Intelligent self-adaptive stall surge recognition method

Abstract

The application belongs to the field of aero-engines, and particularly relates to an intelligent self-adaptive stall surge identification method. The method comprises the following steps: step one, extracting aerodynamic instability characteristic values, wherein the aerodynamic instability characteristic values comprise stall time domain characteristic values, surge time domain characteristic values, waveform drop rate characteristic values, stall power characteristic values and surge power characteristic values; step two, establishing an artificial neural network model, and training the artificial neural network model by taking the aerodynamic instability characteristic values as input; and step three, identifying stall surge based on the trained artificial neural network model. The application has higher accuracy and stronger versatility, achieves the purpose of multi-feature data fusion, improves the accuracy of aerodynamic instability monitoring, constructs a specific structure of the artificial neural network, classifies and identifies instability types, and thus realizes the classification processing of the engine for different instability types.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of aero-engines, and particularly relates to an intelligent self-adaptive stall surge recognition method. BACKGROUND

[0002] The aerodynamic stability of an aero-engine seriously affects the flight safety of an aircraft. How to use a suitable detection method to quickly and accurately detect the stall and surge signals of the engine is a problem that must be considered in the process of developing engines by countries.

[0003] The mature aerodynamic instability detection system in China at present is a differential pressure type aerodynamic instability detection method. This is a mechanical analog type detection system. However, the mechanical analog system does not have advantages in the speed of aerodynamic instability recognition and the anti-interference ability, and shows unstable technical characteristics when being adapted to other types of engines. After years of technical development, a compressor instability detection system based on digital circuit and embedded technology has appeared in China. These systems also contain relatively advanced signal detection technologies such as wavelet analysis, power spectrum calculation, standard deviation analysis, EMD decomposition, and the design level of the aerodynamic instability detection equipment is gradually improving, but there are still deficiencies in intelligent monitoring, universality, instability classification, etc.

[0004] The existing engine aerodynamic instability detection method has three shortcomings:

[0005] a) The current aerodynamic instability detection system basically adopts a threshold trigger judgment mode, and the setting of the threshold parameter is strongly related to the type and state of the engine, which leads to the need for the aerodynamic instability detection system to constantly adjust the parameters according to the aerodynamic characteristics of each type of engine, and the system does not have universality;

[0006] b) The current aerodynamic instability detection system generally adopts a single characteristic value trigger mode, and the single characteristic value recognition method has certain defects, and there is a high risk of misjudgment for a specific compressor test;

[0007] c) The current aerodynamic instability detection system cannot classify and recognize the types of aerodynamic instability, so it cannot realize accurate control according to the instability type.

[0008] Therefore, it is desirable to have a technical solution to overcome or at least alleviate at least one of the above-mentioned deficiencies of the prior art. SUMMARY

[0009] The purpose of the present application is to provide an intelligent self-adaptive stall surge recognition method to solve at least one problem existing in the prior art.

[0010] The technical solution of the present application is:

[0011] An intelligent self-adaptive stall surge identification method, comprising:

[0012] Step one, extracting aerodynamic instability characteristic values, the aerodynamic instability characteristic values including stall time domain characteristic values, surge time domain characteristic values, waveform drop rate characteristic values, stall power characteristic values and surge power characteristic values;

[0013] Step two, establishing an artificial neural network model, training the artificial neural network model by taking the aerodynamic instability characteristic values as input;

[0014] Step three, identifying stall surge based on the trained artificial neural network model.

[0015] In at least one embodiment of the present application, in step one, the extracting of the aerodynamic instability characteristic values comprises:

[0016] Extracting stall time domain characteristic values, specifically:

[0017] Using a band-pass filter with a cutoff frequency f С1 = 40 Hz and f С2 = 200 Hz to digitally filter the compressor pulsating pressure signal P to obtain a first stall time domain characteristic signal P ps .

[0018] Using a low-pass filter with a cutoff frequency f С1 = 40 Hz to digitally filter the compressor pulsating pressure signal P to obtain a second stall time domain characteristic signal P bs .

[0019] According to the first stall time domain characteristic signal P ps and the second stall time domain characteristic signal P bs , a stall time domain characteristic value SA is obtained:

[0020] SA = k sa · |P ps / P bs |

[0021] Wherein, k sa is a stall time domain characteristic value parameter adjustment constant.

[0022] In at least one embodiment of the present application, in step one, the extracting of the aerodynamic instability characteristic values comprises:

[0023] Extracting surge time domain characteristic values, specifically:

[0024] In the previous measurement link, a low-pass filter with a cutoff frequency f С1 = 40 Hz is used to digitally filter the compressor pulsating pressure signal P to obtain a first surge time domain characteristic signal P bc(j-n);

[0025] In the current measurement link, the low-pass filter with the cut-off frequency f С1 = 40 Hz is used to digitally filter the pulsating pressure signal P of the compressor to obtain the second surge time-domain characteristic signal P bc(j) ;

[0026] According to the first surge time-domain characteristic signal P bc(j-n) and the second surge time-domain characteristic signal P bc(j) , a surge time-domain characteristic value CA is obtained:

[0027] CA = 2·k ca · | (P bc(j–n) - P bc(j) ) / (P bc(j–n) + P bc(j) ) |

[0028] Wherein, j is the current data point position, n is the number of data of a certain length, k ca is a surge time-domain characteristic value parameter adjustment constant.

[0029] In at least one embodiment of the present application, in step one, the aerodynamic instability characteristic value is extracted, including:

[0030] The waveform drop rate characteristic value is extracted, specifically:

[0031] In the previous measurement link, the low-pass filter with the cut-off frequency f С1 = 300 Hz is used to digitally filter the pulsating pressure signal P of the compressor to obtain the first waveform drop rate characteristic signal P bd(j-n) ;

[0032] In the current measurement link, the low-pass filter with the cut-off frequency f С1 = 300 Hz is used to digitally filter the pulsating pressure signal P of the compressor to obtain the second waveform drop rate characteristic signal P bd(j) ;

[0033] According to the first waveform drop rate characteristic signal P bd(j-n) and the second waveform drop rate characteristic signal P bd(j) , a waveform drop rate characteristic value PD is obtained:

[0034] PD = k pd · (P bd(j–n) - P bd(j) ) / (Δt·P bd(j–n) )

[0035] Wherein, j is the current data point position, n is the number of data of a certain length, Δt is the detection interval time, k pdThe wave form descending rate characteristic value parameter adjustment constant.

[0036] In at least one embodiment of the present application, in step one, the aerodynamic instability characteristic value extraction includes:

[0037] The stall power characteristic value is extracted, and specifically:

[0038] The compressor pulsation pressure signal P is normalized to obtain a normalized signal P ns .

[0039] According to the normalized signal P ns , the stall power characteristic value ES is obtained:

[0040]

[0041] Wherein, f sl is the lowest stall frequency, f sh is the highest stall frequency, N is the frequency number of the frequency domain signal extraction, k es is the stall power characteristic value parameter adjustment constant.

[0042] In at least one embodiment of the present application, in step one, the aerodynamic instability characteristic value extraction includes:

[0043] The surge power characteristic value is extracted, and specifically:

[0044] The compressor pulsation pressure signal P is normalized to obtain a normalized signal P ns .

[0045] According to the normalized signal P ns , the surge power characteristic value EC is obtained:

[0046]

[0047] Wherein, f cl is the lowest surge frequency, f ch is the highest surge frequency, N is the frequency number of the frequency domain signal extraction, k ec is the surge power characteristic value parameter adjustment constant.

[0048] In at least one embodiment of the present application, in step two, the artificial neural network model establishment includes:

[0049] The artificial neural network model is a BP neural network model, the artificial neural network model includes an input layer, a hidden layer and an output layer, each input is connected to the next layer through a weight w, and the output of the artificial neural network model is:

[0050] y=f(wp+b)

[0051] where f is a transfer function, p is an input vector, and b is a threshold vector.

[0052] In at least one embodiment of the present application, in the artificial neural network model:

[0053] The number of input layer nodes is set to 5, which are stall time domain characteristic value SA, surge time domain characteristic value CA, waveform drop rate characteristic value PD, stall power characteristic value ES, and surge power characteristic value EC, respectively;

[0054] The number of hidden layer nodes is set to 6;

[0055] The number of output layer nodes is set to 3, which contains three data states. When the output is (1, 0, 0), it indicates that surge occurs; when the output is (0, 1, 0), it indicates that stall occurs; and when the output is (0, 0, 1), it indicates that the data is normal;

[0056] The transfer function from the input layer to the hidden layer is a hyperbolic tangent S-shaped function f1:

[0057]

[0058] The transfer function from the hidden layer to the output layer is an exponential S-shaped function f2:

[0059]

[0060] In at least one embodiment of the present application, in the artificial neural network model:

[0061] In the forward transmission of information:

[0062] The output of the i-th neuron of the hidden layer is:

[0063]

[0064] The output of the k-th neuron of the output layer is:

[0065]

[0066] The error function is:

[0067]

[0068] In the backpropagation of error:

[0069] The change of the output layer weight value, from the i-th input to the k-th output, is:

[0070]

[0071] where,

[0072] delta ki = e k f2'

[0073] e k = t k -a2 k

[0074] Similarly, we can get:

[0075]

[0076] The change of the hidden layer weight value, from the jth input to the ith output, is:

[0077]

[0078] Similarly, we can get:

[0079] Delta b1 i = eta ij

[0080] Where, i = 1, 2...s1, k = 1, 2...s2, j = 1, 2...r, t k is the expected value, f1' is the first derivative of the hidden layer transfer function, and f2' is the first derivative of the output layer transfer function.

[0081] In at least one embodiment of the present application, in step three,

[0082] When the aerodynamic instability detection recognition result is normal data, the detection output result is 0V;

[0083] When the aerodynamic instability detection recognition result is stall data, the detection output result is 3V;

[0084] When the aerodynamic instability detection recognition result is surge data, the detection output result is 5V.

[0085] The present application has at least the following beneficial technical effects:

[0086] The intelligent self-adaptive stall and surge recognition method of the present application combines aerodynamic instability multi-feature value extraction and artificial neural network technology, uses the combination pattern of aerodynamic instability feature values for instability recognition, and discards the instability threshold recognition method, without the need to adjust parameters and recognition methods according to the use object, so that the aerodynamic instability recognition system has higher accuracy and stronger versatility; five kinds of aerodynamic instability feature values are extracted, and instability is judged through a neural network, achieving the purpose of multi-feature data fusion and improving the accuracy of aerodynamic instability monitoring; a specific structure artificial neural network is constructed to classify and identify the instability type (stall, surge), so as to realize the classification processing of the engine for different instability types. BRIEF DESCRIPTION OF DRAWINGS

[0087] Figure 1 is a flow chart of an intelligent adaptive stall surge recognition method according to an embodiment of the present application;

[0088] Figure 2 is a schematic diagram of a BP neural network model according to an embodiment of the present application;

[0089] Figure 3 is a schematic diagram of a neural network with one hidden layer according to an embodiment of the present application;

[0090] Figure 4 is a flow chart of a training program of an artificial neural network model according to an embodiment of the present application;

[0091] Figure 5 is a schematic diagram of a recognition result according to an embodiment of the present application. DETAILED DESCRIPTION

[0092] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the embodiments of the present application will be described in more detail below with reference to the drawings. In the drawings, the same or similar notations represent the same or similar elements or elements with the same or similar functions throughout. The described embodiments are some embodiments of the present application, not all embodiments. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application. The embodiments of the present application will be described in detail below with reference to the drawings.

[0093] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the scope of protection of the present application.

[0094] The drawings will be described below in conjunction with the Figures 1 to 5 The present application will be described in further detail.

[0095] The present application provides an intelligent adaptive stall surge recognition method, as shown in FIG. 1, comprising the following steps: Figure 1

[0096] ​Step one, extract the aerodynamic instability characteristic value, the aerodynamic instability characteristic value includes stall time domain characteristic value, surge time domain characteristic value, waveform drop rate characteristic value, stall power characteristic value and surge power characteristic value;

[0097] Step two, establish an artificial neural network model, and train the artificial neural network model by taking the aerodynamic instability characteristic value as input;

[0098] Step three, identify stall and surge based on the trained artificial neural network model.

[0099] The intelligent adaptive stall and surge identification method of the application first extracts five kinds of aerodynamic instability characteristic values, including:

[0100] The stall time domain characteristic value is extracted by using the compressor pulsating pressure signal, and is obtained through time domain band-pass filtering and corresponding signal processing calculation. The pulsating pressure data used comes from the low-pressure or high-pressure compressor outlet. Specifically:

[0101] A band-pass filter with a cutoff frequency f С1 = 40 Hz and f С2 = 200 Hz is used to digitally filter the compressor pulsating pressure signal P to obtain the first stall time domain characteristic signal P ps .

[0102] A low-pass filter with a cutoff frequency f С1 = 40 Hz is used to digitally filter the compressor pulsating pressure signal P to obtain the second stall time domain characteristic signal P bs .

[0103] According to the first stall time domain characteristic signal P ps and the second stall time domain characteristic signal P bs , the stall time domain characteristic value SA is obtained:

[0104] SA = k sa · |P ps / P bs |

[0105] Wherein, k sa is a stall time domain characteristic value parameter adjustment constant, which is set to 20 in this embodiment.

[0106] The surge time domain characteristic value is extracted by using the compressor pulsating pressure signal, and is obtained through time domain low-pass filtering and corresponding signal processing calculation. The pulsating pressure data used comes from the low-pressure or high-pressure compressor outlet. Specifically:

[0107] In the previous measurement link, a low-pass filter with a cutoff frequency f С1A low-pass filter with a frequency of 40Hz is used to digitally filter the compressor pulsating pressure signal P to obtain the first surge time-domain characteristic signal P. bc(j-n) ;

[0108] In the current measurement process, the cutoff frequency f is used. С1 A low-pass filter of 40Hz is used to digitally filter the compressor pulsating pressure signal P to obtain the second surge time-domain characteristic signal P. bc(j) ;

[0109] Based on the first surge time-domain characteristic signal P bc(j-n) and the second surge time-domain characteristic signal P bc(j) The surge time-domain eigenvalue CA is obtained:

[0110] CA = 2·k ca ∣(P bc(j–n) –P bc(j) ) / (P bc(j–n) +P bc(j) )|

[0111] Where j is the location of the currently collected data point, n is the number of data points of a certain length, and K ca The constant for adjusting the surge time-domain characteristic value parameter is set to 2 in this embodiment.

[0112] The waveform descent rate characteristic value is extracted using the compressor pulsating pressure signal, which is obtained through time-domain low-pass filtering and corresponding signal processing. The pulsating pressure data used comes from the low-pressure or high-pressure compressor outlet. Specifically:

[0113] In the previous measurement stage, the cutoff frequency f was used. С1 A low-pass filter with a frequency of 300Hz is used to digitally filter the compressor pulsating pressure signal P, resulting in the first waveform descent rate characteristic signal P. bd(j-n) ;

[0114] In the current measurement process, the cutoff frequency f is used. С1 A low-pass filter with a frequency of 300Hz is used to digitally filter the compressor pulsating pressure signal P, resulting in the second waveform descent rate characteristic signal P. bd(j) ;

[0115] Based on the first waveform's descent rate characteristic signal P bd(j-n) And the second waveform descent rate characteristic signal P bd(j) The waveform descent rate characteristic value PD is obtained:

[0116] PD = k pd ·(P bd(j–n) –P bd(j) ) / (Δt·P bd(j–n) )

[0117] wherein j is the current data point position, n is the number of data points of a certain length, Δt is the detection interval, k pd is a wave form drop rate characteristic value parameter adjustment constant. In this embodiment, the adjustment constant is set to 17.

[0118] The stall power characteristic value is extracted using the normalized pulsation pressure signal of the low pressure or high pressure compressor, and a certain stall frequency data is extracted in the frequency domain signal through fast Fourier transform (FFT), and the root mean square value is calculated to obtain. The pulsation pressure data used comes from the low pressure or high pressure compressor outlet. Specifically:

[0119] The compressor pulsation pressure signal P is normalized to obtain the normalized signal P ns ;

[0120] According to the normalized signal P ns , the stall power characteristic value ES is obtained:

[0121]

[0122] wherein f sl is the lowest stall frequency, f sh is the highest stall frequency, N is the frequency number of the frequency domain signal extracted, k es is a stall power characteristic value parameter adjustment constant. In this embodiment, the adjustment constant is set to 50.

[0123] The FFT calculation uses a data amount length of 15 detection time periods by default for calculation.

[0124] The surge power characteristic value is extracted using the normalized pulsation pressure signal of the low pressure or high pressure compressor, and a certain surge frequency data is extracted in the frequency domain signal through fast Fourier transform (FFT), and the root mean square value is calculated to obtain. The pulsation pressure data used comes from the low pressure or high pressure compressor outlet. Specifically:

[0125] The compressor pulsation pressure signal P is normalized to obtain the normalized signal P ns ;

[0126] According to the normalized signal P ns , the surge power characteristic value EC is obtained:

[0127]

[0128] wherein f cl is the lowest surge frequency, f ch is the highest surge frequency, N is the frequency number of the frequency domain signal extracted, k ec is a surge power characteristic value parameter adjustment constant. In this embodiment, the adjustment constant is set to 10.

[0129] FFT calculations are performed using a data volume of 15 detection time periods by default.

[0130] The intelligent adaptive stall surge identification method of this application uses a BP neural network model, which has a three-layer structure including an input layer, a hidden layer, and an output layer.

[0131] like Figure 2 The diagram shows a basic BP neural network model with R inputs. Each input is connected to the next layer through a weight w. The output of the artificial neural network model is:

[0132] y = f(wp + b)

[0133] Where f is the transfer function, p is the input vector, and b is the threshold vector.

[0134] The BP algorithm consists of two parts: forward propagation of information and backward propagation of error. During forward propagation, the input information is calculated layer by layer from the input layer through the hidden layers and then to the output layer. If the expected output is not obtained in the output layer, the algorithm turns to backward propagation. The error signal is transmitted back along the original connection path through the network to modify the weights of the neurons in each layer until the desired goal is achieved.

[0135] In a preferred embodiment of this application, the number of nodes in the input layer of the artificial neural network model is set to 5, which are the stall time-domain feature value SA, the surge time-domain feature value CA, the waveform descent rate feature value PD, the stall power feature value ES, and the surge power feature value EC, respectively; the number of nodes in the hidden layer is set to 6; and the number of nodes in the output layer is set to 3, which includes three data states: when the output is (1, 0, 0), it indicates that a surge has occurred; when the output is (0, 1, 0), it indicates that a stall has occurred; and when the output is (0, 0, 1), it indicates that the data is normal.

[0136] The transfer function from the input layer to the hidden layer is a hyperbolic tangent sigmoid function f1:

[0137]

[0138] The transfer function from the hidden layer to the output layer is an exponential sigmoid function f2:

[0139]

[0140] like Figure 3 The diagram shows a neural network with one hidden layer. The backpropagation (BP) algorithm is derived using this network.

[0141] In the positive transmission of information:

[0142] The output of the i-th neuron in the hidden layer is:

[0143]

[0144] The output of the kth neuron of the output layer is:

[0145]

[0146] The error function is:

[0147]

[0148] In the back propagation of the error:

[0149] The change of the output layer weight, for the weight from the ith input to the kth output, is:

[0150]

[0151] wherein,

[0152] δ ki = e k f2'

[0153] e k = t k -a2 k

[0154] Similarly, it can be obtained that:

[0155]

[0156] The change of the hidden layer weight, for the weight from the jth input to the ith output, is:

[0157]

[0158] Similarly, it can be obtained that:

[0159] Δb1 i = η ij

[0160] wherein, i = 1, 2...s1, k = 1, 2...s2, j = 1, 2...r, t k is the expected value, f1' is the first-order derivative of the hidden layer transfer function, and f2' is the first-order derivative of the output layer transfer function.

[0161] The intelligent self-adaptive stall surge identification method of the application, after the artificial neural network model is constructed, the model is trained. The training precision of the BP neural network is set to 10-5, the input layer data of the training is five characteristic values extracted, and each five characteristic values are trained as a group. The flow chart of the training program is as follows: Figure 4As shown. Based on the artificial neural network model, when the aerodynamic instability detection recognition result is normal data, the detection output result is 0V; when the aerodynamic instability detection recognition result is stall data, the detection output result is 3V; when the aerodynamic instability detection recognition result is surge data, the detection output result is 5V, as shown in the following table. Figure 5 As shown.

[0162] The intelligent self-adaptive stall surge recognition method of the present application combines aerodynamic instability multi-feature value extraction and artificial neural network technology, uses the combination pattern of aerodynamic instability feature values for instability recognition, and discards the instability threshold recognition mode, so that the aerodynamic instability recognition system has higher accuracy and stronger versatility without adjusting the threshold and recognition method according to the use object.

[0163] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or replacements within the technical scope disclosed by the present application can be easily thought of by those skilled in the art, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A smart adaptive stall surge recognition method, characterized in that, The application relates to a method for detecting and identifying aerodynamic instability, comprising the following steps: Step one, extracting aerodynamic instability characteristic values, wherein the aerodynamic instability characteristic values comprise stall time domain characteristic values, surge time domain characteristic values, waveform drop rate characteristic values, stall power characteristic values and surge power characteristic values; Step two, establishing an artificial neural network model, and training the artificial neural network model by taking the aerodynamic instability characteristic values as inputs; The artificial neural network model is a BP neural network model, and the artificial neural network model comprises an input layer, a hidden layer and an output layer, each input is connected to the next layer through a weight value w, and the output of the artificial neural network model is: y=f(wp+b) wherein f is a transfer function, p is an input vector, and b is a threshold value vector; In the artificial neural network model: the number of nodes in the input layer is set to 5, and the nodes are respectively the stall time domain characteristic values SA, the surge time domain characteristic values CA, the waveform drop rate characteristic values PD, the stall power characteristic values ES and the surge power characteristic values EC; the number of nodes in the hidden layer is set to 6; the number of nodes in the output layer is set to 3, and three data states are contained, wherein when the output is (1, 0, 0), it indicates that surge occurs, when the output is (0, 1, 0), it indicates that stall occurs, and when the output is (0, 0, 1), it indicates that the data is normal; the transfer function from the input layer to the hidden layer is a hyperbolic tangent S-shaped function f1: the transfer function from the hidden layer to the output layer is an exponential S-shaped function f2: Step three, identifying stall and surge based on the trained artificial neural network model. In step one, the aerodynamic instability characteristic values are extracted, comprising:

2. The intelligent adaptive stall surge recognition method of claim 1, wherein, extracting the stall time domain characteristic values, specifically: In step one, the aerodynamic instability characteristic values are extracted, comprising: The pulsation pressure signal P of the compressor is digitally filtered by a band-pass filter with a cut-off frequency f С1 = 40 Hz and f С2 = 200 Hz to obtain a first stall time-domain characteristic signal P ps ; The compressor pulsating pressure signal P is digitally filtered with a low-pass filter having a cut-off frequency f С1 = 40 Hz to obtain a second stall time-domain feature signal P bs ; According to the first stall time domain characteristic signal P ps and the second stall time domain characteristic signal P bs , a stall time domain characteristic value SA is obtained: SA = k sa • |P ps / P bs | where k sa is a stall time domain eigenvalue parameter adjustment constant.

3. The intelligent adaptive stall and surge recognition method of claim 2, wherein, extracting the surge time domain characteristic values, specifically: In step one, the aerodynamic instability characteristic values are extracted, comprising: In the previous measurement link, the low-pass filter with cut-off frequency f С1 = 40 Hz is used to digitally filter the pulsating pressure signal P of the compressor, and the first surge time-domain characteristic signal P bc(j-n) is obtained. In the current measurement link, the low-pass filter with cut-off frequency f С1 = 40 Hz is used to digitally filter the pulsating pressure signal P of the compressor to obtain the second surge time-domain characteristic signal P bc(j) . According to the first surge time-domain characteristic signal P bc(j-n) and the second surge time-domain characteristic signal P bc(j) , a surge time-domain characteristic value CA is obtained: CA = 2-k ca | (P bc(j–n) - P bc(j) ) / (P bc(j–n) + P bc(j) | Wherein, j is the current data point position, n is the data number of a certain length, K ca is the surge time domain characteristic value parameter adjustment constant.

4. The intelligent adaptive stall and surge recognition method of claim 3, wherein, extracting the waveform drop rate characteristic values, specifically: In step one, the aerodynamic instability characteristic values are extracted, comprising: In the previous measurement link, the low-pass filter with cut-off frequency f С1 = 300 Hz is used to filter the pulsating pressure signal P of the compressor to obtain the first waveform droop rate characteristic signal P bd(j-n) ; In the current measurement link, the low-pass filter with cut-off frequency f С1 = 300 Hz is used to digitally filter the pulsating pressure signal P of the compressor to obtain the second waveform droop rate characteristic signal P bd(j) ; According to the first waveform drop rate characteristic signal P bd(j-n) and the second waveform drop rate characteristic signal P bd(j) , a waveform drop rate characteristic value PD is obtained: PD = k pd • (P bd(j–n) – P bd(j) ) / (△t•P bd(j–n) ) Wherein, j is the current data point position, n is the data number of a certain length, △t is the detection interval time, k pd is the waveform drop rate characteristic value parameter adjustment constant.

5. The intelligent adaptive stall and surge recognition method of claim 4, wherein, extracting the stall power characteristic values, specifically: In step one, the aerodynamic instability characteristic values are extracted, comprising: The compressor pulsating pressure signal P is normalized to obtain a normalized signal P ns ; According to the normalized signal P ns , a stall power characteristic value ES is obtained: wherein f sl is the lowest stall frequency, f sh is the highest stall frequency, N is the number of frequencies extracted from the frequency domain signal, k es is a stall power eigenvalue parameter adjustment constant.

6. The intelligent adaptive stall and surge recognition method of claim 5, wherein, extracting the surge power characteristic values, specifically: In the artificial neural network model: The compressor pulsating pressure signal P is normalized to obtain a normalized signal P ns ; According to the normalized signal P ns , a surge power characteristic value EC is obtained: wherein f cl is the lowest surge frequency, f ch is the highest surge frequency, N is the number of frequencies extracted from the frequency domain signal, k ec is a surge power eigenvalue parameter adjustment constant.

7. The intelligent adaptive stall and surge recognition method of claim 6, wherein, in the forward transmission of information: the output of the i-th neuron in the hidden layer is: the output of the k-th neuron in the output layer is: the error function is: in the backward propagation of errors: the change of the output layer weight value is: wherein Similarly, it can be obtained that: δ ki = e k f2' e k = t k - a2 k the change of the hidden layer weight value is: Similarly, it can be obtained that: In step three, Δb1 i = η ij where i = 1, 2...s1, k = 1, 2...s2, j = 1, 2...r, t k E is the expected value, f1' is the first derivative of the hidden layer transfer function, and f2' is the first derivative of the output layer transfer function.

8. The intelligent adaptive stall and surge recognition method of claim 7, wherein, when the aerodynamic instability detection and identification result is normal data, the detection output result is 0V; when the aerodynamic instability detection and identification result is stall data, the detection output result is 3V; when the aerodynamic instability detection and identification result is surge data, the detection output result is 5V. ​

Images