A method for processing generator repair test data

Abstract

The application discloses a kind of generator maintenance test data processing methods, specifically relates to the technical field of aero-generator maintenance test, for solving the problem of fault diagnosis misalignment and maintenance evaluation compliance deficiency caused by data sparsity in prior art;By analyzing the airworthiness test procedure, the working condition parameter boundary condition is extracted;Based on electromagnetic-thermal-force coupling relationship, a physical law constraint matrix is constructed;According to the eigenvalue attenuation index, the constraint matrix is dynamically reconstructed;The measured data and the reconstructed matrix are jointly optimized to generate physical consistency reinforcement data;Disturbance is injected to the boundary of reinforcement data and the high-risk migration area is marked, and finally the airworthiness compliant maintenance evaluation decision is generated based on the marked data, to realize the physical law consistency reinforcement of maintenance test data and safety risk quantization control.

Description

Technical Field

[0001] This invention relates to the field of aircraft generator maintenance and testing technology, and more specifically, to a generator maintenance and testing data processing method. Background Technology

[0002] During aircraft generator maintenance and testing, airworthiness safety regulations impose strict constraints on the scope and frequency of test conditions to prevent irreversible damage to the equipment. This requirement results in extremely sparse and limited-distribution test data, covering only a few safe operating conditions. Meanwhile, generator failure modes are complex and diverse, with critical faults (such as rotor structural damage) accounting for a very small percentage of the measured data. Existing technologies typically rely on general data augmentation algorithms or transfer learning techniques to process such sparse data, but they do not consider the specific constraints of maintenance and testing scenarios.

[0003] Existing data processing methods suffer from insufficient value density in aircraft generator maintenance testing: due to the lack of integration of physical constraints of test procedures and industry knowledge, general algorithms struggle to extract effective features that conform to the actual physical laws of equipment under sparse data conditions, leading to fault diagnosis models producing erroneous conclusions that contradict engineering realities. This deficiency further reduces the reliability of maintenance assessment results and directly affects the compliance of flight safety certification. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a generator maintenance test data processing method to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for processing generator maintenance test data includes the following steps:

[0007] S1. Obtain the actual maintenance test dataset of the aircraft generator and the corresponding airworthiness test procedure text;

[0008] S2. Parse the airworthiness test procedure text and extract the boundary conditions of the operating parameters.

[0009] S3. Construct a physical law constraint matrix based on the electromagnetic-thermal-mechanical coupling physical relationship of the generator, with the boundary conditions of the operating parameters as constraints.

[0010] S4. Perform eigenvalue stability analysis on the physical constraint matrix to generate eigenvalue decay index;

[0011] S5. When the eigenvalue decay index exceeds the critical threshold, reconstruct the physical law constraint matrix;

[0012] S6. Perform joint optimization processing on the measured maintenance test dataset and the reconstructed physical law constraint matrix to generate enhanced data with full-condition physical consistency.

[0013] S7. Inject disturbances into the operating condition boundary of the enhanced data. If the inconsistency rate of generator fault mode identification results exceeds the tolerance, mark the high-risk migration area and perform maintenance assessment decisions based on the marked enhanced data.

[0014] Furthermore, obtain the actual measured maintenance test dataset of the aircraft generator and the corresponding airworthiness test procedure text, including:

[0015] Obtain the actual maintenance test dataset recorded during the maintenance and testing of aircraft generators;

[0016] Obtain the text of the airworthiness testing procedures issued by the airworthiness management authority;

[0017] Establish a mapping relationship between the operating conditions in the measured maintenance test dataset and the maintenance test operation procedures in the airworthiness test procedure text.

[0018] Furthermore, the measured maintenance test dataset includes multiphysics monitoring data that is tied to operating conditions.

[0019] Furthermore, the airworthiness test procedure text defines maintenance test operation procedures in a chapter-based structure.

[0020] Furthermore, the airworthiness test procedure text is analyzed to extract the boundary conditions of the operating parameters, including:

[0021] Based on the operational condition positioning of the airworthiness test procedure text in the actual measured maintenance test dataset, the associated test procedure chapters are determined.

[0022] Analyze the parameter definition table in the corresponding test program chapter and identify the boundary parameter values ​​of the maximum and minimum allowable speed, the maximum and minimum allowable temperature in the table.

[0023] The boundary parameter values ​​of the identified maximum and minimum allowable speed, maximum and minimum allowable temperature are linked to the corresponding operating conditions to establish physical constraint binding relationships, thereby generating the boundary conditions of the operating condition parameters.

[0024] Furthermore, based on the electromagnetic-thermal-mechanical coupling physical relationship of the generator, a physical law constraint matrix is ​​constructed with the boundary conditions of operating parameters as constraints, including:

[0025] Obtain the expressions for electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient as defined in the generator design specifications.

[0026] Based on the maximum and minimum allowable speed, maximum and minimum allowable temperature in the boundary conditions of the working parameters, calculate the allowable range of electromagnetic torque coefficient, thermal strain transmission coefficient and mechanical stiffness attenuation coefficient.

[0027] The allowable ranges of electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient are constructed into a two-dimensional constraint array according to the physical quantity coupling order, thereby generating a physical law constraint matrix.

[0028] Furthermore, eigenvalue stability analysis is performed on the physical constraint matrix to generate eigenvalue decay indices, including:

[0029] Obtain the numerical calculation results of all eigenvalues ​​in the physical law constraint matrix;

[0030] Identify the maximum eigenvalue of the current physical law constraint matrix, and simultaneously retrieve the maximum eigenvalue of the physical law constraint matrix constructed in the historical maintenance cycle;

[0031] Calculate the relative decay of the current maximum eigenvalue and the historical maximum eigenvalue, and use the relative decay as the eigenvalue decay index.

[0032] Furthermore, when the eigenvalue decay index exceeds a critical threshold, the physical law constraint matrix is ​​reconstructed, including:

[0033] Based on the eigenvector components corresponding to the eigenvalue attenuation index, identify the allowable range of electromagnetic torque coefficient, thermal strain transmission coefficient, and mechanical stiffness attenuation coefficient that need to be updated in the physical law constraint matrix.

[0034] Retrieve the technical bulletin document from this maintenance record to obtain the generator design parameter correction value corresponding to the allowable range of the coefficient to be updated;

[0035] The target coefficients in the allowable ranges of electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient are replaced with design parameter correction values ​​to generate the reconstructed physical law constraint matrix.

[0036] Furthermore, the measured maintenance test dataset and the reconstructed physical constraint matrix are jointly optimized to generate enhanced data with full-condition physical consistency, including:

[0037] The allowable ranges of electromagnetic torque coefficient, thermal strain transmission coefficient, and mechanical stiffness attenuation coefficient in the reconstructed physical law constraint matrix are used as constraints.

[0038] The initial sample set is the multiphysics monitoring data from the measured maintenance test dataset;

[0039] New test conditions are generated within the boundaries of the allowable ranges of electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient using a constrained optimization algorithm.

[0040] The initial sample set is expanded based on the new test conditions, and the enhanced data that satisfies physical consistency across all test conditions is output.

[0041] Furthermore, perturbations are injected into the operating condition boundaries of the enhanced data. If the inconsistency rate of generator fault mode identification results exceeds the tolerance, high-risk migration areas are marked. Maintenance assessment decisions are then performed based on the marked enhanced data, including:

[0042] A perturbation test set is generated by superimposing white noise perturbations on the minimum allowable speed boundary, maximum allowable speed boundary, minimum allowable temperature boundary, and maximum allowable temperature boundary of the enhanced data with physical consistency across all operating conditions.

[0043] The pre-trained generator fault diagnosis model is invoked to perform fault mode identification on the disturbance test set, and the inconsistency rate between the fault mode identification results of the original enhanced data and the disturbance test set is calculated.

[0044] When the inconsistency rate exceeds the tolerance threshold, the corresponding operating condition boundary is marked as a high-risk migration area in the enhanced data of physical consistency across all operating conditions.

[0045] Airworthiness compliance maintenance assessment decision reports are generated based on enhanced data marked with high-risk migration areas.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. By deeply integrating the physical boundary constraints of the airworthiness test procedures with the multi-field coupling mechanism of the generator, the dual challenges of sparsity in maintenance test data and distortion of physical laws are solved. By constructing a dynamically updated physical law constraint matrix and implementing boundary disturbance sensitivity analysis, it is ensured that the enhanced data strictly follows the actual working mechanism of the equipment, enabling the fault diagnosis model to extract effective features that conform to engineering reality even under sparse sample conditions. The generated maintenance assessment decisions are directly related to airworthiness clauses and physical risk boundaries, significantly improving the verifiability and compliance of the certification conclusions.

[0048] 2. The test procedure boundary is transformed into a computable dynamic constraint matrix, ensuring that the generated data is always within the airworthiness and safety range; an adaptive reconstruction mechanism based on eigenvalue decay index is used to capture the equipment performance degradation trend in real time; high-risk migration areas are identified through boundary disturbance injection, and key control points for maintenance decisions are accurately located; through synergy, the engineering credibility of fault mode identification and the engineering applicability of maintenance assessment are significantly improved while ensuring mandatory flight safety constraints. Attached Figure Description

[0049] Figure 1 This is a flowchart of a generator maintenance test data processing method according to the present invention. Detailed Implementation

[0050] 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. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0051] Example: Figure 1 This invention provides a generator maintenance test data processing method, which includes the following steps:

[0052] S1. Obtain the actual maintenance test dataset of the aircraft generator and the corresponding airworthiness test procedure text;

[0053] S2. Parse the airworthiness test procedure text and extract the boundary conditions of the operating parameters.

[0054] S3. Construct a physical law constraint matrix based on the electromagnetic-thermal-mechanical coupling physical relationship of the generator, with the boundary conditions of the operating parameters as constraints.

[0055] S4. Perform eigenvalue stability analysis on the physical constraint matrix to generate eigenvalue decay index;

[0056] S5. When the eigenvalue decay index exceeds the critical threshold, reconstruct the physical law constraint matrix;

[0057] S6. Perform joint optimization processing on the measured maintenance test dataset and the reconstructed physical law constraint matrix to generate enhanced data with full-condition physical consistency.

[0058] S7. Inject disturbances into the operating condition boundary of the enhanced data. If the inconsistency rate of generator fault mode identification results exceeds the tolerance, mark the high-risk migration area and perform maintenance assessment decisions based on the marked enhanced data.

[0059] S1. Obtain the actual measured maintenance test dataset of the aircraft generator and the corresponding airworthiness test procedure text. The specific implementation is as follows:

[0060] Operators collect time-domain vibration waveform data using vibration sensors installed in the generator rotor bearing housing. These sensors are piezoelectric accelerometers with a range covering 0 to 10000 Hz to meet engineering requirements for vibration monitoring of aero-engines, and the sampling frequency is set to 10240 Hz. Simultaneously, thermocouples embedded in the stator winding slots collect winding temperature rise sequence data. These thermocouples have a measurement accuracy of ±1.5 degrees Celsius, and the temperature data acquisition interval is 1 second. An insulation detector collects dielectric loss tangent data every 30 seconds at 2500V DC voltage. The above multi-physics monitoring data is transmitted to a data logger via the controller's local area network bus and linked to generator operating parameters, including engine speed, output power, and cooling air pressure. The engine speed is measured in real-time by a photoelectric encoder installed on the spindle. Data binding is achieved by generating a unified timestamp when storing data. For example, vibration data and temperature data collected at a certain time point are associated with the speed value of 15320 revolutions per minute and the output power value of 125 kilowatts recorded at this time, forming a multi-physics field monitoring data set bound to the operating conditions.

[0061] Technicians download the currently effective airworthiness test procedure document from the official website of the aviation authority. This document is stored in portable document format and contains a searchable text layer. The chapter structure of the airworthiness test procedure text follows the format requirements stipulated by aviation regulations. The chapter defining the generator no-load test operation procedure includes detailed operation steps for the test preparation phase, the speed step increase phase, and the steady-state holding phase. The chapter defining the generator load test operation procedure clarifies the operation specifications such as the load cabinet connection sequence and power factor adjustment range. The chapters of the maintenance test operation procedure are organized according to the test type and the type of physical quantity. For example, the chapter specifying the insulation performance test procedure includes the measurement steps of the dielectric loss tangent and the pass / fail criteria.

[0062] When establishing the mapping relationship between operating conditions and maintenance test procedures, the timestamp sequence in the measured maintenance test dataset is first analyzed to extract the operating condition parameter combination corresponding to each data record point. For example, the dataset is identified to contain operating condition combinations with an engine speed of 15,000 rpm and an output power of 0 kW, as well as operating condition combinations with an engine speed of 18,000 rpm and an output power of 150 kW. Simultaneously, the directory structure of the airworthiness test procedure text is analyzed to locate the test procedure chapters corresponding to each operating condition: when an operating condition with an engine speed of 15,000 rpm and an output power of 0 kW is identified, it is associated with the chapter defining the no-load test procedure; when an operating condition with an engine speed of 18,000 rpm and an output power of 150 kW is identified, it is associated with the chapter defining the full-load test procedure. The mapping relationship is established through an operating condition parameter matching table, which contains three columns of data: operating condition parameter combinations from the measured data, airworthiness text chapter numbers, and mapping confidence values. The confidence level is calculated based on the degree of agreement between the operating parameters and the range described in the text. For example, if the measured speed of 15,200 rpm falls within the no-load test range of 15,000 to 15,500 rpm specified in the text, the confidence level is set to 100%. If the measured value exceeds the range specified in the text, the confidence level is reduced proportionally to the deviation. The deviation proportion is calculated by dividing the absolute difference between the measured value and the nominal value in the text by the allowable fluctuation range in the text and multiplying by 100%. Finally, a mapping database containing timestamps, operating parameters, and text chapter numbers is generated.

[0063] In the above implementation process: the upper limit of the vibration sensor range is set based on the rotor dynamics characteristics to ensure coverage of the third harmonic component of the rotor's critical speed frequency; the thermocouples are embedded in the stator winding slots at equal intervals according to the total number of stator slots, for example, one thermocouple is installed every 8 slots for a generator with 48 slots; the unified timestamp for data binding adopts network time protocol synchronization technology to ensure that the time deviation of multi-source data is less than 1 millisecond; when the operating parameters match multiple chapters at the same time, the chapter with the highest priority is selected according to the priority order of the test process; the allowable fluctuation range of the text is given in the header notes of the specification text, for example, a certain no-load test chapter specifies that the allowable fluctuation range of the speed is ±200 rpm of the nominal value.

[0064] S2. Parse the airworthiness test procedure text and extract the boundary conditions of the operating parameters. The specific implementation is as follows:

[0065] Technicians locate the relevant test procedure chapters in the airworthiness test procedure text based on combinations of operating condition parameters from the measured maintenance test dataset. Specifically, they query a pre-established mapping database, matching the corresponding text chapter position based on the current operating condition's engine speed, output power, and cooling air pressure values. For example, when the operating condition is an engine speed of 15,000 rpm and an output power of 0 kW, the system automatically searches the mapping database to retrieve the starting page number and chapter number of the no-load test procedure chapter associated with this operating condition combination. A multi-level matching strategy is employed during the location process: first, a precise match is made between the engine speed and output power values; if a completely identical record exists, the chapter position is directly returned; if no complete match is found, fuzzy matching is initiated, searching for the chapter corresponding to the closest operating condition combination within a range of allowable engine speed deviation ±200 rpm and allowable power deviation ±10 kW.

[0066] When parsing the parameter definition table in the associated test program section, optical character recognition (OCR) technology is used to extract the text content within the table. First, the physical structure of the table is identified, determining that the header row contains columns for parameter name, maximum allowable value, minimum allowable value, and unit. The table content is scanned row by row. When a cell containing the keyword "speed" is identified in the parameter name column, the numeric strings in the maximum and minimum allowable value columns of the same row are extracted and converted to floating-point numbers. For example, in the table of the no-load test section, if a row with the parameter name "rotor speed" is identified, its maximum allowable value column displays 18000, its minimum allowable value column displays 12000, and the unit column is "revolutions per minute," then the maximum allowable speed is recorded as 18000 revolutions per minute, and the minimum allowable speed is recorded as 12000 revolutions per minute. Temperature parameter parsing follows the same principle. When a parameter row containing the keyword "temperature" is identified, the corresponding maximum and minimum allowable values ​​are extracted. For example, in the table of the load test section, the maximum allowable value for "stator winding temperature" is identified as 155 degrees Celsius, and the minimum allowable value is -40 degrees Celsius.

[0067] When establishing physical constraint binding relationships between identified boundary parameter values ​​and operating conditions, a boundary condition data object for operating conditions is created. This object contains three core fields: the operating condition type field stores the description text of the operating condition, such as no-load test condition or full-load test condition; the boundary parameter field stores a set of key-value pairs, with keys named maximum allowable speed, minimum allowable speed, maximum allowable temperature, and minimum allowable temperature, and values ​​containing the corresponding numerical values ​​and units; the constraint relationship field records the application conditions of the boundary parameters. The binding relationship is established as follows: when the operating condition is in a no-load state, the minimum allowable speed of 12,000 rpm and the maximum allowable speed of 18,000 rpm extracted from the no-load test section are bound to this operating condition; when the operating condition is in a full-load state, the minimum allowable temperature of -40 degrees Celsius and the maximum allowable temperature of 155 degrees Celsius extracted from the load test section are bound to this operating condition. After each boundary parameter is bound, a unique constraint identifier is generated. This identifier is composed of the operating condition type code and the parameter type code. For example, the identifier for the maximum speed limit of the no-load condition is NL-RPM-MAX.

[0068] In the above implementation process: the deviation threshold setting of the multi-level matching strategy is based on the tolerance range of the test parameters indicated in the text. For example, if a footnote in a certain chapter indicates that the accuracy of the speed measurement equipment is ±150 rpm, then the fuzzy matching threshold is set to 150 rpm; the preprocessing of optical character recognition includes binarization of the document image and table line detection, and the use of connected component analysis algorithm to locate cell positions; the parameter name keyword matching adopts the string inclusion detection method, and a keyword library is established to include standard terms such as rotor speed, spindle speed, and winding temperature; the constraint identifier generation rule is to take the combination of the English abbreviation of the working condition type and the English abbreviation of the parameter type, with the abbreviation of no-load working condition being NL and the abbreviation of speed parameter being RPM.

[0069] S3. Construct a physical constraint matrix based on the electromagnetic-thermal-mechanical coupling physical relationship of the generator, with the boundary conditions of the operating parameters as constraints. The specific implementation is as follows:

[0070] Operators extracted the electromagnetic torque coefficient expression from the editable text of the generator design specification document. This expression describes the functional relationship between the generator's output torque and armature current and magnetic flux density. Specifically, they searched the document for content containing the definition of "electromagnetic torque coefficient," locating the text paragraph: "Electromagnetic torque coefficient equals the number of pole pairs multiplied by the magnetic flux density multiplied by the effective length of the armature conductor." The thermal strain transfer coefficient expression is defined in the document as: "The thermal strain transfer coefficient equals the linear expansion coefficient of the material multiplied by the rate of change of the elastic modulus," where the linear expansion coefficient corresponds to the physical property value of the silicon steel sheets in the stator core. The mechanical stiffness attenuation coefficient expression is described in the document's table comments as: "The mechanical stiffness attenuation coefficient equals 1 minus the temperature compensation factor multiplied by the difference between the operating temperature and the reference temperature, then divided by the reference temperature," which is explicitly specified as 25 degrees Celsius in the document. Expression extraction was achieved through text pattern matching, identifying statement structures where the left side of the equals sign contains coefficient names and the right side contains mathematical operators.

[0071] Based on the maximum and minimum allowable speeds in the boundary conditions of the operating parameters, the allowable range of the electromagnetic torque coefficient is calculated. Specifically, the maximum allowable speed is substituted into the speed variable in the electromagnetic torque coefficient expression to calculate the upper limit of the electromagnetic torque coefficient; the minimum allowable speed is substituted into the same expression to calculate the lower limit. For example, when the maximum allowable speed is 18,000 revolutions per minute, the unit needs to be converted to radians per second: 18,000 revolutions per minute multiplied by 2, multiplied by pi, and divided by 60 equals 1884.96 radians per second, which is then substituted into the expression to calculate the coefficient value. When calculating the allowable range of the thermal strain transfer coefficient, the maximum allowable temperature is substituted into the operating temperature variable in the expression to calculate the upper limit, and the minimum allowable temperature is substituted into the expression to calculate the lower limit. For example, when the maximum allowable temperature is 155 degrees Celsius, the rate of change of the elastic modulus at that temperature is obtained by consulting the silicon steel sheet material property table, and then multiplied by the material's linear expansion coefficient to obtain the thermal strain transfer coefficient value. The calculation direction for the allowable range of mechanical stiffness attenuation coefficient is reversed: the maximum allowable temperature value is substituted into the expression to calculate the lower limit, and the minimum allowable temperature value is substituted into the expression to calculate the upper limit, because stiffness performance decreases as temperature increases.

[0072] The calculated allowable ranges for electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient are used to construct a two-dimensional constraint array according to the coupling order of physical quantities. The coupling order follows the generator energy conversion path: electromagnetic parameters first, thermal parameters second, and mechanical parameters last. The construction rule is as follows: create a three-row, two-column data structure. The first row corresponds to electromagnetic parameters, with the first column storing the lower limit of the allowable range and the second column storing the upper limit. The second row corresponds to thermal parameters, with the same column structure storing the range of thermal parameters. The third row corresponds to mechanical parameters, with the same column structure storing the range of mechanical parameters. For example, the data structure generated by a certain calculation is as follows:

[0073] First line: Minimum allowable value of electromagnetic torque coefficient, maximum allowable value of electromagnetic torque coefficient;

[0074] Second line: Minimum allowable value of thermal strain transfer coefficient, maximum allowable value of thermal strain transfer coefficient;

[0075] The third line: Minimum allowable value of mechanical stiffness attenuation coefficient, maximum allowable value of mechanical stiffness attenuation coefficient.

[0076] This data structure is stored in a parsable text format and contains a working condition identifier field and a calculation timestamp field, forming a physical law constraint matrix.

[0077] In the above implementation process: when extracting expressions, if there are multiple definition versions in the document, the expression of the latest revised chapter is preferred; in the conversion of rotational speed units, the value of pi is taken as 3.1415926535, and the conversion formula is: radians per second = revolutions per minute × 2 × pi ÷ 60; the material property table is an appendix to the design specification, distributed at equal intervals according to temperature values, with a temperature interval of 10 degrees Celsius, and linear interpolation is used for unlisted temperature points; the temperature compensation factor is determined according to the material type, for example, 0.0035 per degree Celsius for aluminum alloy materials, and this value is listed in the material properties chapter of the document; when storing two-dimensional data structures, the electromagnetic torque coefficient is allowed to retain three decimal places, and other coefficients are retained to four decimal places.

[0078] S4. Perform eigenvalue stability analysis on the physical constraint matrix to generate eigenvalue decay indices. The specific implementation is as follows:

[0079] When operators obtain the eigenvalue calculation results of the physical law constraint matrix, they first convert the two-dimensional constraint array into a square matrix data structure. Specifically, the original three-row, two-column matrix is ​​expanded into a three-row, three-column covariance matrix: for each row of data in the matrix, the average of all elements in that row is calculated; for any two rows of data, the average of the products of corresponding elements in those two rows is subtracted from the product of the averages of those two rows, and the covariance value is filled into the corresponding position in the new matrix. For example, if the original matrix contains three sets of data: the allowable range of electromagnetic torque coefficient, the allowable range of thermal strain transfer coefficient, and the allowable range of mechanical stiffness attenuation coefficient, then the first row and first column of the new square matrix stores the variance value of the electromagnetic torque coefficient data, the first row and second column stores the covariance value of the electromagnetic torque coefficient and the thermal strain transfer coefficient, and so on to complete the matrix construction. Eigenvalue decomposition is performed on this square matrix, and an iterative calculation method is used to cyclically adjust the matrix elements. The iteration stops when the absolute value of all off-diagonal elements is less than 0.000001, and finally three real eigenvalues ​​are output, sorted in descending order to form an eigenvalue list.

[0080] When identifying the largest eigenvalue of the current physical constraint matrix, the first value in the eigenvalue list is selected. Simultaneously, the largest eigenvalue of the physical constraint matrix constructed from historical maintenance cycles is retrieved. Specifically, this is done by accessing the maintenance record database and searching the three most recent maintenance records in ascending order of maintenance time. For example, if the current maintenance record number is M2023-087, the system automatically searches for maintenance records numbered M2022-053, M2021-041, and M2020-032, extracting the stored largest eigenvalue from each record. When multiple historical records are available, the record closest to the current generator's cumulative operating time is selected first, with the cumulative operating time difference controlled within 200 hours.

[0081] When calculating the eigenvalue attenuation index, a historical benchmark value is first determined: if a valid historical record exists, the arithmetic mean of the three most recent maximum eigenvalues ​​is used as the benchmark; for the first maintenance, the theoretical eigenvalue calculated during the generator design phase is used as the benchmark. The eigenvalue attenuation index is calculated through the following steps: subtract the historical benchmark value from the current maximum eigenvalue to obtain the difference, then divide the difference by the historical benchmark value, and finally multiply by 100% to obtain the relative attenuation. For example, if the current maximum eigenvalue is 8.75 and the historical benchmark value is 9.20, then the relative attenuation is equal to 8.75 - 9.20 = -0.45, -0.45 / 9.20 = -0.0489, and multiplying by 100% gives approximately -4.89%. This relative attenuation is output as the eigenvalue attenuation index to the analysis report and simultaneously stored in the current maintenance record database.

[0082] In the above implementation process: the specific process of covariance calculation is as follows: for two sets of data X and Y, first calculate the average value of X and the average value of Y respectively, then calculate the difference between each X element and the average value of X multiplied by the difference between the corresponding Y element and the average value of Y, and finally calculate the average of these products; during the iterative calculation process, the maximum number of iterations is set to 100 times, and the calculation is forcibly terminated when the convergence condition is not met after 100 iterations; when matching historical data, if the time difference between the three most recent maintenance records exceeds 200 hours, the search range is expanded to the previous five maintenance records; when sorting feature values, only the real part is considered, and the influence of the imaginary part is ignored; the benchmark value for the first maintenance is taken from the theoretical calculation result of the feature value in the generator factory test report; the calculation result of the relative attenuation is retained to two decimal places and processed using the rounding rules.

[0083] S5. When the eigenvalue decay index exceeds the critical threshold, the physical law constraint matrix is ​​reconstructed, specifically as follows:

[0084] When the eigenvalue decay index exceeds the critical threshold range specified in the design specifications, operators identify the allowable range of coefficients that need to be updated based on the eigenvalue decomposition results. Specifically, the eigenvector component data corresponding to the eigenvalue decay index is extracted, and this eigenvector is stored synchronously with the eigenvalue calculation results of the physical law constraint matrix. The absolute values ​​of the three components in the eigenvector are analyzed, and the physical quantity type corresponding to the component with the largest absolute value is marked as the allowable range of coefficients that needs to be updated first. For example, when the eigenvector components are [0.85, 0.32, 0.41], the first component, 0.85, has the largest absolute value, and this first component corresponds to the allowable range of the electromagnetic torque coefficient; therefore, the allowable range of the electromagnetic torque coefficient is determined to be the object that needs updating. If the absolute value difference of multiple components is less than 0.1, the allowable range of coefficients corresponding to these components is marked simultaneously.

[0085] When retrieving technical bulletin documents from this maintenance record, the maintenance management system searches for bulletin documents issued within the current maintenance cycle. Technical bulletin documents are stored in Extensible Markup Language (EXPLAIN) format and include a section on design parameter corrections. This section lists the corrected parameter names, correction values, and effective dates in a structured table format. Based on the marked type of coefficient to be updated, the corresponding design parameter correction item is matched in the table: when the allowable range of the electromagnetic torque coefficient needs to be updated, records with "electromagnetic torque" in the parameter name are searched; when the allowable range of the thermal strain transfer coefficient needs to be updated, records with "thermal strain" in the parameter name are searched. For example, in the correction table for bulletin document number TA2023-15, the "electromagnetic torque coefficient compensation factor" correction item is located, and its correction value field displays 1.05.

[0086] When performing the coefficient allowable range replacement operation, the current allowable range values ​​of the target coefficients in the original physical law constraint matrix are first obtained. For the allowable range of the electromagnetic torque coefficient, the lower limit and upper limit values ​​are extracted from the first row of the original matrix; the allowable range of the thermal strain transfer coefficient is extracted from the second row; and the allowable range of the mechanical stiffness attenuation coefficient is extracted from the third row. The correction value is multiplied by the original range boundary value to generate a new allowable range: the new lower limit value is equal to the original lower limit value multiplied by the correction value, and the new upper limit value is equal to the original upper limit value multiplied by the correction value. For example, if the original allowable range of the electromagnetic torque coefficient is [120.5, 185.7] and the correction value is 1.05, then the new lower limit value is 120.5 × 1.05 = 126.525, and the new upper limit value is 185.7 × 1.05 = 194.985. The calculated new lower limit value and new upper limit value are updated in their original positions in the physical law constraint matrix to generate the reconstructed physical law constraint matrix.

[0087] In the above implementation process: Eigenvector component analysis rules: The component number strictly corresponds to the order of physical quantities, and the first component always represents the electromagnetic parameter; The scope of technical bulletin document retrieval is limited to the current maintenance cycle, that is, the bulletins issued between the start date of this maintenance and the current date; Correction value application method: When the correction value is greater than 1, the allowable range is expanded, and when it is less than 1, the allowable range is narrowed; Multi-coefficient update scenario: When multiple coefficients need to be updated simultaneously, they are processed in descending order of the absolute value of the eigenvector components; After the replacement operation, the original matrix backup is retained, and the reconstruction time and the bulletin number on which it is based are recorded in the new matrix metadata; Critical threshold setting basis: The design specification defines that a eigenvalue attenuation rate exceeding 5% requires triggering reconstruction, and this threshold is determined based on the generator design life curve.

[0088] S6. Jointly optimize the measured maintenance test dataset with the reconstructed physical law constraint matrix to generate enhanced data with full-condition physical consistency. The specific implementation is as follows:

[0089] When operators use the allowable ranges of electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient in the reconstructed physical constraint matrix as constraints, they specifically read the three-row, two-column data structure of the physical constraint matrix. The first row stores the lower and upper limits of the electromagnetic torque coefficient, constituting the electromagnetic constraint boundary, for example, a lower limit of 126.525 and an upper limit of 194.985. The second row stores the lower and upper limits of the thermal strain transfer coefficient, constituting the thermal constraint boundary. The third row stores the lower and upper limits of the mechanical stiffness attenuation coefficient, constituting the mechanical constraint boundary. The constraints are expressed as a system of inequalities: the measured value of the electromagnetic torque coefficient must be greater than or equal to the value in the first row and first column of the matrix and less than or equal to the value in the first row and second column; the measured value of the thermal strain transfer coefficient must be greater than or equal to the value in the second row and first column and less than or equal to the value in the second row and second column; the measured value of the mechanical stiffness attenuation coefficient must be greater than or equal to the value in the third row and first column and less than or equal to the value in the third row and second column.

[0090] When using multiphysics monitoring data from the measured maintenance test dataset as the initial sample set, the complete dataset collected during the current maintenance cycle is extracted from the data storage system. This dataset contains multiphysics monitoring data such as vibration waveform data, winding temperature sequences, and dielectric loss tangent values ​​of the generator under different operating conditions. Each data record is associated with operating parameters such as engine speed, output power, and cooling air pressure. The initial sample set is arranged in chronological order, for example, containing 150 data records. Each record contains a 30-second vibration waveform, 60 temperature sampling points, and 2 dielectric loss measurements.

[0091] When generating new test condition points using a constrained optimization algorithm, a sequential quadratic programming algorithm is employed to perform iterative calculations within a three-dimensional constrained space. The electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient are used as optimization variables, with the objective function being to maximize the coverage of test condition points. Initially, three sample points are randomly selected from the initial sample set as the starting point for iteration. In each iteration, the gradient of the objective function is calculated, and new points are searched along the gradient direction. A new point is accepted if it meets the following conditions: the electromagnetic torque coefficient value is between 126.525 and 194.985, the thermal strain transfer coefficient value is within its constraint boundary, and the mechanical stiffness attenuation coefficient value is within its constraint boundary. The iteration terminates when the objective function improvement is less than 0.0001 after five consecutive iterations. For example, after 20 iterations, a new test condition point is generated with the following values: electromagnetic torque coefficient 165.32, thermal strain transfer coefficient 0.85, and mechanical stiffness attenuation coefficient 0.92.

[0092] When expanding the initial sample set based on new test condition points, the following steps are performed: Convert the generated new test condition points to a standard data format and supplement virtual monitoring data fields; calculate the Euclidean distance between the new test condition points and each point in the initial sample set, retaining the point when the minimum distance is greater than a set threshold; add new points that meet the conditions to the end of the initial sample set in batches. For example, if the distance threshold is set to 5.0, a new point is retained when its distance to the nearest sample point is 7.2, and a record containing a timestamp, test condition parameters, and multiphysics monitoring data placeholders is created for that point. The final output is an enhanced dataset containing the original 150 sets of records and 30 newly added virtual records, which meets the full-condition physical consistency requirements.

[0093] In the above implementation process: In the constraint inequality set, the mechanical stiffness attenuation coefficient must additionally meet the engineering constraint that the lower limit of the attenuation coefficient is greater than 0.65; the initial sample set preprocessing includes removing abnormal records that exceed the sensor range, such as records with vibration acceleration exceeding 100g; the step size control parameter of the sequential quadratic programming algorithm is initially set to 0.1, and is reduced to half of the original value every ten iterations; the Euclidean distance calculation uses standardized coefficient values: the electromagnetic torque coefficient is divided by 200, the thermal strain transfer coefficient is divided by 1.0, and the mechanical stiffness attenuation coefficient is divided by 1.0 to eliminate the influence of dimensions; the virtual monitoring data placeholder filling rule is: the vibration data is copied from the most recent measured waveform under the same working condition, and the temperature data is taken as the average of three adjacent points; full working condition physical consistency verification: the characteristic value attenuation index is jointly calculated by the newly added points and the original dataset to ensure that it does not exceed the critical threshold.

[0094] S7. Inject disturbances into the operating condition boundaries of the enhanced data. If the inconsistency rate of generator fault mode identification results exceeds the tolerance limit, mark high-risk migration areas. Perform maintenance assessment decisions based on the marked enhanced data. The specific implementation is as follows:

[0095] When operators inject white noise disturbances into the boundary conditions of the enhanced data with full-condition physical consistency, they specifically perform operations on four key boundary parameters: For the minimum allowable speed boundary, a normally distributed random disturbance with a mean of zero and a standard deviation of 1% of the original value is superimposed on the original boundary value; for the maximum allowable speed boundary, the same disturbance method is used, but the standard deviation is set to 0.8% of the original value; for the minimum allowable temperature boundary, a random disturbance uniformly distributed between -5°C and +5°C is superimposed; for the maximum allowable temperature boundary, a random disturbance uniformly distributed between -3°C and +3°C is superimposed. 100 disturbance samples are generated for each boundary parameter. For example, when the minimum allowable speed boundary is 12000 rpm, a standard deviation of 12000 × 0.01 = 120 rpm is generated, producing disturbance values ​​such as 11985 rpm and 12034 rpm. This ultimately forms a disturbance test set containing 400 samples, with each sample retaining the multiphysics monitoring data from the original enhanced data.

[0096] When using a pre-trained generator fault diagnosis model for fault mode identification, the model is a three-layer feedforward neural network trained based on historical maintenance data. The number of nodes in the input layer corresponds to the feature dimensions of the multi-physics monitoring data. The original enhanced data is input into the model to obtain the baseline identification result, and then the disturbance test set is input line by line to obtain the disturbance identification result. The inconsistency rate is calculated using a sample comparison method: the number of times the fault mode label for the same sample point is inconsistent between the original data and the disturbance data is counted, divided by the total sample size, and then multiplied by 100%. For example, if the original enhanced data has 180 samples and the disturbance test set has 400 samples, and after 580 comparisons, 35 identification results are found to be inconsistent, then the inconsistency rate is (35 / 580) × 100%, which is approximately 6.03%.

[0097] When the inconsistency rate exceeds a preset tolerance threshold, high-risk migration areas are marked in the enhanced data. The tolerance threshold is set according to the generator's safety level; for example, the threshold for Class A safety components is set at 5.0%, and for Class B components at 7.5%. The marking rule is as follows: if the inconsistency rate of a speed boundary disturbance sample exceeds the threshold, the speed boundary point is marked as a high-risk area in the enhanced data; if the inconsistency rate of a temperature boundary disturbance sample exceeds the limit, the corresponding temperature boundary point is marked. The marking information is stored in the form of an additional data layer, including a boundary type field, a boundary value field, and a risk level field. For example, if an inconsistency rate of 6.8% (exceeding the 5.0% threshold) is detected at the maximum allowable speed boundary of 18,000 rpm, then the marking field "Boundary type = maximum speed, risk level = high risk" is added to that operating point.

[0098] When generating a maintenance assessment decision report based on labeled and enhanced data, three core operations are performed: First, boundary point data for all high-risk migration areas are extracted and sorted in descending order of risk level; second, maintenance procedure clauses are associated, for example, when a marked point is a boundary of maximum speed, the "overspeed protection device calibration" clause in the procedure is associated; finally, a decision recommendation table is generated, containing three levels of maintenance instructions: mandatory items, recommended items, and observation items. For example, for high-risk speed boundary points, "check speed sensor calibration status" is included in the mandatory items; for medium-risk temperature boundary points, "clean cooling ducts" is included in the recommended items. The report output is a structured document containing three parts: a risk distribution map of operating condition boundaries, a maintenance priority list, and a declaration of compliance with airworthiness provisions.

[0099] In the above implementation process: the white noise disturbance amplitude is set based on the following: the standard deviation of the rotational speed disturbance is 1% due to the accuracy level of the rotational speed sensor; the temperature disturbance range of ±5 degrees Celsius corresponds to the thermocouple measurement error limit; the number of nodes in the input layer of the fault diagnosis model is fixed at 12, corresponding to 8 characteristic frequency bands of the vibration spectrum + 2 temperature features + 2 insulation features; the inconsistency rate calculation adopts a double-blind verification mechanism: the original data and the disturbance data are diagnosed and processed independently; the risk level classification rules are as follows: an inconsistency rate exceeding twice the threshold is considered extremely high risk, between 1 and 2 times is considered high risk, and between 0.5 and 1 times is considered medium risk; the maintenance instruction mapping rules are as follows: mandatory items correspond to mandatory airworthiness clauses, recommended items correspond to manufacturer service bulletins, and observation items are based on engineering experience; the airworthiness compliance declaration must include verification records of the handling results of all marked points.

[0100] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0101] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0102] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in 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 inventive 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 implementation should not be considered beyond the scope of this application.

[0103] In addition, the functional modules in the various embodiments of this application 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.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, 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.

[0105] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0106] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for processing generator maintenance test data, characterized in that, Includes the following steps: S1. Obtain the actual maintenance test dataset of the aircraft generator and the corresponding airworthiness test procedure text; S2. Parse the airworthiness test procedure text and extract the boundary conditions of the operating parameters. S3. Constructing a physical constraint matrix based on the electromagnetic-thermal-mechanical coupling physical relationship of the generator, with operating parameters and boundary conditions as constraints, including: Obtain the expressions for electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient as defined in the generator design specifications. Based on the maximum and minimum allowable speed, maximum and minimum allowable temperature in the boundary conditions of the working parameters, calculate the allowable range of electromagnetic torque coefficient, thermal strain transmission coefficient and mechanical stiffness attenuation coefficient. The allowable ranges of electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient are constructed into a two-dimensional constraint array according to the physical quantity coupling order, and a physical law constraint matrix is ​​generated. S4. Perform eigenvalue stability analysis on the physical constraint matrix to generate eigenvalue decay indices, including: Obtain the numerical calculation results of all eigenvalues ​​in the physical law constraint matrix; Identify the maximum eigenvalue of the current physical law constraint matrix, and simultaneously retrieve the maximum eigenvalue of the physical law constraint matrix constructed in the historical maintenance cycle; Calculate the relative decay between the current maximum eigenvalue and the historical maximum eigenvalue, and use the relative decay as an indicator of eigenvalue decay. S5. When the eigenvalue decay index exceeds the critical threshold, reconstruct the physical law constraint matrix, including: Based on the eigenvector components corresponding to the eigenvalue attenuation index, identify the allowable range of electromagnetic torque coefficient, thermal strain transmission coefficient, and mechanical stiffness attenuation coefficient that need to be updated in the physical law constraint matrix. Retrieve the technical bulletin document from this maintenance record to obtain the generator design parameter correction value corresponding to the allowable range of the coefficient to be updated; Replace the target coefficients in the allowable range of electromagnetic torque coefficient, allowable range of thermal strain transfer coefficient, and allowable range of mechanical stiffness attenuation coefficient with the design parameter correction values ​​to generate the reconstructed physical law constraint matrix. S6. Perform joint optimization processing on the measured maintenance test dataset and the reconstructed physical law constraint matrix to generate enhanced data with full-condition physical consistency. S7. Inject disturbances into the operating condition boundary of the enhanced data. If the inconsistency rate of generator fault mode identification results exceeds the tolerance, mark the high-risk migration area and perform maintenance assessment decisions based on the marked enhanced data.

2. The generator maintenance test data processing method according to claim 1, characterized in that, Obtain the actual measured maintenance test dataset of the aircraft generator and the corresponding airworthiness test procedure text, including: Obtain the actual maintenance test dataset recorded during the maintenance and testing of aircraft generators; Obtain the text of the airworthiness testing procedures issued by the airworthiness management authority; Establish a mapping relationship between the operating conditions in the measured maintenance test dataset and the maintenance test operation procedures in the airworthiness test procedure text.

3. The generator maintenance test data processing method according to claim 2, characterized in that, The actual maintenance test dataset contains multiphysics monitoring data that is tied to operating conditions.

4. The generator maintenance test data processing method according to claim 2, characterized in that, The airworthiness test procedure text defines the maintenance test operation process in a chapter-based structure.

5. The generator maintenance test data processing method according to claim 2, characterized in that, Parse the airworthiness test procedure text and extract the boundary conditions of the operating parameters, including: Based on the operational condition positioning of the airworthiness test procedure text in the actual measured maintenance test dataset, the associated test procedure chapters are determined. Analyze the parameter definition table in the corresponding test program chapter and identify the boundary parameter values ​​of the maximum and minimum allowable speed, the maximum and minimum allowable temperature in the table. The boundary parameter values ​​of the identified maximum and minimum allowable speed, maximum and minimum allowable temperature are linked to the corresponding operating conditions to establish physical constraint binding relationships, thereby generating the boundary conditions of the operating condition parameters.

6. The generator maintenance test data processing method according to claim 5, characterized in that, The measured maintenance test dataset and the reconstructed physical constraint matrix are jointly optimized to generate enhanced data with full-condition physical consistency, including: The allowable ranges of electromagnetic torque coefficient, thermal strain transmission coefficient, and mechanical stiffness attenuation coefficient in the reconstructed physical law constraint matrix are used as constraints. The initial sample set is the multiphysics monitoring data from the measured maintenance test dataset; New test conditions are generated within the boundaries of the allowable ranges of electromagnetic torque coefficient, thermal strain transfer coefficient, and mechanical stiffness attenuation coefficient using a constrained optimization algorithm. The initial sample set is expanded based on the new test conditions, and the enhanced data that satisfies physical consistency across all test conditions is output.

7. The generator maintenance test data processing method according to claim 6, characterized in that, Perturbations are injected into the operating condition boundaries of the enhanced data. If the inconsistency rate of generator fault mode identification results exceeds the tolerance limit, high-risk migration areas are marked. Maintenance assessment decisions are then performed based on the marked enhanced data, including: A perturbation test set is generated by superimposing white noise perturbations on the minimum allowable speed boundary, maximum allowable speed boundary, minimum allowable temperature boundary, and maximum allowable temperature boundary of the enhanced data with physical consistency across all operating conditions. The pre-trained generator fault diagnosis model is invoked to perform fault mode identification on the disturbance test set, and the inconsistency rate between the fault mode identification results of the original enhanced data and the disturbance test set is calculated. When the inconsistency rate exceeds the tolerance threshold, the corresponding operating condition boundary is marked as a high-risk migration area in the enhanced data of physical consistency across all operating conditions. Airworthiness compliance maintenance assessment decision reports are generated based on enhanced data marked with high-risk migration areas.

Images