Sample set construction method for trend prediction of vibration monitoring data of rotating part of transmission system

Abstract

The invention discloses a sample set construction method for trend prediction of vibration monitoring data of a transmission system rotating part, and belongs to the field of health management of helicopter transmission system rotating parts. Firstly, historical data of helicopter engine power parameters are collected, the starting moment and the ending moment of each task are determined, and task segmentation is carried out. Then constructing a basic data set of each task of the rotating parts through the calculated comprehensive load coefficient of each task and the vibration monitoring data characteristics of each rotating part of the transmission system in each task segment; and constructing a rotating part trend prediction sample set based on the basic data set by taking a task as a step length. And finally, training the trend prediction model through the trend prediction sample set, and using the trend prediction model to predict a new task end moment feature value. According to the method, the data sample is constructed by taking the task as the step length, so that the trained vibration characteristic trend prediction intelligent model can better predict the degradation law of the helicopter transmission system.

Description

Technical Field

[0001] This invention belongs to the field of health management of rotating components in helicopter transmission systems, and relates to a method for constructing a sample set for predicting the trend of vibration monitoring data of rotating components in transmission systems. Background Technology

[0002] The helicopter transmission system is a collective term for the connecting device between the power system and the rotor system, including various rotating components such as gears, bearings, and drive shafts. Numerous vibration sensors are installed within the transmission system to acquire vibration monitoring data in real time for health monitoring. To ensure flight mission safety, trend prediction intelligent models are applied to predict changes in vibration data, confirming that the rotating components in the transmission system remain in good condition during missions.

[0003] Establishing a trend prediction intelligent model first requires converting vibration monitoring data into sample data for model training and testing. In existing technologies, the sample data for prediction models is often derived from direct slicing of helicopter vibration monitoring data, with intervals of individual sampling points. The data at each sampling point within a slice is used to predict the data for the next one or several sampling points. However, helicopter missions are characterized by varying operating conditions and durations, and the rotating components of the transmission system have long lifespans and slow degradation, making it difficult to directly apply vibration monitoring data for real-time trend prediction. Therefore, periodic trend prediction using the mission as the step size has greater applicability. Summary of the Invention

[0004] The purpose of this invention is to construct a sample set using vibration monitoring data of rotating components in a helicopter transmission system, providing a foundation for training an intelligent model to predict vibration trends. Under the background of periodic trend prediction with mission as the step size, the invention divides missions into stages by generating parameters, calculates the comprehensive load coefficient of the mission, extracts features from the vibration data, constructs the sample set, and enables the training and testing of the trend prediction model.

[0005] A method for constructing a sample set for trend prediction of vibration monitoring data of rotating components in a transmission system includes the following steps:

[0006] Step 1: Collect historical data on helicopter engine power parameters, determine the start and end times of each mission, and segment the mission.

[0007] First, historical data on engine power parameters are collected at fixed time intervals over a period of time, and represented as follows:

[0008] (1)

[0009] Representative power parameter The The sampled values ​​at each time point.

[0010] Then, based on the engine's rated power Use 1% as a threshold value to determine the start and end times of the task in the current time period;

[0011] The moment when the collected engine power parameter first reaches this threshold value within the current time period shall be designated as the start time of the mission. The task ends when the collected engine power parameter first falls below this threshold. The discriminant formulas are as follows:

[0012] (2)

[0013] (3)

[0014] Finally, extend the sampling time and repeat the above process to find the start time of all tasks. and the end time of the mission The start times of tasks that are adjacent in time and the end time of the mission Divide the task into one session and record the start and end times of all tasks obtained.

[0015] Step 2: Calculate the overall load factor for each task;

[0016] First, based on the start and end times of each task, the engine power parameter data for each time segment within that task is extracted, as shown below:

[0017] (4)

[0018] In the formula: Representing the Engine power parameter data for each task segment. Representing the The start time of each task segment. Representing the The end time of each task segment.

[0019] Then, the engine power parameters at all times within each task segment are summed to obtain the comprehensive load coefficient for each task.

[0020] The formula for calculating the overall load factor for each task is as follows:

[0021] (5)

[0022] In the formula, Representing the The overall load factor for each task Representing the The first task Each sampling time.

[0023] Step 3: Establish vibration monitoring data characteristics for each rotating component of the transmission system in each task segment;

[0024] Step 3.1: Based on the installation positions of the rotating components of the transmission system and the vibration sensors, select the vibration monitoring data corresponding to the rotating components from the vibration monitoring data.

[0025] For bearings, the relevant vibration monitoring data includes vibration data at shaft frequency, second harmonic, and third harmonic; for gears, the relevant vibration monitoring data includes vibration data at shaft frequency and meshing frequency; for drive shafts, the relevant vibration monitoring data includes vibration data at shaft frequency and second harmonic.

[0026] The general formula for expressing vibration data is:

[0027] (6)

[0028] In the formula, The first part representing the rotating component The first task Vibration monitoring dataset, Representing the Vibration data at the start of each task. Representing the Vibration data at the end of each task.

[0029] Step 3.2: For each type of vibration data of each rotating component, extract the vibration monitoring data of the same duration after the start time of the task and before the end time of the task, calculate the characteristic value at the start time of the task and the characteristic value at the end time of the task to form the vibration monitoring data characteristics.

[0030] For each type of vibration data, a data segment of one minute after the start of the task is extracted, and the mean of the data is calculated as the characteristic value of the task start time for that type of vibration data. The calculation method is as follows:

[0031] (7)

[0032] In the formula, Representing the The first task The characteristic value of the vibration data at the start time of the task. represent Vibration data at any given time This represents the number of sampling points involved in the calculation.

[0033] For each type of vibration data, extract the data segment one minute before the task's end time, calculate the mean of the data, and use it as the characteristic value of the task's end time for that type of vibration data. The calculation method is as follows:

[0034] (8)

[0035] In the formula, Representing the The first task The characteristic value of the task completion time of the vibration data.

[0036] Step 4: Construct the basic dataset for each task of the rotating component based on the comprehensive load coefficient of each task and the characteristic values ​​of the vibration monitoring data of each rotating component for each task;

[0037] Step 5: Using the task as the step size, construct a trend prediction sample set for rotating components based on the basic dataset;

[0038] A single sample for trend prediction includes feature data used as model input. and the label data output by the model Feature data and tag data The determination method is as follows:

[0039] (1) Set the historical step size as Select task numbers i ≥ in the basic dataset The task is to construct individual samples sequentially;

[0040] (2) The construction process of the first trend prediction sample is as follows:

[0041] The feature values ​​of the task completion time for each vibration data point corresponding to task number i=K are used as label data. The vector.

[0042] Set task number i=K and the preceding The characteristic values ​​of the task start time and the comprehensive load coefficient of each vibration data point corresponding to each task number are used as characteristic data. The vector.

[0043] (3) Repeat the above (2) sample construction process in the basic dataset corresponding to the task number within i>K and step size K in sequence to obtain all the samples and number them to form a rotating component trend prediction sample set.

[0044] Step 6: Construct an intelligent trend prediction model and train the model using a trend prediction sample set to predict the feature values ​​at the end of a new task.

[0045] A backpropagation (BP) neural network model is constructed as an intelligent trend prediction model. The constructed trend prediction sample set is randomly divided into a training set and a test set. The model is trained using the training set data, and optimized using the test set data.

[0046] The trained trend prediction model is used to predict the feature values ​​of the vibration data of each rotating component at the end of the task for the new task. The specific process is as follows:

[0047] First, based on the new mission scenario, evaluate the comprehensive load factor of the new mission;

[0048] By predicting the flight path of the new mission, including flight time and flight profile data, the overall payload factor of the new mission is estimated.

[0049] Then, using the recent The input features are constructed from the characteristic values ​​of the start time of each vibration data point for the sub-task, the characteristic values ​​of the end time of each vibration data point for the last task, and the comprehensive load coefficient of the new task. Substitute these parameters into the trend prediction model to obtain the output label parameters. This serves as the prediction result of the characteristic values ​​of the vibration data at the end of the new task.

[0050] Step 7: Based on the prediction results output by the intelligent trend prediction model, the degradation law of the helicopter transmission system is obtained.

[0051] The advantages and positive effects of this invention are as follows:

[0052] (1) The sample set constructed in this invention takes the task as the step size, which is more suitable for predicting the vibration characteristics of helicopters. The constructed sample set is adapted to the characteristics of long life and slow degradation of rotating parts of helicopter transmission system. The intelligent model for predicting vibration characteristics trained by this sample set can be used to ensure the safety of subsequent flight missions.

[0053] (2) The sample set construction method proposed in this invention calculates the comprehensive load coefficient parameter of each task, which can characterize the load borne by the helicopter transmission system in different flight profiles, so that the trained vibration characteristic trend prediction intelligent model can better predict the degradation law of the helicopter transmission system. Attached Figure Description

[0054] Figure 1 This is a flowchart of the sample set construction method for predicting the trend of vibration monitoring data of rotating components in helicopter transmission systems, based on the present invention. Detailed Implementation

[0055] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0056] A method for constructing a sample set for trend prediction of vibration monitoring data of rotating components in helicopter transmission systems, such as... Figure 1 As shown, the specific steps are as follows:

[0057] Step 1: Use the collected historical data of helicopter engine power parameters to divide tasks;

[0058] Step 1.1: Determine the mission start time based on engine power parameters.

[0059] The historical data of helicopter engine power parameters in this embodiment are shown in Table 1:

[0060] Table 1 Historical Data of Engine Power Parameters

[0061]

[0062] In the table, "Time" represents the sampling time of the data point, in the format of year, month, day, hour, minute, and second. 19:01:01:08:00:00 represents 08:00:00 on January 1, 2019. "Value" represents the value of the collected engine power parameter, in kW.

[0063] Engine rated power It is 1000KW. (Based on the engine's rated power) 1%, or 10KW, is used as a threshold value. The moment when the collected engine power parameter first reaches this threshold value is taken as the start time of the mission. .

[0064] (9)

[0065] By substituting the data into the table, we can obtain the task start time. It is 19:01:01:08:00:13.

[0066] Step 1.2: Determine the mission end time based on engine power parameters.

[0067] Based on engine rated power 1%, or 10KW, is used as a threshold value. The moment when the collected engine power parameter first falls below this threshold value is taken as the end time of the task. .

[0068] (10)

[0069] By substituting the data into the table, we can obtain the task's end time. It is 19:01:01:08:39:56.

[0070] Step 1.3: Construct task segments based on the task start and end times.

[0071] Increase the time value and repeat steps 1.1 and 1.2 to find the start time of all tasks. and the end time of the mission A task is defined as a set of tasks whose start and end times are adjacent in time.

[0072] The start and end times of all tasks were recorded, as shown in Table 2.

[0073] Table 2 Task Segment Start and End Time Schedule

[0074]

[0075] Step 2: Calculate the overall load factor for each task.

[0076] Step 2.1: Obtain engine power parameter data for a single task.

[0077] Based on the start and end times of each task, the engine power parameter data for each task is extracted. The extracted engine power parameter data for the first task is shown in Table 3.

[0078] Table 3 Engine Power Parameter Data Table

[0079]

[0080] In the table, "Time" represents the sampling time of the data point, in the format of year, month, day, hour, minute, and second. 19:01:01:08:00:13 represents 08:00:13 on January 1, 2019. "Value" represents the value of the collected engine power parameter, in kW.

[0081] Step 2.2: Sum the engine power parameters at all times within each task segment to obtain the overall task load factor.

[0082] The extracted engine power parameter data for each task are summed to obtain the overall load factor for that task. The overall load factor for the first task is... The calculation results are as follows:

[0083] (11)

[0084] In the formula, The start time of the first task is 19:01:01:08:00:13; This represents the end time of the first task, 19:01:01:08:39:56.

[0085] Step 2.3: Record the overall task load factor.

[0086] For each task, steps 2.1 and 2.2 are used to obtain the overall task load coefficient.

[0087] Record the comprehensive load coefficients for each task, as shown in Table 4.

[0088] Table 4. Overall Task Load Factor Table

[0089]

[0090] Step 3: Establish vibration monitoring data characteristics for each task of the rotating component.

[0091] Vibration monitoring data for rotating components includes: for bearings, relevant vibration monitoring data includes shaft frequency, second harmonic, and third harmonic vibration data; for gears, relevant vibration monitoring data includes shaft frequency and meshing frequency vibration data; for drive shafts, relevant vibration monitoring data includes shaft frequency and second harmonic vibration data.

[0092] Step 3.1: Select the rotating component and determine the relevant vibration monitoring data.

[0093] The selected rotating component is the tail drive shaft. Based on the location of the tail drive shaft and the installation position of the vibration sensor, the vibration monitoring data corresponding to the tail drive shaft is selected from the vibration monitoring data, including the fundamental frequency and the second harmonic of the tail drive shaft.

[0094] The fundamental frequency vibration monitoring data and the second harmonic frequency vibration monitoring data of the tail drive shaft for the first task are shown in Table 5.

[0095] Table 5 Vibration monitoring data of tail drive shaft

[0096]

[0097] In the table, "Time" represents the sampling time of the data point, in the format of year, month, day, hour, minute, and second. 19:01:01:08:00:13 represents 08:00:13 on January 1, 2019. "Fundamental frequency vibration value" represents the value of the collected fundamental frequency vibration monitoring data of the tail drive shaft, in g. "Second harmonic vibration value" represents the value of the collected second harmonic vibration monitoring data of the tail drive shaft, in g.

[0098] Step 3.2: Extract vibration monitoring data at the start of the task and calculate the characteristics.

[0099] For each type of vibration data, a data segment one minute after the start of the task is extracted, and the mean of the data is calculated as the task start time characteristic value for that type of vibration data. The task start time characteristic value of the tail drive shaft fundamental frequency vibration monitoring data for the first task is shown below. The calculation results are as follows:

[0100] (12)

[0101] In the formula, The start time of the first task is 19:01:01:08:00:13; The time representing one minute after the start time of the first task is 19:01:01:08:01:13; represent The fundamental frequency vibration data of the tail drive shaft at a given time.

[0102] The characteristic value of the tail drive shaft second harmonic vibration monitoring data for the first task at the start time is... The calculation results are as follows:

[0103] (13)

[0104] In the formula, represent The second harmonic vibration data of the tail drive shaft at a given moment.

[0105] Step 3.3: Extract vibration monitoring data at the end of the task and calculate the characteristics.

[0106] For each type of vibration data, a segment of data one minute before the end of the task is extracted, and the mean of the data is calculated as the characteristic value of the task end time for that type of vibration data. The characteristic value of the task end time for the tail drive shaft fundamental frequency vibration monitoring data of the first task is shown below. The calculation results are as follows:

[0107] (14)

[0108] In the formula, The end time of the first task is 19:01:01:08:39:56; The time representing one minute before the end of the first task is 19:01:01:08:38:56; represent The fundamental frequency vibration data of the tail drive shaft at a given time.

[0109] The characteristic values ​​of the tail drive shaft second harmonic vibration monitoring data at the end of the first task. The calculation results are as follows:

[0110] (15)

[0111] In the formula, represent The second harmonic vibration data of the tail drive shaft at a given moment.

[0112] Step 3.4: Summarize and form a vibration monitoring data characteristic table.

[0113] For each vibration data point of the rotating component for each task, the characteristic values ​​at the start and end of the task are calculated. The summarized results are shown in Table 6.

[0114] Table 6 Characteristics of Vibration Monitoring Data for Tail Driveshaft

[0115]

[0116] Step 4: Construct the basic dataset for each task of the rotating component.

[0117] Based on the data tables from step 3.4 and step 2.3, the basic dataset for the tail drive shaft is formed by merging them. The dataset construction results are shown in Table 7.

[0118] Table 7 Tail Drive Shaft Basic Data Set

[0119]

[0120] Step 5: Construct a sample set for predicting the trend of rotating components.

[0121] Step 5.1: Construct a single sample for trend prediction of rotating components.

[0122] (1) Set the historical step size to =3. First select For task number 003 in section 7, construct the first sample;

[0123] (2) Use the characteristic value of the end time of each vibration data corresponding to the task number as the tag data. Vectors. Tag data. [1.020, 1.205];

[0124] (3) Construct characteristic data by taking the task number and the characteristic values ​​of the vibration data corresponding to the first two task numbers at the task start time and the comprehensive load coefficient. The vector. Feature data. [1.017, 1.205, 2478614.250, 1.018, 1.205, 4523834.240, 1.020, 1.203, 5310048.320];

[0125] (4) Thus, the first sample is constructed, sample number 001, feature data [1.017, 1.205, 2478614.250, 1.018, 1.205, 4523834.240, 1.020, 1.203, 5310048.320], Tag Data [1.020, 1.205].

[0126] (5) Select the task with task number 004 in Table 7 and construct the second sample; and so on, until the last task with task number 130 is selected and the last sample is constructed.

[0127] Step 5.2: Form a trend prediction sample set for rotating components.

[0128] The tail drive shaft trend prediction samples obtained from the above steps are summarized to form a tail drive shaft trend prediction sample set. The results of the trend prediction sample set construction are shown in Table 8.

[0129] Table 8. Tail Drive Shaft Trend Prediction Sample Set

[0130]

[0131] Step 6, Application of the sample set

[0132] Based on the above sample set, a corresponding trend prediction model can be constructed, and the model can be trained and tested.

[0133] A backpropagation (BP) neural network model was constructed as an intelligent trend prediction model. The input layer dimension was set to 9, the output layer dimension to 2, and the hidden layer dimension to 16. The constructed trend prediction sample set was randomly divided into a training set and a test set in a 7:3 ratio. The model was trained using the training set data, and optimized using the test set data.

[0134] For the trained trend prediction model, the comprehensive load coefficient of the new task is evaluated according to the new task scenario. Input features are constructed using the task start time feature values ​​and comprehensive load coefficients of the vibration data corresponding to the two most recent tasks, the task end time feature values ​​of the vibration data corresponding to the last task, and the comprehensive load coefficient of the new task. Substitute these parameters into the model to obtain the output label parameters. This serves as the prediction result of the characteristic values ​​of the vibration data at the end of the new task.

[0135] The last task refers to the task preceding the new task, that is, the last task among the completed tasks; the task end time feature value of each vibration data corresponding to the last task is used as the start time feature value of the new task.

Claims

1. A method for constructing a sample set for trend prediction of vibration monitoring data of rotating components in a transmission system, characterized in that, Includes the following steps: Step 1: Collect historical data on helicopter engine power parameters, determine the start and end times of each mission, segment the missions, and calculate the comprehensive load factor for each mission. The start and end times of each task are determined based on the engine's rated power. 1% as the threshold value; The overall load factor for each task is the sum of the engine power parameters at each moment within each task segment; Step 2: Establish vibration monitoring data characteristics for each rotating component of the transmission system in each task segment; First, based on the installation positions of the rotating components in the transmission system and the vibration sensors, the vibration monitoring data corresponding to the rotating components is selected from the vibration monitoring data. Then, for each type of vibration data of each rotating component, the vibration monitoring data of the same duration after the start time of the task and before the end time of the task are extracted respectively. The characteristic value at the start time of the task and the characteristic value at the end time of the task are obtained by calculating the average value, which constitutes the vibration monitoring data characteristics. Step 3: Construct the basic dataset for each task of the rotating component based on the comprehensive load coefficient of each task and the characteristic values ​​of the vibration monitoring data of each rotating component for each task; Step 4: Using the task as the step size, construct a trend prediction sample set for rotating components based on the basic dataset; A single sample for trend prediction includes feature data used as model input. and the label data output by the model Feature data and tag data The determination method is as follows: (1) Set the historical step size as Select task numbers i ≥ in the basic dataset The task is to construct individual samples sequentially; (2) The construction process of the first trend prediction sample is as follows: The feature values ​​of the task completion time for each vibration data point corresponding to task number i=K are used as label data. ; Set task number i=K and the preceding The characteristic values ​​of the task start time and the comprehensive load coefficient of each vibration data point corresponding to each task number are used as characteristic data. ; (3) Repeat the above (2) sample construction process in the basic dataset corresponding to the task number within i>K and step size K in sequence to obtain all the samples and number them to form a rotating component trend prediction sample set.

2. The method for constructing a sample set for trend prediction of vibration monitoring data of rotating components in a transmission system according to claim 1, characterized in that, The specific process for determining the start and end times of each task is as follows: The moment when the collected engine power parameter first reaches this threshold value within the current time period shall be designated as the start time of the mission. The task ends when the collected engine power parameter first falls below this threshold. The discriminant formulas are as follows: （2） （3） Representative power parameter The The sampled values ​​at each moment; Finally, extend the sampling time and repeat the above process to find the start time of all tasks. and the end time of the mission The start times of tasks that are adjacent in time and the end time of the mission Divide the task into one session and record the start and end times of all tasks obtained.

3. A method for constructing a sample set for trend prediction of vibration monitoring data of rotating components in a transmission system, as described in claim 1 or 2, characterized in that, The calculation process for the comprehensive load factor of each task is as follows: First, based on the start and end times of each task, the engine power parameter data for each time segment within that task is extracted, as shown below: （4） In the formula: Representing the Engine power parameter data for each task segment. Representing the The start time of each task segment. Representing the The end time of each task segment; Then, the engine power parameters at all times within each task segment are summed to obtain the comprehensive load coefficient for each task. The formula for calculating the overall load factor for each task is as follows: （5） In the formula, Representing the The overall load factor for each task Representing the The first task Each sampling time.

4. The method for constructing a sample set for trend prediction of vibration monitoring data of rotating components in a transmission system according to claim 1, characterized in that, The vibration monitoring data for each rotating component of the transmission system are as follows: For bearings, the relevant vibration monitoring data includes vibration data at shaft frequency, second harmonic, and third harmonic; for gears, the relevant vibration monitoring data includes vibration data at shaft frequency and meshing frequency; for drive shafts, the relevant vibration monitoring data includes vibration data at shaft frequency and second harmonic.

5. The method for constructing a sample set for trend prediction of vibration monitoring data of rotating components in a transmission system according to claim 1, characterized in that, The feature values ​​at the start and end of the task are as follows: For each type of vibration data, a data segment of one minute after the start of the task is extracted, and the mean of the data is calculated as the characteristic value of the task start time for that type of vibration data. The calculation method is as follows: （7） In the formula, Representing the The first task The characteristic value of the vibration data at the start time of the task. represent Vibration data at any given time This represents the number of sampling points involved in the calculation; For each type of vibration data, extract the data segment one minute before the task's end time, calculate the mean of the data, and use it as the characteristic value of the task's end time for that type of vibration data. The calculation method is as follows: （8） In the formula, Representing the The first task The characteristic value of the task completion time of the vibration data.

6. The method for constructing a sample set for trend prediction of vibration monitoring data of rotating components in a transmission system according to claim 1, characterized in that, The intelligent trend prediction model is optimized by using a trend prediction sample set and then used to predict the feature values ​​at the end of a new mission, thereby deriving the degradation law of the helicopter transmission system. The specific process is as follows: First, the constructed trend prediction sample set is randomly divided into a training set and a test set to train and test the intelligent trend prediction model based on the BP neural network model, thus completing the optimal selection of the model. Then, the trained intelligent trend prediction model is used to predict the feature values ​​of the vibration data of each rotating component at the end of the task for the new task. Finally, the degradation law of the helicopter transmission system is derived based on the prediction results output by the intelligent trend prediction model.

7. The method for constructing a sample set for predicting the trend of vibration monitoring data of rotating components in a transmission system according to claim 6, characterized in that, The process of predicting the feature values ​​at the end of the new task is as follows: First, based on the new mission scenario, evaluate the comprehensive load factor of the new mission; By predicting the flight path of the new mission, including flight time and flight profile data, the overall payload factor of the new mission is estimated. Then, using the recent The input features are constructed from the characteristic values ​​of the start time of each vibration data point for the sub-task, the characteristic values ​​of the end time of each vibration data point for the last task, and the comprehensive load coefficient of the new task. Substitute these parameters into the trend prediction model to obtain the output label parameters. This serves as the prediction result of the characteristic values ​​of the vibration data at the end of the new task.

Images