Aero-engine intermediate bearing fault diagnosis method independent of casing measuring point position

Abstract

The invention belongs to the technical field of aero-engine fault diagnosis, and particularly relates to an aero-engine intermediate bearing fault diagnosis method independent of a cartridge receiver measuring point position, which comprises the following steps of: establishing an intermediate bearing fault diagnosis data set; preprocessing the data set; extracting features; model training; and fault diagnosis. According to the method, whole machine measuring point modeling is completed in the early model training stage, fault diagnosis of different fault types of the intermediate bearing of the aero-engine can be completed by inputting measuring point vibration data on any single case in subsequent fault diagnosis application, and the problem that in an existing method, dependence on the fixing position of a sensor is caused is solved; and the flexibility and robustness in practical engineering application are enhanced.

Description

Technical Field

[0001] The invention belongs to the technical field of aircraft engine fault diagnosis, and in particular relates to an aircraft engine intermediate bearing fault diagnosis method that is independent of casing measuring point positions, thereby realizing aircraft engine intermediate bearing fault diagnosis that is agnostic to casing measuring points. Background Art

[0002] Aircraft engines are core components of modern aircraft, and the reliability and safety of their operation are directly related to flight safety. In aircraft engines, the intermediate bearing is one of the key mechanical components, located between the high-pressure rotor and the low-pressure rotor, and plays a role in supporting and transmitting power. The operating state of the intermediate bearing has a significant impact on the dynamic characteristics of the rotor and the overall performance of the aircraft engine. However, due to its complex operating environment, the bearing is subjected to high temperature, high speed and high load conditions, and is prone to various faults such as wear and cracks on the inner ring, outer ring and rolling elements. If these faults cannot be detected and diagnosed in a timely and accurate manner, they may lead to engine failure and bring serious consequences.

[0003] In order to monitor the health status of the intermediate bearing, multiple sensors are usually installed on the outside of the engine casing to collect vibration signals. However, due to the compact space design of aircraft engines, the available installation positions on the casing surface are limited. Some areas may not be able to install sensors due to interference from other components, or the mechanical vibration characteristics of aircraft engines under different working conditions (such as cruising, takeoff, landing, etc.) may change significantly. Some positions are suitable as signal collection points under certain working conditions, but may perform poorly under other working conditions. It is necessary to flexibly adjust the sensor position. After the aircraft engine has been running for a long time, the casing may be mechanically deformed due to temperature and vibration, and some positions that were originally suitable for installing sensors may no longer be stable. The fixed position of the sensor limits the flexibility of traditional fault diagnosis methods. Once the sensor position changes, the diagnostic model will completely fail. Therefore, it is of great engineering significance to study a method for fault diagnosis of intermediate bearings of aircraft engines that does not rely on the position of the casing measurement points.

[0004] In recent years, fault diagnosis methods based on deep learning have become a research hotspot. Compared with traditional physical model-driven methods, deep learning methods can automatically extract fault features from vibration signals of complex aviation engine casing measurement points, and have the characteristics of high precision and high robustness. For example, convolutional neural networks (CNNs) are widely used in feature extraction of vibration signals. Through local perception and weight sharing mechanisms, they can capture local fault features in vibration signals. However, due to the above-mentioned limitations on the installation position of external sensors in aircraft engine casings, existing methods perform poorly when directly applied to aircraft engine fault diagnosis. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides an aircraft engine intermediate bearing fault diagnosis method that is independent of the position of the casing measuring points. The modeling of the whole machine measuring points is completed in the early model training stage. In the subsequent fault diagnosis application, only the vibration data of any single measuring point on the casing is required to complete the fault diagnosis of different fault types of the aircraft engine intermediate bearing, thereby solving the dependence on the fixed position of the sensor and the limitation of the measuring point position change in the existing method, and enhancing the flexibility and robustness in actual fault diagnosis.

[0006] The present invention is implemented by providing an aircraft engine intermediate bearing fault diagnosis method that does not rely on the position of the casing measurement point, comprising the following steps: Step 1: Establish an intermediate bearing fault diagnosis dataset: determine the number and location of casing measurement points, use sensors to collect the original vibration signals of the experimental test of the corresponding number and location measurement points of the real aircraft engine casing, classify the original vibration signals into different fault types, and slice and group the original vibration signals to obtain the intermediate bearing fault diagnosis dataset; Step 2: Dataset preprocessing: The intermediate bearing fault diagnosis dataset obtained in step 1 is analyzed by extracting the signal amplitude envelope, and the vibration signal in the intermediate bearing fault diagnosis dataset is normalized; Step 3: Feature extraction: A three-layer one-dimensional convolutional neural network is used to extract the vibration signal features of the intermediate bearing fault diagnosis dataset normalized in step 2. Batch normalization and activation function ReLU are connected after each convolution layer, and the maximum pooling operation is used to reduce the feature length of the vibration signal; Step 4: Model training: Randomly select the vibration signal features collected by the position sensor of the casing measurement point as input, build an input scenario with unknown position, use the Adam optimizer to train the model, and obtain the optimal diagnostic model; Step 5: Fault diagnosis: Input the actual vibration data of any single measuring point on the casing into the optimal diagnosis model to diagnose different fault types of the aircraft engine intermediate bearing.

[0007] Preferably, in step 1, K casing measuring points are provided, which are distributed at K different positions of the casing, and the collected original vibration signal is expressed as:

[0008] in, Indicates K The vibration signal collected by the sensor, R represents the set of real numbers, L is the signal length.

[0009] Preferably, in step 1, the fault types include normal state, outer ring fault, inner ring type 1 fault, and inner ring type 2 fault.

[0010] Further preferably, in step 1, data slicing is specifically: the original vibration signal According to the signal length L Slice to generate multiple fixed-length samples that meet the computational requirements of a three-layer one-dimensional convolutional neural network, expressed as:

[0011] in, Indicates i samples, t A time variable representing a vibration signal.

[0012] Preferably, in step 1, data grouping specifically includes: dividing the original vibration signal data after data slicing into a training set, a validation set and a test set according to a ratio.

[0013] Further preferably, in step 2, the high-frequency carrier signal of each vibration signal in the intermediate bearing fault diagnosis data set is extracted, and noise and irrelevant components are suppressed at the same time, and the pass range of the bandpass filter is selected to be a frequency band containing the characteristic frequency and harmonic components of the bearing fault type, and the filtered vibration signal is expressed as:

[0014] in, is a low frequency modulation signal, It is a high frequency carrier signal; The filtered vibration signal The signal amplitude envelope is extracted by constructing the analytical signal of the vibration signal to extract the signal amplitude envelope using the Hilbert transform method.

[0015] Further preferably, in step 2, each vibration signal after extracting the signal amplitude envelope is normalized to eliminate the influence of amplitude difference on model training, and the normalization method is:

[0016] in m is the mean value of the vibration signal: , s is the signal standard deviation: .

[0017] Further preferably, in step 4, the Adam optimizer is used to perform model training and update model parameters:

[0018] in, is the learning rate, are model parameters, is the number of iterations; The initial learning rate is 0.001 and the momentum parameter is , The number of training iterations is 50 epochs, and the learning rate is halved every 10 epochs. During the training process, the training set is used to complete the modeling of the entire casing measurement points, and the validation set is used to select the optimal diagnostic model.

[0019] Compared with the prior art, the advantages of the present invention are: The problem of existing methods relying on fixed positions of sensors is solved, and problems such as limited sensor installation positions, adjustment of installation positions due to actual needs, and unstable positions of originally installed sensors are addressed. The flexibility and robustness of fault diagnosis are improved. In subsequent fault diagnosis applications, the method of the present invention only requires vibration data of any single measuring point on the casing to complete fault diagnosis of different types of intermediate bearing faults of aircraft engines, thereby reducing resource consumption for simultaneous acquisition of multiple sensors and providing the possibility of redundant sensor design. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram of sensor installation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to more clearly illustrate the technical solution of the specific implementation of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It is obvious that the described embodiment is one parameter setting of the present invention, and all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] This embodiment provides an aircraft engine intermediate bearing fault diagnosis method that does not rely on the position of the casing measurement point, and the specific steps are as follows: Step 1: Establish the intermediate bearing fault diagnosis data set, such as Figure 1 As shown, in this embodiment, a total of 6 sensor collection positions are arranged on the outer casing of the aircraft engine, namely the first sensor 1, the second sensor 2, the third sensor 3, the fourth sensor 4, the fifth sensor 5 and the sixth sensor 6. The length of the original signal recorded by each sensor is 20480 points, the sampling frequency is 25kHz, the sample collection time is 15 seconds, and the data are classified into four fault types: normal state, outer ring fault, inner ring type 1 fault, and inner ring type 2 fault.

[0023] Data slicing: Each original vibration signal is divided into 10 slices, and the number of data samples at each sensor position is increased to 10 times the original number of samples. The length of each slice is 2048 points, generating smaller samples to adapt to the computing needs of the neural network; Data grouping: The vibration signals are divided into training set (70%), validation set (10%) and test set (20%) according to the proportion to ensure the generalization ability of the model. The training set is used for model optimization in the training phase, the best model parameters are selected on the validation set, and finally the model performance is evaluated on the test set; Step 2: Dataset preprocessing: Extract the high-frequency carrier signal of each vibration signal in the intermediate bearing fault diagnosis data set, suppress noise and irrelevant components, and select the pass range of the bandpass filter to be the frequency band containing the characteristic frequency and harmonic components of the bearing fault type. The filtered vibration signal is expressed as:

[0024] in, is a low frequency modulation signal, is a high frequency carrier signal, t is the time variable of the signal; The filtered vibration signal The signal amplitude envelope is extracted by constructing the analytical signal of the vibration signal to extract the signal amplitude envelope using the Hilbert transform method.

[0025] After extracting the signal amplitude envelope, each vibration signal is normalized to eliminate the influence of amplitude difference on model training. The normalization method is:

[0026] in x i Indicates i Samples ,m is the mean value of the vibration signal: , L is the signal length, s is the signal standard deviation: .

[0027] Step 3: Feature extraction: A three-layer one-dimensional convolutional neural network (1D-CNN) is used to extract signal features: First convolution layer: kernel size: 5, step size: 1, number of output channels: 16, activation function: ReLU, subsequent operations: batch normalization and max pooling, pooling window size of 4, step size of 4. Second convolution layer: kernel size: 5, step size: 1, number of output channels: 32, activation function: ReLU, subsequent operations: batch normalization and max pooling, pooling window size of 4, step size of 4. Third convolution layer: kernel size: 5 step size: 1, number of output channels: 64, activation function: ReLU, subsequent operations: batch normalization and max pooling, pooling window size of 4, step size of 4.

[0028] Step 4: Model training: Data randomization arbitrary input means that in each training process, a signal is randomly selected from the 6 sensor positions as input, with a shape of 1×2048, to construct an input scenario where the position is agnostic. Through random selection, the model is forced to learn the difference characteristics of signals at different positions, thereby enhancing its generalization ability; By randomly selecting the signal at the sensor position as input, the Adam optimizer is used to train the model and update the model parameters:

[0029] in, is the learning rate, are model parameters, is the number of iterations; The initial learning rate is 0.001 and the momentum parameter is , The number of training iterations is 50 epochs, and the learning rate is halved every 10 epochs. During the training process, the training set is used to complete the modeling of the entire casing measurement points, and the validation set is used to select the optimal diagnostic model.

[0030] Step 5: Fault diagnosis: Input the vibration data of any single measuring point on the casing (without providing the sensor position) to complete the fault diagnosis of different fault types of the intermediate bearing of the aircraft engine. Table 1 provides a comparison of the test accuracy of the measuring point position of any casing of different measuring point models. It can be seen that the test accuracy of the measuring point position of the model provided by the present invention is much higher than that of other models: Table 1

[0031] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. An aircraft engine intermediate bearing fault diagnosis method that does not rely on the position of the casing measurement point, characterized in that: The steps include: Step 1: Establish an intermediate bearing fault diagnosis dataset: determine the number and location of casing measurement points, use sensors to collect the original vibration signals of the experimental test of the corresponding number and location measurement points of the real aircraft engine casing, classify the original vibration signals into different fault types, and slice and group the original vibration signals to obtain the intermediate bearing fault diagnosis dataset; Step 2: Dataset preprocessing: The intermediate bearing fault diagnosis dataset obtained in step 1 is analyzed by extracting the signal amplitude envelope, and the vibration signal in the intermediate bearing fault diagnosis dataset is normalized; Step 3: Feature extraction: A three-layer one-dimensional convolutional neural network is used to extract the vibration signal features of the intermediate bearing fault diagnosis dataset normalized in step 2. Batch normalization and activation function ReLU are connected after each convolution layer, and the maximum pooling operation is used to reduce the feature length of the vibration signal; Step 4: Model training: Randomly select the vibration signal features collected by the position sensor of the casing measurement point as input, build an input scenario with unknown position, use the Adam optimizer to train the model, and obtain the optimal diagnostic model; Step 5: Fault diagnosis: Input the actual vibration data of any single measuring point on the casing into the optimal diagnosis model to diagnose different fault types of the aircraft engine intermediate bearing.

2. The method for diagnosing the intermediate bearing fault of an aircraft engine that is independent of the position of the casing measuring point according to claim 1 is characterized in that: In step 1, K casing measuring points are provided, which are distributed at K different positions of the casing, and the collected original vibration signal is expressed as: ; in, Indicates K The vibration signal collected by the sensor, R represents the set of real numbers, L is the signal length.

3. The method for diagnosing the intermediate bearing fault of an aircraft engine that is independent of the position of the casing measuring point according to claim 1 is characterized in that: In step 1, the fault types include normal state, outer ring fault, inner ring type 1 fault, and inner ring type 2 fault.

4. The method for diagnosing the intermediate bearing fault of an aircraft engine that is independent of the position of the casing measuring point according to claim 2 is characterized in that: In step 1, data slicing is specifically: the original vibration signal According to the signal length L Slice to generate multiple fixed-length samples that meet the computational requirements of a three-layer one-dimensional convolutional neural network, expressed as: {x}_{i}=\left [ {x\left ( {t} \right ), x\left ( {t+1} \right )\cdot \cdot \cdot x\left ( {t+L-1} \right )} \right ] in, Indicates i samples, t A time variable representing a vibration signal.

5. The method for diagnosing the intermediate bearing fault of an aircraft engine that is independent of the position of the casing measuring point according to claim 1 is characterized in that: In the step 1, data grouping specifically includes: dividing the original vibration signal data after data slicing into a training set, a validation set and a test set according to a ratio.

6. The method for diagnosing the intermediate bearing fault of an aircraft engine that is independent of the position of the casing measuring point according to claim 4 is characterized in that: In step 2, the high-frequency carrier signal of each vibration signal in the intermediate bearing fault diagnosis data set is extracted, and noise and irrelevant components are suppressed at the same time. The pass range of the bandpass filter is selected to be a frequency band containing the characteristic frequency and harmonic components of the bearing fault type. The filtered vibration signal is expressed as: ; in, is a low frequency modulation signal, It is a high frequency carrier signal; The filtered vibration signal The signal amplitude envelope is extracted by constructing the analytical signal of the vibration signal to extract the signal amplitude envelope using the Hilbert transform method.

7. The method for diagnosing the intermediate bearing fault of an aircraft engine that is independent of the position of the casing measuring point according to claim 6 is characterized in that: In step 2, each vibration signal after extracting the signal amplitude envelope is normalized to eliminate the influence of amplitude difference on model training. The normalization method is: in μ is the mean value of the vibration signal: , σ is the signal standard deviation: .

8. The method for diagnosing the fault of the intermediate bearing of an aircraft engine independent of the position of the casing measuring point according to claim 2 is characterized in that: In step 4, the Adam optimizer is used to train the model and update the model parameters: in, is the learning rate, are model parameters, is the number of iterations; The initial learning rate is 0.001 and the momentum parameter is , The number of training iterations is 50 epochs, and the learning rate is halved every 10 epochs. During the training process, the training set is used to complete the modeling of the entire casing measurement points, and the validation set is used to select the optimal diagnostic model.

Images