Generator rotor turn-to-turn short circuit fault diagnosis method based on multi-modal data fusion

Abstract

The invention relates to the technical field of power equipment fault diagnosis, and provides a generator rotor turn-to-turn short circuit fault diagnosis method based on multi-modal data fusion, and the method comprises the steps: synchronously collecting multi-source data, including electrical parameters and image data, during the operation of a generator rotor having a turn-to-turn short circuit fault; extracting electrical signal features in the electrical parameters and image features in the image data; the image features are preprocessed, mismatching points in the image features are removed, the electrical signal features are preprocessed, and the electrical signal feature time sequence is aligned; calculating generator rotor turn-to-turn short circuit fault distribution based on the preprocessed image features, the electrical signal features and a weight coefficient dynamic adjustment method corresponding to the image features and the electrical signal features; and outputting a fault positioning result based on the generator rotor turn-to-turn short circuit fault distribution. According to the invention, the early fault detection rate is effectively improved, the fault positioning precision is effectively improved, the detection time is effectively shortened, and online real-time monitoring is supported.

Description

Technical Field

[0001] This invention relates to the field of power equipment fault diagnosis technology, and in particular to a method for diagnosing generator rotor turn-to-turn short-circuit faults based on multimodal data fusion. Background Technology

[0002] As the core component of a synchronous generator, the generator rotor is most commonly affected by inter-turn short circuits in its windings, which is a common electrical defect in power systems. According to the IEEE 2022 Fault Statistics Report, rotor faults account for 34.7% of all faults in large generators, and mainly manifest as follows: I. Impact of inter-turn short-circuit faults: 1. Magnetic field weakening and voltage fluctuation Reduced effective turns: Short circuits cause some windings to be bypassed, reducing rotor magnetomotive force. If the original excitation current is maintained, the generator output voltage may decrease.

[0003] Increased excitation demand: The excitation current needs to be increased to compensate for magnetic field loss, which may cause overload or overheating of the excitation system and accelerate equipment aging with long-term operation.

[0004] 2. Risks of localized overheating and insulation Short-circuit circulating current: A circulating current forms at the short-circuit point, causing local high temperature, which may burn the insulation, trigger a grounding fault, or expand the short-circuit range.

[0005] Thermal stress accumulation: Repeated thermal expansion and contraction may deform the windings, exacerbate insulation deterioration, and even cause fire risk.

[0006] 3. Mechanical vibration and noise Magnetic field asymmetry: Uneven magnetic pull of the rotor causes radial force fluctuations, which can lead to abnormal vibrations and potentially damage components such as bearings and couplings.

[0007] Harmonic resonance: If the vibration frequency is close to the natural frequency of the structure, it may cause resonance, resulting in serious mechanical damage.

[0008] 4. Efficiency and Loss Issues Increased copper losses: Short-circuit current causes additional resistance losses, reducing generator efficiency and increasing operating costs.

[0009] Harmonic loss: Magnetic field distortion may generate harmonics in the stator, leading to increased iron loss and additional heating.

[0010] 5. Increased burden on the excitation system Overload risk: Continuous high excitation current may exceed the rated capacity of equipment such as rectifiers and excitation transformers, leading to failure.

[0011] Decreased dynamic response: Reduced excitation system regulation margin affects the generator's ability to respond quickly to grid voltage fluctuations.

[0012] 6. Threats to power system stability Decreased voltage regulation capability: This may lead to grid voltage instability, especially during sudden load changes, which can easily cause voltage collapse.

[0013] Risk of subsynchronous oscillation: Magnetic field asymmetry may induce subsynchronous resonance, threatening the safety of long-distance power transmission systems. Summary of the Invention

[0014] The purpose of this invention is to solve at least one technical problem in the background art and to provide a generator rotor turn-to-turn short-circuit fault diagnosis method based on multimodal data fusion.

[0015] To achieve the above objectives, this invention provides a method for diagnosing inter-turn short-circuit faults in generator rotors based on multi-modal data fusion, comprising: Simultaneously acquire multi-source data during generator rotor operation with inter-turn short circuit fault, including electrical parameters and image data; Extract electrical signal features from electrical parameters and image features from image data; An improved SIFT algorithm is used to preprocess image features and remove mismatched points in the image features. The electrical signal features are preprocessed based on the dynamic time warping algorithm to align the timing of the electrical signal features. The distribution of inter-turn short-circuit faults in the generator rotor is calculated based on the preprocessed image features, electrical signal features, and the corresponding weight coefficients of the image features and electrical signal features dynamically adjusted. Based on the distribution of short-circuit faults between generator rotor turns, the fault location results are output.

[0016] According to one aspect of the present invention, the electrical parameters include: DC resistance, AC impedance, voltage distribution to ground, and shaft voltage waveform; The image data includes: infrared thermal imaging images, ultraviolet discharge images, and high-resolution visible light images.

[0017] According to one aspect of the present invention, an improved SIFT algorithm is used to preprocess image features and remove mismatch points from the image features, including: (1) Coordinate transformation: For the cylindrical structure of the generator rotor, the infrared thermal imaging image is transformed into polar coordinates to map the curved surface image to the plane coordinate system, so as to eliminate the geometric distortion caused by rotation shooting and establish a unified coordinate basis for subsequent feature matching. (2) Feature extraction and description: On the image after coordinate transformation, feature descriptors with rotation invariant properties (such as ORB) are used to detect key points and generate their feature vectors to characterize the local structural information of the image; (3) Feature matching and optimization: First, the feature vectors generated in step (2) are initially matched using the bidirectional nearest neighbor matching method; then, the random sampling consensus algorithm is used to iteratively optimize the initial matching results, eliminate mismatched points caused by noise, occlusion or deformation, and finally obtain accurate and robust feature matching pairs.

[0018] According to one aspect of the present invention, the method for dynamically adjusting the weight coefficients of the preprocessed image features, electrical signal features, and corresponding image features and electrical signal features to calculate the distribution of generator rotor inter-turn short-circuit faults is as follows: P(x,y,z)=α·ΣE_i+β·ΣI_j+γ·ΣU_k; Where P(x,y,z) represents the fault probability value at position (x,y,z) in the rotor three-dimensional coordinate system, α is the weight of the electrical signal feature, β is the weight of the infrared thermal imaging image, γ is the weight of the ultraviolet discharge image, and α+β+γ=1; E_i is the electrical signal feature, I_j is the image feature, and U_k is the feature value of the k-th ultraviolet discharge image. The method for dynamically adjusting the weight coefficients of the corresponding image features and electrical signal features includes: α=0.4+0.1×sigmoid(E_accuracy+0.2×V asymmetry +0.15×DI); β = 0.3 tanh(I_quality); ; Where E_accuracy is the electrical parameter reliability, I_quality is the image quality score, and V asymmetry Indicates the voltage asymmetry to ground; DI represents the axis voltage waveform distortion index, reflecting the degree of dynamic waveform distortion; N UV The number of ultraviolet photons.

[0019] According to one aspect of the invention, the extracted electrical signal features include: The third harmonic current content is considered to be an inter-turn short circuit when the proportion of the third harmonic current exceeds 5 times the baseline value. Temperature rise rate: When the local temperature rise rate is >15K / min, a short circuit warning is triggered. The magnetic field asymmetry coefficient is calculated based on the rotor magnetic field distribution. When the magnetic field asymmetry coefficient is greater than 0.2, it is determined to be an inter-turn short circuit.

[0020] The relative rate of change of DC resistance triggers an early warning when the relative rate of change of DC resistance is greater than 5%. The AC impedance ratio is considered abnormal when it exceeds 15%. The threshold for voltage asymmetry to ground is set at 10%. The extracted image features include: Areas with abnormal temperatures in infrared thermal images; Detecting discharge points with photon counts > 500 cps in ultraviolet images; Identify areas of insulation damage in visible light images; In infrared thermal images, local temperature gradients > 5 K / cm² are marked as short-circuit hotspots; When the ultraviolet discharge pulse frequency is >10Hz, it is determined to be a continuous discharge.

[0021] According to one aspect of the invention, it further includes: Vibration signals of the generator rotor during operation are collected by vibration sensors, and ultraviolet discharge signals are monitored in real time. Extract vibration signal features from vibration signals and extract ultraviolet discharge signal features from ultraviolet discharge signals; Based on at least two of the extracted electrical signal features, image features, and vibration signal features, one or more cross-validation parameters are calculated. The cross-validation parameters are used to quantify the consistency between multi-source data and, in the step of calculating the distribution of generator rotor inter-turn short-circuit faults, are used to correct the confidence level of the preliminary fault probability distribution or as a criterion for triggering manual review. The distribution of short-circuit faults between rotor turns of the generator is calculated based on the preprocessed image features, electrical signal features, vibration signal features, ultraviolet discharge signal features, and the corresponding weight coefficients of the image features, electrical signal features, vibration signal features, and ultraviolet discharge signal features.

[0022] According to one aspect of the present invention, the method for dynamically adjusting the weight coefficients of preprocessed image features, electrical signal features, vibration signal features, and ultraviolet discharge signal features, as well as the corresponding image features, electrical signal features, vibration signal features, and ultraviolet discharge signal features, calculates the distribution of inter-turn short-circuit faults in the generator rotor as follows: P(x,y,z)=α·ΣE_i+β·ΣI_j+γ·ΣU_k+δ·ΣV_m+ε·ΣD_n; Where δ is the weight of the vibration signal feature, ε is the weight of the ultraviolet discharge signal feature; α+β+γ+δ+ε=1; V_m is the m-th vibration signal feature value, and D_n is the n-th ultraviolet discharge signal feature value; The method for dynamically adjusting the weighting coefficients corresponding to vibration signal characteristics and ultraviolet discharge signal characteristics includes: δ=0.1×sigmoid(F_vib / 12%), ε=0.05×tanh(E_UV / 1000); Where F_vib is the proportion of high-frequency vibration energy, and E_UV is the cumulative energy of ultraviolet discharge.

[0023] To achieve the above objectives, the present invention also provides a generator rotor inter-turn short-circuit fault diagnosis system based on multimodal data fusion, comprising: The multi-source data acquisition module synchronously acquires multi-source data during generator rotor operation when an inter-turn short-circuit fault occurs, including electrical parameters and image data; The data feature extraction module extracts electrical signal features from electrical parameters and image features from image data; The data feature preprocessing module uses an improved SIFT algorithm to preprocess image features, removes mismatched points in the image features, and uses a dynamic time warping algorithm to preprocess electrical signal features and align the timing of electrical signal features. The inter-turn short-circuit fault distribution calculation module calculates the generator rotor inter-turn short-circuit fault distribution based on preprocessed image features, electrical signal features, and a dynamic adjustment method for the weight coefficients of the corresponding image features and electrical signal features. The results output module outputs fault location results based on the distribution of short-circuit faults between generator rotor turns.

[0024] To achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the generator rotor turn-to-turn short-circuit fault diagnosis method based on multimodal data fusion as described above.

[0025] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the generator rotor turn-to-turn short-circuit fault diagnosis method based on multimodal data fusion as described above.

[0026] According to the present invention, the present invention achieves millisecond-level synchronous acquisition of electrical parameters (DC resistance, AC impedance, shaft voltage) and multispectral images (infrared thermography, ultraviolet discharge, visible light) and other fault criteria through the Precision Clock Synchronization Protocol (PTPv2); establishes a rotor pole-slot-axis three-dimensional coordinate system, and uses an improved SIFT-ORB fusion algorithm for spatiotemporal registration to compensate for coordinate offsets caused by thermal expansion and centrifugal deformation; constructs a dynamic weight model, adaptively adjusts the contribution weights of multi-source data according to parameters such as electrical reliability, image quality, and operating condition fluctuations, and calculates the fault probability distribution using a Bayesian probability field, finally achieving sub-slot-level positioning (±0.5 slots) and confidence assessment through a three-dimensional heat map. Compared with traditional methods, the present invention effectively improves the early fault detection rate, effectively enhances positioning accuracy, effectively reduces detection time, and supports online real-time monitoring. Attached Figure Description

[0027] Figure 1 The flowchart illustrates a generator rotor inter-turn short-circuit fault diagnosis method based on multimodal data fusion according to an embodiment of the present invention. Detailed Implementation

[0028] The invention will now be discussed with reference to exemplary embodiments. It should be understood that the described embodiments are merely intended to enable those skilled in the art to better understand and thus implement the invention, and are not intended to imply any limitation on the scope of the invention.

[0029] As used herein, the term "comprising" and its variations are to be interpreted as open-ended terms meaning "including but not limited to". The term "based on" is to be interpreted as "at least partially based on". The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment".

[0030] Figure 1 This diagram illustrates a flowchart of a generator rotor inter-turn short-circuit fault diagnosis method based on multimodal data fusion according to an embodiment of the present invention. Figure 1 As shown, in this embodiment, the generator rotor inter-turn short-circuit fault diagnosis method based on multi-modal data fusion includes: Simultaneously acquire multi-source data during generator rotor operation with inter-turn short circuit fault, including electrical parameters and image data; Extract electrical signal features from electrical parameters and image features from image data; An improved SIFT algorithm is used to preprocess image features and remove mismatched points in the image features. The electrical signal features are preprocessed based on the dynamic time warping algorithm to align the timing of the electrical signal features. The distribution of inter-turn short-circuit faults in the generator rotor is calculated based on the preprocessed image features, electrical signal features, and the corresponding weight coefficients of the image features and electrical signal features dynamically adjusted. Based on the distribution of short-circuit faults between generator rotor turns, the fault location results are output.

[0031] Furthermore, according to one embodiment of the present invention, the electrical parameters include: DC resistance, AC impedance, voltage distribution to ground, and shaft voltage waveform; The image data includes: infrared thermal imaging images, ultraviolet discharge images, and high-resolution visible light images.

[0032] Furthermore, according to one embodiment of the present invention, an improved SIFT algorithm is used to preprocess image features and remove mismatch points in the image features, including: (1) Coordinate transformation: For the cylindrical structure of the generator rotor, the infrared thermal imaging image is transformed into polar coordinates to map the curved surface image to the plane coordinate system, so as to eliminate the geometric distortion caused by rotation shooting and establish a unified coordinate basis for subsequent feature matching. (2) Feature extraction and description: On the image after coordinate transformation, feature descriptors with rotation invariant properties (such as ORB) are used to detect key points and generate their feature vectors to characterize the local structural information of the image; (3) Feature matching and optimization: First, the feature vectors generated in step (2) are initially matched using the bidirectional nearest neighbor matching method; then, the random sampling consensus algorithm is used to iteratively optimize the initial matching results, eliminate mismatched points caused by noise, occlusion or deformation, and finally obtain accurate and robust feature matching pairs.

[0033] Furthermore, according to one embodiment of the present invention, the distribution of inter-turn short-circuit faults in the generator rotor is calculated based on preprocessed image features, electrical signal features, and a dynamic adjustment method for the weight coefficients of the corresponding image features and electrical signal features: P(x,y,z)=α·ΣE_i+β·ΣI_j+γ·ΣU_k; Where P(x, y, z) represents the fault probability value at position (x, y, z) in the rotor three-dimensional coordinate system, α is the weight of the electrical signal feature, β is the weight of the infrared thermal imaging image, γ is the weight of the ultraviolet discharge image, and α+β+γ=1; E_i is the electrical signal feature, I_j is the image feature, and U_k is the feature value of the k-th ultraviolet discharge image. The methods for dynamically adjusting the weight coefficients of corresponding image features and electrical signal features include: α=0.4+0.1×sigmoid(E_accuracy+0.2×V asymmetry +0.15×DI); β = 0.3 tanh(I_quality); ; Where E_accuracy is the electrical parameter reliability, I_quality is the image quality score, and V asymmetry Indicates the voltage asymmetry to ground; DI represents the axis voltage waveform distortion index, reflecting the degree of dynamic waveform distortion; N UV The number of ultraviolet photons.

[0034] Furthermore, according to one embodiment of the present invention, the extracted electrical signal features include: The third harmonic current content (I_3rd) is determined to be an inter-turn short circuit when the proportion of the third harmonic current exceeds 5 times the baseline value (i.e., I_3rd / I_1st > 15%). Temperature rise rate (dT / dt): When the local temperature rise rate > 15K / min, a short circuit warning is triggered. The magnetic field asymmetry coefficient (M_asym) is calculated based on the rotor magnetic field distribution. When the magnetic field asymmetry coefficient > 0.2, it is determined to be an inter-turn short circuit.

[0035] The relative change rate of DC resistance ΔR / R0 triggers an early warning when the relative change rate of DC resistance > 5%. The AC impedance ratio Z'' / Z' is considered abnormal when it is greater than 15%. The threshold for voltage asymmetry to ground is set at 10%. The extracted image features include: Temperature anomaly regions in infrared thermal images (ΔT > 8K and area > 5cm²); Discharge points with photon counts > 500 cps were detected in ultraviolet images; Identify areas of insulation damage (damage length > 2 mm) in visible light images; When the local temperature gradient (▽T) in an infrared thermal image is greater than 5K / cm², it is marked as a short-circuit hotspot; When the ultraviolet discharge pulse frequency (f_UV) is greater than 10Hz, it is determined to be a continuous discharge.

[0036] Furthermore, according to one embodiment of the present invention, the present invention further includes: Vibration signals of the generator rotor during operation are collected by vibration sensors, and ultraviolet discharge signals are monitored in real time. Extract vibration signal features from vibration signals and extract ultraviolet discharge signal features from ultraviolet discharge signals; Based on at least two of the extracted electrical signal features, image features, and vibration signal features, one or more cross-validation parameters are calculated. The cross-validation parameters are used to quantify the consistency between multi-source data and, in the step of calculating the distribution of generator rotor inter-turn short-circuit faults, are used to correct the confidence level of the preliminary fault probability distribution or as a criterion for triggering manual review. The distribution of short-circuit faults between rotor turns of the generator is calculated based on the preprocessed image features, electrical signal features, vibration signal features, ultraviolet discharge signal features, and the corresponding weight coefficients of the image features, electrical signal features, vibration signal features, and ultraviolet discharge signal features.

[0037] Furthermore, according to one embodiment of the present invention, the distribution of inter-turn short-circuit faults in the generator rotor is calculated based on preprocessed image features, electrical signal features, vibration signal features, ultraviolet discharge signal features, and a dynamic adjustment method for the weighting coefficients of the corresponding image features, electrical signal features, vibration signal features, and ultraviolet discharge signal features: P(x,y,z) =α·ΣE_i+β·ΣI_j+γ·ΣU_k+δ·ΣV_m+ε·ΣD_n; Where P(x, y, z) represents the fault probability value at position (x, y, z) in the rotor three-dimensional coordinate system, δ is the weight of the vibration signal feature, ε is the weight of the ultraviolet discharge signal feature; α+β+γ+δ+ε=1; V_m is the m-th vibration signal feature value, and D_n is the n-th ultraviolet discharge signal feature value; The method for dynamically adjusting the weighting coefficients corresponding to vibration signal characteristics and ultraviolet discharge signal characteristics includes: δ=0.1×sigmoid(F_vib / 12%), ε=0.05×tanh(E_UV / 1000); Where F_vib is the proportion of high-frequency vibration energy, and E_UV is the cumulative energy of ultraviolet discharge.

[0038] Furthermore, to achieve the above objectives, the present invention also provides a generator rotor inter-turn short-circuit fault diagnosis system based on multimodal data fusion, comprising: The multi-source data acquisition module synchronously acquires multi-source data during generator rotor operation when an inter-turn short-circuit fault occurs, including electrical parameters and image data; The data feature extraction module extracts electrical signal features from electrical parameters and image features from image data; The data feature preprocessing module uses an improved SIFT algorithm to preprocess image features, removes mismatched points in the image features, and uses a dynamic time warping algorithm to preprocess electrical signal features and align the timing of electrical signal features. The inter-turn short-circuit fault distribution calculation module calculates the generator rotor inter-turn short-circuit fault distribution based on preprocessed image features, electrical signal features, and a dynamic adjustment method for the weight coefficients of the corresponding image features and electrical signal features. The results output module outputs fault location results based on the distribution of short-circuit faults between generator rotor turns.

[0039] The generator rotor inter-turn short circuit fault diagnosis system based on multimodal data fusion according to the present invention can realize the generator rotor inter-turn short circuit fault diagnosis method based on multimodal data fusion. The specific process steps are as described above and will not be repeated here.

[0040] Furthermore, to achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the generator rotor turn-to-turn short-circuit fault diagnosis method based on multimodal data fusion as described above.

[0041] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the generator rotor turn-to-turn short-circuit fault diagnosis method based on multimodal data fusion as described above.

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of the invention and are only used to explain the invention. They do not limit the scope of protection of the invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0043] Example 1 This embodiment provides a generator rotor inter-turn short-circuit fault diagnosis method based on multimodal data fusion, including the following steps: Step 1: Synchronous Acquisition of Multi-Source Data Operation: The electrical acquisition equipment and image acquisition equipment are synchronously triggered through the Precision Clock Synchronization Protocol (PTPv2) to acquire multi-source data on the generator rotor during inter-turn short circuit fault operation in milliseconds.

[0044] Electrical data acquisition equipment, such as the NI PXIe-4300 data acquisition card, acquires DC resistance, AC impedance, and shaft voltage waveforms at a sampling rate of 2 MS / s; and acquires the voltage distribution to ground through a distributed voltage sensor (HBM S9M) network.

[0045] Image acquisition equipment: such as the FLIR A8580sc infrared thermal imager, which acquires images of the rotor winding temperature field; and the OFILSuperB+ ultraviolet imager, which simultaneously acquires ultraviolet discharge photon counts and visible light images.

[0046] Output and Input: Under the trigger of a synchronous clock, all the above acquisition devices will output the acquired raw electrical signals and raw image data in real time through a high-speed data bus (such as PXIe bus) and input them into the multi-source data receiving and buffering module built into this diagnostic system to form raw data packets with strict time alignment and spatial information association for subsequent steps to call.

[0047] Step 2: Multimodal Feature Extraction In this embodiment, the three weights (α, β, γ) are dynamically adjusted using a nonlinear function; The weight α of electrical signal characteristics increases rapidly with the confidence index and dominates fault diagnosis.

[0048] The weight β of infrared thermal imaging images increases smoothly with image quality, aiding in the localization of high-temperature areas.

[0049] The weight γ of the ultraviolet discharge image is positively correlated with the discharge intensity, but its upper limit is limited to avoid false alarms.

[0050] The weight δ of vibration signal characteristics: When the proportion of high-frequency vibration energy exceeds the threshold, δ increases significantly, enhancing the contribution of mechanical abnormalities to the failure probability.

[0051] The weight ε of ultraviolet discharge signal characteristics: The higher the cumulative discharge energy, the larger the ε value, reflecting the continuous risk of insulation degradation.

[0052] The sum of the above five weights is always 1, ensuring the rationality of weight allocation under different working conditions.

[0053] Step 3: Spatiotemporal registration and fusion To achieve accurate fusion of multi-source data within a unified spatial framework, this step establishes a spatial reference system for the rotor and corrects coordinate errors caused by physical deformation.

[0054] 1. Establishment of the rotor three-dimensional coordinate system: Define the rotor three-dimensional coordinate system for fault location: θ (circumferential angle): The angle along the direction of rotation, starting from a reference point on the rotor (such as the keyway), in degrees.

[0055] z (axial position): the length coordinate along the rotor axis, usually with a certain end face as the origin, in meters.

[0056] r (radial offset): The radial distance from the fault point to the rotor shaft center, in meters. This coordinate system provides a unified absolute spatial reference for all image features, electrical features, and the final fault distribution.

[0057] 2. Deformation compensation and coordinate correction: Under high-speed rotation and temperature rise conditions, the rotor will undergo centrifugal expansion and thermal deformation, causing the directly measured characteristic coordinates (θ_raw, z_raw) to deviate from their true positions (θ_aligned, z_aligned) in the three-dimensional coordinate system. Therefore, a deformation compensation transformation matrix is ​​introduced for correction. This matrix focuses on correcting the two dimensions θ and z, which are significantly affected by deformation. Its mathematical expression and physical meaning are as follows: = × ; Where θ_raw and z_raw are the original measured circumferential angle (unit: degrees) and axial position (unit: meters), respectively; θ_aligned and z_aligned are the angle and axial position after registration correction, respectively. Physical meaning of parameters in the formula: (1) Angle correction factor (1.02): This compensates for the circumferential tensile deformation of the rotor caused by centrifugal force. When the rotational speed reaches the rated value (e.g., 3000 rpm), the rotor diameter expands by approximately 0.2% due to centrifugal force. This coefficient is determined through experimental calibration (e.g., by measuring deformation data using a laser rangefinder).

[0058] (2) Axial coupling coefficient (0.015): The axial-circumferential deformation coupling effect caused by thermal expansion is characterized. For every 10°C increase in rotor temperature, axial expansion introduces an angular shift of approximately 0.15° (verified by finite element thermodynamic simulation).

[0059] (3) Axial compressibility (0.98): This counteracts the centrifugal compression effect on the rotor core. Actual measurement data shows that the rotor axial length shortens by approximately 2% when operating at full load (resulting from the balance between material yield strength and centrifugal force).

[0060] (4) Off-diagonal zero term (0): Based on the rotor's symmetrical design, it is assumed that changes in the circumferential angle have no direct impact on the axial position (simplified model error <0.1mm). In this embodiment, feature matching includes: 1. Feature extraction: Bidirectional Nearest Neighbor matching is used to filter feature point pairs, and Random Sampling Consensus Algorithm (RANSAC) is used to remove mismatched points, thereby improving the robustness of registration.

[0061] 2. Mismatch Removal: Through iterative optimization using the Random Sampling Consensus (RANSAC) algorithm, feature point pairs with high spatial consistency are selected, and abnormal matches are removed.

[0062] 3. Deformation compensation: In polar coordinate transformation, a centrifugal expansion correction factor (1.02) and a thermal expansion coupling coefficient (0.015) are introduced to dynamically correct the circumferential stretching and axial offset caused by high-speed rotor rotation (rated speed 3000 rpm) and temperature rise (ΔT>50K).

[0063] Step 4: Dynamic Weight Fault Calculation 1. Preparation of dynamic weight calculation parameters: The following parameters are calculated as inputs to the weighting formula: electrical accuracy (E_accuracy) (0-1), image quality score (I_quality) (0-1), and operating condition parameters such as voltage asymmetry to ground (V_asymmetry), axial voltage waveform distortion index (DI), ultraviolet photon count (N_UV), high-frequency vibration energy percentage (F_vib), and ultraviolet discharge cumulative energy (E_UV).

[0064] 2. Dynamic weight calculation: Substituting the above parameters into the dynamic weight formula, the weights α, β, γ, δ, ε of each modal feature are calculated in real time. All weights satisfy the normalization condition (α+β+γ+δ+ε=1).

[0065] 3. Failure probability distribution calculation: The modal feature values ​​extracted and preprocessed in step 2 are weighted and fused with the corresponding weights calculated in step 2. The fault probability value at each position in the rotor three-dimensional space is calculated using the formula P(x, y, z) = α·ΣE_i + β·ΣI_j + γ·ΣU_k + δ·ΣV_m + ε·ΣD_n, generating a fault probability data field.

[0066] 4. Output: Fault probability heatmap: The fault probability data field is visualized and rendered, outputting a fault probability heatmap. This heatmap uses a 3D rotor model as a base and color mapping to visually display the fault probability at each location, with a resolution set to 1° circumferentially and 5mm axially.

[0067] A dynamic weighting model is used, and the formula is as follows: 1. Weight α of electrical signal characteristics: α = 0.4 + 0.1 × sigmoid(E_accuracy + 0.2 × V) asymmetry +0.15×DI) The electrical parameter accuracy, E_accuracy, is calculated using the relative rate of change of DC resistance, ΔR / R0; V asymmetryThe sigmoid function represents the voltage asymmetry to ground (threshold 10%), which enhances abnormal signals through a sigmoid function. DI represents the axis voltage waveform distortion index (threshold 0.1), reflecting the degree of dynamic waveform distortion. When DI > 0.1, the sigmoid function output increases significantly, boosting the electrical weight α. Simultaneously, DI reduces the ultraviolet weight γ through a linear attenuation factor, avoiding interference from multiple abnormal signals. Coefficients 0.2 and 0.15 are experimental calibration values, balancing the contribution weights of different parameters. The sigmoid function maps electrical changes to a confidence value between 0 and 1. When electrical changes exceed a preset threshold (e.g., 5%), the confidence level increases significantly, reflecting an increased likelihood of abnormal electrical parameters. Adjusting the coefficients controls the steepness of the curve, ensuring a clear response near the threshold. Third harmonic weighting factor: If the proportion of the third harmonic exceeds the threshold, the electrical weight α increases by 0.1. 2. Weight β of infrared thermal imaging image: β = 0.3×tanh(I_quality); Based on the signal-to-noise ratio (SNR) calculation of infrared images, the SNR is converted into an image quality score I_quality between 0 and 1 using the hyperbolic tangent function (Tanh). The higher the SNR, the closer the image quality score I_quality is to 1, ensuring that clear thermal image data receives higher weight.

[0068] Temperature rise rate correction: When dT / dt > 15K / min, the thermal image weight β is increased by 0.05; 3. Weight γ of the ultraviolet discharge image: ; N UV N represents the number of ultraviolet photons. UV At >500 cps, γ increases with the number of photons, while suppressing excessive response through DI.

[0069] 4. Weight δ of vibration signal characteristics: δ = 0.1 × sigmoid(F_vib / 12%); When the proportion of high-frequency vibration energy F_vib exceeds the threshold, δ increases significantly, enhancing the contribution of mechanical abnormalities to the failure probability.

[0070] 5. Weight ε of ultraviolet discharge signal characteristics: ε = 0.05 × tanh(E_UV / 1000); The higher the cumulative energy of ultraviolet discharge, E_UV, the larger the ε value, reflecting the continuous risk of insulation degradation.

[0071] Step 5: Failure Probability Calculation 1. Multi-dimensional distance metric 1) Electrical Mahalanobis distance: measures the statistical deviation of the current electrical parameters (ΔR / R0, Z'' / Z') from historical normal data, taking into account the correlation between the parameters.

[0072] 2) Image Euclidean distance: Calculate the spatial consistency of infrared and ultraviolet features to identify the superposition effect of local anomalous regions.

[0073] 3) Historical data divergence: KL divergence is used to measure the difference between the current failure mode and historical cases, thereby enhancing the model's generalization ability.

[0074] 2. Probability Field Generation and Update 1) Initial Calculation: A three-dimensional probability distribution is generated based on the weighted distance. The initial three-dimensional fault probability distribution is generated based on the weighted sum of the three distances mentioned above. This weighted sum can be expressed as: Where Delectric, Dimage, and Dhistory represent the standardized values ​​of the electrical Mahalanobis distance, the image Euclidean distance, and the historical data KL divergence, respectively; For the corresponding initial weight coefficients, satisfying 2) Online learning: During system operation, the weight parameters are adjusted in reverse based on the fault location results verified manually. By minimizing the error between the model's predicted location and the manually confirmed location through optimization algorithms such as gradient descent, the model can dynamically adapt to different units or operating conditions, gradually improving positioning accuracy.

[0075] Input: Manually verified fault location coordinates (x) true ,y true ,z true ) as a monitoring signal Optimization objective: Minimize the loss function Among them, the predicted location Determined by the region with the highest probability in the current probability field Update process: Calculate the loss function with respect to the weight parameters The gradient is calculated and updated in the reverse direction of the gradient: (i=1,2,3) in, This is the learning rate. After the update... Perform normalization to ensure that the sum is 1.

[0076] Output: Optimized weight parameters This will be used for probability field calculations in subsequent data batches, forming a closed-loop optimization. 3) Parameter adjustment based on cross-validation After generating the preliminary failure probability field P_preliminary(x, y, z), cross-validation parameters are introduced to correct the results for reasonableness: 1. Consistency Enhancement: If the cross-validation parameter (such as M_asym) exceeds the threshold and spatially matches the fault region indicated by the probability field, the final fault probability value P_final of the region is appropriately increased according to certain rules (such as multiplying by a confidence coefficient C_boost greater than 1, ranging from 1.0 to 1.2).

[0077] 2. Conflict Alarm: If the probability field shows a high failure probability in a certain area, but key cross-validation parameters do not show any abnormalities (e.g., there is no corresponding temperature rise or magnetic field distortion verification for that area), the system marks the diagnostic result for that area with a "low confidence" flag and suggests manual review or initiating more refined specialized testing. The final output failure probability distribution P_final(x, y, z) is the result after such data consistency verification and correction.

[0078] Step 6: 3D Visualization and Early Warning The WebGL engine renders the rotor model, and the fault area is rendered using RGB heat mapping. Red (>80% probability): Immediately stop the machine for inspection; Yellow (50–80%): Planned maintenance; Green (<50%): Continuous monitoring; Add an axial periodic compression effect (frequency 0.5Hz, amplitude 3%) to the 3D model to simulate a "breathing" motion effect and enhance the visual perception of the fault area in human-computer interaction.

[0079] System hardware configuration: Edge computing nodes: equipped with dedicated FPGA chips to achieve real-time spatiotemporal registration, electrical signal processing units with a sampling rate of ≥1MS / s, image processing units supporting OpenCL accelerated computing, and data fusion coprocessors using KaIman filtering to optimize multi-source data; This embodiment achieves millisecond-level synchronous acquisition of electrical parameters (DC resistance, AC impedance, shaft voltage) and multispectral images (infrared thermography, ultraviolet discharge, visible light) and other fault criteria through the Precision Clock Synchronization Protocol (PTPv2). A rotor pole-slot-axial three-dimensional coordinate system is established, and an improved SIFT-ORB fusion algorithm is used for spatiotemporal registration to compensate for coordinate offsets caused by thermal expansion and centrifugal deformation (circumferential error <0.9°, axial error <7mm). A dynamic weighting model is constructed to adaptively adjust the contribution weights of multi-source data based on parameters such as electrical reliability, image quality, and operating condition fluctuations. Combined with a Bayesian probability field, the fault probability distribution is calculated, and finally, sub-slot-level positioning (±0.5 slots) and confidence assessment are achieved through a three-dimensional heat map. Compared to traditional methods, this invention increases the early fault (3-turn short circuit) detection rate from 38% to 92%, improves positioning accuracy by 4 times (circumferential ±5.6°, axial ±5mm), reduces detection time from 72 hours to 8 minutes, and supports online real-time monitoring. Verified by a 1000MW unit, it can reduce losses from a single downtime by more than 3 million yuan and reduce annual maintenance costs by more than 30%.

[0080] Compared with the prior art, the beneficial effects that this embodiment can achieve are as follows: 1. Precise diagnosis: By integrating multimodal data and dynamic weight model, the positioning accuracy is improved to ±0.5 groove (5.6°), the axial error is ≤5mm, and the early fault (3-turn short circuit) detection rate is increased from 38% to 92%, avoiding the risk of "small faults leading to big accidents".

[0081] 2. Efficient Operation and Maintenance: Online detection replaces traditional downtime detection, reducing the single detection time from 72 hours to 8 minutes. Based on a 1000MW unit, this reduces power generation losses by more than 20 million yuan per year.

[0082] 3. Intelligent upgrade: Build an adaptive diagnostic system to reduce the false alarm rate to 7.3% under fluctuating operating conditions, and support fault development prediction (early warning 3-6 months in advance), promoting the transformation of power plant operation and maintenance from "planned maintenance" to "condition-based maintenance".

[0083] 4. High reliability: By introducing cross-validation parameters based on multi-source feature calculation, the preliminary diagnostic results are internally consistent and the confidence level is corrected, which effectively reduces the false alarm rate caused by single sensor failure or data noise, and improves the overall robustness and trustworthiness of the diagnostic system.

[0084] Those skilled in the art will recognize that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0085] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and equipment can be referred to the corresponding process in the foregoing method implementation, and will not be repeated here.

[0086] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0087] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0088] In addition, the functional modules in the embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0089] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the sending / receiving methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0090] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this application.

[0091] It should be understood that the sequence number of each step in the invention and its embodiments does not absolutely 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 the present invention.

Claims

1. A method for diagnosing inter-turn short-circuit faults in generator rotors based on multi-modal data fusion, characterized in that, include: Simultaneously acquire multi-source data during generator rotor operation with inter-turn short circuit fault, including electrical parameters and image data; Extract electrical signal features from electrical parameters and image features from image data; An improved SIFT algorithm is used to preprocess image features and remove mismatched points in the image features. The electrical signal features are preprocessed based on the dynamic time warping algorithm to align the timing of the electrical signal features. The distribution of inter-turn short-circuit faults in the generator rotor is calculated based on the preprocessed image features, electrical signal features, and the corresponding weight coefficients of the image features and electrical signal features dynamically adjusted. Based on the distribution of short-circuit faults between generator rotor turns, the fault location results are output.

2. The generator rotor inter-turn short-circuit fault diagnosis method based on multi-modal data fusion according to claim 1, characterized in that, The electrical parameters include: DC resistance, AC impedance, voltage distribution to ground, and shaft voltage waveform; The image data includes: infrared thermal imaging images, ultraviolet discharge images, and high-resolution visible light images.

3. The method according to claim 1, characterized in that, The process of preprocessing image features using the improved SIFT algorithm to remove mismatched points in the image features includes: (1) Coordinate transformation: For the cylindrical structure of the generator rotor, the infrared thermal imaging image is transformed into polar coordinates to map the curved surface image to the plane coordinate system, so as to eliminate the geometric distortion caused by rotation shooting and establish a unified coordinate basis for subsequent feature matching. (2) Feature extraction and description: On the image after coordinate transformation, feature descriptors with rotation invariant properties are used to detect key points and generate their feature vectors to characterize the local structural information of the image; (3) Feature matching and optimization: First, the feature vectors generated in step (2) are initially matched using the bidirectional nearest neighbor matching method; then, the random sampling consensus algorithm is used to iteratively optimize the initial matching results, eliminate mismatched points caused by noise, occlusion or deformation, and finally obtain accurate and robust feature matching pairs.

4. The generator rotor inter-turn short-circuit fault diagnosis method based on multi-modal data fusion according to claim 1, characterized in that, The method for dynamically adjusting the weight coefficients of the preprocessed image features, electrical signal features, and corresponding image and electrical signal features to calculate the generator rotor inter-turn short circuit fault is described. P(x,y,z)=α·ΣE_i+β·ΣI_j+γ·ΣU_k; Where P(x, y, z) represents the fault probability value at position (x, y, z) in the rotor three-dimensional coordinate system; α is the weight of the electrical signal feature, β is the weight of the infrared thermal imaging image, γ is the weight of the ultraviolet discharge image, and α+β+γ=1; E_i is the electrical signal feature, I_j is the image feature, and U_k is the feature value of the k-th ultraviolet discharge image. The method for dynamically adjusting the weight coefficients of the corresponding image features and electrical signal features includes: α=0.4+0.1×sigmoid(E_accuracy+0.2×V asymmetry +0.15×DI); β = 0.3 tanh(I_quality); ； Where E_accuracy is the electrical parameter reliability, I_quality is the image quality score, and V asymmetry Indicates the voltage asymmetry to ground; DI represents the axis voltage waveform distortion index, reflecting the degree of dynamic waveform distortion; N UV The number of ultraviolet photons.

5. The generator rotor inter-turn short-circuit fault diagnosis method based on multimodal data fusion according to claim 4, characterized in that, The extracted electrical signal features include: The third harmonic current content is considered to be an inter-turn short circuit when the proportion of the third harmonic current exceeds 5 times the baseline value. Temperature rise rate: When the local temperature rise rate is >15K / min, a short circuit warning is triggered. The magnetic field asymmetry coefficient is calculated based on the rotor magnetic field distribution. When the magnetic field asymmetry coefficient > 0.2, it is determined to be an inter-turn short circuit. The relative rate of change of DC resistance triggers an early warning when the relative rate of change of DC resistance is greater than 5%. The AC impedance ratio is considered abnormal when it exceeds 15%. The threshold for voltage asymmetry to ground is set at 10%. The extracted image features include: Areas with abnormal temperatures in infrared thermal images; Discharge points with photon counts > 500 cps were detected in ultraviolet images; Identify areas of insulation damage in visible light images; In infrared thermal images, local temperature gradients > 5 K / cm² are marked as short-circuit hotspots; When the ultraviolet discharge pulse frequency is >10Hz, it is determined to be a continuous discharge.

6. The generator rotor inter-turn short-circuit fault diagnosis method based on multi-modal data fusion according to claim 5, characterized in that, Also includes: Vibration signals of the generator rotor during operation are collected by vibration sensors, and ultraviolet discharge signals are monitored in real time. Extract vibration signal features from vibration signals and extract ultraviolet discharge signal features from ultraviolet discharge signals; Based on at least two of the extracted electrical signal features, image features, and vibration signal features, one or more cross-validation parameters are calculated. The cross-validation parameters are used to quantify the consistency between multi-source data and, in the step of calculating the distribution of generator rotor inter-turn short-circuit faults, are used to correct the confidence level of the preliminary fault probability distribution or as a criterion for triggering manual review. The distribution of short-circuit faults between rotor turns of the generator is calculated based on the preprocessed image features, electrical signal features, vibration signal features, ultraviolet discharge signal features, and the corresponding weight coefficients of the image features, electrical signal features, vibration signal features, and ultraviolet discharge signal features.

7. The generator rotor inter-turn short-circuit fault diagnosis method based on multimodal data fusion according to claim 6, characterized in that, The method for dynamically adjusting the weight coefficients of preprocessed image features, electrical signal features, vibration signal features, and ultraviolet discharge signal features, as well as the corresponding image features, electrical signal features, vibration signal features, and ultraviolet discharge signal features, calculates the distribution of inter-turn short-circuit faults in the generator rotor as follows: P(x,y,z) =α·ΣE_i+β·ΣI_j+γ·ΣU_k+δ·ΣV_m+ε·ΣD_n; Where δ is the weight of the vibration signal feature, ε is the weight of the ultraviolet discharge signal feature; α+β+γ+δ+ε=1; V_m is the m-th vibration signal feature value, and D_n is the n-th ultraviolet discharge signal feature value; The method for dynamically adjusting the weighting coefficients corresponding to vibration signal characteristics and ultraviolet discharge signal characteristics includes: δ=0.1×sigmoid(F_vib / 12%), ε=0.05×tanh(E_UV / 1000); Where F_vib is the proportion of high-frequency vibration energy, and E_UV is the cumulative energy of ultraviolet discharge.

8. A generator rotor inter-turn short-circuit fault diagnosis system based on multi-modal data fusion, characterized in that, include: The multi-source data acquisition module synchronously acquires multi-source data during generator rotor operation when an inter-turn short-circuit fault occurs, including electrical parameters and image data; The data feature extraction module extracts electrical signal features from electrical parameters and image features from image data; The data feature preprocessing module uses an improved SIFT algorithm to preprocess image features, removes mismatched points in the image features, and uses a dynamic time warping algorithm to preprocess electrical signal features and align the timing of electrical signal features. The inter-turn short-circuit fault distribution calculation module calculates the generator rotor inter-turn short-circuit fault distribution based on preprocessed image features, electrical signal features, and a dynamic adjustment method for the weight coefficients of the corresponding image features and electrical signal features. The results output module outputs fault location results based on the distribution of short-circuit faults between generator rotor turns.

9. An electronic device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements the generator rotor turn-to-turn short-circuit fault diagnosis method based on multimodal data fusion as described in any one of claims 1-3.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the generator rotor turn-to-turn short-circuit fault diagnosis method based on multimodal data fusion as described in any one of claims 1-3.

Images