Search

CN-122014422-A - Prediction and repair method for performance degradation trend of aeroengine

CN122014422ACN 122014422 ACN122014422 ACN 122014422ACN-122014422-A

Abstract

The invention relates to the technical field of aviation engineering and discloses a prediction and repair method for performance degradation trend of an aero-engine, which comprises the steps of 1, establishing an intelligent prediction model, monitoring engine key data in real time through a sensor, predicting degradation trend of each component by combining a machine learning algorithm, providing early warning of repair time for maintainers, 2, applying a local repair technology, identifying the component to be repaired based on the intelligent prediction model, accurately repairing the damaged component by adopting a laser repair technology and a metal spraying technology, reducing the repair cost, prolonging the service life of the component, 3, selecting the accurate repair time, monitoring the state of each component of the engine in real time, and automatically sending repair early warning according to the degradation trend.

Inventors

  • ZHOU SONG
  • AN JINLAN
  • Qian Zhaoxing
  • WANG XIANGMING
  • CHENG RONGHUI
  • SUN YUMIN
  • PAN XIN
  • LI XIAODAN
  • XING BENDONG
  • YANG LINQING

Assignees

  • 沈阳航空航天大学

Dates

Publication Date
20260512
Application Date
20251231

Claims (10)

  1. 1. The method for predicting and repairing the performance degradation trend of the aeroengine is characterized by comprising the following steps of: Step 1, an intelligent prediction model is established, key data of an engine is monitored in real time through a sensor, and the fading trend of each part is predicted by combining a machine learning algorithm, so that early warning of early repair time is provided for maintenance personnel; Step 2, applying a local repair technology, identifying a part to be repaired based on an intelligent prediction model, accurately repairing the damaged part by adopting laser repair and metal spraying technologies, reducing the maintenance cost and prolonging the service life of the part; Step 3, selecting accurate repairing time, wherein the system monitors the states of all parts of the engine in real time, automatically sends out repairing early warning according to the declining trend, dynamically adjusts the repairing period, ensures accurate repairing time and avoids premature and too late repairing; Step 4, the decision support system gathers the engine health data and maintenance history information to the cloud platform, supports maintenance decision, and is in butt joint with the flight management system and the aircraft health monitoring system to coordinate flight tasks and maintenance plans, so that downtime is reduced; And 5, evaluating and optimizing the repair effect, wherein after each repair, the system evaluates the repair effect, ensures that the repair quality meets the expectations, continuously optimizes the prediction model and the repair strategy based on the repair effect, and improves the accuracy of future repair.
  2. 2. The method for predicting and repairing an aeroengine performance degradation trend according to claim 1, wherein the step 1 comprises: Step 1.1, real-time data acquisition, namely, real-time monitoring temperature, pressure, vibration and rotating speed key operation data of a key component of an engine by installing a plurality of sensors on the key component; step 1.2, preprocessing data, namely denoising, normalizing and filling missing values of the original data acquired by the sensor, and ensuring the accuracy and the effectiveness of the data: The denoising formula is: , is the original signal which is then used to determine, Is the impulse response of the filter, Is the filtered signal; the normalization formula is: , Is the original data of the data set, And Respectively, a minimum value and a maximum value of the data set; The missing value filling formula is: formissingvalues, is the mean value of the data set, Is padded data.
  3. 3. The method for predicting and repairing an aeroengine performance degradation trend according to claim 1, wherein the step 1 further comprises: Step 1.3, machine learning algorithm prediction, namely inputting the preprocessed data into a machine learning model, wherein the machine learning model is a deep learning model based on historical data and real-time data, and predicting the fading trend of each part of an engine by using the model; step 1.3.1 update formula of cell state and hidden state in long-short-term memory network: forgetting the door: , The output of the door is forgotten, The weight matrix of the forgetting gate, The hidden state of the previous moment is adopted, The current input data is used to determine the current input data, The bias term is used for the bias of the bias term, Sigmoid activating function; an input door: , the output of the input gate is input, The weight matrix of the gate is input, Bias term Cell status update: , The state of the cell at the current moment, A hyperbolic tangent activation function; Output door: , The door is output to the outside of the device, Outputting a weight matrix of the gate; Final hidden state output: , a hidden state at the current moment; the loss function is used for measuring the difference between the predicted result and the actual value, and the formula is as follows: , The output of the model prediction is used to determine, The actual output value of the device is calculated, Sample number; Step 1.3.2, a convolutional neural network is commonly used for feature extraction and data preprocessing, and particularly can extract key features for prediction when pattern recognition is performed in high-dimensional time sequence data, and the formula is as follows: , The method comprises the steps of carrying out convolution kernel, The data is input and the data is input, The bias term is used for the bias of the bias term, The function is activated and the function is activated, Convolution operation; Analyzing the fading trend and early warning according to the prediction result output by the machine learning model, analyzing the fading rate and the residual life of each component, and calculating the optimal repairing time, thereby providing early warning of repairing time for maintenance personnel; step 1.4.1 the decay trend is often fitted using an exponential decay model, the basic formula of which is: , At the time of The state of health of the component at the time, The initial state of health of the component, The rate of decay is set to be, Time is taken; Estimating decay rate by fitting historical data And predicting future fading trend; Step 1.4.2 the remaining service life refers to the time that the component can continue to operate in the future, usually predicted by a regression model, with the formula: , the residual service life of the device is longer, The initial state of health of the component, The health state at the current moment is shown, The decay rate can provide the expected failure time of the component for maintenance personnel by calculating RUL; The gradient elevator in step 1.4.3 is an efficient integrated algorithm for regression and classification tasks, and is suitable for component fault prediction, and the formula is as follows: , The model predicts the output of the model, The m-th base learner, The weight of the mth base learner; Step 1.5, notifying the repairing opportunity, namely automatically notifying maintenance personnel to perform necessary maintenance operation or repairing when the predicted declining trend reaches a preset threshold value; Step 1.5.1 when the decay trend reaches a preset threshold, the system automatically triggers a notification to remind maintenance personnel to repair, and the system compares the predicted value of the decay rate or the residual life with the preset threshold according to the predicted value of the decay rate or the residual life, if Below a certain preset value, a maintenance notification is triggered: If ,thennotifymaintenance, the residual service life of the device is longer, If the RUL is predicted to be 50 hours and the set threshold is 100 hours, the system can send out repair early warning when the part reaches the remaining 50 hours.
  4. 4. The method for predicting and repairing an aeroengine performance degradation trend according to claim 1, wherein the step 2 comprises: step 2.1, monitoring equipment operation data in real time through an intelligent prediction model, analyzing the health state and the fading trend of each component, and automatically identifying the components needing to be repaired; step 2.1.1 health index is a quantitative indicator of the health condition of the component, which can be calculated by the following formula: , Component health index at time t, The observation of the ith sensor or feature at time t, Weights of the ith feature, The number of sensors or the number of features; the index decay model formula of step 2.1.2 is: , Component performance values at time t, The initial performance of the component is shown as such, The rate of decay is set to be, Time is taken; 2.2, accurately repairing the damaged part by adopting a laser repairing technology for the part which is identified to be repaired, and melting and re-synthesizing the surface material by a high-precision laser beam to recover the performance of the part; step 2.2.1, the laser energy density calculation formula is as follows: , e, laser energy density, P, laser power and A, laser irradiation area; step 2.2.2, the depth of laser repair is in direct proportion to the laser energy density, and the calculation formula is as follows: , d, repairing depth, eta, laser repairing efficiency, P, laser power, t, laser irradiation time and A, laser irradiation area; step 2.3, repairing the seriously damaged part by using a metal spraying technology, and spraying a metal material to restore the shape and strength of the part and improve the wear resistance and corrosion resistance of the part; step 2.3.1 spraying efficiency refers to the ratio of the volume of metal actually sprayed to the surface of the part to the total sprayed amount, and the formula is as follows: , the spraying efficiency is that the paint is high, The volume of metal actually sprayed onto the surface of the component, The total metal volume sprayed; step 2.3.2 the thickness of the sprayed layer influences the performance recovery degree of the component, and the calculation formula is as follows: , The thickness of the spray coating is equal to that of the spray coating, The volume of metal actually sprayed onto the surface of the component, Surface area of the spray coating.
  5. 5. The method for predicting and repairing an aeroengine performance degradation trend according to claim 1, wherein the step 3 comprises: Step 3.1, dynamically adjusting the maintenance period is calculated based on the declining trend of the component, the environmental condition and the working condition change, and the adjustment formula of the maintenance period is as follows: , the maintenance period after dynamic adjustment is adopted, The standard maintenance period is adopted, The weight of the environmental factor is that, Deviation between the decay rate of the current part and the expected decay rate ), Initial preset part decay rate; Step 3.2 in order to ensure that the repair opportunity of the component is not too early or too late, the following optimization model is used to calculate the optimal repair opportunity: , The best repairing time is provided with a time for repairing, The cost of repair is a function of time, Risk weight coefficient, balancing relationship between maintenance cost and fault risk The function of the risk of failure as a function of time increases is typically derived based on a prediction of remaining useful life.
  6. 6. The method for predicting and repairing an aeroengine performance degradation trend according to claim 1, wherein the step 3 further comprises: step 3.3 to dynamically adjust the repair cycle and determine the repair opportunity, the system typically incorporates a "threshold" model whose repair opportunity threshold adjustment formula is: , The maintenance period after the adjustment is carried out, The reference maintenance period is adopted, Adjusting the coefficient to reflect the influence of the decay rate change on the maintenance period, The decay rate of the current component, Initially setting a decay rate; Step 3.4 when multiple components need to be repaired, the system usually performs task sequencing according to the risk factors, and preferentially repairs the most critical components, wherein the calculation formula of the repair task priority is as follows: , The priority of the repair task is given to the user, The residual service life of the device is longer, Risk factors for component failure, Task weighting coefficients, representing the impact of maintenance costs on priority, The maintenance cost is that, The maintenance period after adjustment; Step 3.5 to ensure accuracy of repair opportunities, the selection of repair opportunities generally requires a combination of factors to determine an optimal repair window, and an optimization model of the repair window may be expressed by the following formula: , , , And Representing the minimum and maximum repair times of the repair window respectively, The triggering time point of the early warning is given, The optimal repair time is set up in advance, Reserved time for preventing premature or late repair.
  7. 7. The method for predicting and repairing an aeroengine performance degradation trend according to claim 1, wherein the step 4 comprises: step 4.1, providing suggestions for maintenance decisions through a decision support system based on data in the cloud platform, and supporting arrangement and optimization of maintenance tasks; Step 4.1.1 optimizing maintenance tasks by the decision support system with the goal of minimizing the sum of maintenance costs and downtime, the formula is as follows: , a maintenance cost function representing maintenance costs generated according to different maintenance time points t, Weight coefficient, balance the relationship between maintenance cost and downtime, A downtime function, representing downtime for component repair, typically associated with repair time and man-hour resources; step 4.1.2, predicting the maintenance time of the component by a decision tree algorithm, wherein the formula is as follows: , The predicted maintenance time or maintenance proposal, The weight coefficient of the decision tree node, The decision function of each decision node based on the input characteristics ; Step 4.2, coordinating a maintenance plan according to the maintenance decision and the requirements of the flight mission, reducing the downtime and ensuring the effective connection between the maintenance operation and the flight mission; Step 4.2.1 there are multi-maintenance tasks and flight tasks to be scheduled, the optimization problem can be described by the following multi-objective optimization formula: , First of all The downtime of the individual maintenance tasks, First of all The time delay of individual flight tasks due to maintenance delays, Weight coefficient for balancing the coordination degree of maintenance and flight tasks, The number of maintenance tasks is counted up, Number of flight missions Step 4.2.2 to ensure maintenance and flight mission engagement, the system may schedule based on the idea of optimal time allocation, with the formula: , the maintenance time is preset, and the maintenance time is preset, The actual maintenance start or end time is taken, The weight coefficient of the scheduling delay, First of all Delay time of individual flight tasks.
  8. 8. The method for predicting and repairing an aeroengine performance degradation trend according to claim 1, wherein the step 4 further comprises: step 4.3, interfacing the decision support system with the flight management system, acquiring flight mission data in real time, and adjusting a maintenance plan based on the flight mission data; step 4.3.1 a priority scheduling model is set to dynamically adjust the schedule between the maintenance schedule and the flight mission with the aim of minimizing delays in the flight mission and downtime of the maintenance mission, the formula being as follows: , First of all The delay time of the individual flight tasks, First of all The downtime of the individual maintenance tasks, Weight coefficient for balancing the relationship between flight mission delay and maintenance mission downtime, The number of the flight tasks is calculated, The number of maintenance tasks; Step 4.4, the decision support system is in butt joint with the aircraft health monitoring system to acquire the whole health state data of the aircraft, and the whole health state data is used for assisting in accurate selection of maintenance decisions; step 4.4.1 the overall health status of the aircraft is assessed by a plurality of health parameters, and a weighted average formula may be used to calculate an overall health index, which formula is: , the overall health index of the aircraft, Each health parameter Reflecting the degree of influence of the parameter on the overall health status, First of all A personal health parameter; Step 4.4.2, selecting an optimal maintenance time point through a predictive maintenance model based on the overall health index of the aircraft, and using a residual service life prediction model, wherein the formula is as follows: , The expected remaining useful life, which is indicative of the time that the aircraft component is still in normal use, The overall health index of the patient is shown, Regression coefficients, learned based on historical data, Error terms representing uncertainty of model prediction; step 4.5, dynamically adjusting the arrangement between maintenance and flight tasks, ensuring coordination and compatibility of maintenance plans and flight tasks, and reducing downtime; Step 4.5.1 to dynamically adjust the schedule between maintenance and flight tasks, a constraint optimization model may be used to minimize downtime and delays by adjusting maintenance opportunities and flight plans, the formula of which is: , , , First of all The downtime of the individual maintenance tasks, First of all The delay time of the individual flight tasks, Maximum acceptable downtime for each maintenance task, The maximum acceptable delay time for each flight mission, The number of maintenance tasks is counted up, Number of flight missions.
  9. 9. The method for predicting and repairing an aeroengine performance degradation trend according to claim 1, wherein the step 5 comprises: step 5.1, after each maintenance is completed, the system evaluates the maintenance effect, and confirms whether the repair quality meets the expectations or not by comparing the state data before and after the maintenance with the maintenance targets, and generates an evaluation report; Step 5.1.1, calculating the health state difference before and after maintenance by using the Euclidean distance, wherein the formula is as follows: , The state difference metric is used to determine the state difference, Post-repair item The value of the individual health parameter is calculated, First before maintenance The value of the individual health parameter is calculated, Estimating the number of the related health parameters; Step 5.1.2 for whether the repair meets the expectations, a repair quality score model may be used to generate a composite quality score by comparing the repaired parameters to the expected targets, with the formula: , A repair quality score, representing whether the repair meets expectations, The i-th health parameter value after repair, The expected target value of the ith health parameter, The weight of the ith health parameter, reflecting the importance of that parameter in the assessment, The total estimated health parameter quantity; Step 5.2, determining whether the repair quality reaches the standard according to the repair effect evaluation result, and providing corresponding quality control measures according to the evaluation report to ensure the reliability and accuracy of the repair operation; step 5.2.1, setting a repair quality threshold, and comparing the current quality score with a preset standard to determine whether quality control measures need to be taken or not, wherein the formula is as follows: , The quality score of the repair is calculated, A preset repair quality score threshold value is adopted, Whether repair quality control measures are adopted or not, wherein 1 is to be controlled, and 0 is not to be controlled; step 5.3, feeding back the evaluation result and the quality control measure to a maintenance system for guiding the adjustment and optimization of the subsequent maintenance task, and improving the accuracy and reliability of the repair operation; Step 5.3.1, adjusting a maintenance strategy based on a maintenance effect evaluation result by introducing a feedback learning algorithm, and optimizing a subsequent maintenance task by using a reward function in reinforcement learning, wherein the formula is as follows: , Feedback of the adjusted prize value, The current quality score of the repair is calculated, The quality score of the intended repair is calculated, The value of the healthful parameter after the repair is carried out, The value of the target health parameter is calculated, Weights of all health parameters are adopted, The number of health parameters to be assessed, And adjusting the factors, and setting according to task requirements.
  10. 10. The method for predicting and repairing an aeroengine performance degradation trend according to claim 1, wherein the step 5 further comprises: step 5.4, based on the repair effect evaluation result, adjusting and optimizing key parameters in the prediction model to improve the prediction accuracy and ensure the accurate arrangement of future maintenance tasks; Step 5.4.1, assuming that the prediction model adopts linear regression, the restoring effect evaluation result influences the regression coefficient in the prediction model, and optimizing the regression coefficient by a least square method, wherein the formula is as follows: , the optimization process minimizes the prediction error by adjusting the regression coefficients, the loss function is: , the real maintenance data are used for the maintenance of the vehicle, The predicted maintenance data is used to determine the maintenance data, Regression coefficients, representing the influence of each parameter on the prediction result, Training the sample number of the data set; Step 5.5, continuously optimizing a maintenance strategy, adjusting a maintenance plan and an operation flow according to the repair effect and feedback information of the prediction model, ensuring that similar problems can be more effectively treated in a future maintenance process, and reducing unnecessary downtime; Step 5.5.1 is implemented by a multi-objective optimization model, taking multiple objectives into account for comprehensive scheduling. The objective function can be expressed as: , , , First of all The downtime of the individual maintenance tasks, First of all The delay time of the individual maintenance tasks, Weight coefficients for balancing the effects of downtime and delay time, Maximum acceptable downtime for each maintenance task, The maximum acceptable delay time for each flight mission, The number of maintenance tasks is counted up, The number of flight tasks; step 5.6, the system gradually improves maintenance decision and maintenance strategy by continuously learning historical maintenance data and maintenance effect, forms a closed-loop optimization mechanism, and improves the overall efficiency and quality of the maintenance process; The reward function in reinforcement learning in step 5.6.1 can quantify the execution effect of the maintenance task, and improve the efficiency by continuously adjusting the strategy, and the strategy optimization is performed by using the Q-learning algorithm, and the reward function can be expressed as: , in the current state s, the quality value of the action a is executed, In the current state s, the instant rewards obtained by the action a are executed, A discount factor, for measuring the impact of future rewards, Learning rate, controlling the speed of policy updating, The new state after the action a is performed, An optional next action.

Description

Prediction and repair method for performance degradation trend of aeroengine Technical Field The invention relates to the technical field of aviation engineering, in particular to a prediction and repair method for performance degradation trend of an aeroengine. Background During maintenance of an aeroengine, performance degradation is inevitable, particularly in the case of long-term operation or high-load operation, and as the aeroengine ages, the performance of its various components gradually decreases, resulting in a decrease in the overall efficiency of the engine. In order to ensure aviation safety and improve service life of an engine, it is very important to perform performance detection and maintenance regularly, however, many existing repair methods currently have certain limitations, mainly: the method for repairing the engine is characterized in that the traditional repairing method cannot accurately identify the degradation of each part of the engine, generally depends on the traditional maintenance and part replacement modes, and lacks a local repairing technology aiming at specific degradation problems; The method has the advantages that the method is inaccurate in repairing opportunity selection, most of repairing schemes at present depend on periodic inspection or experience judgment, the decline trend of each part of the engine is difficult to accurately grasp, and the repairing opportunity can have the problems of early or late, so that the repairing cost is increased and the shutdown time is unnecessarily prolonged. Accordingly, those skilled in the art provide a method for predicting and repairing the performance degradation trend of an aeroengine to solve the above-mentioned problems. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a prediction and repair method for the performance degradation trend of an aeroengine, so as to solve the problems in the background art. In order to achieve the purpose, the invention is realized by the following technical scheme that the prediction and repair method for the performance degradation trend of the aeroengine comprises the following steps: Step 1, an intelligent prediction model is established, key data of an engine is monitored in real time through a sensor, and the fading trend of each part is predicted by combining a machine learning algorithm, so that early warning of early repair time is provided for maintenance personnel; Step 2, applying a local repair technology, identifying a part to be repaired based on an intelligent prediction model, accurately repairing the damaged part by adopting laser repair and metal spraying technologies, reducing the maintenance cost and prolonging the service life of the part; Step 3, selecting accurate repairing time, wherein the system monitors the states of all parts of the engine in real time, automatically sends out repairing early warning according to the declining trend, dynamically adjusts the repairing period, ensures accurate repairing time and avoids premature and too late repairing; Step 4, the decision support system gathers the engine health data and maintenance history information to the cloud platform, supports maintenance decision, and is in butt joint with the flight management system and the aircraft health monitoring system to coordinate flight tasks and maintenance plans, so that downtime is reduced; And 5, evaluating and optimizing the repair effect, wherein after each repair, the system evaluates the repair effect, ensures that the repair quality meets the expectations, continuously optimizes the prediction model and the repair strategy based on the repair effect, and improves the accuracy of future repair. Preferably, the step 1 comprises: Step 1.1, real-time data acquisition, namely, real-time monitoring temperature, pressure, vibration and rotating speed key operation data of a key component of an engine by installing a plurality of sensors on the key component; step 1.2, preprocessing data, namely denoising, normalizing and filling missing values of the original data acquired by the sensor, and ensuring the accuracy and the effectiveness of the data: The denoising formula is: , is the original signal which is then used to determine, Is the impulse response of the filter,Is the filtered signal; the normalization formula is: , Is the original data of the data set, AndRespectively, a minimum value and a maximum value of the data set; The missing value filling formula is: formissingvalues, is the mean value of the data set, Is padded data. Preferably, the step 1 further includes: Step 1.3, machine learning algorithm prediction, namely inputting the preprocessed data into a machine learning model, wherein the machine learning model is a deep learning model based on historical data and real-time data, and predicting the fading trend of each part of an engine by using the model; step 1.3.1 update formula of cell state and hidden state in long-short-