Gas turbine multi-disc rotor imbalance fault diagnosis method and related device

Abstract

The invention belongs to the technical field of rotor dynamics, and discloses a gas turbine multi-disc rotor imbalance fault diagnosis method and a related device. The method comprises the following steps: extracting a feature vector based on acquired abnormal vibration data; on the basis of the feature vectors, a rotor unbalance fault diagnosis model is used for identifying and positioning a fault source, and a multi-disc rotor unbalance fault online identification and tracing result is obtained; the rotor imbalance fault diagnosis model is composed of a low-fidelity model network and a high-fidelity model network, and the low-fidelity model network is used for predicting a low-fidelity result under the condition of given input data after training. And the high-fidelity model network is used for acquiring a high-fidelity prediction result by learning a mapping relationship between the low-fidelity output result and the high-fidelity output result. According to the invention, accurate identification and online traceability of the unbalance fault parameters of the multi-disc rotor are realized.

Description

Technical Field

[0001] This invention belongs to the field of rotor dynamics technology, and specifically relates to a method and related device for diagnosing unbalanced faults in multi-disc rotors of gas turbines. Background Technology

[0002] Multi-disc rotor structures are widely used in critical rotating machinery fields such as gas turbines and steam turbines due to their high-efficiency energy transmission and compact design. However, due to manufacturing errors, material inhomogeneity, or operational wear, the discs of a multi-disc rotor can become eccentric, causing shaft imbalance. This not only severely affects the operating accuracy of the equipment and reduces torque transmission efficiency but also accelerates component wear and may even lead to equipment failure or malfunction. Therefore, accurate and efficient diagnosis of multi-disc rotor imbalance faults is crucial.

[0003] Currently, existing rotor imbalance diagnosis solutions mainly employ spectrum analysis, influence coefficient method, and data-driven artificial intelligence methods. Among these, spectrum analysis relies on the frequency domain characteristics of vibration signals, but its ability to identify multiple coupled fault conditions is limited. The influence coefficient method requires extensive on-site dynamic balancing tests, and the calibration process is complex, costly, and unsuitable for multi-disc rotor multi-parameter identification. While artificial intelligence methods (such as neural networks and support vector machines) have certain recognition capabilities, they generally rely on a large number of high-quality labeled samples, which are difficult to obtain in actual engineering projects. Furthermore, the models are often "black boxes" and lack physical interpretability, resulting in low reliability of diagnostic results and insufficient support for maintenance decisions. Summary of the Invention

[0004] The purpose of this invention is to provide a method and related apparatus for diagnosing multi-disc rotor imbalance faults in gas turbines, thereby solving one or more of the aforementioned technical problems. Specifically, the technical solution disclosed in this invention is a multi-fidelity data fusion-based fault diagnosis scheme for multi-disc rotor imbalance in gas turbines. It establishes a rotor fault diagnosis model based on multi-fidelity data fusion, achieving accurate identification and online tracing of multi-disc rotor imbalance fault parameters.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for diagnosing imbalance faults in multi-disc rotors of gas turbines, comprising the following steps: Feature vectors are extracted based on the acquired abnormal vibration data; Based on the feature vector, the rotor imbalance fault diagnosis model is used to identify and locate the fault source, and the online identification and source tracing results of multi-disc rotor imbalance faults are obtained. The rotor imbalance fault diagnosis model consists of a low-fidelity model network and a high-fidelity model network. The low-fidelity model network is used to predict low-fidelity results given input data after training. The high-fidelity model network is used to obtain high-fidelity prediction results by learning the mapping relationship between low-fidelity output results and high-fidelity output results. The mapping relationship between low-fidelity output and high-fidelity output is expressed as follows: ; In the formula, It is a high-fidelity output result; It is a mapping network; X is the input data, which is the displacement, velocity, or acceleration signal of the multi-disc rotor system; It is a low-fidelity output result; It is a network hyperparameter.

[0006] A further improvement to the technical solution of the present invention lies in that, during the training process of the rotor imbalance fault diagnosis model, The low-fidelity model network is trained using low-fidelity data to obtain low-fidelity prediction results, which are expressed as follows: ; In the formula, These are low-fidelity prediction results, including predicted values ​​for unbalanced mass, radius, and disk number. The mapping relationship for low-fidelity model networks; It is the input low-fidelity data, which is the displacement, velocity or acceleration result obtained from the dynamic analysis of multi-disc rotors; For the hyperparameters of the low-fidelity model network; The high-fidelity model network uses the learned low-fidelity prediction results and high-fidelity data to obtain high-fidelity prediction results, expressed as follows: ; In the formula, It provides high-fidelity prediction results, including predicted values ​​for unbalanced mass, radius, and disk number; The mapping relationship for the high-fidelity model network; It is the input high-fidelity data, which is the displacement, velocity or acceleration signal measured in the experiment; These are the hyperparameters of the high-fidelity model network; During overall training, the low-fidelity data and high-fidelity data datasets are aligned and scaled preprocessed. The network is iteratively optimized through gradient backpropagation to reduce the composite empirical loss function. The experimental measurement results are used as high-fidelity data, and the rotor dynamics analysis results are used as low-fidelity data.

[0007] A further improvement to the technical solution of this invention lies in the following specific implementation steps: performing data alignment and scaling preprocessing on the two datasets of low-fidelity data and high-fidelity data, and iteratively optimizing the network through a gradient backpropagation mechanism to reduce the composite empirical loss function. The min-max normalization method is used to normalize all input and output data to the range [-1, 1]. Automatic differentiation is used to calculate the gradient information required for hyperparameter optimization during backpropagation. The Adam optimizer is used to optimize the network's hyperparameters to minimize the following loss function: ; In the formula, For the composite empirical loss function of the rotor imbalance fault diagnosis model; The mean square error of the low-fidelity model network; The mean square error of the high-fidelity model network; This is a weighting factor used to adjust the degree of contribution to high-fidelity data; ; ; In the formula, For low-fidelity data scale; For high-fidelity data scale.

[0008] A further improvement to the technical solution of this invention lies in that the experimental measurement results are obtained based on a rotor imbalance testing device; the rotor imbalance testing device includes: a base platform, a main shaft, a perforated disk, a servo motor, a support bearing, a bearing housing, a coupling, an eccentric mass block, a centrifugal mass block, a triaxial accelerometer, an infrared sensor, and a data acquisition unit; wherein, The base platform provides support and vibration isolation. The main shaft is rotatably mounted on the base platform via the bearing housing and the supporting bearing. A servo motor is fixedly mounted on the base platform, and its output is connected to the main shaft via a coupling. Multiple perforated discs are detachably mounted on the main shaft. The servo motor drives the main shaft and the perforated discs to rotate at a specified speed. Multiple imbalance adjustment holes are radially provided on the perforated discs for mounting the eccentric mass blocks to simulate a rotor system with imbalance faults. Multiple mounting holes are provided on the circumferential surface of the perforated discs for mounting the centrifugal mass blocks to simulate equivalent centrifugal loads on the blades. The triaxial accelerometers are mounted on bearing seats at both ends of the spindle, and are used to independently output displacement, velocity and acceleration signals in the three measurement axes. The infrared sensor is mounted on the bearing seats of each centrifugal mass block on the peripheral edge of the perforated disk, and is used to generate a voltage pulse when the centrifugal mass block passes directly in front of the infrared sensor. The rotational speed and phase angle of the axis are measured by calculating the number of pulses. The data acquisition unit is used to amplify the analog signals input from each sensor and convert them into high-sampling-rate digital signals. Noise is removed by adding a low-pass filter to obtain the final experimental measurement results.

[0009] A further improvement to the technical solution of the present invention is that the steps for obtaining test measurement results based on the rotor imbalance test device specifically include: The fault feature space is designed based on the unbalanced state of multi-disc rotors, and the combination of unbalanced fault test parameters is selected using the partial factorial experimental design method; the parameters include eccentric disk number, eccentric mass and eccentric radius. Multiple tests were conducted by installing an eccentric mass block according to the combination of unbalanced fault parameters. After the rotational speed stabilized, the acceleration and phase angle were recorded and obtained using a triaxial acceleration sensor. x , y , z The original directional acceleration signal is processed by adding a Hanning window to the periodically segmented data. The time-domain acceleration signal is converted to the frequency domain using Fourier transform to obtain the measured acceleration signal value. The fault feature vector, including the response amplitude and phase at each harmonic, is extracted from the frequency domain signal to obtain the final experimental measurement result.

[0010] A further improvement of the technical solution of the present invention is that the steps for obtaining the rotor dynamics analysis results include: establishing a multi-disc rotor dynamics finite element analysis model corresponding to the rotor unbalanced test device, constructing the dynamic equation of the unbalanced multi-disc rotor system according to the fault characteristic space, reducing the dynamic equation by the degree of freedom reduction method, solving it by the complex exponential harmonic balance method, and obtaining the rotor dynamics analysis results.

[0011] A further improvement of the technical solution of the present invention is that the training process of the rotor imbalance fault diagnosis model further includes: using the hold-out method or cross-validation to verify the diagnostic accuracy of the trained model with experimental measurement results data that were not involved in the training.

[0012] In a second aspect, the present invention provides a fault diagnosis system for multi-disc rotor imbalance in gas turbines, comprising: The data acquisition unit is used to extract feature vectors based on the acquired abnormal vibration data; The identification and positioning unit is used to identify and locate the fault source based on the feature vector using the rotor imbalance fault diagnosis model, and obtain online identification and source tracing results of multi-disc rotor imbalance faults. The rotor imbalance fault diagnosis model consists of a low-fidelity model network and a high-fidelity model network. The low-fidelity model network is used to predict low-fidelity results given input data after training. The high-fidelity model network is used to obtain high-fidelity prediction results by learning the mapping relationship between low-fidelity output results and high-fidelity output results. The mapping relationship between low-fidelity output and high-fidelity output is expressed as follows: ; In the formula, It is a high-fidelity output result; It is a mapping network; X is the input data, which is the displacement, velocity, or acceleration signal of the multi-disc rotor system; It is a low-fidelity output result; It is a network hyperparameter.

[0013] In a third aspect, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the gas turbine multi-disc rotor imbalance fault diagnosis method as described in any one of the first aspects of the present invention.

[0014] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the gas turbine multi-disc rotor imbalance fault diagnosis method as described in any one of the first aspects of the present invention.

[0015] Compared with the prior art, the present invention has the following beneficial effects: In the technical solution provided by the embodiments of the present invention, the fault source is identified and located based on the established multi-fidelity data fusion rotor fault diagnosis model, and the online identification and tracing results of multi-disc rotor imbalance faults can be obtained. The rotor fault diagnosis model is composed of a low-fidelity model network and a high-fidelity model network. While ensuring accuracy, it can achieve one-time accurate identification of the characteristic parameters of multiple imbalance fault diagnosis models, and can also have good performance under noise interference and variable operating conditions.

[0016] In the preferred embodiment of this invention, a high-fidelity test signal is obtained through a rotor imbalance test device, and combined with low-fidelity numerical simulation data obtained from rotor dynamics analysis. Furthermore, a specific implementation method for establishing a rotor fault diagnosis model based on multi-fidelity data fusion is provided, achieving accurate identification and online tracing of multi-disc rotor imbalance fault parameters (eccentric disc number, eccentric radius, and eccentric mass). The addition of a Hanning window to the signal primarily reduces spectral leakage in the Fourier transform, thereby obtaining more accurate frequency components and amplitudes, and improving the identification effect of imbalance faults. The complex exponential harmonic balance method is used to solve the steady-state response of the nonlinear multi-disc rotor system, transforming the original time-domain dynamic equations to the frequency domain for solution. By combining this with the Newton-Raphson method, the steady-state response within the unstable speed range of the nonlinear multi-disc rotor system can be obtained, facilitating the handling of multiple solutions caused by nonlinearity.

[0017] In summary, the rotor fault diagnosis model of this invention utilizes high-fidelity but costly experimental data to correct and calibrate a low-fidelity but inexpensive numerical model, thereby constructing a high-precision and efficient multi-fidelity fusion model for accurate identification and fault diagnosis of multi-disc rotor imbalance. The proposed multi-fidelity data fusion model has advantages such as low data requirements, high diagnostic accuracy, strong interpretability, and support for online fault tracing, effectively solving problems such as strong data dependence, poor engineering adaptability, and ambiguous diagnostic information in existing technologies. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a method for diagnosing imbalance faults in a multi-disc rotor of a gas turbine, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the rotor imbalance test device in an embodiment of the present invention; Figure 3 This is a flowchart illustrating a method for diagnosing multi-disc rotor imbalance faults in gas turbines using multi-fidelity data fusion, as described in a specific embodiment of the present invention. Figure 4 This is a schematic diagram of the finite element analysis model of multi-disc rotor dynamics in an embodiment of the present invention; Figure 5 This is a schematic diagram of a rotor imbalance fault diagnosis model based on multi-fidelity data fusion in an embodiment of the present invention; Figure 6This is a schematic diagram of the eccentric disc number diagnosis results in an embodiment of the present invention; Figure 7 This is a schematic diagram of the eccentricity quality diagnosis results in an embodiment of the present invention; Figure 8 This is a schematic diagram of the eccentricity radius diagnosis results in an embodiment of the present invention; Figure 9 This is a schematic diagram of a gas turbine multi-disc rotor imbalance fault diagnosis system in an embodiment of the present invention; The explanations of the reference numerals in the figures are as follows: 1. Basic platform; 2. Support bearings; 3. Triaxial accelerometer; 4. Bearing housing; 5. Perforated disc; 6. Centrifugal mass block; 7. Unbalance adjustment hole; 8. Eccentric mass block; 9. Spindle; 10. Coupling; 11. Servo motor; 12. Infrared sensor; 13. Data acquisition unit. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention; obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Based on the technical solutions disclosed in the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0022] Please see Figure 1 The present invention provides a method for diagnosing the imbalance fault of a multi-disc rotor in a gas turbine, comprising the following steps: Step 1: Extract feature vectors based on the acquired abnormal vibration data; Step 2: Based on the feature vector, the rotor imbalance fault diagnosis model is used to identify and locate the fault source, and the online identification and source tracing results of multi-disc rotor imbalance faults are obtained. The rotor imbalance fault diagnosis model consists of a low-fidelity model network and a high-fidelity model network. The low-fidelity model network is used to predict low-fidelity results given input data after training. The high-fidelity model network is used to obtain high-fidelity prediction results by learning the mapping relationship between low-fidelity output results and high-fidelity output results. The mapping relationship between low-fidelity output and high-fidelity output is expressed as follows: ; In the formula, It is a high-fidelity output result; It is a mapping network; X is the input data, which is the displacement, velocity, or acceleration signal of the multi-disc rotor system; It is a low-fidelity output result; It is a network hyperparameter.

[0023] Existing spectrum analysis relies on the frequency domain characteristics of vibration signals, which has limited ability to identify multi-fault coupled conditions. In the technical solution of this invention, the rotor imbalance fault diagnosis model consists of a low-fidelity model network and a high-fidelity model network. After training, the low-fidelity model network can predict low-fidelity results, while the high-fidelity model network learns the mapping relationship between low-fidelity and high-fidelity outputs to obtain high-fidelity prediction results. This multi-fidelity data fusion method does not simply rely on a single frequency domain feature, but integrates multiple aspects of information, enabling a more comprehensive analysis of abnormal vibration data and uncovering characteristic patterns under different fault couplings. This effectively improves the ability to identify multi-fault coupled conditions and solves the shortcomings of spectrum analysis in this regard.

[0024] The influence coefficient method requires extensive on-site dynamic balancing tests, and the calibration process is complex, costly, and unsuitable for multi-parameter identification of multi-disc rotors. The technical solution of this invention extracts feature vectors from acquired abnormal vibration data and uses a rotor imbalance fault diagnosis model for identification and location. Unlike the influence coefficient method, it eliminates the need for extensive on-site testing, reducing cost and operational complexity. Furthermore, the model design of this invention can accurately identify multiple characteristic parameters of the imbalance fault diagnosis model in a single step, making it particularly suitable for multi-disc rotors and solving the problem of multi-parameter identification inherent in the influence coefficient method.

[0025] Artificial intelligence methods rely on a large number of high-quality labeled samples, which are difficult to obtain in actual engineering. Furthermore, many models are "black boxes," lacking physical interpretability, leading to low reliability of diagnostic results and insufficient support for maintenance decisions. The technical solution of this invention constructs a rotor fault diagnosis model based on multi-fidelity data fusion. On the one hand, multi-fidelity data fusion can reduce the dependence on a large number of labeled samples to some extent, utilizing the relationships between data of different fidelities for learning and prediction. On the other hand, the model structure is not completely "black box"; the division of labor and cooperation between low-fidelity and high-fidelity model networks gives the model a certain degree of logic and interpretability. Diagnostic results obtained based on such a model are more reliable and can provide stronger support for maintenance decisions.

[0026] In summary, the technical solution of this invention can obtain online identification and source tracing results of multi-disc rotor imbalance faults. While ensuring accuracy, it can achieve one-time accurate identification of the characteristic parameters of various imbalance fault diagnosis models, and also has good performance under noise interference and variable operating conditions.

[0027] Please see Figure 2 In this embodiment of the invention, the rotor imbalance testing device includes: a base platform 1, a main shaft 9, a perforated disc 5, a servo motor 11, a support bearing 2, a bearing housing 4, a coupling 10, an eccentric mass block 8, a centrifugal mass block 6, a triaxial accelerometer 3, an infrared sensor 12, and a data acquisition unit 13, etc.; wherein, Basic platform 1 is used to provide support and vibration isolation environment; Bearing housing 4 and support bearing 2 support the rotating parts in the rotor imbalance test device, while bearing the load and reducing friction; main shaft 9 detachably connects multiple perforated disks 5 into one unit and is supported by two support bearings; the perforated disks 5 have multiple imbalance adjustment holes 7 radially to simulate a rotor system with imbalance faults; servo motor 11 drives main shaft 9 and perforated disks 5 to rotate at a specified speed; coupling 10 connects servo motor 11 and main shaft 9; eccentric mass block 8 simulates the imbalance and can be connected to different positions of perforated disk 5 via bolts and nuts; centrifugal mass... Block 6 is used to simulate the equivalent centrifugal load of the blade and is added to each opening on the side of the perforated disk 5. The triaxial accelerometer 3 is installed on the bearing seats at both ends of the shaft. Its internal microelectromechanical system can detect the capacitance change caused by the deformation of the small cantilever structure due to acceleration and is used to independently output displacement, velocity and acceleration signals in the three measurement axes. The infrared sensor 12 is installed on the bearing seat of the centrifugal mass block with openings on the side edge of the perforated disk. When the centrifugal mass block passes directly in front of the sensor, a voltage pulse is generated. The rotational speed and phase angle of the shaft can be measured by calculating the number of pulses. The data acquisition unit 13 (DAU) is used to amplify the analog signal input from the sensor and convert it into a high sampling rate digital signal, and remove noise by adding a low-pass filter.

[0028] In a specific exemplary technical solution, the main shaft is rotatably mounted on the base platform via bearing seats and supporting bearings. The servo motor is fixedly mounted on the base platform, and the output end of the servo motor is connected to the main shaft via a coupling. Multiple perforated discs are detachably mounted on the main shaft. The servo motor is used to drive the main shaft and the perforated discs to rotate at a specified speed. Multiple imbalance adjustment holes are radially formed on the perforated discs to simulate a rotor system with imbalance faults. The eccentric mass block is used to simulate the imbalance and can be connected to different positions of the perforated discs via bolts and nuts. The centrifugal mass block is used to simulate the equivalent centrifugal load of the blades and is added to each opening on the side of the perforated discs.

[0029] In a specific example, the unbalance fault parameters include the eccentric disk number, eccentric radius, and eccentric mass. Each disk has multiple unbalance adjustment holes radially spaced, with holes positioned at radii of 5 cm, 7.5 cm, and 10 cm from the disk center. The eccentric mass blocks have masses of 10 g, 15 g, and 20 g, respectively. Centrifugal mass blocks simulate the equivalent centrifugal load of the blades. By machining centrifugal mass blocks with a certain mass deviation and installing them in the perforated disks, the influence of blade mistuning on the multi-disc rotor system is investigated.

[0030] Please see Figure 3 The present invention provides a method for diagnosing multi-disc rotor imbalance faults in gas turbines based on multi-fidelity data fusion, the main processes of which are as follows: Step S1: Design the fault feature space based on the unbalanced state of the multi-disc rotor. Use the partial factorial design method to select the combination of unbalanced fault test parameters (including eccentric disc number, eccentric mass and eccentric radius) to ensure that the test samples can cover the fault feature space to the maximum extent.

[0031] Step S2: Multiple sets of tests were conducted by installing eccentric mass blocks according to the unbalanced fault parameter combinations. After the rotor speed stabilized, the acceleration and phase angle were recorded, and data acquisition began for 5 seconds to obtain the raw signal set. x , y , z The raw directional acceleration signal is processed by applying a Hanning window to the periodically segmented data. A Fourier transform is then used to convert the time-domain acceleration signal to the frequency domain, obtaining the measured acceleration signal values. Fault feature vectors, including the response amplitude and phase at each harmonic, are extracted from the frequency domain signal. Interpretably, the main purpose of applying the Hanning window to the signal is to reduce spectral leakage in the Fourier transform, thereby obtaining more accurate frequency components and amplitudes, and improving the identification effect of imbalance faults.

[0032] Step S3: Establish a finite element analysis model of the multi-disc rotor dynamics corresponding to the multi-disc rotor imbalance fault diagnosis system. Based on the fault characteristic space, construct the dynamic equations of the unbalanced multi-disc rotor system. Reduce the dynamic equations using the degree-of-freedom reduction method, and then solve them using the complex exponential harmonic balance method to obtain the rotor dynamic response results. Explainingly, the complex exponential harmonic balance method is used to solve the steady-state response of the nonlinear multi-disc rotor system. The original time-domain dynamic equations are transformed to the frequency domain for solution. By combining this with the Newton-Raphson method, the steady-state response within the unstable speed range of the nonlinear multi-disc rotor system can be obtained, facilitating the handling of multiple solutions caused by nonlinearity. Using the complex exponential harmonic balance method to solve the rotor dynamic equations avoids the product and difference operations required by the sine and cosine harmonic balance method, simplifying formula derivation and improving program conciseness. It is particularly suitable for accurately solving complex rotor dynamic systems that require consideration of higher-order harmonics and strong nonlinear characteristics. Methods for reducing degrees of freedom include substructure synthesis (Craig-Bampton method) and modal synthesis, which greatly improve computational efficiency while maintaining computational accuracy by reducing the full-degree-of-freedom finite element model to a smaller number of degrees of freedom.

[0033] Step S4: Using the reduced model analysis results as low-fidelity data and the experimental measurement results as high-fidelity data, construct a rotor imbalance fault diagnosis model with multi-fidelity data fusion. Perform data alignment and scaling preprocessing on the two datasets. Iteratively optimize the network through gradient backpropagation mechanism to reduce the composite empirical loss function. Use hold-out method or cross-validation to verify the diagnostic accuracy of the model with experimental data that did not participate in the training. Furthermore, it also includes: step S5, when new abnormal vibration data is detected, the feature vector is extracted in real time and input into the model, so that the constructed multi-fidelity fault diagnosis model can be used to identify and locate the fault source, and realize the online identification and source tracing of multi-disc rotor imbalance faults.

[0034] Please see Figure 4 The example provides a flowchart of the finite element analysis model and solution method for multi-disc rotor dynamics.

[0035] In this embodiment of the invention, the dynamic equation of the rotor in the rotating coordinate system can be expressed as: (1) in, Here is the rotor mass matrix. The rotor Coriolis force matrix, Here is the rotor static stiffness matrix. For rotor rotation softening matrix, It is a linear excitation force vector; It is a nonlinear excitation force vector. Let ω represent the rotor's displacement, velocity, and acceleration, respectively, and ω be the rotor speed.

[0036] The effect of the bearing on the rotor is considered as additional stiffness and additional damping. An eight-coefficient bearing model is adopted, with the eight coefficients including: principal stiffness coefficients. and Cross stiffness coefficient and Principal damping coefficient and and cross-damping coefficient and Assuming the bearing damping is isotropic and neglecting the cross-damping coefficient, that is... , .

[0037] Support force of bearings in fixed and rotating coordinate systems and They are respectively: (2) (3) in, Here is the bearing stiffness matrix in a fixed coordinate system. Here is the bearing damping matrix in a fixed coordinate system. Here is the bearing stiffness matrix in the rotating coordinate system. Let be the bearing damping matrix in the rotating coordinate system; and the rotating coordinate system and the fixed coordinate system have the following relationship: (4) (5) Substituting equations (4) and (5) into equations (2) and (3), we can obtain the bearing matrix in the rotating coordinate system as follows: (6) (7) in: (8) and (9) Adding the bearing matrix (6) to equation (1) yields: (10) Substructure Modal Synthesis (CMS) transforms the rotor's displacement coordinates into principal modal coordinates and generalized coordinates corresponding to the interface degrees of freedom by constructing coordinate transformation relationships. In a fixed-interface CMS, the generalized coordinates corresponding to the interface degrees of freedom are the displacement coordinates of the interface degrees of freedom, which can be directly used to establish a connection with the bearing. Therefore, CMS can be implemented by first reducing and then assembling. Ignoring damping, equation (1) can be written as: (11) in, Here is the rotor stiffness matrix. , For internal displacement coordinates, The coordinates are the interface displacement coordinates, with subscripts. i Indicates the interior of the rotor. b Indicates the boundary.

[0038] The coordinate transformation equation for a fixed-interface CMS can be expressed as: (12) in, The modal cutoff number is The principal mode matrix, For the constraint mode matrix, For a fixed-interface CMS, Principal mode coordinate vector, The coordinate vector reserved for a fixed-interface CMS.

[0039] The principal mode vector can be obtained by the following formula: (13) in, For the first 1 eigenvalue, For the first Eigenvectors. Equation (13) is the eigenvalue equation after fixing the interface coordinates of the rotor.

[0040] The constraint mode matrix can be represented as: (14) Substitute equation (12) into equation (10) and multiply both sides of the equation by the left side. We can obtain: (15) in, (16) Equation (15) only reduces the degrees of freedom of the rotor structure; it is still necessary to establish the connection relationship between the rotor and the bearing.

[0041] Arrange the bearing matrix according to the degree of freedom order in equation (12). Taking the rotor dynamics system with two supporting bearings in the embodiment of the present invention as an example, let the last 4 degrees of freedom be the bearing degrees of freedom, then the bearing matrix of the reduced model can be expressed as: (17) in, These are the stiffness matrices of the first and second bearings, respectively. These are the damping matrices for the first and second bearings, respectively.

[0042] By combining equation (17) and equation (15), we can obtain the dynamic equation of the rotor system after the fixed interface CMS reduction: (18) The above formula uses "^" to represent a shrunk matrix.

[0043] The reduced model has the following degrees of freedom and modal cutoff numbers. The sum of the bearing degrees of freedom is generally much smaller than the total degrees of freedom of the complete model, thus achieving model reduction. Furthermore, during this reduction process, no reduction was applied to the bearing parameters, ensuring the accuracy of the calculations after reduction.

[0044] Furthermore, the dynamic response of the nonlinear rotor system is solved using the complex exponential harmonic balance method. This is based on Euler's formula, i.e. ,but: (19) in, The imaginary unit, .

[0045] Therefore, equation (19) can be written as: (20) in, (twenty one) Based on the complex exponential Fourier series, the steady-state response, linear force, and nonlinear force can be expressed as follows: (twenty two) in, For the displacement of the first First harmonic coefficient; For the first linear force First harmonic coefficient; The first nonlinear force The first harmonic coefficient. Substituting the expansions of the steady-state response, linear force, and nonlinear force in equation (22) into equation (20), and based on the fact that the coefficients of the same exponent terms are equal, we can obtain: (twenty three) (twenty four) (25) (26) (27) (28) in, The harmonic number is; the dimension of the matrix on the left side of equation (23) is ,in Let be the number of degrees of freedom of the system.

[0046] The complex exponential harmonic balance method used in this invention transforms the dynamic equations with periodic time-varying coefficients in the time domain into a system of equations with finite dimensions in the frequency domain. Although this increases the number of unknowns, the system of equations can be solved directly to obtain the response, avoiding time-domain integration and improving computational efficiency.

[0047] As a special case, if there are no nonlinear forces in the rotor system, and it is only excited by gravity and unbalanced forces, then equation (23) is linear. The matrix on the left side of the equation... It can be determined in advance; only a linear excitation force vector exists on the right side of the equation. The equation degenerates into: In this case, the equation in equation (23) can be solved directly. However, if there are nonlinear forces, equation (23) cannot be solved directly. This is because nonlinear forces such as crack force and rubbing force are directly related to the system response. In this case, the solution of the equation requires the use of iterative algorithms, such as Newton's method and arc-length extension method. After obtaining the solution of equation (23), the harmonic coefficients of the response are substituted into equation (22), and the dynamic response in the fixed coordinate system can be obtained by using the coordinate transformation equation in equation (4).

[0048] Please see Figure 5 This paper exemplifies a rotor imbalance fault diagnosis model based on multi-fidelity data fusion. The multi-fidelity model proposed in this invention consists of two networks: NN L A network, after training, can predict low-fidelity results y given system input X. L ; and NN H The network, through learning y L and y H The mapping relationship between them is used to approximate the prediction of high-fidelity results y. H .

[0049] Using low-fidelity data Training NN L Network, to obtain low-fidelity prediction results , can be represented as:

[0050] in, It is the input low-fidelity data, which can be the displacement, velocity or acceleration results obtained from rotor dynamics analysis; These are low-fidelity prediction results, including predicted values ​​for unbalanced mass, radius, and disk number. For low-fidelity networks NN L The mapping relationship, For low-fidelity networks NN L The hyperparameters. The loss function for the low-fidelity model is:

[0051] in, The mean squared error of the low-fidelity model. For low-fidelity data scale.

[0052] To train the NN H The network uses the learned low-fidelity prediction results. and high-fidelity data Thus, high-fidelity prediction results are obtained. , can be represented as:

[0053] in, It is the input high-fidelity data, which can be displacement, velocity or acceleration signals measured in the experiment; These are high-fidelity prediction results, including predicted values ​​for unbalanced mass, radius, and disk number. For high-fidelity networks NN H The mapping relationship, For high-fidelity networks NN H Hyperparameters.

[0054] The loss function for the high-fidelity model is: ; In the formula, The mean square error of the high-fidelity model network; For high-fidelity data scale.

[0055] During the overall training process, the max-min normalization method is used to normalize all input and output data to the range [-1, 1], eliminating differences in dimensions and orders of magnitude between data and avoiding training instability. Automatic differentiation is used to calculate the gradient information required for network hyperparameter optimization during backpropagation. The Adam optimizer is then used to optimize the network's hyperparameters to minimize the following loss function:

[0056] in, For the composite empirical loss function of the multifidelity model, λ As a weighting factor, it can be adjusted to determine the degree of contribution of high-fidelity data.

[0057] Please see Figures 6 to 8 The example provides fault diagnosis results, with regression analyses of eccentric disk number, eccentric mass, and eccentric radius position detection shown in the figures. The horizontal axis represents the true value, and the vertical axis represents the detected value. As shown in the figures, all three possible locations of the three types of imbalance faults can be accurately detected, with detection errors within ±5%. This verifies that the multi-fidelity data fusion fault diagnosis model proposed in this embodiment of the invention has the ability to accurately identify and locate imbalance faults.

[0058] In summary, this invention discloses a multi-disc rotor imbalance fault diagnosis scheme for gas turbines based on multi-fidelity data fusion. The scheme designs a multi-disc rotor imbalance fault feature space including eccentric disk number, eccentric mass, and eccentric radius. It conducts multi-disc rotor imbalance fault simulation experiments to obtain measured values ​​of the response signals. A multi-disc rotor dynamics finite element analysis model corresponding to the multi-disc rotor imbalance fault diagnosis system is established. The dynamic equations are reduced using the degree-of-freedom reduction method, and further solved using the complex exponential harmonic balance method to obtain the rotor dynamic response calculation results. Using the reduced model analysis results as low-fidelity data and the experimental measurement results as high-fidelity data, a multi-fidelity data fusion rotor imbalance fault diagnosis model is constructed. Feature vectors are extracted in real time and input into the model. The constructed multi-fidelity fault diagnosis model is used to identify and locate the fault source, achieving accurate identification and online source tracing of multi-disc rotor imbalance faults.

[0059] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0060] Please see Figure 9 In this embodiment of the invention, a gas turbine multi-disc rotor imbalance fault diagnosis system is provided, comprising: The data acquisition unit is used to extract feature vectors based on the acquired abnormal vibration data; The identification and positioning unit is used to identify and locate the fault source based on the feature vector using the rotor imbalance fault diagnosis model, and obtain online identification and source tracing results of multi-disc rotor imbalance faults. The rotor imbalance fault diagnosis model consists of a low-fidelity model network and a high-fidelity model network. The low-fidelity model network is used to predict low-fidelity results given input data after training. The high-fidelity model network is used to obtain high-fidelity prediction results by learning the mapping relationship between low-fidelity output results and high-fidelity output results. The mapping relationship between low-fidelity output and high-fidelity output is expressed as follows: ; In the formula, It is a high-fidelity output result; It is a mapping network; X is the input data, which is the displacement, velocity, or acceleration signal of the multi-disc rotor system; It is a low-fidelity output result; It is a network hyperparameter.

[0061] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used to perform the operation of a gas turbine multi-disc rotor imbalance fault diagnosis method.

[0062] In one embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the operating system of the terminal. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the gas turbine multi-disc rotor imbalance fault diagnosis method in the above embodiments.

[0063] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0064] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0065] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for diagnosing imbalance faults in multi-disc rotors of gas turbines, characterized in that, Includes the following steps: Feature vectors are extracted based on the acquired abnormal vibration data; Based on the feature vector, the rotor imbalance fault diagnosis model is used to identify and locate the fault source, and the online identification and source tracing results of multi-disc rotor imbalance faults are obtained. The rotor imbalance fault diagnosis model consists of a low-fidelity model network and a high-fidelity model network. The low-fidelity model network is used to predict low-fidelity results given input data after training. The high-fidelity model network is used to obtain high-fidelity prediction results by learning the mapping relationship between low-fidelity output results and high-fidelity output results. The mapping relationship between low-fidelity output and high-fidelity output is expressed as follows: ； In the formula, It is a high-fidelity output result; It is a mapping network; X is the input data, which is the displacement, velocity, or acceleration signal of the multi-disc rotor system; It is a low-fidelity output result; It is a network hyperparameter.

2. The method for diagnosing imbalance faults in a multi-disc rotor of a gas turbine according to claim 1, characterized in that, During the training process of the rotor imbalance fault diagnosis model, The low-fidelity model network is trained using low-fidelity data to obtain low-fidelity prediction results, which are expressed as follows: ； In the formula, These are low-fidelity prediction results, including predicted values ​​for unbalanced mass, radius, and disk number. The mapping relationship for low-fidelity model networks; It is the input low-fidelity data, which is the displacement, velocity or acceleration result obtained from the dynamic analysis of multi-disc rotors; For the hyperparameters of the low-fidelity model network; The high-fidelity model network uses the learned low-fidelity prediction results and high-fidelity data to obtain high-fidelity prediction results, expressed as follows: ； In the formula, It provides high-fidelity prediction results, including predicted values ​​for unbalanced mass, radius, and disk number; The mapping relationship for the high-fidelity model network; It is the input high-fidelity data, which is the displacement, velocity or acceleration signal measured in the experiment; These are the hyperparameters of the high-fidelity model network; During overall training, the low-fidelity data and high-fidelity data datasets are aligned and scaled preprocessed. The network is iteratively optimized through gradient backpropagation to reduce the composite empirical loss function. The experimental measurement results are used as high-fidelity data, and the rotor dynamics analysis results are used as low-fidelity data.

3. The method for diagnosing imbalance faults in a multi-disc rotor of a gas turbine according to claim 2, characterized in that, The specific implementation steps for aligning and scaling the two datasets (low-fidelity and high-fidelity data) and iteratively optimizing the network using gradient backpropagation to reduce the composite empirical loss function are as follows: The min-max normalization method is used to normalize all input and output data to the range [-1, 1]. Automatic differentiation is used to calculate the gradient information required for hyperparameter optimization during backpropagation. The Adam optimizer is used to optimize the network's hyperparameters to minimize the following loss function: ； In the formula, For the composite empirical loss function of the rotor imbalance fault diagnosis model; The mean square error of the low-fidelity model network; The mean square error of the high-fidelity model network; This is a weighting factor used to adjust the degree of contribution to high-fidelity data; ； ； In the formula, For low-fidelity data scale; For high-fidelity data scale.

4. The method for diagnosing imbalance faults in a multi-disc rotor of a gas turbine according to claim 2, characterized in that, The test measurement results are obtained based on a rotor imbalance test device; the rotor imbalance test device includes: a base platform, a main shaft, a perforated disk, a servo motor, support bearings, bearing housings, a coupling, an eccentric mass block, a centrifugal mass block, a triaxial accelerometer, an infrared sensor, and a data acquisition unit; wherein... The base platform provides support and vibration isolation. The main shaft is rotatably mounted on the base platform via the bearing housing and the supporting bearing. A servo motor is fixedly mounted on the base platform, and its output is connected to the main shaft via a coupling. Multiple perforated discs are detachably mounted on the main shaft. The servo motor drives the main shaft and the perforated discs to rotate at a specified speed. Multiple imbalance adjustment holes are radially provided on the perforated discs for mounting the eccentric mass blocks to simulate a rotor system with imbalance faults. Multiple mounting holes are provided on the circumferential surface of the perforated discs for mounting the centrifugal mass blocks to simulate equivalent centrifugal loads on the blades. The triaxial accelerometers are mounted on bearing seats at both ends of the spindle, and are used to independently output displacement, velocity and acceleration signals in the three measurement axes. The infrared sensor is mounted on the bearing seats of each centrifugal mass block on the peripheral edge of the perforated disk, and is used to generate a voltage pulse when the centrifugal mass block passes directly in front of the infrared sensor. The rotational speed and phase angle of the axis are measured by calculating the number of pulses. The data acquisition unit is used to amplify the analog signals input from each sensor and convert them into high-sampling-rate digital signals. Noise is removed by adding a low-pass filter to obtain the final experimental measurement results.

5. The method for diagnosing imbalance faults in a multi-disc rotor of a gas turbine according to claim 4, characterized in that, The specific steps for obtaining test measurement results based on the rotor imbalance test device include: The fault feature space is designed based on the unbalanced state of multi-disc rotors, and the combination of unbalanced fault test parameters is selected using the partial factorial experimental design method; the parameters include eccentric disk number, eccentric mass and eccentric radius. Multiple tests were conducted by installing an eccentric mass block according to the combination of unbalanced fault parameters. After the rotational speed stabilized, the acceleration and phase angle were recorded and obtained using a triaxial acceleration sensor. x , y , z The original directional acceleration signal is processed by adding a Hanning window to the periodically segmented data. The time-domain acceleration signal is converted to the frequency domain using Fourier transform to obtain the measured acceleration signal value. The fault feature vector, including the response amplitude and phase at each harmonic, is extracted from the frequency domain signal to obtain the final experimental measurement result.

6. The method for diagnosing imbalance faults in a multi-disc rotor of a gas turbine according to claim 5, characterized in that, The steps for obtaining the rotor dynamics analysis results include: establishing a multi-disc rotor dynamics finite element analysis model corresponding to the rotor unbalanced test device; constructing the dynamic equations of the unbalanced multi-disc rotor system based on the fault characteristic space; reducing the dynamic equations using the degree-of-freedom reduction method; solving the equations using the complex exponential harmonic balance method; and obtaining the rotor dynamics analysis results.

7. The method for diagnosing imbalance faults in a multi-disc rotor of a gas turbine according to claim 2, characterized in that, The training process of the rotor imbalance fault diagnosis model also includes: using the hold-out method or cross-validation to verify the diagnostic accuracy of the trained model with experimental measurement results data that were not involved in the training.

8. A fault diagnosis system for multi-disc rotor imbalance in a gas turbine, characterized in that, include: The data acquisition unit is used to extract feature vectors based on the acquired abnormal vibration data; The identification and positioning unit is used to identify and locate the fault source based on the feature vector using the rotor imbalance fault diagnosis model, and obtain online identification and source tracing results of multi-disc rotor imbalance faults. The rotor imbalance fault diagnosis model consists of a low-fidelity model network and a high-fidelity model network. The low-fidelity model network is used to predict low-fidelity results given input data after training. The high-fidelity model network is used to obtain high-fidelity prediction results by learning the mapping relationship between low-fidelity output results and high-fidelity output results. The mapping relationship between low-fidelity output and high-fidelity output is expressed as follows: ； In the formula, It is a high-fidelity output result; It is a mapping network; X is the input data, which is the displacement, velocity, or acceleration signal of the multi-disc rotor system; It is a low-fidelity output result; It is a network hyperparameter.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the gas turbine multi-disc rotor imbalance fault diagnosis method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the gas turbine multi-disc rotor imbalance fault diagnosis method as described in any one of claims 1 to 7.

Images