectral image sequence that eliminates motion blur and filters out specular reflection interference. Specifically, the following steps are included: S11. Obtain the real-time rotational frequency of the coupling. (e.g., 100Hz); and set the number of phase slices, which is selected as a positive integer in the Fibonacci sequence, to construct a non-harmonic overlapping sampling interval; S12. The vernier strobe frequency is obtained by dividing the number of phase slices by the sum of that number and one, and then multiplying by the real-time rotation frequency. S13. The acquisition is triggered by the stroboscopic frequency of the vernier, so that the acquired quasi-static spectral image sequence is presented in the time domain as a super-resolution reconstructed image that rotates slowly according to a preset slip rate.

[0025] This embodiment selects the Fibonacci sequence as the number of phase slices, utilizing its golden ratio approximation property to create a non-harmonic overlap relationship between the sampling frequency and the rotation frequency. This avoids the "visual standing wave" or "sampling dead point" that may occur in conventional stroboscopic sampling at specific rotation speeds, ensuring that the reconstructed image sequence can cover the entire circumferential surface of the coupling in an ergodic manner, achieving full-field-of-view super-resolution reconstruction without dead angles.

[0026] In some possible embodiments, the coupling rotates at a speed during the startup phase. Rapidly changing, fixed-frequency flickering can cause image synchronization issues. Edge computing devices read the encoder rotation speed in real time and dynamically adjust the vernier flicker frequency. Even during acceleration from 0 to 10,000 rpm, the system maintains... The locking relationship is established. Regardless of the speed fluctuations, the acquired images always maintain a stable visual effect of slow rotation, ensuring that fault monitoring during critical start-up and shutdown phases is not lost.

[0027] Step S2: Pre-set a linear transformation matrix based on offline big data computation in the edge computing device. The matrix is ​​used to directly map high-dimensional raw spectral data to a fault-sensitive low-dimensional feature subspace to obtain a manifold mapping sparse spectral projection operator. The process of obtaining the manifold mapping sparse spectral projection operator: During the offline phase, full-band spectral data containing several fault types is acquired and fed into a spectral pre-screening strategy to obtain a set of high-confidence candidate bands. The spectral pre-screening strategy includes calculating the spectral information entropy of each band in the full band. This embodiment eliminates invalid bands with information entropy below a preset noise threshold, avoiding wasting computational resources on bands with no or redundant information and addressing the "curse of dimensionality" in spectral data. For the remaining bands, it calculates the spectral cross-correlation matrix, identifies and eliminates collinear bands with correlation coefficients greater than a preset redundancy threshold, thereby generating a high-confidence candidate band set. This embodiment introduces a spectral information entropy elimination mechanism to automatically filter out low-information bands caused by dark current noise or overexposure, preventing them from interfering with subsequent model construction. It also introduces a spectral cross-correlation matrix deduplication mechanism to decouple collinearity between bands, preventing overfitting or weight oscillations in the regression model due to high feature correlation, ensuring the numerical stability of the generated projection operator under harsh conditions.

[0028] The sparse regression algorithm is applied to the high-confidence candidate band set for topological feature selection, and the sparse band combination that is most sensitive to the microcrack texture and abnormal temperature rise on the metal surface is selected. Construct the optimal projection relationship from the full-band space to the low-dimensional manifold space defined by the sparse band combination, obtain the optimal weight matrix by the least squares method, and solidify the weight matrix into the manifold mapping sparse spectral projection operator.

[0029] This embodiment achieves physical-level compression of data dimensions by using sparse regression screening in the offline stage to eliminate redundant bands and retain only the physical characteristic bands most sensitive to faults (such as the bands corresponding to iron oxide spectral peaks). By solidifying the complex manifold mapping relationship into an optimal weight matrix, the online calculation process possesses strict mathematical determinism, avoiding the randomness risks and wasted computational resources associated with online neural network training, and significantly improving the industrial reliability of the system.

[0030] Step S3: Calculate the transient health state feature vector: Input the single-frame quasi-static spectral image data into the manifold mapping sparse spectral projection operator, and output the transient health state feature vector containing the texture entropy component and the heat distribution component through a single-step linear operation; The process of obtaining the transient health state feature vector includes: performing matrix multiplication on the collected image pixel gray value vector and the manifold mapping sparse spectral projection operator; and outputting the result vector, wherein the first dimension is defined as the spectral texture entropy index, which is used to characterize the physical integrity of the material surface; and the second dimension is defined as the thermal field coupling strength index, which is used to characterize the degree of abnormal aggregation of the local temperature field.

[0031] Edge device execution This reduces the computational complexity of online computing from Reduced to Matrix multiplication enables embedded chips (such as STM32 or FPGA) to complete feature extraction in microseconds.

[0032] By outputting a two-dimensional feature vector through matrix multiplication, the health status of the coupling is clearly separated into two orthogonal dimensions: physical integrity (texture entropy) and thermodynamic state (thermal coupling). This allows the system to accurately locate the root cause of the fault (e.g., whether it is material damage or installation deviation), avoiding the shortcomings of vague alarms from single indicators. Moreover, the calculation process only involves linear algebra operations, resulting in a fast response speed.

[0033] In some possible embodiments, harsh operating conditions exist where oil mist is present. The oil mist alters the optical path transmission characteristics, leading to an overall decrease in spectral intensity across the entire band. The thermal coupling intensity index is calculated using normalization (e.g., the green channel / red channel ratio), rather than absolute intensity. Simultaneously, a pre-screening strategy automatically removes bands severely absorbed by oil mist, retaining only characteristic bands with strong penetration (e.g., specific near-infrared windows). Even if the lens is slightly contaminated with oil or the environment contains oil mist, the diagnostic results remain robust.

[0034] Step S4: Construct a driving adaptive inertial memory recursive state model, configured to input the transient health state feature vector into a recursive model with a time memory decay mechanism, and calculate the dynamic reference vector at the current moment and the nonlinear residual vector relative to the reference. The construction process of the driving adaptive inertial memory recursive state model includes: Step S41: Perform cold reference clamp initialization, construct a two-dimensional state space containing spectral texture entropy components and thermal field coupling strength components. Within the preset stable window at the initial stage of coupling startup, collect multiple frames of transient health state feature vectors and calculate their arithmetic mean. Solidify the arithmetic mean as the initial dynamic reference vector at time zero to eliminate the uncertainty caused by random initialization. Step S42: Construct the fault evolution memory decay matrix A two-dimensional diagonal matrix is ​​established as the system inertia operator. The first diagonal element corresponding to the spectral texture entropy component is set as the texture inertia factor. The surface texture (physical integrity) of the coupling is extremely stable under normal conditions; setting a very small factor implies that the model has great inertia, "refusing" to believe that the texture will abruptly change. Therefore, its value is preset to a minimum based on the signal-to-noise ratio of the imaging system to suppress high-frequency noise. The second diagonal element corresponding to the thermal field coupling strength component is set as the thermal inertia factor. Based on the specific heat capacity of the metal, temperature changes are lag-dependent. A moderate factor is set to simulate the physical heat conduction process; therefore, its value is preset based on the specific heat capacity and thermal conductivity of the coupling's metal material to simulate the lag effect of physical heat conduction. Using the texture inertia factor, once the diaphragm breaks, the true value... Dramatic changes, and model values Being "dragged" by inertia, the difference between the two (residual) widens instantly, enabling the keen detection of sudden faults.

[0035] By filtering out instantaneous electromagnetic interference noise through thermal inertia factors, it only responds to temperature trends that conform to physical laws.

[0036] Step S43: Update the dynamic reference vector in real time according to the discrete time step. The dynamic reference vector at the current moment is composed of two linearly superimposed parts, including the transient health state feature vector of the previous moment multiplied by the fault evolution memory decay matrix and the difference matrix of the dynamic reference vector of the previous moment multiplied by the identity matrix and the fault evolution memory decay matrix. The formula is expressed as follows: ; It is the current theoretical benchmark value, i.e., the dynamic benchmark, which represents the system's prediction of the coupling's current state based on past experience; It was the true state of things a moment ago. This represents the theoretical baseline value at the previous moment; Represents the identity matrix; When texture inertia factor When set to the minimum value, because the coupling is made of metal, its surface texture (whether it has cracks) is extremely stable under normal circumstances and hardly changes; because Very small, the formula is mainly composed of Dominant; at this point, the model is very "stubborn," even Because the reflection caused the data to flicker, the model didn't believe it and thought the baseline shouldn't change. But if If the model remains at the old value due to a sudden change caused by a break, the difference between the two (residual) will increase instantly, thus triggering an alarm.

[0037] When the thermal inertia factor is set to a moderate value, the temperature change is lag-dependent due to the specific heat capacity of the metal, but it will still drift slowly with the environment or operating conditions (temperature drift). The temperature will rise, allowing the model to absorb new data appropriately; if the ambient temperature rises slowly, the model will gradually catch up with this pace through iteration, acknowledging that this is the normal baseline, thereby avoiding false alarms.

[0038] Step S44: Calculate the difference vector between the transient health state feature vector at the current moment and the updated dynamic reference vector at the current moment, and take the absolute value of each component of the difference vector to obtain the nonlinear residual vector used for final fault determination. The formula is expressed as follows: ; This embodiment eliminates the zero-point drift of the sensor itself by cold-state reference clamping and establishes a unified comparison benchmark.

[0039] The inertial operator, set based on specific heat capacity and signal-to-noise ratio, essentially simulates the physical heat conduction hysteresis effect of the coupling in the algorithm. This gives the model a natural low-pass filtering characteristic, which can automatically filter out high-frequency electromagnetic interference and at the same time keenly capture abnormal changes that violate physical inertia (such as diaphragm rupture), thus realizing anomaly detection based on physical essence.

[0040] In online diagnostic scenarios for couplings, deep learning is often used in the field for accurate diagnosis. However, in the application scenario of this embodiment, a "white-box" mechanism model has advantages over the "black-box" approach of deep learning, as shown in the following aspects: Computing power and real-time limitations: Embedded edge computing devices struggle to complete tasks within milliseconds, while deep learning models (such as LSTM) have high inference time at the edge. However, the recursive formula of this model only involves a few multiplication and addition operations, which can be completed in microseconds, perfectly solving the latency problem.

[0041] The problem of small sample size: Diaphragm breakage in couplings is an extremely low-probability event, making it difficult to collect enough fault samples in industrial settings to train a deep learning model. However, the model in this embodiment is based on physical rules (residual mutations), which can directly detect anomalies without the need for fault sample training.

[0042] Explainability and Trustworthiness: Deep learning is difficult to explain, but in this embodiment, by driving an adaptive inertial memory recursive state model, the reason can be directly pointed out as: "Because the current value deviates from the theoretical value based on thermal inertia by more than the threshold."

[0043] Step S5: Based on the amplitude abrupt change characteristics and time-series accumulation characteristics of the nonlinear residual vector, determine whether the coupling experiences transient breakage or gradual misalignment. Specific steps include: Step S51: Extract the spectral texture entropy exponential residual component from the nonlinear residual vector; Step S52: If the amplitude of the residual component of the spectral texture entropy index in a single frame is greater than the preset brittle fracture threshold, it is determined that the coupling has experienced diaphragm breakage or surface peeling, and an instantaneous stop command is generated. Step S53: Extract the thermal field coupling intensity exponential residual component from the nonlinear residual vector; Step S54: Perform time-domain integration on the residual component of the thermal field coupling strength index. If the integral value is greater than the preset thermal accumulation failure threshold, it is determined that the coupling has misalignment or lubrication failure, and a maintenance warning signal is generated.

[0044] Setting an instantaneous threshold for the spectral texture entropy residual can capture millisecond-level brittle fracture signals, directly triggering shutdown and preventing catastrophic accidents.

[0045] By employing time-domain integration to address thermal field residuals, the cumulative effect of integration is used to amplify subtle temperature rise trends, enabling early warnings of misalignment faults and achieving a shift from "reactive post-event maintenance" to "predictive pre-event maintenance."

[0046] The method also includes an anti-interference step based on decoupling of virtual and real features: When the increase in the thermal coupling strength index within a preset time window exceeds a preset increase threshold, the spectral texture entropy index under the same spatial coordinates is invoked. If the spectral texture entropy index falls entirely within the spectral texture entropy index benchmark range within the preset time window, then the current thermal anomaly is determined to be artifact noise caused by environmental thermal radiation reflection, and the process returns to step S4 to suppress the weight of the region in the adaptive inertial memory recursive state model.

[0047] By utilizing multimodal interlocking logic, namely "thermal anomalies must be accompanied by texture anomalies", artifact noise caused by environmental thermal radiation (such as reflections from bypass steam pipes) can be effectively identified and eliminated.

[0048] By dynamically suppressing weights, the system is endowed with adaptive survivability in harsh industrial environments (high reflectivity, strong interference), which significantly reduces the false alarm rate of the system.

[0049] The problem of false positives under complex photothermal environments has been solved.

[0050] In some possible embodiments, strong environmental thermal radiation interference, such as in the bypass of a high-temperature steam pipe, can cause traditional infrared thermal imagers to receive reflected heat from the steam pipe on the surface of a shiny coupling, falsely reporting overheating of the coupling (false positive). To address this, the system triggers a virtual-real feature decoupling strategy. When an increase in the thermal field index is detected, the spectral texture entropy of that coordinate point is immediately indexed. If the texture entropy does not change significantly, it indicates that the surface material has not oxidized or been damaged. Based on multimodal interlocking logic, the system determines that the hot spot is a false artifact.

[0051] In the recursive model, the weight matrix for this region Dynamic suppression (setting to zero) was implemented to prevent it from participating in residual calculation, thus achieving zero false alarm operation under complex photothermal environments.

[0052] The threshold and weight settings involved in this embodiment can be set by default according to the present invention, or can be set by those skilled in the art.

[0053] Example 2 This embodiment provides an electronic device, including: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor executes the aforementioned intelligent diagnostic method for couplings based on spectral imaging by calling the computer program stored in memory.

[0054] The electronic device can vary considerably depending on its configuration or performance. It may include one or more Central Processing Units (CPUs) and one or more memories, wherein the memory stores at least one computer program, which is loaded and executed by the processor to implement the intelligent coupling diagnostic method based on spectral imaging provided in the above-described embodiment. The electronic device may also include other components for implementing its functions; for example, it may have wired or wireless network interfaces and input / output interfaces for data input and output. Further details are omitted in this embodiment.

[0055] Example 3 This embodiment proposes a computer-readable storage medium 200 on which an erasable and rewritable computer program is stored. When the computer program runs on the computer device, it causes the computer device to perform the above-mentioned intelligent diagnostic method for couplings based on spectral imaging.

[0056] For example, the computer-readable storage medium 200 can be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0057] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0058] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0060] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0062] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A coupling intelligent diagnostic method based on spectral imaging, characterized in that: Includes the following steps: Step S1: Establish a phase-locked vernier acquisition environment, monitor the rotational speed and frequency of the coupling under test in real time, and construct a phase-locked trigger signal to drive a multispectral stroboscopic light source and cooperate with an orthogonal polarization imaging optical path to perform non-continuous discrete-time sampling on the high-speed rotating coupling, thereby obtaining a quasi-static spectral image sequence. Step S2: Pre-set a linear transformation matrix based on offline big data computation in the edge computing device. The matrix is ​​used to directly map high-dimensional raw spectral data to a fault-sensitive low-dimensional feature subspace to obtain a manifold mapping sparse spectral projection operator. Step S3: Input the single-frame quasi-static spectral image data into the manifold mapping sparse spectral projection operator, and output the transient health state feature vector containing texture entropy components and heat distribution components through single-step linear operation; Step S4: Construct a driving adaptive inertial memory recursive state model, configured to input the transient health state feature vector into a recursive model with a time memory decay mechanism, and calculate the dynamic reference vector at the current moment and the nonlinear residual vector relative to the reference. Step S5: Based on the amplitude mutation characteristics and time-series accumulation characteristics of the nonlinear residual vector, determine the transient fracture fault and the gradual misalignment fault of the coupling.

2. The intelligent diagnostic method for couplings based on spectral imaging according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11. Obtain the real-time rotational frequency of the coupling; and set the number of phase slices, wherein the number of phase slices is selected as a positive integer in the Fibonacci sequence to construct a non-harmonic overlapping sampling interval; S12. The vernier strobe frequency is obtained by dividing the number of phase slices by the sum of that number and one, and then multiplying by the real-time rotation frequency. S13. The acquisition is triggered by the stroboscopic frequency of the vernier, so that the acquired quasi-static spectral image sequence is presented in the time domain as a super-resolution reconstructed image that rotates slowly according to a preset slip rate.

3. The intelligent diagnostic method for couplings based on spectral imaging according to claim 1, characterized in that: The process of obtaining the manifold mapping sparse spectral projection operator: In the offline phase, full-band spectral data containing several types of faults are acquired and fed into the spectral pre-screening strategy to obtain a set of high-confidence candidate bands. The sparse regression algorithm is applied to the high-confidence candidate band set for topological feature selection, and the sparse band combination that is most sensitive to the microcrack texture and abnormal temperature rise on the metal surface is selected. Construct the optimal projection relationship from the full-band space to the low-dimensional manifold space defined by the sparse band combination, obtain the optimal weight matrix by the least squares method, and solidify the weight matrix into the manifold mapping sparse spectral projection operator.

4. The intelligent diagnostic method for couplings based on spectral imaging according to claim 3, characterized in that: The spectral pre-screening strategy includes calculating the spectral information entropy of each band in the entire band, removing invalid bands whose information entropy is lower than a preset noise threshold; calculating the spectral cross-correlation matrix of the remaining bands, identifying and removing collinear bands whose correlation coefficient is greater than a preset redundancy threshold, thereby generating a set of high-confidence candidate bands.

5. The intelligent diagnostic method for couplings based on spectral imaging according to claim 1, characterized in that: The process of obtaining the transient health state feature vector includes: performing matrix multiplication on the collected image pixel gray value vector and the manifold mapping sparse spectral projection operator; and outputting the result vector, wherein the first dimension is defined as the spectral texture entropy index, which is used to characterize the physical integrity of the material surface; and the second dimension is defined as the thermal field coupling strength index, which is used to characterize the degree of abnormal aggregation of the local temperature field.

6. The intelligent diagnostic method for couplings based on spectral imaging according to claim 1, characterized in that: The construction process of the driving adaptive inertial memory recursive state model includes: Step S41: Perform cold reference clamp initialization, construct a two-dimensional state space containing spectral texture entropy components and thermal field coupling strength components, collect multiple frames of transient health state feature vectors within the preset stable window at the initial stage of coupling startup and calculate their arithmetic mean, and solidify the arithmetic mean as the initial dynamic reference vector at time zero. Step S42: Construct a fault evolution memory decay matrix and establish a two-dimensional diagonal matrix as the system inertia operator. The first diagonal element corresponding to the spectral texture entropy component is set as the texture inertia factor, and its value is preset to a minimum value based on the signal-to-noise ratio of the imaging system to suppress high-frequency noise. The second diagonal element corresponding to the thermal field coupling intensity component is set as the thermal inertia factor, and its value is preset based on the specific heat capacity and thermal conductivity of the coupling metal material to simulate the hysteresis effect of physical heat conduction. Step S43: Update the dynamic reference vector in real time according to the discrete time step. The dynamic reference vector at the current time is composed of the transient health state feature vector at the previous time multiplied by the fault evolution memory decay matrix and the difference matrix between the dynamic reference vector at the previous time multiplied by the identity matrix and the fault evolution memory decay matrix. Step S44: Calculate the difference vector between the transient health state feature vector at the current moment and the updated dynamic reference vector at the current moment, and take the absolute value of each component of the difference vector to obtain the nonlinear residual vector used for final fault determination.

7. The intelligent diagnostic method for couplings based on spectral imaging according to claim 6, characterized in that: The specific steps of step S5 include: Step S51: Extract the spectral texture entropy exponential residual component from the nonlinear residual vector; Step S52: If the amplitude of the residual component of the spectral texture entropy index in a single frame is greater than the preset brittle fracture threshold, it is determined that the coupling has experienced diaphragm breakage or surface peeling, and an instantaneous stop command is generated. Step S53: Extract the thermal field coupling intensity exponential residual component from the nonlinear residual vector; Step S54: Perform time-domain integration on the residual component of the thermal field coupling strength index. If the integral value is greater than the preset thermal accumulation failure threshold, it is determined that the coupling has misalignment or lubrication failure, and a maintenance warning signal is generated.

8. The intelligent diagnostic method for couplings based on spectral imaging according to claim 1, characterized in that: The method also includes an anti-interference step based on decoupling of virtual and real features: When the increase in the thermal coupling strength index within a preset time window exceeds a preset increase threshold, the spectral texture entropy index under the same spatial coordinates is invoked. If all the spectral texture entropy indices fall within the spectral texture entropy index benchmark range within the preset time window, the current thermal anomaly is determined to be artifact noise caused by environmental thermal radiation reflection, and the process returns to step S4 to suppress the weights of the corresponding regions in the adaptive inertial memory recursive state model.

9. An electronic device, characterized in that, include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor executes the intelligent diagnostic method for couplings based on spectral imaging as described in any one of claims 1-8 by calling the computer program stored in the memory.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instruction is executed by the processor, it implements the intelligent diagnostic method for couplings based on spectral imaging as described in any one of claims 1-8.

Images