A sensor set selection method for aero-engine fault diagnosis performance

Abstract

The application discloses a kind of sensor set selection methods for aero-engine fault diagnosis performance, comprising the following steps: obtaining initial data obtained by aero-engine component level model simulation, establish engine knowledge base according to systematized sensor selection strategy;According to the engine knowledge base established, using the current sensor set to judge the detectability of fault using sensor root mean square value over-limit method, after detecting fault, establish inverse model to isolate fault and calculate diagnosis performance value;According to the fault diagnosis performance value of sensor set and the expected sensor factor, calculate the optimal value of sensor set, use genetic algorithm to optimize iterative calculation, optimize and select sensor set;Considering the influence of uncertain factors, select the optimal sensor set from the optimized and selected sensor set.The application can consistently and quantitatively evaluate the fault diagnosis performance of sensor set, select the aero-engine sensor set with better comprehensive performance, and obtain the best monitoring and diagnosis effect.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of system control and simulation, and relates to a sensor set selection method for the fault diagnosis performance of an aero-engine. BACKGROUND

[0002] Currently, engine sensors are mainly based on control requirements. In order to further meet the requirements of fault diagnosis and health management, more sensors need to be added on the basis of the existing ones. With the development of sensor technology, it is possible to add new sensors. Traditional sensor selection methods mainly consider the performance characteristics of measurement, and little consideration is given to health management. The selection process is largely based on the experience of field experts, and there is a lack of consistent quantitative methods to evaluate the influence of measurement selection on diagnostic performance. In recent years, there have been many in-depth studies on the performance evaluation and selection of sensors. However, in the field of aero-engines, especially in the field of fault diagnosis and health management of aero-engine gas path components, there has been no application research based on systematic sensor selection strategies. Due to the limitations of its own characteristics, it is difficult to add a large number of sensors to an aero-engine, so it is of great value and great engineering significance to select the optimal sensor set. SUMMARY

[0003] The purpose of the present application is to solve the problems in the prior art and provide a sensor set selection method for the fault diagnosis performance of an aero-engine, which can consistently and quantitatively evaluate the fault diagnosis performance of the sensor set and realize the optimal installation of the sensor set of the aero-engine.

[0004] To achieve the above purpose, the technical scheme is adopted as follows:

[0005] A sensor set selection method for the fault diagnosis performance of an aero-engine, comprising the following steps:

[0006] Obtain initial data obtained by aero-engine component-level model simulation, and establish an engine knowledge base according to a systematic sensor selection strategy;

[0007] According to the established engine knowledge base, use the current sensor set to judge the detectability of faults using the sensor root mean square value over-limit method. After detecting the fault, an inverse model is established to isolate the fault and calculate the diagnostic performance value;

[0008] According to the fault diagnosis performance value of the sensor set and the expected sensor factor, the optimal value of the sensor set is calculated, and the genetic algorithm is used for optimization and iterative calculation to optimize and select the sensor set;

[0009] Considering the influence of uncertain factors, the optimal sensor set is selected from the optimized and selected sensor set.

[0010] Further, the knowledge base comprises system simulation models and system health information, the system simulation models are an aero-engine model and a fault diagnosis model, and the system health information is simulation fault data.

[0011] Further, in the process of establishing the knowledge base, the input health parameters of the component-level models under the nominal condition are all set to 1 to generate a nominal number, and the generation method of the fault data is as follows:

[0012] The fault is simulated by adjusting the changes of the two health parameters, i.e., the efficiency coefficient e and the flow coefficient f of the component, while the health parameters of other components maintain the nominal values, the adjustment amount of the efficiency and the flow when each component fails is determined by adjusting the formula, and the formula is as follows:

[0013]

[0014] f adj =e adj ×(f:e)

[0015] Wherein, f:e represents the failure rate, F mag represents the fault amplitude, e adj and f adj represent the component health parameter adjustment value.

[0016] Further, in the process of generating the knowledge base, the simulation data are processed according to the following formula:

[0017]

[0018] Wherein, y i represents the processed sensor value, y fault,i represents the measured value of the sensor when the fault occurs, y nominal,i represents the nominal value of the sensor, and σ i represents the standard deviation of the noise.

[0019] Further, in the process of judging the detectability of the fault by using the root mean square value over-limit method, the fault can be diagnosed if the over-limit rule is met, and the fault cannot be diagnosed if the over-limit rule is not met, and the diagnosis performance value is set to 0, and the root mean square calculation formula is as follows:

[0020]

[0021] Wherein, RMS k represents the root mean square of the kth candidate sensor set, and m k represents the number of sensors in the kth candidate sensor set.

[0022] Further, the inverse model is as follows:

[0023]

[0024] Where n represents the number of faults, m represents the number of sensors, and q represents the number of interpolation nodes in each fault j. Indicates fault x j The value at the interpolation node q, The constant coefficient influence factor is used to represent the impact of fault j on the measured parameters. This represents the estimated value of fault j. This represents the sensor's estimated value.

[0025] In the interval middle, The nonlinear optimization function lsqnonlin in Matlab is used to calculate the minimum value of the objective function. As a result of fault diagnosis.

[0026] Furthermore, the formula for calculating the diagnostic performance value is as follows:

[0027]

[0028] Among them, D k d represents the diagnostic performance value of sensor set k. j C represents the diagnostic performance value of fault j. j Z represents the weighting coefficient of fault j. j This represents the diagnostic performance metric for fault j.

[0029] Furthermore, the formula for calculating the sensor's feature value is as follows:

[0030] M k =P k D k U k

[0031]

[0032]

[0033] Among them, M k Merit represents the figure of merit of sensor set k. k P k This indicates that when the number of sensors in sensor set k deviates from the expected number N... d Decrease in time-optimal value; N K,actual N represents the actual number of sensors in the cascaded sensor array. desired W represents the desired number of sensors. penalty U represents the weighting coefficient. k This represents the average measurable effectiveness of m sensors in sensor set k.

[0034] Further, the genetic algorithm iterative optimization process is: first, the diagnostic performance value D of all candidate sensor sets is calculated k into the genetic algorithm program, and then the genetic algorithm program calculation is started, and the objective function is:

[0035]

[0036] Wherein, u i represents the characteristics of each sensor, N d represents the number of sensors in the expected sensor set, N k represents the number of sensors in the actual sensor set, m represents the number of sensors, C j represents the weight coefficient of fault j, Z jk represents the diagnostic performance measure of fault j, W k represents the weight coefficient.

[0037] Further, the expected sensor factors include engine component faults, fault weight coefficients, sensor costs, weights, power requirements and installation difficulty; the uncertain factors include sensor measurement noise, different engine operating points and different fault points.

[0038] Compared with the prior art, the present application has the following beneficial effects:

[0039] The present application provides a sensor set selection method for the fault diagnosis performance of an aero-engine. By comprehensively and quantitatively evaluating the diagnostic performance of the sensor set, a sensor set with better comprehensive performance, especially better fault diagnosis performance of the aero-engine gas path components, can be selected, so that the aero-engine can be installed with fewer sensors under multiple restrictions, achieving effective monitoring and optimal diagnosis effect. Moreover, all the calculation processes can be realized through the modularized program of the software, and have strong portability for other mechanical systems, which can consistently and quantitatively evaluate the fault diagnosis performance of the sensor set, realize the optimal sensor set installation of the aero-engine, and obtain the optimal monitoring and diagnosis effect. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0041] Figure 1 The flow chart of the sensor set selection method for the fault diagnosis performance of the aero-engine of the present application.

[0042] Figure 2Strategy flow chart for sensor selection of the present application.

[0043] Figure 3 Iteration flow chart for sensor selection of the present application.

[0044] Figure 4 Genetic algorithm iteration calculation flow chart of the present application.

[0045] Figure 5 Merit of each group of optimal sensor set calculated by genetic algorithm of the present application k Figure.

[0046] Figure 6 Sensor measurement parameter with added random noise of the present application.

[0047] Figure 7 Verification result figure of the benchmark sensor set of the present application.

[0048] Figure 8 Verification result figure of the benchmark sensor set with P25 sensor of the present application.

[0049] Figure 9 Verification result figure of the benchmark sensor set with P25, T5 sensor of the present application. DETAILED DESCRIPTION

[0050] The exemplary embodiments of the present application will be described hereinafter with reference to the accompanying drawings, in which the various specific details of the embodiments of the present application are given to provide a thorough understanding of the present application. It will be apparent, however, to those skilled in the art that the present application can be practiced without such specific details. In other instances, well-known structures and functions have not been described in detail in order to avoid obscuring the present application. The following description is presented to enable any person skilled in the art to make and use the present application.

[0051] It is apparent that the described embodiments are only part of the embodiments of the present application, and not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0052] It should be noted that the terminal involved in the embodiments of the present application can include, but is not limited to, a mobile phone, a personal digital assistant (PDA), a wireless handheld device, a tablet computer, a personal computer (PC), an MP3 player, an MP4 player, a wearable device (for example, smart glasses, a smart watch, a smart bracelet, etc.), a smart home device, and the like.

[0053] In addition, the term "and / or" used in the present document is only used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in the present document generally represents an "or" relationship between the front and rear associated objects.

[0054] The present application will be further described in detail below with reference to the accompanying drawings:

[0055] Referring to Figure 1 , the present application provides a sensor set selection method for aero-engine fault diagnosis performance, comprising the following steps:

[0056] A turbofan engine knowledge base is established, the knowledge base including a system simulation model and system health information, the system simulation model being an aero-engine model and a fault diagnosis model, providing data generation support for subsequent work; the health information mainly being simulation fault data, which can reflect fault characteristics, fault development, component / sensor reliability and sensor characteristics (such as noise, etc.);

[0057] Fault diagnosis is performed using the current sensor set and the diagnosis performance value is calculated, the root mean square value of the sensor is used to determine the detectability of the fault, and after the fault is detected, the fault is isolated and the diagnosis performance value D k is calculated, when the fault is not detectable, D k = 0; after the fault is detected, an inverse model is established to isolate the fault and output the diagnosis performance value, the principle of the inverse model to isolate the fault is to match the sensor output through parameter optimization, so as to determine the fault type and amplitude, wherein the fault characteristic that best matches the fault hypothesis is considered as the real fault;

[0058] The genetic algorithm is used to optimize and select the sensor set, the fault diagnosis result of the sensor set, the measurement residual and the fault weight coefficient are used to calculate the diagnosis performance value D k , then the number of sensors, cost, weight, power requirement, installation difficulty and other factors are taken into account in the calculation of the merit Merit k , the merit Merit k is taken as the optimization objective function, and the genetic algorithm is used for optimization and iterative calculation to select several better sensor sets;

[0059] The final selection is made by comprehensively considering various uncertain factors, including sensor measurement noise, different engine operating points and different fault points, the optimal sensor set is selected from the several better sensor sets selected by the genetic algorithm before, and the effectiveness and rationality of the sensor set optimization selection under different conditions are also verified.

[0060] Example 1:

[0061] The research object of the embodiment is a certain type of high-bypass-ratio, split-exhaust, dual-rotor civil turbofan engine. A method for selecting sensors for the performance of turbofan engine fault diagnosis is programmed for calculation by a modular program. A software program module structure diagram is shown in Figure 1 The detailed establishment steps are as follows:

[0062] Step 1: Establish a turbofan engine knowledge base

[0063] The knowledge base is the basis for iterative calculation and final selection. Initial data is obtained by simulation of an aero-engine component-level model based on GasTurb / Matlab. The entire program architecture is in accordance with the principles of a systematic sensor selection strategy, as shown in Figure 2 .

[0064] 1) Generation of nominal numbers: component faults are defined as the degradation of health parameters, and the input health parameters of the component-level model under nominal conditions are all 1.

[0065] 2) Generation of fault data: faults are simulated by adjusting the changes of the efficiency coefficient e and the flow coefficient f of the health parameters of the components, while the health parameters of other components remain at the nominal values. The formula is used to determine the adjustment amount of the efficiency and flow when each component fails. The adjustment formula is as follows:

[0066]

[0067] f adj =e adj ×(f:e) (1)

[0068] In the formula, f:e represents the failure rate, F mag represents the failure amplitude, e adj and f adj represent the component health parameter adjustment values.

[0069] 3) Generation of the knowledge base: to improve numerical stability, the simulation data is processed according to the following formula:

[0070]

[0071] In the formula, y i represents the processed sensor value, y fault,i represents the measured value of the sensor when a fault occurs, y nominal,i represents the nominal value of the sensor, and σ i represents the standard deviation of the noise.

[0072] Steps 2 and 3 are both sensor iterative selection processes, and the overall principle diagram is shown in Figure 3 .

[0073] Step 2: Fault diagnosis using current sensor set:

[0074] 1) Diagnosability judgment: Diagnosability is judged by using the root mean square (RMS) method. If the RMS is over the limit, the fault is diagnosable, otherwise, the fault is not diagnosable, and the diagnosis performance value is set to 0. The formula of RMS is as follows:

[0075]

[0076] where, RMSk k represents the root mean square of the kth candidate sensor set, m k represents the number of sensors in the kth candidate sensor set.

[0077] 2) Fault diagnosis: After diagnosability is judged, an inverse model is established to isolate the fault. For fault j, some interpolation nodes are selected The value of sensor i at each interpolation node q is obtained by engine simulation, and the inverse model is as follows:

[0078]

[0079] where, n represents the number of faults, m represents the number of sensors, q represents the number of interpolation nodes in each fault j, represents the value of fault x j at interpolation node q. represents the influence of fault j on the measured parameter using a constant coefficient influence factor, represents the estimate of fault j, represents the estimate of sensor.

[0080] In the interval , the value of is calculated as the fault diagnosis result by using the nonlinear optimization function lsqnonlin in Matlab to calculate the minimum value of the objective function.

[0081] Diagnosis performance value calculation: The calculation formula is as follows:

[0082]

[0083] where, D k represents the diagnosis performance value of sensor set k, d j represents the diagnosis performance value of fault j, C j represents the weight coefficient of fault j, Z j represents the diagnosis performance measure of fault j.

[0084] Step 3: Selecting better sensor set using genetic algorithm:

[0085] 1) Goodness-of-Mouth Calculation: Compared to the diagnostic performance value, the goodness-of-mouth value incorporates considerations of both the expected and actual number of sensors. The calculation formula is as follows:

[0086] M k =P k D k U k (6)

[0087]

[0088] In the formula, M k Merit represents the figure of merit of sensor set k. k P k Its function is to address the issue of the number of sensors in sensor set k deviating from the expected number N. d When, the merit value decreases, U k N represents the mean measurable utility of m sensors in sensor set k. desired N represents the actual number of sensors in the cascaded sensor array. desired W represents the desired number of sensors. penalty This represents the weighting coefficient; in this paper, we take W. penalty It is 0.1.

[0089]

[0090] In the formula, U k u represents the mean measurable utility of m sensors in sensor set k. i This represents the characteristics of each sensor; the magnitude of the value is determined by the user and may include factors such as cost, weight, power requirements, and ease of installation. In the initial verification phase, it is assumed that each u... i The values ​​are all 1.

[0091] 2) Genetic Algorithm Iterative Optimization: First, calculate the diagnostic performance value D of all candidate sensor sets. k Input the data into the program, and then start the genetic algorithm calculation. The genetic algorithm calculation interface is as follows: Figure 4 As shown.

[0092] Encoding Rules: In this example, there are 7 baseline sensors: T25, T3, PS3, WF, N1, N2, and T45, and 5 candidate sensors: T13, P13, P25, P5, and T5. Therefore, [a1a2 a3 a4 a5] can be used to represent the sensor status, where 1 represents selection and 0 represents non-selection. For example, [0 0 1 0 1] indicates that sensors P25 and T5 are selected, resulting in a total of 32 sensor sets. The objective function is:

[0093]

[0094] Figure 5 For the optimal value of each set of sensor sets calculated by the genetic algorithm, the two sets of better sensor sets selected by the genetic algorithm are the benchmark sensor set plus P25, the benchmark sensor set plus P25 and T5.

[0095] Step 4: Make a final selection by comprehensively considering:

[0096] The purpose of the final selection is to verify the effectiveness and rationality of the performance of the several sets of sensor sets selected in the previous genetic iteration selection process. The verification of the iteration selection process does not consider the influence of some uncertain factors, while the final selection process mainly considers these uncertain factors. The main uncertain factors are sensor measurement noise.

[0097] The method of verifying by considering noise in the final verification is to superimpose 500 times of normally distributed noise on the original knowledge base data to obtain an augmented data matrix, and then verify the effectiveness and rationality of the optimal sensor set under the conditions of denoising and not denoising. The augmentation method is as shown in Figure 6 .

[0098] Figure 7 For the verification results of the benchmark sensor set under the condition of denoising, including six precision charts of sensor fault detection, sensor set fault isolation results, fan fault amplitude estimation relative error, high-pressure compressor fault amplitude estimation relative error, high-pressure turbine fault amplitude estimation relative error, and low-pressure turbine fault amplitude estimation relative error.

[0099] Figure 8 For the verification results of the benchmark sensor set plus P25 sensor under the condition of denoising, including six precision charts of sensor fault detection, sensor set fault isolation results, fan fault amplitude estimation relative error, high-pressure compressor fault amplitude estimation relative error, high-pressure turbine fault amplitude estimation relative error, and low-pressure turbine fault amplitude estimation relative error.

[0100] Figure 9 For the verification results of the benchmark sensor set plus P25, T5 sensors under the condition of denoising, including six precision charts of sensor fault detection, sensor set fault isolation results, fan fault amplitude estimation relative error, high-pressure compressor fault amplitude estimation relative error, high-pressure turbine fault amplitude estimation relative error, and low-pressure turbine fault amplitude estimation relative error.

[0101] From Figure 7 , Figure 8 , Figure 9The detection curves of fan fault, high-pressure turbine fault and low-pressure turbine fault are all above the threshold value in the fault detection of the benchmark sensor set, but the HPC curves of high-pressure compressor fault are all below the threshold value, that is, fan fault, high-pressure turbine fault and low-pressure turbine fault can be detected, and high-pressure compressor fault cannot be detected; in the case of adding P25 and adding P25 and T5, all the detection curves including the detection curves of high-pressure compressor fault are above the threshold value, that is, all faults can be detected. Therefore, the detection performance of the sensor set after adding P25 and adding P25 and T5 is greatly improved.

[0102] From Figure 7 、 Figure 8 、 Figure 9 It can be obtained that the relative estimation errors of the fan fault amplitudes of the three sensor sets are not much different, and the average values are all at the level of 3%-5%; the average value of the relative estimation errors of the high-pressure compressor fault amplitudes of the benchmark sensor set is about 8%, and the average values of the relative errors after adding P25 and P25 and T5 are both reduced to the level of about 1.5%; the average value of the relative estimation errors of the high-pressure turbine fault amplitudes of the benchmark sensor set is about 2%, and is increased to about 5% after adding P25, and is reduced to about 0.5% after adding P25 and T5; the average value of the relative estimation errors of the low-pressure turbine fault amplitudes of the benchmark sensor set is about 3.75%, and is increased to about 12% after adding P25, and is increased to about 7% after adding P25 and T5.

[0103] The optimal value data obtained is finally verified and given in the following table. Table 1 is the optimal value verification result of the benchmark sensor set, Table 2 is the optimal value verification result of the sensor set after adding P25, and Table 3 is the optimal value verification result of the sensor set after adding P25 and T5.

[0104] Table 1 Optimal value verification of the benchmark sensor set

[0105]

[0106] Table 2 Optimal value verification of the sensor set after adding P25

[0107]

[0108] Table 3 Optimal value verification of the sensor set after adding P25 and T5

[0109]

[0110] The comparison of the average results of the reference sensor set, the sensor set with P25, the sensor set with P25 and T5, and the average results without denoising, wavelet denoising and no denoising can obtain that the reference sensor set and the sensor set with P25 and T5 are the optimal sensor sets in the comprehensive diagnosis performance considering the uncertain factors such as noise.

[0111] The preferred embodiments of the present application have been described above with the preferred embodiments, but the present application is not limited to the above examples, and various modifications and changes can be made by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for selecting a sensor set for aeroengine fault diagnosis performance, characterized in that, The method comprises the following steps: Obtaining initial data obtained by simulating an aero-engine component level model, and establishing an engine knowledge base according to a systematic sensor selection strategy; According to the established engine knowledge base, using the current sensor set to judge the detectability of the fault by using the sensor root mean square value over-limit method, detecting the fault, establishing an inverse model to isolate the fault and calculating a diagnosis performance value; According to the fault diagnosis performance value of the sensor set and an expected sensor factor, calculating a sensor set optimization value, using a genetic algorithm to perform iterative optimization calculation, and optimizing and selecting a sensor set; Considering the influence of uncertain factors, selecting an optimal sensor set from the optimized and selected sensor set; The calculation formula of the diagnosis performance value is as follows: wherein represents a diagnostic performance value of the sensor set , represents a diagnostic performance value of the fault , represents a weight coefficient of the fault , represents a diagnostic performance value of the fault ; The calculation formula of the sensor set optimization value is as follows: wherein represents the merit of the sensor set , , represents a merit decrease value when the number of sensors in the sensor set deviates from the expected number ; represents the number of sensors in the actual sensor set represents the expected number of sensors represents the weight coefficient represents the mean of the measurable utility of the m sensors in the sensor set k; The iterative optimization process of the genetic algorithm is as follows: first, the diagnostic performance values ​​of all candidate sensor sets are... The input is fed into the genetic algorithm program, and then the genetic algorithm program starts calculating. The objective function is: wherein represents a characteristic of each sensor, represents a number of sensors in a desired sensor set, represents a number of sensors in an actual sensor set, m represents a number of sensors, represents a fault a weight coefficient, represents a fault a diagnostic performance measure, represents a weight coefficient.

2. The method of claim 1, wherein, The knowledge base comprises a system simulation model and system health information, the system simulation model is an aero-engine model and a fault diagnosis model, and the system health information is simulation fault data.

3. The method of claim 1, wherein, During the process of establishing the knowledge base, the input end health parameters of the component level model under the nominal condition are all set to 1 to generate a nominal number, and the generation method of the fault data is as follows: The failure is simulated by adjusting the variation of two health parameters of the component, efficiency coefficient and flow coefficient , while the health parameters of other components maintain nominal values. The adjustment amount of efficiency and flow when each component fails is determined by adjusting formula as follows: wherein, represents a failure rate, represents a failure amplitude, and represents a component health parameter adjustment value.

4. The method of claim 1, wherein, During the process of establishing the knowledge base, the simulation data are processed according to the following formula: wherein represents the processed sensor value, represents the measured value of the sensor when malfunctioning, represents the nominal value of the sensor, represents the standard deviation of the noise.

5. The method of claim 1, wherein, During the process of judging the detectability of the fault by using the sensor root mean square value over-limit method, if the over-limit rule is diagnosable, and if the over-limit rule is not diagnosable, the diagnosis performance value is set to 0, and the root mean square calculation formula is as follows: wherein, RMSkdenotes the root mean square of the kth candidate sensor set, Sdenotes the processed sensor value, Nkdenotes the number of sensors in the kth candidate sensor set.

6. The method of claim 1, wherein, The inverse model is as follows: wherein represents the number of faults, represents the number of sensors, represents the value of each fault at the interpolation node, represents the value of the fault at the interpolation node , represents the influence of the fault on the measured parameter with a constant coefficient influence factor, represents the estimate of the fault , represents the estimate of the sensor; In the interval , , the nonlinear optimization function lsqnonlin in Matlab is used to calculate the minimum of the objective function as the fault diagnosis result.

7. The method of claim 1, wherein, The expected sensor factor comprises an engine component fault, a fault weight coefficient, a sensor cost, a weight, a power requirement and an installation difficulty; and the uncertain factors comprise sensor measurement noise, different engine working points and different fault points.

Images