A method and system for comprehensive detection and evaluation of a generator without extracting a rotor

Abstract

This invention discloses a method and system for comprehensive testing and evaluation of generators without rotor removal, belonging to the field of condition monitoring and fault diagnosis technology for large rotating electrical machines. The method includes: collecting multi-source detection data; preprocessing and extracting features from the multi-source detection data to obtain multi-dimensional feature vectors; inputting the multi-dimensional feature vectors into a pre-constructed multi-level information fusion model for comprehensive evaluation; the multi-level information fusion model includes data layer normalization fusion, feature layer principal component analysis dimensionality reduction fusion, and decision layer fuzzy comprehensive evaluation fusion, outputting a quantitative score and health level of the generator's overall health status; and generating a test report containing recommendations on the necessity of rotor removal maintenance based on the quantitative score and health level. This invention can integrate multi-dimensional detection information of generators without rotor removal, realize multi-source data fusion evaluation, reduce maintenance risks and costs, provide data support for condition-based maintenance, and has engineering applicability.

Description

Technical Field

[0001] This invention belongs to the field of large rotating electric motor condition monitoring and fault diagnosis technology, specifically relating to a comprehensive testing and evaluation method and system for generators without removing the rotor. Background Technology

[0002] The long-term operational reliability of large generators is directly related to power grid security. As the units age, defects such as stator winding insulation aging, slot wedge loosening, short circuits between lamellar plates, and short circuits between rotor winding turns gradually emerge. Regular rotor overhaul is currently the main means of ensuring the safe operation of generators. However, this method has relatively high operating costs. During rotor removal, the air gap between the stator and rotor is small, and careless operation can lead to significant economic losses. Furthermore, rotor removal overhaul requires the removal of a large number of components such as end covers, hydrogen coolers, and fans, resulting in a long overall cycle and significant downtime losses. It also suffers from a high degree of randomness, as the traditional mandatory overhaul model lacks specificity. Many units, although meeting the service life requirements, are in good internal condition, and unnecessary rotor removal inspections result in wasted resources.

[0003] As an improvement, current technologies have been used to research and apply generator non-rotor inspection techniques, such as the development of air gap inspection robots and generator inspection robots. These technologies enable visual inspection of the stator bore, detection of slot wedge tightness, and detection of electromagnetic defects in the core, making non-rotor evaluation possible.

[0004] However, currently, in the testing of generators without rotor removal, the various testing methods are independent of each other, lacking a unified data platform and comprehensive evaluation model. Testing results are mostly isolated judgments, making it difficult to form quantitative conclusions about the overall health status of the generator. The detection of rotor electrical faults (such as inter-turn short circuits) still relies on traditional tests after rotor removal, without effective integration with in-machine testing, resulting in a lack of evaluation dimensions. Maintenance decisions lack objective basis and may still rely on expert experience, failing to achieve true condition-based maintenance. Therefore, in summary, current generator testing without rotor removal relies on independent testing methods without effective integration, resulting in relatively singular evaluation dimensions and difficulty in forming a comprehensive decision-making basis. Summary of the Invention

[0005] This invention provides a comprehensive testing and evaluation method and system for generators without rotor removal. The purpose is to solve the problem that in the current testing of generators without rotor removal, the various testing methods are independent of each other and there is no effective integration of the testing, resulting in a relatively single evaluation dimension and difficulty in forming a comprehensive decision-making basis.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a comprehensive testing and evaluation method for a generator without removing the rotor, comprising the following steps: S1. Simultaneously collect multi-source detection data through the detection platform; the multi-source detection data includes at least visual image data, slot wedge vibration data, iron core electromagnetic induction data, air gap distance data, and rotor electrical pulse response data; S2. Preprocess and extract features from the multi-source detection data to obtain multi-dimensional feature vectors including at least the visual anomaly index, slot wedge loosening index, core anomaly index, air gap non-uniformity, and rotor turn-to-turn short-circuit difference. S3. Input the multi-dimensional feature vectors into a pre-built multi-level information fusion model for comprehensive evaluation; Among them, the multi-level information fusion model includes at least data layer normalization fusion, feature layer principal component analysis dimensionality reduction fusion and decision layer fuzzy comprehensive evaluation fusion, and outputs the overall health status quantitative score and health level of the generator; S4. Based on the overall health status quantitative score and health level, generate an inspection report that includes suggestions on the necessity of rotor removal and maintenance, and complete the comprehensive inspection and evaluation of the generator without removing the rotor.

[0007] In some implementations, in S1, visual image data is acquired using a high-resolution industrial camera and a light source. After preprocessing the acquired visual image data, a visual anomaly index is calculated. The formula for calculating the visual anomaly index is as follows: ; in, The defect area is... For reference area, This is the defect type weighting factor.

[0008] In some implementations, in S1, the groove wedge vibration data is obtained by an accelerometer after the groove wedge is excited by an electromagnetically excited probe. The collected groove wedge vibration data is then subjected to spectral analysis to calculate the groove wedge loosening index. The formula for calculating the groove wedge loosening index is as follows: ; in, The loosening index of a single slot wedge. This represents the dominant frequency offset of a single slotted wedge vibration signal. The threshold constant is This represents the ratio of the energy of a single slotted wedge vibration signal to the total energy within the frequency band. The energy ratio reference value for a normal slot wedge. is a threshold constant; the maximum value of all slot wedge loosening indices is taken as the slot wedge loosening index in the multidimensional feature vector.

[0009] In some implementations, in S1, the electromagnetic induction data of the iron core is collected by the electromagnetic induction detection unit to obtain the effective value sequence of the induced voltage. After processing the effective value sequence of the induced voltage, the iron core anomaly index is calculated. The calculation formula of the iron core anomaly index is as follows: ; in, This refers to the core anomaly index. This represents the maximum deviation value of the induced voltage. For reference voltage, This represents the length of the abnormal region due to electromagnetic defects in the iron core. This is a reference length.

[0010] In some implementations, in S1, the air gap distance data is obtained by measuring the distance between the stator and rotor point by point using a laser displacement sensor. The air gap non-uniformity is calculated based on all the measured air gap distance values. The formula for calculating the air gap non-uniformity is as follows: ; in, For air gap non-uniformity, This represents the maximum value of the measured air gap distance. This is the minimum value of the measured air gap distance. This represents the average value of the measured air gap distance.

[0011] In some implementations, in S1, the rotor electrical pulse response data is obtained by acquiring the response waveforms of the positive and negative poles of the rotor through a repetitive pulse method detection device. After processing the response waveforms, the rotor inter-turn short-circuit difference is calculated. The calculation formula for the rotor inter-turn short-circuit difference is as follows: ; ; in, The difference in short circuit between rotor turns, In response to the length of the waveform, This is the response waveform of the rotor's positive pole. The response waveform is for the negative pole of the rotor.

[0012] In some implementations, in S4, data layer fusion involves normalizing the multidimensional feature vectors to eliminate the influence of dimensions; feature layer fusion involves using principal component analysis to reduce the dimensionality of the normalized feature vectors, calculating the sample covariance matrix of the feature vectors and solving the characteristic equation, selecting a preset number of principal components to form a transformation matrix, ensuring that the cumulative variance contribution rate of the preset number of principal components is not lower than a preset threshold, and obtaining the fused feature vectors through the transformation matrix.

[0013] In some implementations, S3 specifically includes: S31. Divide the generator into three key components: stator winding, stator core, and rotor winding. Map the features in the multidimensional feature vector to each key component according to their physical correlation to form the feature vector of each component. S32. Design a membership function for the health level for each component’s feature vector. The health level shall include at least three levels: healthy, attentive, and severe. S33. Use the analytic hierarchy process (AHP) to determine the weights of the internal features of each component and the weights between the key components. S34. Calculate the membership vector of each component based on the weights and membership functions of the internal features of each component. S35. The membership vectors of each component are weighted and fused according to the weights between each key component to obtain the comprehensive membership vector of the overall health status of the generator. The level corresponding to the largest membership degree in the comprehensive membership vector is taken as the overall health level of the generator. S36. Calculate the quantitative score of the overall health status of the generator based on the comprehensive membership vector.

[0014] In some implementations, the formula for calculating the quantitative score in S4 is as follows: ; in, A quantitative score for the overall health status of the generator. The degree of membership in the health level. To ensure the membership degree of the hierarchy, The degree of membership for severity level, The weighting coefficient for health level. To pay attention to the weighting coefficients of the grades, This represents the weighting factor for the severity level.

[0015] This invention also provides a comprehensive testing and evaluation system for generators without rotor extraction, to implement the aforementioned comprehensive testing and evaluation method for generators without rotor extraction. The system includes a multi-source data acquisition module, a multi-dimensional feature vector extraction module, a comprehensive evaluation module, and a testing and evaluation module, wherein: The multi-source data acquisition module is used to: synchronously acquire multi-source detection data through the detection platform; the multi-source detection data includes at least visual image data, slot wedge vibration data, iron core electromagnetic induction data, air gap distance data, and rotor electrical pulse response data; The multidimensional feature vector extraction module is used to: preprocess and extract features from multi-source detection data to obtain multidimensional feature vectors including at least the visual anomaly index, slot wedge loosening index, core anomaly index, air gap non-uniformity, and rotor turn-to-turn short-circuit difference. The comprehensive evaluation module is used to: input multi-dimensional feature vectors into a pre-built multi-level information fusion model for comprehensive evaluation; Among them, the multi-level information fusion model includes at least data layer normalization fusion, feature layer principal component analysis dimensionality reduction fusion and decision layer fuzzy comprehensive evaluation fusion, and outputs the overall health status quantitative score and health level of the generator; The detection and evaluation module is used to: generate a detection report containing recommendations on the necessity of rotor removal and maintenance based on the overall health status quantitative score and health level, and complete the comprehensive detection and evaluation of the generator without removing the rotor.

[0016] Compared with the prior art, the comprehensive testing and evaluation method and system for generators without removing the rotor of the present invention has the following beneficial effects: This invention presents a comprehensive generator inspection and evaluation method that does not require rotor removal. The method offers a relatively comprehensive range of inspection dimensions and strong information complementarity. By integrating visual, slot wedge, core, air gap, and rotor electrical inspection methods, it covers the main potential fault types of generator stators and rotors, avoiding the limitations of single inspection methods. This invention achieves multi-source information fusion evaluation. Through a multi-level fusion model at the data layer, feature layer, and decision layer, it transforms scattered inspection results into unified quantitative scores and maintenance recommendations, overcoming the drawbacks of relying on expert experience and reducing maintenance risks and costs. Comprehensive inspection can be completed without rotor removal, relatively improving the operational risks associated with rotor removal. Experimental tests show that the maintenance period of this invention can be shortened from the traditional 20-30 days to 5-7 days, saving power generation companies relatively significant downtime losses. The evaluation results are traceable and iterative. Inspection data and evaluation reports are digitally stored, facilitating historical data comparison and trend analysis. It supports full life-cycle management of equipment, is easy to promote, and is applicable to various large generator sets with air gaps of 10mm or more, including thermal power, nuclear power, and synchronous condenser sets, demonstrating promising industrial application prospects. Attached Figure Description

[0017] The accompanying drawings are provided to further understand the invention and constitute a part of this invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0018] Figure 1 This is a flowchart illustrating a comprehensive testing and evaluation method for a generator without removing the rotor, according to the present invention.

[0019] Figure 2 This is a schematic diagram of the structure of a movable testing platform in an embodiment of a generator comprehensive testing and evaluation method without removing the rotor according to the present invention; Figure 3 This is a schematic diagram of the multi-source information fusion evaluation model architecture provided in an embodiment of the generator comprehensive detection and evaluation method without rotor extraction according to the present invention. Figure 4 This is a schematic diagram of a sample comprehensive test report generated as an embodiment of the comprehensive test and evaluation method for a generator without removing the rotor according to the present invention; Figure 5 This is a schematic diagram of the RSO offline test results, which is an embodiment of the comprehensive testing and evaluation method for generators without rotor extraction according to the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0022] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0023] It should be noted that the apparatus and methods disclosed in the embodiments herein can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments herein. In this regard, each block in a flowchart or block diagram may represent a module, program, or part of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system to perform the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0024] In addition, the functional modules in the various embodiments of this article can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0025] How to provide a comprehensive testing method that can integrate multi-dimensional detection information, achieve multi-source data fusion evaluation, and provide a scientific basis for decision-making regarding the necessity of generator rotor removal.

[0026] Based on this, the present invention provides a comprehensive testing and evaluation method for generators without removing the rotor, comprising the following steps: S1. Simultaneously collect multi-source detection data through the detection platform; the multi-source detection data includes at least visual image data, slot wedge vibration data, iron core electromagnetic induction data, air gap distance data, and rotor electrical pulse response data; S2. Preprocess and extract features from the multi-source detection data to obtain multi-dimensional feature vectors including at least the visual anomaly index, slot wedge loosening index, core anomaly index, air gap non-uniformity, and rotor turn-to-turn short-circuit difference. S3. Input the multi-dimensional feature vectors into a pre-built multi-level information fusion model for comprehensive evaluation; Among them, the multi-level information fusion model includes at least data layer normalization fusion, feature layer principal component analysis dimensionality reduction fusion and decision layer fuzzy comprehensive evaluation fusion, and outputs the overall health status quantitative score and health level of the generator; S4. Based on the overall health status quantitative score and health level, generate an inspection report that includes suggestions on the necessity of rotor removal and maintenance, and complete the comprehensive inspection and evaluation of the generator without removing the rotor.

[0027] This invention deploys a mobile inspection platform into the air gap between the generator stator and rotor. It simultaneously collects multi-source data from within the rotor housing using integrated sensors including vision, slot wedge loosening detection, ELCID, air gap ranging, and RSO. The collected data undergoes preprocessing and feature extraction to construct a multi-dimensional feature vector containing visual anomaly indices, slot wedge loosening indices, core anomaly indices, air gap non-uniformity, and rotor turn-to-turn short-circuit differences. This feature vector is then input into a pre-constructed multi-level information fusion model, which includes data layer normalization, feature layer principal component analysis dimensionality reduction, and decision layer fuzzy comprehensive evaluation. The model outputs a quantitative score and health level for the overall generator health status. Finally, based on the score, an inspection report is generated, including recommendations on the necessity of rotor removal. This achieves a multi-dimensional comprehensive assessment of the generator without rotor removal, significantly reducing maintenance risks and costs, and providing a scientific basis for condition-based maintenance decisions. like Figure 1 As shown, this invention provides a comprehensive testing and evaluation method for generators without removing the rotor. This method is applied to the scenario of comprehensive testing and evaluation of large synchronous generators without removing the rotor, and is performed according to the following steps: Step S100: Deployment of the detection platform and entry into the chamber.

[0028] After the generator end cover is removed, a mobile testing platform is deployed at the air gap inlet. The platform adopts a wall-climbing or guide rail structure and moves between the stator and rotor air gaps through magnetic adsorption or guide rail guidance to enter the testing area inside the stator bore.

[0029] Step S200: Synchronous acquisition of multi-source detection data.

[0030] The testing platform travels along a preset path, carrying the following sensor modules sequentially or simultaneously to collect multi-dimensional data from the stator bore and rotor surface.

[0031] Visual inspection: A high-resolution industrial camera (resolution ≥1920×1080) equipped with an LED ring light source is used to acquire images of the stator winding surface, ends, slot wedges, ventilation ducts, and rotor surface. The camera's focal length is adjustable, enabling clear imaging within a 10mm wide air gap. The acquired image data is indexed according to slot number and axial position to form an image sequence set. ; in, The number of stator slots The number of images acquired per slot.

[0032] Slot wedge loosening detection: An electromagnetically excited probe is used to excite the slot wedge to vibrate in a non-contact manner. The excitation signal is a swept-frequency sine wave with a frequency range of 500Hz to 5kHz. The vibration response is picked up by an accelerometer and then conditioned by a charge amplifier before acquisition. For the first... i Troughk A slotted wedge is used to obtain the time-domain waveform of the vibration signal. The sampling rate is not less than 50kHz and the sampling length is not less than 0.2s.

[0033] Core electromagnetic defect detection: Electromagnetic induction method (ELCID) detection technology is employed. The detection unit consists of an excitation winding and a Chattock coil. A 50Hz alternating current is applied to the excitation winding, generating magnetic flux along the inner circumference of the stator; the Chattock coil moves along the stator axis to measure the local magnetomotive force difference. (Regarding axial position...) The induced voltage signal was measured. Its effective value is denoted as The unit is Abnormal core loss is caused by The local elevation is characterized.

[0034] Air gap parameter detection: Equipped with a laser displacement sensor (measuring range 5~50mm, accuracy ±0.1mm), the distance between the stator and rotor is measured point by point along the inner wall of the stator. For the first... Trough At each axial position, the air gap value is obtained. Calculate the air gap non-uniformity index based on all measurements: ; in, For air gap non-uniformity, This represents the maximum value of the measured air gap distance. This is the minimum value of the measured air gap distance. This represents the average value of the measured air gap distance.

[0035] Rotor electrical condition detection: After in-bores testing, a repetitive pulse (RSO) detection device is used to apply repetitive pulses (pulse rise time ≤ 100ns, amplitude 5~20V) while the rotor is stationary, and the positive and negative responses are collected. The sampling rate is not less than 500MHz, and the recorded waveform length corresponds to the entire length of the rotor.

[0036] Step S300: Preprocessing and feature extraction of multi-source detection data.

[0037] The various types of data collected in step S200 are preprocessed and feature quantized to form input feature vectors in a unified format.

[0038] S310: Visual image processing and feature extraction.

[0039] For each frame of image Perform the following processing: Image enhancement: Histogram equalization and adaptive contrast enhancement are employed to improve details in low-light areas.

[0040] Defect detection: Semantic segmentation models based on convolutional neural networks (CNNs), such as U-Net, identify defect regions such as insulation damage, electrolytic corrosion, and discharge traces. Let the... i The area of ​​the defect detected in the image is (Number of pixels), converted to actual area based on camera calibration. (mm²). Define the visual anomaly index: ; in ， This is a reference area (10 mm² can be used). The defect type weighting factor is set (e.g., 1.5 for insulation damage, 1.0 for electrolytic corrosion, and 1.2 for discharge marks).

[0041] Finally, extract from all images The maximum value is used as a visual detection feature. .

[0042] S320: Groove wedge vibration signal processing and feature extraction.

[0043] (1) Vibration signal for each slotted wedge Perform spectral analysis to extract the following features: (2) Main frequency offset Calculate the difference between the peak frequency of the spectrum and the average peak frequency of the normal slot wedge.

[0044] (3) Energy ratio The ratio of energy to total energy in the 1~3kHz frequency band.

[0045] (4) Define the slot wedge loosening index: ; in, and It is a threshold constant (determined by statistical analysis of historical data). This is the baseline value for the normal slot wedge energy ratio.

[0046] Finally, the maximum value of all loosening indices is taken as the groove wedge feature. .

[0047] S330: ELCID signal processing and feature extraction.

[0048] For the effective value sequence of induced voltage Median filtering is performed to remove noise, and the differential signal is calculated. Searching A continuous region, recording the region length and peak value. Define the core anomaly index: ; in, This is the reference voltage (50μV can be used). The length of the abnormal region. This is a reference length (10mm is acceptable).

[0049] Finally, As a characteristic of iron core .

[0050] S340: Air Gap Data Processing and Feature Extraction Based on all air gap measurements ,calculate: (1) Air gap non-uniformity ; (2) Minimum air gap ,like If the value is less than the minimum allowable value (e.g., 8mm), an alarm will be triggered.

[0051] Finally, take As a characteristic of the air gap.

[0052] S350: RSO waveform processing and feature extraction.

[0053] For waveforms and Perform time-domain alignment and calculate the differential waveform. Let the waveform length be... Define the difference index: ; in, The difference in short circuit between rotor turns, In response to the length of the waveform, This is the response waveform of the rotor's positive pole. The response waveform is for the negative pole of the rotor.

[0054] like (The threshold can be set to 0.1), indicating the presence of an inter-turn short circuit. This is determined based on the timing of the differential pulse occurrence. Calculate the distance to the fault: in The pulse propagation speed can be taken as 150 m / μs.

[0055] Finally, take As a rotor electrical characteristic.

[0056] The above features are combined into an initial feature vector: .

[0057] Step S400: Comprehensive evaluation based on the multi-level information fusion model. The feature vector extracted in step S300 is input into the pre-constructed multi-level information fusion model, which includes three levels: data layer fusion, feature layer fusion, and decision layer fusion.

[0058] S410: Data layer fusion; The initial eigenvectors are normalized to eliminate the influence of dimensions.

[0059] S420: Feature layer fusion; Principal Component Analysis (PCA) is used to reduce the dimensionality of the normalized features and extract principal components. Let the normalized feature vector after data fusion be... (Corresponding to 5 original features). First, based on historical sample sets or multiple detection samples collected on-site, the sample covariance matrix of the feature vectors is calculated. : ; in, For the sample size, Let be the sample mean vector. Solve for the characteristic equation of the covariance matrix: ; Obtain eigenvalues and the corresponding unit orthogonal eigenvectors Each eigenvector represents a principal component direction, and the eigenvalues ​​represent the variance contribution of the corresponding principal component.

[0060] Before selection One principal component is selected to ensure that the cumulative variance contribution rate reaches a preset threshold (usually ≥90%): .

[0061] Constructing the transformation matrix The column vectors are the selected feature vectors. Then the fused feature vectors (i.e., principal component scores) are: ; in, The components are the first to the second. Principal component score. In this embodiment, it is typically taken as... =3, which allows the dimensionality-reduced feature vectors to retain most of the information of the original features and facilitate subsequent fusion processing by the decision layer.

[0062] S430: Integration of Decision-Making Levels Based on feature layer dimensionality reduction and fusion, this step uses fuzzy comprehensive evaluation method to independently assess the health status of each key component of the generator, and then obtains a quantitative score and level determination of the overall health status of the generator through weighted fusion.

[0063] S431: Component division and feature allocation.

[0064] The generator system is divided into three key components: stator winding (C1), stator core (C2), and rotor winding (C3). Based on physical correlations, the original features before feature layer fusion (or principal components after PCA dimensionality reduction) are mapped to each component, with the specific mapping relationships as follows: Stator winding C1: Primarily associated with visual anomaly index and air gap non-uniformity (Uneven air gap may cause winding vibration and wear). Stator core C2: Mainly related slot wedge loosening index and core anomaly index ; Rotor winding C3: Primarily related to the difference in rotor inter-turn short circuits .

[0065] For each component, its associated features are selected to form a component feature vector, denoted as . .

[0066] S432: Membership function design.

[0067] For each component, membership functions are defined for three health levels (Healthy H, Attention A, Severe S). The membership functions employ trapezoidal or triangular distributions, with thresholds set based on historical data or expert experience. Taking stator core C2 as an example, membership functions are constructed accordingly.

[0068] For features : ; ; ; For features : ; ; ; Similarly, define the membership functions for each level of other components. (The threshold for calculating the membership degree can be adjusted according to actual working conditions and needs.)

[0069] S433: Determine the weight within the component.

[0070] For each component, the contributions of its internal features to its health status differ, requiring weight assignment. The Analytic Hierarchy Process (AHP) is used to determine these weights. Taking stator core C2 as an example, pairwise comparison matrices are constructed, and the normalized eigenvector corresponding to the largest eigenvalue is calculated to obtain the feature weight vector. The sum of the elements is 1. Similarly, we can obtain... and .

[0071] S434: Calculation of component membership vector.

[0072] For components Its relation to level The membership degree is obtained by weighted average of the membership degrees of each feature: ; in =1 corresponds to health (H) =2 corresponds to note (A), =3 corresponds to severity (S). Normalize the calculation results (optional) to obtain the component membership vector. .

[0073] S435: Determining the weights between components.

[0074] Different components have varying degrees of importance to the overall health of the generator, necessitating the determination of weight vectors among these components. The importance of components can be determined by pairwise comparisons using the AHP method.

[0075] S436: Comprehensive membership degree and overall health level.

[0076] By weighting the membership vectors of each component according to their respective component weights, the comprehensive membership vector of the generator's overall health status is obtained. : ; The level corresponding to the highest membership degree is taken as the overall health level of the generator, with corresponding levels (1-healthy, 2-caution, 3-serious).

[0077] S437: Overall Health Score.

[0078] To provide more refined quantitative results, a comprehensive health score is defined. : ; in, These are the weighting coefficients for different levels of health, attention, and severity, respectively. The degree of membership in the health level. To ensure the membership degree of the hierarchy, This represents the membership degree of the severity level. In this embodiment, we take... =1.0, =0.6, =0.2. The selection of this set of coefficients is based on the following considerations: The coefficient values ​​ensure that the overall score naturally aligns with the decision threshold: when the membership vector is completely uncertain (i.e. = = When =1 / 3), =100×(1 / 3×1.0+1 / 3×0.6+1 / 3×0.2)=60, which falls exactly at the lower limit of the "attention" range; When health and attention are equally balanced ( = =0.5, When =0), =100×(0.5×1.0+0.5×0.6)=80, corresponding to the dividing line between "healthy" and "attention". Therefore, the score varies continuously between 0 and 100, with a high score indicating good health and a low score indicating serious illness.

[0079] Step S500: Maintenance Decision Support and Report Generation. Based on the comprehensive score and thresholds for each indicator, generate an inspection report containing the following: (1) Quantitative scoring of the health status of each component and a list of defects; (2) Overall health level of the generator (e.g., "Good", "Caution", "Severe"); (3) Recommendations on the necessity of rotor removal and maintenance (e.g., "No need to remove the rotor, it can continue to operate for X years", "It is recommended to arrange rotor removal and maintenance within Y years", "Arrange rotor removal and maintenance immediately"). (4) Attach the original detection map and images of abnormal areas for expert review.

[0080] The decision-making rules are as follows: (1) If If the score is ≥80 and all sub-targets are not "severe", it is judged as "good". It is recommended that there is no need to remove the rotor and it can continue to operate for more than 5 years. (2) If 60≤ If the value is less than 80 or any sub-target is marked "Caution" but not "Serious", it is judged as "Caution" and it is recommended to arrange rotor inspection and maintenance within 3 years. (3) If If the value is less than 60 or any sub-target is "serious", it is deemed "serious" and it is recommended to arrange for rotor removal and maintenance immediately.

[0081] This invention also provides a comprehensive testing and evaluation system for generators without rotor extraction, comprising a multi-source data acquisition module, a multi-dimensional feature vector extraction module, a comprehensive evaluation module, and a testing and evaluation module, wherein: The multi-source data acquisition module is used to: synchronously acquire multi-source detection data through the detection platform; the multi-source detection data includes at least visual image data, slot wedge vibration data, iron core electromagnetic induction data, air gap distance data, and rotor electrical pulse response data; The multidimensional feature vector extraction module is used to: preprocess and extract features from multi-source detection data to obtain multidimensional feature vectors including at least the visual anomaly index, slot wedge loosening index, core anomaly index, air gap non-uniformity, and rotor turn-to-turn short-circuit difference. The comprehensive evaluation module is used to: input multi-dimensional feature vectors into a pre-built multi-level information fusion model for comprehensive evaluation; Among them, the multi-level information fusion model includes at least data layer normalization fusion, feature layer principal component analysis dimensionality reduction fusion and decision layer fuzzy comprehensive evaluation fusion, and outputs the overall health status quantitative score and health level of the generator; The detection and evaluation module is used to: generate a detection report containing recommendations on the necessity of rotor removal and maintenance based on the overall health status quantitative score and health level, and complete the comprehensive detection and evaluation of the generator without removing the rotor.

[0082] In some practical operating conditions, the system deployment of the present invention can be selected to include: Mobile inspection platform: Used to carry multiple sensor modules and move between the stator and rotor air gaps. The mobile inspection platform includes an arc-shaped main frame, a magnetic track drive unit, an embedded controller, a wireless communication module, and a power management unit. The minimum thickness is ≤30mm, and it can pass through a minimum air gap of 10mm.

[0083] Multi-sensor integrated module: including vision module, slot wedge detection module, ELCID detection module, air gap ranging module and RSO detection unit, each module adopts standardized interface for quick replacement.

[0084] Data acquisition and preprocessing unit: used for sensor signal conditioning, analog-to-digital conversion and preliminary feature extraction, with sampling rates meeting the requirements of each sensor (real-time compressed transmission of visual images, vibration signals ≥50kHz, ELCID signals ≥1kHz, RSO signals ≥500MHz).

[0085] Information fusion and evaluation server: Deployed with a multi-level information fusion model, it performs comprehensive evaluation and decision support for steps S300 to S500. The server is configured with a high-performance CPU / GPU and runs data processing and fusion algorithms developed in MATLAB or Python.

[0086] Human-computer interaction terminal: used for monitoring the detection process, setting parameters and viewing reports, including portable computers or tablets with dedicated monitoring software installed.

[0087] The present invention will be further described in detail below through specific embodiments.

[0088] This embodiment takes a 600MW water-hydrogen-cooled steam turbine generator in a power plant as an example.

[0089] 1) Detection background The generator has been in operation for 10 years and, according to regulations, requires a major overhaul with rotor removal. To assess the actual necessity of the overhaul, the power plant first conducted a comprehensive inspection without rotor removal using the system of this invention. The generator's air gap width is 25mm, meeting the entry requirements of the inspection platform.

[0090] 2) Deployment of the testing platform and data acquisition After removing the end covers at both ends of the generator and some fan components, a wall-climbing inspection robot was deployed at the air gap inlet at the steam end. The robot travels one revolution along the inner wall of the stator, sequentially carrying a vision module, a slot wedge detection module, and an ELCID module to collect data. After the data collection is completed, an RSO detector is connected at the rotor end junction box to collect the rotor winding pulse response waveform.

[0091] Example of data collection results: Visual inspection: After processing, images acquired from the end of the upper layer bar in the 15th slot of phase A were used to detect suspected electro-corrosion defect areas, with an area of ​​[missing information]. =18mm², Defect type weight =1.2, reference area =10 mm², calculated to =(18 / 10)×1.2=2.16.

[0092] Slot wedge testing: A total of 252 slot wedges in 42 slots were tested, among which the dominant frequency of the vibration spectrum of the second slot wedge in the 8th slot was shifted. =25Hz, energy ratio R=0.72, reference value =10Hz, =0.55, =0.10, calculate the loosening index L = 25 / 10 + |0.72 - 0.55| / 0.10 = 4.2. Maximum loosening index of the full-groove wedge. =4.2.

[0093] ELCID detection: in axial position The effective value of the induced voltage was measured at a distance of 1.25m. =82μV, average background voltage 35μV, threshold =15μV, length of abnormal region =35mm, reference value =50μV, =10mm, calculate the core anomaly index =((82-35) / 50)×(35 / 10)=(47 / 50)×3.5=3.29, take =3.29.

[0094] Air gap detection: Maximum deviation of air gap measurement value =4.5mm, average value =24.8 mm, non-uniformity =4.5 / 24.8=0.181, take =0.181.

[0095] RSO detection: The difference between the positive and negative waveforms is calculated. =0.18, threshold =0.1, indicating an inter-turn short circuit. The fault distance is calculated as follows: =150m / μs×22.5μs / 2=1687.5mm (corresponding to the 4th coil of the rotor body, RSO offline test result is as follows) Figure 5 As shown), take =0.18.

[0096] 3. Multi-source data fusion evaluation.

[0097] 3.1 First, normalize the features and set the maximum and minimum values ​​for each feature (based on historical statistics):

[0098] ; .

[0099] After normalization, we obtain the normalized vector. = .

[0100] 3.2 PCA dimensionality reduction is used. Based on the historical sample covariance matrix, the cumulative variance contribution rate of the first three principal components is calculated to be 91%. Taking m=3, and assuming the principal component transformation matrix is ​​known, the principal component scores are obtained. (Example value).

[0101] 3.3 The decision-making level adopts fuzzy comprehensive evaluation. Taking the stator core as an example, for normalized features... Substitute into the membership function: ; For normalized features Substitute into the membership function: ; Take the weight within the component ; Therefore, the membership vector of the stator core .

[0102] Similar to calculating the membership vectors of the stator and rotor windings. Assume the weights of each sub-objective are determined by AHP: [0.3, 0.5, 0.2]. Assuming we obtain: ; Then the overall membership degree =0.3×[0.6, 0.3, 0.1]+0.5×[0.0, 0.3227, 0.6773]+0.2×[0.2, 0.7, 0.1]=[0.22,0.3913,0.3887].

[0103] Maximum membership degree: 0.3913.

[0104] Overall score calculation: =100*(1.0*0.22 + 0.6*0.3913 + 0.2*0.3887) =53.252.

[0105] 4. Maintenance Decision and Report Generation (S500) Overall score: 53.252 <60), the system determines the overall level as "serious" and recommends "immediately arrange rotor inspection and repair". The report outputs the following key defects: local abnormality of stator core (ELCID abnormality index 3.29), short circuit between rotor turns (difference degree 0.18), and severe loosening of slot wedge in slot 8 (loosening index 4.2).

[0106] 5. Verification The power plant adopted the suggestion and scheduled a rotor inspection two years later. After the rotor was inspected, it was found that there were indeed two inter-laminar short circuits in the 12th slot core, the insulation between the turns of the 4th coil of the rotor was partially carbonized, and the slot wedge of the 8th slot was obviously loose. These findings were highly consistent with the detection results of this invention, verifying the accuracy of the fusion evaluation model.

[0107] The construction process of the fusion model of this invention includes feature selection, membership function design, and weight determination.

[0108] 1. Establishment of the feature sample database To construct a universally applicable integrated evaluation model, we first collected non-rotor inspection data from multiple generators of the same type (capacity 300MW~1000MW) during various maintenance operations, along with corresponding actual condition records after rotor removal maintenance. The sample library contains the following two types of data: (1) Detection feature data: 5-dimensional normalized feature vectors extracted in steps S200~S300: ; Each sample corresponds to one test record.

[0109] (2) Real status label: Based on the results of rotor inspection, experts will classify the overall health status of the generator into three levels: healthy (H), alert (A), and serious (S), which will be used as sample labels.

[0110] The sample library contains at least 200 sets of valid data, covering various health states and defect types, to ensure the model's generalization ability.

[0111] 2. Calibrating the threshold of the membership function Stator core characteristics For example, the method for determining the threshold (0.3, 0.6, 0.9) in the membership function is as follows: Extract all samples labeled "healthy" from the sample library and perform statistical analysis. The distribution is used, and the 85th percentile is taken as the boundary between health and attention (0.3 in this example), that is, 85% of the healthy samples... Less than 0.3.

[0112] Extract all samples labeled "severe" and perform statistical analysis. The distribution is used, and the 15th percentile is taken as the boundary between attention and severity (0.6 in this example), that is, 85% of the severe samples... Greater than 0.6.

[0113] Taking 0.9 as the saturation threshold, when A value >0.9 falls entirely in the severe category; this value is determined by experts based on extreme operating conditions.

[0114] The thresholds for other features are calibrated using a similar method, ultimately forming a complete membership function parameter table.

[0115] 3. Weight Determination (1) Weights between components The importance of the stator winding (C1), stator core (C2), and rotor winding (C3) is compared pairwise, and a judgment matrix A is constructed using a 1-9 scale method: ; Calculate the largest eigenvalue of the judgment matrix =3.0092, the corresponding normalized eigenvector is the weight vector, which, after rounding, is: =[0.3,0.5,0.2].

[0116] 3.2 Component Intra-component Weights Taking stator core C2 as an example, its characteristics are: and The expert judgment matrix is ​​as follows: ; Calculated weights =[0.6,0.4]. Similarly, the stator winding weights are obtained. =[0.7,0.2], which corresponds to and Rotor winding weight =[1], corresponding to .

[0117] 4. PCA Principal Component Analysis Parameter Settings Based on all normalized feature vectors in the sample database Calculate the covariance matrix S. Solve for the eigenvalues. and corresponding feature vectors .

[0118] Select the first m principal components whose cumulative variance contribution rate is ≥0.90. For this sample database, the cumulative contribution rate of the first 3 principal components is calculated to be 91.2%, so m=3 is chosen. Save the transformation matrix. This is used for dimensionality reduction of subsequent real-time detection data.

[0119] 5. Determination of the weighting coefficients for the overall scoring level Comprehensive scoring formula The weighting coefficients for the ratings were determined through an optimization method. Using the true labels of the samples in the sample library as a benchmark, a grid search method was employed to maximize the correlation between the ratings and expert experience. In this embodiment, the optimization result is as follows: α 1 = 1.0, α 2 = 0.6, α =3=0.2, this set of coefficients makes the score continuously distributed between 0 and 100.

[0120] Through the steps described above, the method of the present invention, based on a pre-built stable and reliable multi-source information fusion evaluation model, can be directly deployed on an information fusion and evaluation server for comprehensive evaluation of actual generator rotor-free testing, and has good effectiveness and engineering applicability.

[0121] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Anyone skilled in the art can readily implement the present invention according to the description and above. Any modifications, alterations, or equivalent variations made using the technical content disclosed above are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A comprehensive testing and evaluation method for a generator without removing the rotor, characterized in that, Includes the following steps: S1. Simultaneously collect multi-source detection data through the detection platform; the multi-source detection data includes at least visual image data, slot wedge vibration data, iron core electromagnetic induction data, air gap distance data, and rotor electrical pulse response data; S2. Preprocess and extract features from the multi-source detection data to obtain multi-dimensional feature vectors including at least the visual anomaly index, slot wedge loosening index, core anomaly index, air gap non-uniformity, and rotor turn-to-turn short-circuit difference. S3. Input the multi-dimensional feature vectors into a pre-built multi-level information fusion model for comprehensive evaluation; Among them, the multi-level information fusion model includes at least data layer normalization fusion, feature layer principal component analysis dimensionality reduction fusion and decision layer fuzzy comprehensive evaluation fusion, and outputs the overall health status quantitative score and health level of the generator; S4. Based on the overall health status quantitative score and health level, generate an inspection report that includes suggestions on the necessity of rotor removal and maintenance, and complete the comprehensive inspection and evaluation of the generator without removing the rotor.

2. The comprehensive testing and evaluation method for generators without rotor extraction according to claim 1, characterized in that, In step S1, visual image data is acquired using a high-resolution industrial camera and a light source. After preprocessing the acquired visual image data, a visual anomaly index is calculated. The formula for calculating the visual anomaly index is as follows: ； in, The defect area is... For reference area, This is the defect type weighting factor.

3. The comprehensive testing and evaluation method for generators without rotor extraction according to claim 1, characterized in that, In step S1, the slot wedge vibration data is obtained by an accelerometer after the slot wedge is excited by an electromagnetic excitation probe. The collected slot wedge vibration data is then subjected to spectral analysis to calculate the slot wedge loosening index. The formula for calculating the slot wedge loosening index is as follows: ； in, The loosening index of a single slot wedge. This represents the dominant frequency offset of a single slotted wedge vibration signal. The threshold constant is This represents the ratio of the energy of a single slotted wedge vibration signal to the total energy within the frequency band. The energy ratio reference value for a normal slot wedge. is a threshold constant; the maximum value of all slot wedge loosening indices is taken as the slot wedge loosening index in the multidimensional feature vector.

4. The comprehensive testing and evaluation method for generators without rotor extraction according to claim 1, characterized in that, In step S1, the electromagnetic induction data of the iron core is collected by the electromagnetic induction detection unit to obtain the effective value sequence of the induced voltage. After processing the effective value sequence of the induced voltage, the iron core anomaly index is calculated. The calculation formula of the iron core anomaly index is as follows: ； in, This refers to the core anomaly index. This represents the maximum deviation value of the induced voltage. For reference voltage, This represents the length of the abnormal region due to electromagnetic defects in the iron core. This is a reference length.

5. The comprehensive testing and evaluation method for generators without rotor extraction according to claim 1, characterized in that, In step S1, the air gap distance data is obtained by measuring the distance between the stator and rotor point by point using a laser displacement sensor. The air gap non-uniformity is calculated based on all the measured air gap distance values. The formula for calculating the air gap non-uniformity is as follows: ； in, For air gap non-uniformity, This represents the maximum value of the measured air gap distance. This is the minimum value of the measured air gap distance. This represents the average value of the measured air gap distance.

6. The comprehensive testing and evaluation method for generators without rotor extraction according to claim 1, characterized in that, In step S1, the rotor electrical pulse response data is obtained by acquiring the response waveforms of the positive and negative poles of the rotor through a repetitive pulse method detection device. After processing the response waveforms, the rotor inter-turn short-circuit difference is calculated. The calculation formula for the rotor inter-turn short-circuit difference is as follows: ； ； in, The difference in short circuit between rotor turns, In response to the length of the waveform, This is the response waveform of the rotor's positive pole. The response waveform is for the negative pole of the rotor.

7. The comprehensive testing and evaluation method for generators without rotor extraction according to claim 1, characterized in that, In S3, data layer fusion involves normalizing the multidimensional feature vectors to eliminate the influence of dimensions; feature layer fusion involves using principal component analysis to reduce the dimensionality of the normalized feature vectors, calculating the sample covariance matrix of the feature vectors and solving the characteristic equation, selecting a preset number of principal components to form a transformation matrix, ensuring that the cumulative variance contribution rate of the preset number of principal components is not lower than a preset threshold, and obtaining the fused feature vectors through the transformation matrix.

8. The comprehensive testing and evaluation method for generators without rotor extraction according to claim 1, characterized in that, S3 specifically includes: S31. Divide the generator into three key components: stator winding, stator core, and rotor winding. Map the features in the multidimensional feature vector to each key component according to their physical correlation to form the feature vector of each component. S32. Design a membership function for the health level for each component’s feature vector. The health level shall include at least three levels: healthy, attentive, and severe. S33. Use the analytic hierarchy process (AHP) to determine the weights of the internal features of each component and the weights between the key components. S34. Calculate the membership vector of each component based on the weights and membership functions of the internal features of each component. S35. The membership vectors of each component are weighted and fused according to the weights between each key component to obtain the comprehensive membership vector of the overall health status of the generator. The level corresponding to the largest membership degree in the comprehensive membership vector is taken as the overall health level of the generator. S36. Calculate the quantitative score of the overall health status of the generator based on the comprehensive membership vector.

9. The comprehensive testing and evaluation method for generators without rotor extraction according to claim 1, characterized in that, In S4, the formula for calculating the quantitative score is as follows: ； in, A quantitative score for the overall health status of the generator. The degree of membership in the health level. To ensure the membership degree of the hierarchy, The degree of membership for severity level, The weighting coefficient for health level. To pay attention to the weighting coefficients of the levels, This represents the weighting factor for the severity level.

10. A comprehensive testing and evaluation system for a generator without removing the rotor, used to implement the comprehensive testing and evaluation method for a generator without removing the rotor as described in any one of claims 1-9, characterized in that, It includes a multi-source data acquisition module, a multi-dimensional feature vector extraction module, a comprehensive evaluation module, and a detection and evaluation module, among which: The multi-source data acquisition module is used to: synchronously acquire multi-source detection data through the detection platform; the multi-source detection data includes at least visual image data, slot wedge vibration data, iron core electromagnetic induction data, air gap distance data, and rotor electrical pulse response data; The multidimensional feature vector extraction module is used to: preprocess and extract features from multi-source detection data to obtain multidimensional feature vectors including at least the visual anomaly index, slot wedge loosening index, core anomaly index, air gap non-uniformity, and rotor turn-to-turn short-circuit difference. The comprehensive evaluation module is used to: input multi-dimensional feature vectors into a pre-built multi-level information fusion model for comprehensive evaluation; Among them, the multi-level information fusion model includes at least data layer normalization fusion, feature layer principal component analysis dimensionality reduction fusion and decision layer fuzzy comprehensive evaluation fusion, and outputs the overall health status quantitative score and health level of the generator; The detection and evaluation module is used to: generate a detection report containing recommendations on the necessity of rotor removal and maintenance based on the overall health status quantitative score and health level, and complete the comprehensive detection and evaluation of the generator without removing the rotor.

Images