Intelligent ignition device and method for aviation turbine engine with redundancy

Abstract

The invention discloses an aviation turbine engine intelligent ignition device and method with redundancy. The device comprises a plurality of ignition electric nozzles arranged in parallel, an ignition sensor, a parameter optimization module and a control module. A plurality of ignition electric nozzles are mounted between the flame tube and the outer casing in the length direction, and each ignition electric nozzle is numbered; the parameter optimization module is used for obtaining an ignition prediction model based on the training data set by using an integrated learning algorithm, predicting ignition result data of each ignition electric nozzle by using the ignition prediction model, feeding back a prediction result to a reinforcement learning algorithm, carrying out iterative optimization on ignition control data of the ignition prediction model, and outputting the ignition control data of the ignition prediction model. And controlling the corresponding ignition electric nozzle to ignite according to the optimal ignition control parameter until the optimal ignition control parameter is obtained. According to the invention, intelligent ignition control based on machine learning is realized, the design of ignition redundancy is realized, the effectiveness of ignition is improved, and the practicability is relatively good.

Description

Technical Field

[0001] The invention belongs to the technical field of aviation turbine engine ignition devices, and in particular relates to an aviation turbine engine intelligent ignition device with redundancy and a method thereof. Background Art

[0002] At present, the ignition method used in most aviation turbine engines adopts a fixed single ignition nozzle for ignition. The disadvantages of this ignition method are: narrow ignition boundary, low ignition success rate, and impact on ground starting of aircraft engines and flameout and re-ignition in the air.

[0003] The Chinese invention patent with application number 202310114578.9 discloses a nozzle-mobile aviation turbine engine ignition device, which includes: an outer casing, an ignition nozzle, a flame tube, a nozzle moving mechanism, an ignition sensor and a controller. The nozzle moving mechanism is connected to one end of the ignition nozzle and is used to control the ignition nozzle to move in the direction of the flame tube, and to make the other end of the ignition nozzle extend into the flame tube through the ignition nozzle hole or withdraw from the flame tube along the ignition nozzle hole; the controller is connected to the ignition sensor and is used to receive the fuel combustion situation in the flame tube transmitted by the ignition sensor, and the controller is connected to the nozzle moving mechanism and is used to control the movement of the ignition nozzle according to the data transmitted by the ignition sensor. The present invention also discloses an ignition method using the ignition device. The present invention quickly moves the ignition nozzle to the optimal ignition position according to different working states, thereby improving the problem of narrow ignition boundary in the fixed ignition method. However, there is only a single ignition nozzle in the technical solution, and the ignition effect can only be adjusted by controlling the insertion depth of the ignition nozzle in the fixed position. There are many factors that affect the ignition effect of the ignition device. Based on the service life and engine running quality, the optimal ignition position is different, and it is impossible to avoid the situation where a single ignition nozzle fails and delays its use in time. Summary of the invention

[0004] The object of the present invention is to provide an intelligent ignition device and method for an aviation turbine engine with redundancy, aiming to solve the above-mentioned problems.

[0005] The present invention is mainly achieved through the following technical solutions:

[0006] An intelligent ignition device for an aviation turbine engine with redundancy comprises a flame tube, an outer casing, a plurality of ignition nozzles arranged in parallel, an ignition sensor, a parameter optimization module and a control module;

[0007] A plurality of ignition nozzles are installed between the flame tube and the outer casing along the length direction, and each ignition nozzle is numbered; a plurality of ignition holes are correspondingly opened on the side wall of the flame tube along the length direction, and a linear drive mechanism is provided on one side of the ignition nozzle for driving the ignition nozzle to telescopically move in the ignition hole;

[0008] An ignition sensor is provided in the combustion area inside the flame tube to detect the fuel combustion condition;

[0009] The parameter optimization module is used to train an ignition prediction model based on a training data set using an integrated learning algorithm, predict the ignition result data of each ignition nozzle using the ignition prediction model, and feed the prediction results back to the reinforcement learning algorithm to iteratively optimize the ignition control data of the ignition prediction model until the optimal ignition control parameters are obtained;

[0010] The training data set includes numerical ignition control data and ignition result data, wherein the ignition control data includes the ignition nozzle number, the movement distance of the ignition nozzle, the ignition control voltage and frequency signal, and the ignition result data includes the ignition result, the fuel combustion condition, the engine speed and load signal, and the ignition nozzle operation state, and the ignition nozzle operation state is the fluctuation deviation value of the ignition control voltage and frequency signal compared with the voltage and frequency signal inside the ignition nozzle;

[0011] The control module is used to control the ignition of the corresponding ignition nozzle according to the optimal ignition control parameters.

[0012] In order to better implement the present invention, further, a preprocessing module is included, and the preprocessing module is used to collect historical data, preprocess the data and form a training set. The data preprocessing includes sequentially performing numerical processing and normalization processing on the data.

[0013] In order to better realize the present invention, further, a reinforcement learning model is built based on the reinforcement learning algorithm, and the prediction results are fed back to the reinforcement learning model so that the reinforcement learning model updates the neural network strategy according to the prediction results; new ignition control data is selected based on the updated neural network strategy and input into the ignition prediction model; the ignition control data of the ignition prediction model is iteratively updated using the reinforcement learning model until the detection result is optimal and the corresponding ignition control data is the optimal ignition control parameter.

[0014] In order to better implement the present invention, further, in the neural network strategy updated according to the prediction results, a reward function is used to enhance the probability of selection of each ignition control data that helps to output a positive prediction result in the previous prediction result, and to weaken the probability of selection of the ignition control data that helps to output a reverse prediction result.

[0015] In order to better realize the present invention, further, the linear drive mechanism includes a mounting seat, an electromagnetic seat, a spring and a magnetic suction head. The outer casing is provided with a mounting seat, the ignition nozzle is slidably connected to the mounting seat, the inner side wall of the mounting seat is installed with an electromagnetic seat, the sliding end of the ignition nozzle is correspondingly installed with a magnetic suction head, and a spring is installed between the electromagnetic seat and the magnetic suction head.

[0016] In order to better realize the present invention, further, a plurality of ignition nozzles are staggeredly arranged at the front end of the flame tube along the length direction, and a plurality of ignition holes are correspondingly staggeredly opened at the side wall of the flame tube along the length direction.

[0017] The present invention is mainly achieved through the following technical solutions:

[0018] A redundant intelligent ignition method for an aviation turbine engine is based on the above device and comprises the following steps:

[0019] Step S1: collecting historical data and digitizing the historical data to form a training data set and a test data set, wherein the training data set includes digitized ignition control data and ignition result data; the ignition control data includes the ignition nozzle number, the movement distance of the ignition nozzle, the ignition control voltage and frequency signal, and the ignition result data includes the ignition result, fuel combustion status, engine speed and load signal, and ignition nozzle operation status;

[0020] Step S2: training the data set and obtaining an ignition prediction model by using an integrated learning algorithm;

[0021] Boostrap sampling with replacement is used to obtain several training subsets; a classification algorithm is used to independently train several base classifiers on each training subset; for each test sample in the test data set, several base classifiers are used to obtain several prediction values, and for each test sample, the final prediction result is obtained by majority voting;

[0022] Step S3: using the ignition prediction model to predict the ignition result data of each ignition nozzle, and feeding the prediction results back to the reinforcement learning algorithm, iteratively optimizing the ignition control data of the ignition prediction model until the optimal ignition control parameters are obtained, that is, the optimal ignition nozzle number, the movement distance of the ignition nozzle, the ignition control voltage and the frequency signal are obtained;

[0023] Step S4: Perform ignition control based on the optimal ignition control parameters.

[0024] In order to better implement the present invention, further, in the step S1, after the historical data is digitized, the data is normalized.

[0025] The beneficial effects of the present invention are as follows:

[0026] In the ignition process, the present invention numbers several specific ignition positions designed, and then obtains prediction results of various ignition positions, driving movement distances, and ignition control voltages and frequency signals based on an ignition prediction model formed by an integrated learning algorithm. The prediction results take into account whether the ignition is successful or not, ignition stability, fuel combustion conditions, and engine operation conditions, and comprehensively evaluate the ignition performance; then, based on the reinforcement learning algorithm, the best ignition control data for ignition is obtained, that is, the ignition nozzle with the corresponding number is controlled to move according to the movement distance, and the ignition intensity is controlled based on the ignition control voltage and frequency signal, thereby realizing intelligent ignition control based on machine learning, realizing the design of ignition redundancy, improving the effectiveness of ignition, and having good practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0028] Embodiment 1:

[0029] A redundant intelligent ignition method for an aviation turbine engine, firstly, a plurality of ignition nozzles are staggered along the length direction at the front end of the flame tube, and a plurality of ignition holes are staggered along the length direction on the side wall of the flame tube. An ignition nozzle is installed between the flame tube and the outer casing, and a linear drive mechanism is provided on one side of the ignition nozzle for driving the ignition nozzle to telescopically move in the ignition hole. Specifically, the linear drive mechanism includes a mounting seat, an electromagnetic seat, a spring and a magnetic suction head, a mounting seat is provided on the outer casing, the ignition nozzle is slidably connected to the mounting seat, an electromagnetic seat is installed on the inner side wall of the mounting seat, a magnetic suction head is correspondingly installed at the sliding end of the ignition nozzle, and a spring is installed between the electromagnetic seat and the magnetic suction head.

[0030] Then, the historical data is collected and processed numerically, and then normalized to obtain ignition control data and ignition result data to form a training data set and a test data set. The ignition control data includes the ignition nozzle number, the movement distance of the ignition nozzle, the ignition control voltage and frequency signal, and the ignition result data includes the ignition result, the fuel combustion situation, the engine speed and load signal, and the ignition nozzle operation state. The operation state of the ignition nozzle is the fluctuation deviation value of the ignition control voltage and frequency signal compared with the voltage and frequency signal inside the ignition nozzle.

[0031] Based on the training data set and using the integrated learning algorithm to train the ignition prediction model, the ignition prediction model is used to predict the ignition result data of each ignition nozzle, and the prediction results are fed back to the reinforcement learning algorithm to iteratively optimize the ignition control data of the ignition prediction model until the optimal ignition control parameters are obtained, that is, the optimal ignition nozzle number, ignition nozzle movement distance, ignition control voltage and frequency signal are obtained. The corresponding ignition nozzle is controlled to ignite based on the optimal ignition control parameters.

[0032] Preferably, if Figure 1 As shown, the present invention specifically comprises the following steps:

[0033] Step S1: collecting historical data, and digitizing and normalizing the historical data to form a training data set and a test data set, wherein the training data set includes digitized ignition control data and ignition result data; the ignition control data includes the ignition nozzle number, the movement distance of the ignition nozzle, the ignition control voltage and frequency signal, and the ignition result data includes the ignition result, fuel combustion condition, engine speed and load signal, and the ignition nozzle operation state;

[0034] Step S2: training the data set and obtaining an ignition prediction model by using an integrated learning algorithm;

[0035] Boostrap sampling with replacement is used to obtain several training subsets; a classification algorithm is used to independently train several base classifiers on each training subset; for each test sample in the test data set, several base classifiers are used to obtain several prediction values, and for each test sample, the final prediction result is obtained by majority voting;

[0036] The ensemble learning algorithm itself is not a separate machine learning algorithm, but it completes the learning task by building and combining multiple machine learning machines. It can be said to be a combination of the strengths of many schools of thought, and can have a higher accuracy rate in machine learning algorithms. Adaboost is a more representative algorithm in boosting. The basic idea is to construct a classifier based on the distribution of training data, and then find the weight of this weak classifier through the error rate, and iterate by updating the distribution of training data until the number of iterations is reached or the loss function is less than a certain threshold.

[0037] Step S3: using the ignition prediction model to predict the ignition result data of each ignition nozzle, and feeding the prediction results back to the reinforcement learning algorithm, iteratively optimizing the ignition control data of the ignition prediction model until the optimal ignition control parameters are obtained, that is, the optimal ignition nozzle number, the movement distance of the ignition nozzle, the ignition control voltage and the frequency signal are obtained;

[0038] Specifically, a reinforcement learning model is built based on a reinforcement learning algorithm, and the prediction results are fed back to the reinforcement learning model so that the reinforcement learning model updates the neural network strategy according to the prediction results; new ignition control data is selected based on the updated neural network strategy and input into the ignition prediction model; the ignition control data of the ignition prediction model is iteratively updated using the reinforcement learning model until the detection result is optimal and the corresponding ignition control data is the optimal ignition control parameter. Preferably, in the updating of the neural network strategy according to the prediction results, the probability of each ignition control data that helps to output a positive prediction result in the last prediction result being selected is enhanced by using a reward function, and the probability of the ignition control data that helps to output a negative prediction result being selected is weakened.

[0039] Step S4: Perform ignition control based on the optimal ignition control parameters.

[0040] In the ignition process, the present invention numbers several specific ignition positions designed, and then obtains prediction results of various ignition positions, driving movement distances, and ignition control voltages and frequency signals based on an ignition prediction model formed by an integrated learning algorithm. The prediction results take into account whether the ignition is successful or not, ignition stability, fuel combustion conditions, and engine operation conditions, and comprehensively evaluate the ignition performance; then, based on the reinforcement learning algorithm, the best ignition control data for ignition is obtained, that is, the ignition nozzle with the corresponding number is controlled to move according to the movement distance, and the ignition intensity is controlled based on the ignition control voltage and frequency signal, thereby realizing intelligent ignition control based on machine learning, realizing the design of ignition redundancy, improving the effectiveness of ignition, and having good practicality.

[0041] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. An intelligent ignition device for an aviation turbine engine with redundancy, characterized in that: It includes a flame tube, an outer casing, a plurality of ignition nozzles arranged in parallel, an ignition sensor, a parameter optimization module and a control module; A plurality of ignition nozzles are installed between the flame tube and the outer casing along the length direction, and each ignition nozzle is numbered; a plurality of ignition holes are correspondingly opened on the side wall of the flame tube along the length direction, and a linear drive mechanism is provided on one side of the ignition nozzle for driving the ignition nozzle to telescopically move in the ignition hole; An ignition sensor is provided in the combustion area inside the flame tube to detect the fuel combustion condition; The parameter optimization module is used to train an ignition prediction model based on a training data set using an integrated learning algorithm, predict the ignition result data of each ignition nozzle using the ignition prediction model, and feed the prediction results back to the reinforcement learning algorithm to iteratively optimize the ignition control data of the ignition prediction model until the optimal ignition control parameters are obtained; The training data set includes numerical ignition control data and ignition result data, wherein the ignition control data includes the ignition nozzle number, the movement distance of the ignition nozzle, the ignition control voltage and frequency signal, and the ignition result data includes the ignition result, the fuel combustion condition, the engine speed and load signal, and the ignition nozzle operation state, and the ignition nozzle operation state is the fluctuation deviation value of the ignition control voltage and frequency signal compared with the voltage and frequency signal inside the ignition nozzle; The control module is used to control the ignition of the corresponding ignition nozzle according to the optimal ignition control parameters.

2. The intelligent ignition device for an aviation turbine engine with redundancy according to claim 1, characterized in that: It also includes a preprocessing module, which is used to collect historical data, preprocess the data and form a training set. The data preprocessing includes sequentially performing numerical processing and normalization processing on the data.

3. The intelligent ignition device for an aviation turbine engine with redundancy according to claim 2, characterized in that: A reinforcement learning model is built based on the reinforcement learning algorithm, and the prediction results are fed back to the reinforcement learning model so that the reinforcement learning model updates the neural network strategy according to the prediction results; new ignition control data is selected based on the updated neural network strategy and input into the ignition prediction model; the ignition control data of the ignition prediction model is iteratively updated using the reinforcement learning model until the detection result is optimal and the corresponding ignition control data is the optimal ignition control parameter.

4. The intelligent ignition device for an aviation turbine engine with redundancy according to claim 3, characterized in that: In the strategy of updating the neural network according to the prediction results, the reward function is used to enhance the probability of selecting each ignition control data that helps output a positive prediction result in the previous prediction result, and to reduce the probability of selecting the ignition control data that helps output a negative prediction result.

5. The intelligent ignition device for an aviation turbine engine with redundancy according to claim 1, characterized in that: The linear drive mechanism includes a mounting seat, an electromagnetic seat, a spring and a magnetic suction head. The outer casing is provided with a mounting seat, the ignition nozzle is slidably connected to the mounting seat, the inner side wall of the mounting seat is installed with an electromagnetic seat, the sliding end of the ignition nozzle is correspondingly installed with a magnetic suction head, and a spring is installed between the electromagnetic seat and the magnetic suction head.

6. An intelligent ignition device for an aviation turbine engine with redundancy according to any one of claims 1 to 5, characterized in that: A plurality of ignition nozzles are staggeredly arranged at the front end of the flame tube along the length direction, and a plurality of ignition holes are correspondingly staggeredly opened at the side wall of the flame tube along the length direction.

7. A method for intelligent ignition of an aviation turbine engine with redundancy, based on the device according to any one of claims 1 to 6, characterized in that: The following steps are involved: Step S1: collecting historical data and digitizing the historical data to form a training data set and a test data set, wherein the training data set includes digitized ignition control data and ignition result data; the ignition control data includes the ignition nozzle number, the movement distance of the ignition nozzle, the ignition control voltage and frequency signal, and the ignition result data includes the ignition result, fuel combustion status, engine speed and load signal, and ignition nozzle operation status; Step S2: training the data set and obtaining an ignition prediction model by using an integrated learning algorithm; Boostrap sampling with replacement is used to obtain several training subsets; a classification algorithm is used to independently train several base classifiers on each training subset; for each test sample in the test data set, several base classifiers are used to obtain several prediction values, and for each test sample, the final prediction result is obtained by majority voting; Step S3: using the ignition prediction model to predict the ignition result data of each ignition nozzle, and feeding the prediction results back to the reinforcement learning algorithm, iteratively optimizing the ignition control data of the ignition prediction model until the optimal ignition control parameters are obtained, that is, the optimal ignition nozzle number, the movement distance of the ignition nozzle, the ignition control voltage and the frequency signal are obtained; Step S4: Perform ignition control based on the optimal ignition control parameters.

8. The intelligent ignition method for an aviation turbine engine with redundancy according to claim 7, characterized in that: In the step S1, after the historical data is digitized, the data is normalized.

Images