A method and related apparatus for intelligent in-situ balancing of aero-engines based on LSTM

Abstract

This invention provides an LSTM-based intelligent in-situ balancing method and related apparatus for aero-engines. The method includes acquiring time-series data of the aero-engine's rotor speed and the original vibration phase and amplitude; inputting the time-series data of the rotor speed and the original vibration phase and amplitude into a pre-trained LSTM prediction model to obtain the weighting angle and magnitude for intelligent in-situ balancing of the aero-engine; wherein the LSTM prediction model is trained based on the LSTM algorithm, taking the time-series data of the rotor speed and the original vibration phase and amplitude as input, and outputting the weighting angle and magnitude for intelligent in-situ balancing of the aero-engine. This invention utilizes LSTM to uncover the nonlinear relationship between vibration magnitude and imbalance, thereby accurately supporting the intelligent dynamic balancing of the engine's core rotor.

Description

Technical Field

[0001] This invention belongs to the fields of aero-engine vibration and fault diagnosis technology, specifically relating to an intelligent in-situ balancing method and related device for aero-engines based on LSTM. Background Technology

[0002] Currently, a certain type of aero-engine is experiencing a sudden increase in vibration due to imbalance. Traditional methods for addressing this problem, such as the three-circle method and the influence coefficient method, suffer from drawbacks including numerous trial weights and high assembly and verification costs. Furthermore, after multiple trial weight runs, the influence coefficients obtained from whole-engine vibration measurement points do not converge, making it difficult to guide the engine's balancing.

[0003] When performing dynamic balancing, the first step is to obtain data such as the magnitude, phase, and weight increment of the rotor vibration. Vibration magnitude refers to the amplitude of the vibration, phase is the degree reference of the vibration waveform, and weight increment is the size of the mass block added at the reference position. Accurately determining the relationship between the amplitude of the rotor's unbalanced vibration and the amount of unbalance is crucial to the effectiveness of rotor dynamic balancing. Due to the complexity of aero-engines and the existence of empirical contact angles, there is currently no effective model that reflects the relationship between vibration magnitude and unbalance, making it difficult to accurately achieve intelligent dynamic balancing of the engine's core rotor. Summary of the Invention

[0004] In view of this, the present invention aims to provide an intelligent in-situ balancing method and related device for aero-engines based on LSTM. By using an LSTM model to mine the nonlinear relationship between aero-engine vibration and imbalance, the method can accurately support the intelligent dynamic balancing of the engine's core rotor.

[0005] To achieve the above objectives, the technical solution provided by the present invention is as follows:

[0006] The first aspect of this invention provides an intelligent in-situ balancing method for aero-engines based on LSTM, comprising the following steps:

[0007] Acquire time-series data of the rotor speed, raw vibration phase, and amplitude of the aero-engine;

[0008] The time series data of rotor speed and original vibration phase and amplitude are input into a pre-trained LSTM prediction model to obtain the weighting angle and weighting magnitude for intelligent in-flight balancing of aero-engines.

[0009] The LSTM prediction model is trained based on the LSTM algorithm. It takes time series data of rotor speed and original vibration phase and amplitude as input, and outputs the weighting angle and weighting magnitude of the intelligent in-flight balancing of the aero-engine.

[0010] Furthermore, the time series data of rotor speed and original vibration phase and amplitude are obtained based on key phase signal and vibration acceleration signal. The key phase signal is a speed signal with key phase function obtained by measuring the modified rotor blade with added low tooth blade using a key phase speed sensor. The position of the low tooth blade on the modified rotor blade is the reference position for balancing the aero engine.

[0011] Furthermore, when obtaining time-series data of rotor speed and original vibration phase and amplitude based on the key phase signal and vibration acceleration signal, the key phase signal is also conditioned, including:

[0012] The bond phase signal is normalized using the maximum-minimum normalization method;

[0013] The location of the low-tooth signal is found from the normalized bond phase signal based on the low-tooth signal threshold.

[0014] The signal at the low-tooth signal location is set to a high level of 1, while the signals at other locations are set to 0, thereby obtaining the conditioned bond phase signal.

[0015] Furthermore, the formula for calculating the original vibration phase is as follows:

[0016]

[0017] The phase corresponding to the fundamental frequency of the phase signal's rotational speed. This represents the phase of the vibration acceleration signal at the fundamental frequency position of the rotational speed, obtained through FFT analysis.

[0018] Furthermore, FFT analysis of the bond phase signal and vibration acceleration signal includes:

[0019] Perform an FFT on the original key phase signal to calculate the rotational speed corresponding to the current signal, and obtain the rotor speed;

[0020] Then, an FFT is performed on the conditioned key phase signal to calculate the phase corresponding to the rotational speed fundamental frequency;

[0021] Perform an FFT on the vibration acceleration signal to calculate the phase of each frequency within a set range to the left and right of the rotational speed base frequency, and take the average value of all phases as the phase of the vibration acceleration signal at the rotational speed base frequency position.

[0022] Furthermore, the network architecture of the LSTM prediction model includes: an input layer, hidden layers, and an output layer;

[0023] The input layer is used to receive time-series data of the rotor speed of the aero-engine, as well as the raw vibration phase and amplitude.

[0024] Hidden layers are used to mine the nonlinear relationship between vibration and imbalance in time series data through LSTM structures;

[0025] The output layer is used to output the information mined by the hidden layer for iterative prediction, and outputs the predicted weighting angle and weighting magnitude.

[0026] Furthermore, the LSTM prediction model uses the error backpropagation method for model training.

[0027] A second aspect of the present invention provides an LSTM-based intelligent in-situ balancing device for aero-engines, comprising:

[0028] The data acquisition module is used to acquire time-series data of the rotor speed, raw vibration phase, and amplitude of the aero-engine.

[0029] The balancing module is used to input the time series data of rotor speed and original vibration phase and amplitude into a pre-trained LSTM prediction model to obtain the weighting angle and weighting magnitude for intelligent in-flight balancing of aero-engines.

[0030] The LSTM prediction model is trained based on the LSTM algorithm. It takes time series data of rotor speed and original vibration phase and amplitude as input, and outputs the weighting angle and weighting magnitude of the intelligent in-flight balancing of the aero-engine.

[0031] Accordingly, the present invention also provides a computer device, the device including a processor and a memory:

[0032] The memory is used to store computer programs and send the instructions of the computer programs to the processor;

[0033] The processor executes, according to the instructions of the computer program, a smart local balancing method for aero-engines based on LSTM, as described in the first aspect.

[0034] Accordingly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements an LSTM-based intelligent in-situ balancing method for aero-engines as described in the first aspect.

[0035] In summary, this invention provides an LSTM-based intelligent in-situ balancing method and related apparatus for aero-engines. The method includes acquiring time-series data of the aero-engine's rotor speed and the original vibration phase and amplitude; inputting the time-series data of the rotor speed and the original vibration phase and amplitude into a pre-trained LSTM prediction model to obtain the weighting angle and magnitude for intelligent in-situ balancing of the aero-engine; wherein the LSTM prediction model is trained based on the LSTM algorithm, taking the time-series data of the rotor speed and the original vibration phase and amplitude as input, and outputting the weighting angle and magnitude for intelligent in-situ balancing of the aero-engine. This invention utilizes LSTM to uncover the nonlinear relationship between vibration magnitude and imbalance, thereby accurately supporting the intelligent dynamic balancing of the engine's core rotor. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 A flowchart of an LSTM-based intelligent in-situ balancing method for aero-engines provided in an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of the data testing scheme provided in an embodiment of the present invention;

[0039] Figure 3 This is a real biomarker signal diagram provided for an embodiment of the present invention;

[0040] Figure 4 Vibration acceleration signal diagram provided in an embodiment of the present invention;

[0041] Figure 5 A schematic diagram illustrating the principle of vibration phase calculation provided in this embodiment of the invention;

[0042] Figure 6 This is a conditioning phase signal diagram provided in an embodiment of the present invention;

[0043] Figure 7 This is a high-voltage rotor vibration phase calculation process provided in an embodiment of the present invention;

[0044] Figure 8 A graph showing the relationship between the empirical weighting angle and rotational speed provided for embodiments of the present invention;

[0045] Figure 9 A threshold configuration diagram provided for an embodiment of the present invention;

[0046] Figure 10 This is a diagram of the LSTM hidden layer cell structure provided in an embodiment of the present invention;

[0047] Figure 11 This is a diagram of the original LSTM structure provided in an embodiment of the present invention;

[0048] Figure 12 This is a flowchart illustrating the overall framework of the LSTM prediction model provided in this embodiment of the invention.

[0049] Figure 13 The prediction result diagram provided in the embodiment of the present invention;

[0050] Figure 14 This is a block diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0052] Currently, the original weighting phase angle calculated based on vibration data and fiber bonded phase data is not the true weighting angle and still needs to be combined with empirical angles. Determining these empirical angles requires extensive human experience. This invention provides an intelligent in-situ balancing method and related device for aero-engines based on LSTM. Based on a large amount of historical empirical data from in-situ balancing tests of a certain type of aero-engine core rotor, it employs artificial intelligence methods to deeply mine the relevant patterns hidden within the large amount of data from previous successful balancing tests, obtaining an intelligent in-situ balancing model for the rotor. Finally, based on this model, intelligent dynamic balancing of the engine's core rotor is achieved.

[0053] Please see Figure 1 This embodiment provides an intelligent in-situ balancing method for aero-engines based on LSTM, including the following steps:

[0054] S11: Acquire time-series data of the rotor speed of the aero-engine, as well as the original vibration phase and amplitude.

[0055] It should be noted that this step first involves collecting time-series data of key parameters from the actual operation of the aero-engine, including rotor speed, initial vibration phase, and vibration amplitude. These data reflect the dynamic behavior of the engine under different operating conditions and form the basis for balancing analysis.

[0056] In some possible implementations, the acquired data also needs to be preprocessed, that is, the data is normalized according to the min-max normalization method.

[0057] S12: Input the time series data of rotor speed and original vibration phase and amplitude into the pre-trained LSTM prediction model to obtain the weighting angle and weighting magnitude for intelligent in-flight balancing of aero-engines;

[0058] The LSTM prediction model is trained based on the LSTM algorithm. It takes time series data of rotor speed and original vibration phase and amplitude as input, and outputs the weighting angle and weighting magnitude of the intelligent in-flight balancing of the aero-engine.

[0059] It should be noted that this step uses the collected time-series data of rotor speed, vibration phase, and amplitude as input to the LSTM model. This time-series data contains information about the engine's operating state over time, especially the vibration amplitude, which directly reflects the unbalanced state, while the speed and phase provide context for the vibration characteristics.

[0060] LSTM is a special type of recurrent neural network (RNN) that is particularly well-suited for processing time series data. Through its unique gating mechanism, it can effectively capture long-distance time dependencies, thereby learning the complex nonlinear relationship between vibration magnitude and imbalance.

[0061] During the pre-training phase, the LSTM model is trained using a large amount of labeled historical data. These datasets should contain matching instances of vibration characteristics under different working conditions and known balance adjustments (weighting angle and weighting magnitude), enabling the model to learn how to predict the most suitable balance adjustment parameters from the input vibration characteristics.

[0062] The model outputs intelligent in-machine balancing recommendations for a specific engine condition, specifying the exact weighting angle and magnitude. The weighting angle indicates where the counterweight should be adjusted on the rotor, while the magnitude indicates how much mass needs to be adjusted.

[0063] This embodiment provides an intelligent in-flight balancing method for aero-engines based on LSTM. This method utilizes an LSTM model to process and learn nonlinear relationships, which is crucial for aero-engine balancing. This is because the relationship between vibration magnitude and imbalance is often not a simple linear one, but a complex interaction influenced by multiple factors (such as engine speed, vibration phase, and structural characteristics). Furthermore, this method enables real-time analysis based on time-series data. The LSTM model can provide real-time balancing adjustment suggestions during engine operation, helping to quickly respond to and optimize engine performance, reduce vibration, and improve operational efficiency and safety.

[0064] In some embodiments, the time-series data of the rotor speed and the original vibration phase and amplitude of the aero-engine are extracted from the key phase signal and the vibration acceleration signal. The key phase signal is a speed signal with key phase function obtained by measuring the modified rotor blades with added low-tooth blades using a key phase speed sensor. The position of the low-tooth blades on the modified rotor blades is the reference position for balancing the aero-engine.

[0065] Please see Figure 2 , Figure 2 A testing scheme is presented, which provides the relationship between vibration phase, weighting angle, unbalance angle, and vibration phase lag angle based on a reference position. In this scheme, the rotor blades of the 4th and 5th stages of the compressor are structurally modified to obtain modified blades with low-characteristic tooth signals. A key-phase speed sensor is used to test and obtain the speed signal with key-phase functionality. This low-tooth blade is used as the reference position for vibration phase and weighting angle. The weighting angle θ for adding unbalance is defined by the rotation direction and the position of the key-phase blade. Figure 2 The test scheme shown was used to install and test the sensor, obtaining the bond phase signal and vibration acceleration signal at different rotational speeds. The actual bond phase signal and vibration acceleration signal collected are as follows: Figure 3 and Figure 4 As shown.

[0066] In this embodiment, the low-tooth blades are designed as a signal source, and their position is set as a reference point for machine balancing. These characteristic structures (low teeth) can be identified by external sensors as the rotor rotates, forming specific signal characteristics. The modified rotor blades are tested using a key-phase speed sensor. This type of sensor is specifically designed to capture the passing events of specific markers on rotating components (in this case, the low-tooth blades), thereby generating a speed signal containing "key phase" information. This signal carries information about the rotor rotation phase, which is crucial for subsequent analysis. By analyzing the signals acquired by the sensors, the precise moment when the low-tooth blades pass the sensor can be clearly identified, thus determining their phase position in the rotation cycle. The position of the low-tooth blades is used as a reference point for the vibration phase, i.e., the start or zero point of vibration analysis, and also as an angular reference during weighting (counterweighting) operations. Based on the engine's rotation direction and the determined reference position (low-tooth blades), the exact angle θ of the required unbalance (e.g., through counterweight) can be defined.

[0067] In some embodiments, during the balancing of an aero-engine, it is also necessary to calculate the vibration phase and vibration phase lag angle of the rotor blades based on the key phase signal and vibration acceleration signal. Based on the structural characteristic that the key phase low-tooth signal generates one low tooth per revolution, the vibration phase corresponding to the fundamental frequency of the high-voltage rotor is obtained by performing cross-spectral calculations between the fundamental frequency component of the vibration signal and the key phase signal. The principle is as follows: Figure 5 As shown. The formulas for calculating the vibration phase and the vibration phase lag angle are as follows:

[0068]

[0069] In the formula, This represents the vibration phase corresponding to the fundamental frequency of the high-voltage rotor. This represents the phase corresponding to the fundamental frequency of the bond phase signal after FFT analysis. α represents the phase of the vibration acceleration signal at the fundamental frequency position of the rotational speed after FFT analysis; α is the vibration phase lag angle, and θ is the weighting angle.

[0070] During the balancing process, the vibration characteristics of the rotor blades must first be accurately measured. This includes obtaining the vibration phase of the blades, i.e., the instantaneous position of the blades within their vibration cycle, and the vibration phase lag angle. The vibration phase lag angle refers to the angle of deviation of the actual vibration response from the theoretically ideal vibration position due to various factors (such as rotor dynamics, load conditions, etc.). To accurately measure the vibration characteristics of the high-voltage rotor, cross-spectral analysis is performed using the fundamental frequency component of the vibration signal and the key phase signal. Cross-spectral analysis is a signal processing technique that reveals the phase relationship between two signals. Here, one signal is the fundamental frequency component of the rotor vibration, which reflects the frequency of the rotor's main vibration modes; the other signal is the low-tooth signal captured by the key phase sensor, providing an accurate time reference. Through the above cross-spectral calculation, the phase difference between the fundamental frequency vibration of the high-voltage rotor and the key phase signal, i.e., the vibration phase, can be determined. This result is crucial for understanding the dynamic characteristics of rotor vibration because the vibration phase is directly related to the location and magnitude of the rotor unbalance. Once the vibration phase is determined, the location of the unbalanced mass can be inferred, and it can be calculated where and how weight needs to be added or removed to achieve balance.

[0071] In some embodiments, when obtaining the time series data of the rotor speed and the original vibration phase and amplitude based on the key phase signal and the vibration acceleration signal, it is also necessary to condition the key phase signal, including:

[0072] S21: Normalize the bond phase signal according to the maximum-minimum normalization method.

[0073] It should be noted that the purpose of normalization is to map the amplitude range of the original bond phase signal to a fixed interval, typically [0,1]. This process helps to eliminate the influence of signal amplitude differences on subsequent analysis, ensuring that the model or algorithm has the same sensitivity to all signals.

[0074] The formula for max-min normalization is as follows:

[0075]

[0076] In the formula, x is the input signal, x o This is the output signal.

[0077] By calculating the maximum and minimum values ​​in the signal, a linear transformation is performed on each signal sample to ensure that the maximum value of the processed signal is 1 and the minimum value is 0.

[0078] S22: Find the location of the low-tooth signal from the normalized bond phase signal based on the low-tooth signal threshold.

[0079] It should be noted that the low-tooth signal is a specific, recurring feature in the key phase signal, representing a fixed position during rotor rotation. Identifying these positions provides a precise time reference point for subsequent vibration phase analysis. This step involves setting a threshold, typically based on the signal's statistical characteristics (such as mean plus or minus standard deviation) or a previously known low-tooth signal intensity, then identifying the peak points of the current signal. The threshold is then used to determine whether each peak point corresponds to a low-tooth signal position. These peaks correspond to the instant the low-tooth signal passes through the sensor.

[0080] S23: Set the signal at the low-tooth signal position to a high level of 1, and set the signals at the other positions to 0, thereby obtaining the conditioned bond phase signal.

[0081] It should be noted that the signal value at all detected low-tooth signal locations is set to a high level of 1, indicating that a low-tooth signal exists at these times; the signal value at other times is set to a low level of 0, indicating that there is no low-tooth signal.

[0082] This embodiment converts the original key phase signal into a series of easily analyzable binary pulse signals, each pulse corresponding to the precise moment when a specific low-tooth blade on the rotor passes through a detection point. This conditioned signal not only highlights key phase information but also eliminates unnecessary noise and interference, providing clear and efficient input data for subsequent prediction of the aircraft engine's internal balance empirical angle using deep learning models or other algorithms. The conditioned key phase signal is as follows: Figure 6 As shown.

[0083] In some embodiments, FFT analysis of the bond phase signal and vibration acceleration signal includes:

[0084] S31: Perform FFT on the original key phase signal to calculate the rotational speed corresponding to the current signal.

[0085] It should be noted that this step is for preliminary identification of the rotational speed. The original key phase signal contains periodic information about the rotor's rotation. By performing FFT analysis on it, the frequency components of the signal can be identified, with the most significant frequency typically corresponding to the rotor's rotational frequency (i.e., the fundamental frequency of the rotational speed). This step is mainly used for preliminary verification and calculation of the engine's actual rotational speed.

[0086] S32: Then perform FFT on the conditioned key phase signal to calculate the phase corresponding to the rotational speed fundamental frequency.

[0087] It should be noted that this step precisely calculates the phase of the fundamental speed frequency. After preprocessing the key phase signal (such as normalization and low-tooth signal position identification), the fundamental speed frequency component in the signal becomes clearer. Performing FFT analysis again allows for a more accurate determination of the phase information corresponding to the fundamental speed frequency. Phase information is crucial for subsequent balance analysis because it is directly related to the specific position of the rotor and the location of the imbalance. In practice, an FFT is performed on the key phase signal, and the frequency corresponding to the maximum spectral value within the 0:300Hz range is the fundamental speed frequency of the current signal.

[0088] S33: Perform FFT on the vibration acceleration signal, calculate the phase of each frequency within a set range to the left and right of the rotational speed base frequency, and take the average value of all phases as the phase of the vibration acceleration signal at the rotational speed base frequency position.

[0089] It should be noted that this step analyzes vibration characteristics and calculates the average phase. An FFT is performed on the vibration acceleration signal to identify the vibration frequencies and their distribution related to engine speed. Specifically, within a defined frequency range around the engine's fundamental frequency (e.g., within 10 Hz to the left and right of the fundamental frequency), the phases of these frequency components are calculated, and then the average of all these phases is taken. This average phase value reflects the overall phase characteristics of the vibration acceleration signal near the engine's fundamental frequency, which is crucial for understanding and predicting engine vibration behavior. Calculating the average phase reduces the influence of noise and random fluctuations, providing a more robust vibration phase estimate. The process of signal conditioning and FFT analysis of the key phase signal and vibration acceleration to calculate the vibration phase is as follows: Figure 7 As shown.

[0090] In some embodiments, referring to the previously calculated engine balance results, the relationship between the empirical weighting angle and engine speed for a certain engine model is summarized as follows: Figure 8 As shown, the empirical formula obtained is as follows:

[0091]

[0092] In the formula, S is the rotational speed.

[0093] LSTM (Laser-Sensitive Time Measuring) neural networks are a type of time-recursive neural network, primarily used for predicting long-delayed events in time series. LSTM is a variant of RNN (Recurrent Neural Network), its algorithm adding cells to determine whether information is useful and which is retained. LSTM introduces the concept of a "gate," such as... Figure 9 As shown, the threshold filtering information consists of a Sigmoid neural network layer and a dot product algorithm. The Sigmoid neural network layer is responsible for outputting probability values ​​between 0 and 1. Three types of thresholds are placed in a single cell: the input threshold, the forget threshold, and the output threshold.

[0094] By increasing the input threshold, forget threshold, and output threshold, the weights of the self-loop are variable. Therefore, during the training of the LSTM model, even if the model parameters are fixed, the integral scale of the weights changes dynamically at different times, thus avoiding the gradient vanishing or gradient inflation problems that occur in RNNs.

[0095] The cells in the LSTM model, acting as hidden layer cells, possess long-term memory capabilities and have evolved to become... Figure 10 As shown. Its forward computation method can be expressed as:

[0096] f t =σ(W f [h t-1 ,x t ]+b f (1)

[0097] i t =σ(W i [h t-1 ,x t ]+b i (2)

[0098]

[0099] o t =σ(W o [h t-1 ,x t ]+b o (5)

[0100] h t =o t *tanh(C t (6)

[0101] In the formula: f t Indicates the forgetting threshold; i t Indicates the input threshold; o t Indicates the output threshold; C~C tThe cell state at the previous time step is represented by the candidate vector; C t This represents the current cell state (where the current loop is occurring); W represents the weight coefficient matrix (e.g., W...). h b represents the weight coefficient matrix of the hidden layer; b represents the bias vector (e.g., b0). h (represents the bias vector of the hidden layer); h t-1 h is the output of the unit at the previous time step. t σ represents the output of the current cell; σ is the Sigmoid function; tanh is the hyperbolic tangent activation function; the subscript t indicates time.

[0102] Information that meets the algorithm's authentication criteria is retained; otherwise, it is forgotten by the forget gate. This one-in-two-out working principle can improve the training precision and accuracy of the algorithm through repeated calculations.

[0103] The training of LSTM neural networks uses the BP algorithm, which is the backpropagation algorithm for error. Based on the BP algorithm, the error is propagated back along the time series, and then the parameters are adjusted using a gradient descent strategy along the negative gradient direction of the target. The network itself can be represented by a graph structure with loops, such as... Figure 11 As shown.

[0104] The BPTT (Back-propagation Through Time) algorithm is employed. The algorithm consists of four steps: First, the cell output values ​​of the LSTM1 are obtained based on the previous method; second, the error of each cell is calculated using the BP algorithm; third, the gradient value of the weight corresponding to each error term is calculated; finally, the weights are updated based on the gradient optimization algorithm, such as... Figure 12 As shown. The main training and prediction process of the model is as follows:

[0105] 1) Using the prepared training sample set, the obtained N2 rotational speed (i.e., high-pressure rotor speed), original vibration phase and amplitude are used as inputs to the LSTM method, and the balanced engine balance angle is used as the output to complete the LSTM model parameter training and obtain the intelligent dynamic balance model.

[0106] 2) In actual prediction, the original vibration and fiber optic signals are collected, and the current rotational speed, vibration phase and amplitude are obtained from the collected signals. The obtained input data is input into the trained intelligent dynamic balancing model, and the prediction results are output, namely the angle position and the magnitude of the increase that need to be applied for this dynamic balancing.

[0107] The proposed local intelligent fusion balancing method was validated. The validation results are as follows: Figure 13 As shown in the results, the prediction accuracy after fusion is better than that of the previous single method. This is mainly reflected in:

[0108] Using data from the first 21 engines for training, the prediction of the last engine's data was more stable, with a prediction of 188.6 degrees.

[0109] Prediction was performed using 30%, 20%, 10%, and 5% of randomly selected data, with the remaining data serving as training data. The mean absolute errors were 6.46, 6.17, 6.8, and 4.81 degrees, respectively. Compared to single-method approaches, the accuracy was improved in all cases.

[0110] Based on the same inventive concept, this application also provides an LSTM-based intelligent on-board balancing device for implementing the aforementioned LSTM-based intelligent on-board balancing method for aero-engines. The solution provided by this system is similar to the implementation described in the above method. Therefore, the specific limitations in the embodiments of the LSTM-based intelligent on-board balancing device for aero-engines provided below can be found in the limitations of the LSTM-based intelligent on-board balancing method for aero-engines described above, and will not be repeated here.

[0111] This embodiment provides an LSTM-based intelligent in-situ balancing device for aero-engines, comprising:

[0112] The data acquisition module is used to acquire time-series data of the rotor speed, raw vibration phase, and amplitude of the aero-engine.

[0113] The balancing module is used to input the time series data of rotor speed and original vibration phase and amplitude into a pre-trained LSTM prediction model to obtain the weighting angle and weighting magnitude for intelligent in-flight balancing of aero-engines.

[0114] The LSTM prediction model is trained based on the LSTM algorithm. It takes time series data of rotor speed and original vibration phase and amplitude as input, and outputs the weighting angle and weighting magnitude of the intelligent in-flight balancing of the aero-engine.

[0115] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0116] Reference Figure 14 The present invention also provides a computer device, including: a memory and a processor, and a computer program stored in the memory, wherein when the computer program is executed on the processor, it implements the LSTM-based intelligent in-situ balancing method for aero-engines as described in any of the above methods.

[0117] The computer device may be a desktop computer, laptop, handheld computer, or cloud server, etc. This computer device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that... Figure 14 The examples of computer devices are merely examples and do not constitute a limitation on computer devices. They may include more or fewer components than shown in the illustration, or combinations of certain components, or different components. For example, they may also include input / output devices, network access devices, etc.

[0118] The processor referred to can be a Central Processing Unit (CPU), but it can also be 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. A general-purpose processor can be a microprocessor or any conventional processor.

[0119] In some embodiments, the memory may be an internal storage unit of the computer device, such as a hard drive or RAM. In other embodiments, the memory may be an external storage device of the computer device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory may include both internal and external storage units of the computer device. The memory is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory can also be used to temporarily store data that has been output or will be output.

[0120] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the LSTM-based intelligent in-situ balancing method for aero-engines as described in any of the above methods.

[0121] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0122] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0123] Those skilled in the art will recognize that the units 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 design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0124] In the embodiments disclosed in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0125] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart in-situ balancing method for aero-engines based on LSTM, characterized in that, Includes the following steps: Acquire time-series data of the rotor speed, raw vibration phase, and amplitude of the aero-engine; The time series data of the rotor speed and the original vibration phase and amplitude are input into the pre-trained LSTM prediction model to obtain the weighting angle and weighting magnitude for intelligent in-flight balancing of aero-engines. The LSTM prediction model is trained based on the LSTM algorithm, taking the time series data of rotor speed and original vibration phase and amplitude as input, and the weighting angle and weighting magnitude of the intelligent in-flight balance of the aero-engine as output. The time-series data of rotor speed and original vibration phase and amplitude are obtained based on key phase signal and vibration acceleration signal; When obtaining the time series data of the rotor speed and the original vibration phase and amplitude based on the key phase signal and the vibration acceleration signal, the key phase signal is also conditioned, including: The bond phase signal is normalized according to the maximum-minimum normalization method; The location of the low-tooth signal is found from the normalized bond phase signal based on the low-tooth signal threshold. The signal at the found low-tooth signal position is set to a high level of 1, and the signals at other positions are set to 0, thereby obtaining the conditioned bond phase signal; The network architecture of the LSTM prediction model includes: an input layer, a hidden layer, and an output layer; The input layer is used to receive the rotor speed of the aero-engine and the time series data of the original vibration phase and amplitude; The hidden layer is used to mine the nonlinear relationship between vibration and imbalance in the time series data through the LSTM structure; The output layer is used to output the information mined by the hidden layer for iterative prediction, and output the predicted weighting angle and weighting magnitude.

2. The intelligent in-situ balancing method for aero-engines based on LSTM according to claim 1, characterized in that, The key phase signal is a speed signal with key phase function obtained by measuring the modified rotor blade with added low-tooth blade using a key phase speed sensor. The position of the low-tooth blade on the modified rotor blade is the reference position for balancing the aero engine.

3. The intelligent in-situ balancing method for aero-engines based on LSTM according to claim 1, characterized in that, The formula for calculating the original vibration phase is as follows: ； In the formula, This represents the vibration phase corresponding to the fundamental frequency of the high-voltage rotor. The phase corresponding to the fundamental frequency of the bond phase signal after FFT analysis is the rotational speed. The phase of the vibration acceleration signal at the fundamental frequency position of the rotational speed, as determined by FFT analysis.

4. The LSTM-based intelligent in-situ balancing method for aero-engines according to claim 3, characterized in that, The FFT analysis of the bond phase signal and the vibration acceleration signal includes: Perform an FFT on the original key phase signal to calculate the rotational speed corresponding to the current signal, and obtain the rotor speed; Then, an FFT is performed on the conditioned key phase signal to calculate the phase corresponding to the rotational speed fundamental frequency; Perform an FFT on the vibration acceleration signal to calculate the phase of each frequency within a set range to the left and right of the rotational speed base frequency, and take the average value of all phases as the phase of the vibration acceleration signal at the rotational speed base frequency position.

5. The intelligent in-situ balancing method for aero-engines based on LSTM according to claim 1, characterized in that, The LSTM prediction model is trained using the backpropagation method.

6. A smart in-situ balancing device for aero-engines based on LSTM, characterized in that, include: The data acquisition module is used to acquire time-series data of the rotor speed, raw vibration phase, and amplitude of the aero-engine. The balancing module is used to input the time series data of the rotor speed and the original vibration phase and amplitude into the pre-trained LSTM prediction model to obtain the weighting angle and weighting magnitude for intelligent in-flight balancing of the aero-engine. The LSTM prediction model is trained based on the LSTM algorithm, taking the time series data of rotor speed and original vibration phase and amplitude as input, and the weighting angle and weighting magnitude of the intelligent in-flight balance of the aero-engine as output. The time-series data of rotor speed and original vibration phase and amplitude are obtained based on key phase signal and vibration acceleration signal; When obtaining the time series data of the rotor speed and the original vibration phase and amplitude based on the key phase signal and the vibration acceleration signal, the key phase signal is also conditioned, including: The bond phase signal is normalized according to the maximum-minimum normalization method; The location of the low-tooth signal is found from the normalized bond phase signal based on the low-tooth signal threshold. The signal at the found low-tooth signal position is set to a high level of 1, and the signals at other positions are set to 0, thereby obtaining the conditioned bond phase signal; The network architecture of the LSTM prediction model includes: an input layer, a hidden layer, and an output layer; The input layer is used to receive the rotor speed of the aero-engine and the time series data of the original vibration phase and amplitude; The hidden layer is used to mine the nonlinear relationship between vibration and imbalance in the time series data through the LSTM structure; The output layer is used to output the information mined by the hidden layer for iterative prediction, and output the predicted weighting angle and weighting magnitude.

7. A computer device, characterized in that, The device includes a processor and a memory: The memory is used to store computer programs and send the instructions of the computer programs to the processor; The processor executes, according to the instructions of the computer program, an LSTM-based intelligent in-situ balancing method for aero-engines as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements an LSTM-based intelligent in-situ balancing method for aero-engines as described in any one of claims 1-5.

Images