Aero-engine pipeline damage mode identification method and system thereof

Abstract

The application belongs to the technical field of aero-engine pipeline damage mode identification, and discloses an aero-engine pipeline damage mode identification method and system. The method proposes secondary improvement of damage identification parameters, and eliminates the differential influence brought by different rotating speed vibration noise excitation in aero-engine operation. The application proposes a neural network method constructed by pipeline damage samples to carry out identification. The application solves the problem of inconsistent curvature magnitudes of each measuring point, the influence of parameter values affected by different working conditions of the aero-engine and different damage conditions of the pipeline, and the influence of curvature mutation caused by damage, and comprehensively judges the damage identification error of the samples by three new evaluation standards.

Description

Technical Field

[0001] This invention belongs to the field of aero-engine pipeline damage mode recognition technology, and particularly relates to an aero-engine pipeline damage mode recognition method and system. Background Technology

[0002] Common structural damage types in pipelines include surface abrasion, crack damage, and structural fracture. According to relevant literature analysis, surface abrasion often occurs at pipeline connections or fixed locations. Initially, it may appear as minor scratches, but with vibration and noise, wear pits will form on the pipeline surface, significantly reducing the pipeline's safe lifespan. Cracks are a more severe form of surface abrasion; due to the relatively thin pipeline, surface damage easily develops into structural damage. Fracture is a more serious form of structural damage, resulting from the development of surface abrasion and crack damage to a certain extent, and is positively correlated with engine vibration amplitude and duration. Therefore, it is evident that pipeline structural damage, from its initial stage to its final stage, is a gradual expansion of surface damage.

[0003] According to literature review, commonly used parameters for identifying pipeline structural damage include natural frequencies and modal displacements. These parameters are derived from data calculated using the structural vibration excitation equation. The biggest drawbacks of this method are as follows: the data is easily affected by vibration and noise generated by the different speeds of the aero-engine; the pipeline damage identification error is large and the accuracy is low, especially when multiple surface damages occur in the structure; and the identification capabilities of the relevant parameters vary significantly: modifying the natural frequency parameter to a natural frequency variation ratio parameter can identify the location of damage in the pipeline, but its ability to identify the degree of damage is poor; positioning the modal displacement as the first-order displacement with the largest vibration change also has a high ability to identify the location of damage but a poor ability to identify the degree of damage. Summary of the Invention

[0004] To overcome the problems existing in related technologies, the present invention discloses an embodiment of a method and system for identifying damage modes of aero-engine pipelines, specifically involving a method for identifying damage modes of aero-engine pipelines based on normalized interval curvature difference modal parameters.

[0005] The technical solution is as follows: A method for identifying the damage modes of aero-engine pipelines, which involves secondary improvement of damage identification parameters to eliminate the differential effects of vibration and noise excitation at different engine speeds during operation, and constructing a neural network using pipeline damage samples for aero-engine pipeline damage mode identification, including the following steps:

[0006] S1. Utilizing the structural characteristic that the first-order principal vibration mode of the pipeline is a bend, the displacement of the first-order bending mode is improved. After improvement by measuring the curvature mode value, the normalized curvature mode value, and the normalized curvature difference of the measuring points before and after damage, the normalized interval curvature difference of the measuring points of the pipeline sample is obtained as a parameter for identifying pipeline structural damage.

[0007] S2, using the normalized characteristic function interval curvature value MI before and after damage at adjacent measuring points. r As parameters for identifying pipeline structural damage, a BP neural network identification model for pipeline damage is constructed.

[0008] S3, based on the constructed pipeline damage BP neural network identification model, uses the identification error, average identification error, and error variance of the damage sample as error accuracy evaluation indicators to determine the pipeline damage sample identification indicators;

[0009] S4. Comparative analysis of different identification parameters: The accuracy of pipeline damage identification is compared and analyzed using the first-order mode displacement mode and curvature mode as parameters and the normalized interval curvature difference parameter.

[0010] In step S1, the improvement of the curvature mode normalization value includes:

[0011] Normalized interval curvature difference is used as the identification parameter for structural damage. The curvature values ​​of the pipeline measuring points are normalized according to the following formula. The identification error, average identification error, and identification error variance of structural damage are used as new evaluation standards for identification accuracy. This addresses the inconsistency in the curvature magnitude of measuring points under different operating conditions and vibration excitations. The normalization of the curvature of each measuring point is expressed as follows:

[0012]

[0013] In the formula: (D r ) * α is the normalized curvature value; α is the balance factor, which addresses the calculation bias caused by a parameter value of 0 during parameter value normalization. α is set to 0.9; D r Let D be the curvature of the pipeline structure damage characteristic function at measurement point r. r ) max For the minimum curvature of different damage models at the measurement point, (D) r ) max This represents the maximum curvature of different damage models at the measurement point.

[0014] In step S1, the normalized curvature difference of the measuring points before and after damage is improved to obtain the normalized interval curvature difference of the measuring points in the pipeline sample, which serves as a parameter for identifying pipeline structural damage, including:

[0015] The curvature difference identification parameter is used to characterize the feature changes of different types of damage in pipelines. The curvature difference identification parameter is expressed as follows:

[0016]

[0017] In the formula, (ΔD) r ) * The difference. Let be the curvature-normalized value of the characteristic function of the vibration response of the damaged pipeline at node r. The curvature normalized value of the characteristic function of the vibration response of the undamaged pipeline at node r;

[0018] Subtracting the difference in curvature of the normalized characteristic function before and after damage at adjacent measuring points reduces and eliminates the error caused by the sudden change in curvature at measuring points after pipeline damage. The expression for subtracting the difference in curvature of the normalized characteristic function before and after damage at adjacent measuring points is as follows:

[0019] MI r =Δ(D) r+1 ) * -Δ(D r ) *

[0020] In the formula, MI r The normalized characteristic function interval curvature value before and after damage at adjacent measuring points represents the value of this value. The larger this value is, the more significant the abrupt change in the difference between adjacent measuring points r and r+1 before and after damage, indicating a higher probability of pipeline damage; Δ(D r+1 ) * Let Δ(D) be the normalized curvature difference of the structural model before and after damage at measurement point r+1 along the axial direction. r ) * This represents the normalized curvature difference of the structural model before and after damage at the r-point along the axial direction.

[0021] In step S2, a pipeline damage BP neural network identification model is constructed, including:

[0022] The intermediate layer uses the function Transig(), and both the input and output layers use the sigmoid transfer function logsig(), with output values ​​in the range (0,1). The number of nodes in the intermediate layer is determined empirically using the following formula:

[0023]

[0024] In the formula, l is the number of nodes in the intermediate layer or hidden layer, n is the number of nodes in the input layer, m is the number of nodes in the output layer, and a is an empirical value;

[0025] The network is set to a maximum training iteration count of 1000, a sum of squared errors of 0.0001, and a learning rate of 0.5.

[0026] Furthermore, in the pipeline damage BP neural network identification model,

[0027] Let the number of input nodes be n, and the input value of the nth node be x. n The number of output nodes is m, and the expected output value of the m-th node is t. m The calculation method for the pipeline damage BP neural network identification model is as follows:

[0028] The output of the hidden layer node is:

[0029]

[0030] In the formula, Z l For hidden layer nodes, ω ln Let f() be the weight between the input node and the hidden node, and θ be the output function. l The threshold value is the output parameter, and n is the number of input nodes in the neural network.

[0031] The output node outputs:

[0032]

[0033] In the formula, y m For the output node, ω ml is the weight between the hidden layer node and the output node, and l is the number of intermediate or hidden layer nodes in the neural network.

[0034] In step S3, the expression for the identification error of the damaged sample is:

[0035]

[0036] In the formula, E u For the identification error of damaged samples, This corresponds to the actual network output of the u-th damage model sample. This represents the network's expected output for the u-th damage model sample, where T is the transpose sign and p is the vector dimension.

[0037] The expression for the average recognition error is:

[0038]

[0039] In the formula, The average identification error is represented by c, where c is the total number of pipeline surface damage model samples and u is the u-th pipeline surface damage model sample.

[0040] The expression for the error variance is:

[0041]

[0042] In the formula, S represents the error variance.

[0043] In step S4, the accuracy of pipeline damage identification is compared and analyzed using the first-order mode displacement mode and curvature mode as parameters, and the normalized interval curvature difference parameter is analyzed. This includes:

[0044] Step 1: Model the pipeline to be tested, and use a hammer to excite the measuring points to complete the vibration modal test of the pipeline, or use ANSYS simulation calculation to continuously obtain the vibration modal parameter values ​​of the pipeline throughout its use process; among which, the vibration modal parameter values ​​include: the natural frequency of the pipeline structure and the mode displacement of the pipeline.

[0045] Step 2: Using the central difference method, calculate the curvature mode value of each measuring point based on the obtained first-order mode displacement data of each measuring point in the pipeline;

[0046] Step 3: Introduce a balance factor α to normalize the curvature mode values ​​of each measuring point, eliminate the data magnitude difference caused by noise excitation from different working states of the pipeline, and make the curvature values ​​of the structural damage characteristic function comparable under different working conditions.

[0047] Step 4: To fully characterize the changes in features before and after pipeline damage, normalized interval curvature identification parameters are used instead of curvature parameters;

[0048] Step 5: In view of the problem that there may be abrupt changes in the curvature curves of various measuring points in the pipeline after the pipeline structure is damaged, in order to eliminate the influence of curvature abrupt change error caused by structural damage, the interval curvature parameter is improved to the difference of normalized interval curvature difference.

[0049] Step 6: By comparing the changes in the pipe modal parameter curvature and normalized interval curvature difference before and after damage at different locations and to different degrees in the pipe, it is confirmed that the normalized interval curvature difference modal parameter significantly characterizes the pipe damage status.

[0050] Step 7: Set the input and ideal output matrices for pipeline damage based on the damage location and degree of damage in the pipeline damage model;

[0051] Step 8: Determine the input layer and output layer data based on the input and ideal output matrices of pipeline damage, and estimate the hidden layer data based on empirical formulas;

[0052] Step 9: Construct an input-output network model for pipeline damage identification using a BP neural network, and perform simulation training to obtain the damage output of different damage samples, i.e., the actual output matrix of pipeline damage.

[0053] Step 10: Calculate the identification error, average identification error, and error variance of the damaged sample based on the formulas for identification error, average identification error, and identification error variance.

[0054] Step 11: Using curvature mode and first-order mode displacement as identification parameters, construct a BP neural network damage identification model according to the same idea, and obtain the identification error, average identification error, and error variance of the damage sample under different identification parameters;

[0055] Step 12: Analyze the identification capabilities of the normalized interval curvature difference, curvature mode, and first-order mode displacement identification parameters for single and double damage in pipeline structures from three perspectives: identification error, average identification error, and error variance. Confirm the damage identification capability of the normalized interval curvature difference parameter.

[0056] In step 12, the damage identification capability of the normalized interval curvature difference parameter is confirmed, including:

[0057] Using the non-destructive piping model M 00 And single-damage pipeline model M ij Given i = 1, 2, 3; j = 1, 2, 3, 4, 5, construct a damage identification network model. Use the normalized interval curvature difference as the identification parameter to obtain the input matrix p1 of the pipeline single damage identification BP neural network, which is expressed as:

[0058] p1 = [MI MI] 11 …MI 15 MI 21 …MI 25 MI 31 …MI 35 ] 11×16

[0059] In the formula, MI is the non-destructive pipeline structure model. 11 Based on the outer ring radius of the pipeline, the damage location is at the fan-shaped structure of the pipeline segment 23 with angles of (24.9°, 155.1°), and the radius thickness is reduced by 0.1 mm. MI 15 Based on the outer ring radius of the pipeline, the damage location is at the fan-shaped structure of the pipeline segment 23 with angles of (24.9°, 155.1°), and the radius thickness is reduced by 0.5 mm. MI 21 Based on the outer ring radius of the pipeline, the damage location is at a sector-shaped structure with angles of (24.9°, 155.1°) in the 45-degree segment of the pipeline, and the radius thickness is reduced by 0.1 mm. MI 25 Based on the outer ring radius of the pipeline, the damage location is at a sector-shaped structure with angles of (24.9°, 155.1°) in the 45-degree segment of the pipeline, and the radius thickness is reduced by 0.5 mm. MI 31 Based on the outer ring radius of the pipeline, the damage location is at the fan-shaped structure of the pipeline segment with angles of (24.9°, 155.1°) in segment 78, with the radius and thickness reduced by 0.1, 0.2, 0.3, 0.4, and 0.5 mm, respectively. MI 35 This is a single-damage pipeline model with the outer ring radius of the pipeline as the reference, the damage location being at the pipeline sector structure with angles of (24.9°, 155.1°) in segment 78 of the pipeline, and the radius and thickness reduced by 0.5 mm.

[0060] The output vector of the recognition network is defined as [xy]. T x is the length from the midpoint of the damaged location to the fixed end of the pipeline, and y is the ratio of the damaged thickness to the undamaged pipeline thickness.

[0061] M 11 The output vector corresponding to the model is [0.0745 0.1]. T M 35 The output vector corresponding to the model is [0.43000.5]. T Set M 00 The corresponding output vector is [0 0] T Define the ideal output matrix t1 of the single-damage recognition BP neural network as follows:

[0062]

[0063] The neural network based on curvature mode structural damage has input matrix p2 and output matrix t1, and a 12×13×2 network structure is established, expressed as follows:

[0064] p2 = [MI MI] 11 L MI 15 MI 21 …MI 25 MI 31 …MI 35 ] 12×16

[0065] The neural network for a single-damage pipeline, with the first-order modal displacement change rate as the identification parameter, uses p3 as the input and t1 as the output matrix, and establishes a 10×12×2 network structure. The expression is as follows:

[0066] p3 = [MI MI] 11 …MI 15 MI 21 …MI 25 MI 31 …MI 35 ] 10×16

[0067] The output for both of the above parameters is t1.

[0068] Furthermore, the method also includes:

[0069] The input matrix p4 of the BP neural network for dual-damage recognition is obtained using the normalized interval curvature difference as a parameter, and its expression is:

[0070] p4 = [MI MI] 10 …MI 15 MI20 …MI 25 ] 11×13

[0071] In the formula, MI 10 This is a pipeline damage model, with the damage location based on the outer ring radius of the pipeline. The damage thickness is reduced by 0.2 mm in both the 78th and 12th pipeline segments. (MI) 20 This is a pipeline damage model. The damage location is based on the outer ring radius of the pipeline, and the damage thickness decreases by 0.2 mm in the 78th segment and by 0.1 mm in the 12th segment.

[0072] The ideal output matrix t2 of the damage identification BP neural network for the dual-damage pipeline model is defined as follows:

[0073]

[0074] Based on curvature mode, a BP neural network for structural dual-damage recognition is established using input matrices p5 and t2, with a 12×13×4 network structure. The expression is as follows:

[0075] p5 = [MI MI] 10 …MI 15 MI 20 …MI 25 ] 12×13

[0076] The neural network input matrices p6 and t2 for a single-damage pipeline, with the first-order mode displacement change rate as the identification parameter, are established, and a 10×12×4 network structure is constructed. The expression is as follows:

[0077] p6 = [MI MI] 10 …MI 15 MI 20 …MI 25 ] 10×13 .

[0078] Another object of the present invention is to provide an aero-engine piping damage mode identification system, which implements the aero-engine piping damage mode identification method, and the system includes:

[0079] The pipeline structure damage identification parameter acquisition module utilizes the structural characteristic that the first-order principal vibration mode of the pipeline is a bend, improves the displacement of the first-order bending vibration mode, and obtains the normalized interval curvature difference of the pipeline sample measurement points through the measurement point curvature mode value, the normalized curvature mode value, and the normalized curvature difference of the measurement points before and after damage, as the pipeline structure damage identification parameter.

[0080] The pipeline damage BP neural network identification model construction module uses the normalized feature function interval curvature value MI before and after damage at adjacent measuring points. r A BP neural network identification model for pipeline damage was constructed using parameters for pipeline structural damage identification.

[0081] The pipeline damage sample identification index determination module, based on the constructed pipeline damage BP neural network identification model, uses the identification error, average identification error, and error variance of the damage sample as error accuracy evaluation indicators to determine the pipeline damage sample identification index.

[0082] The module compares and analyzes different identification parameters, using the first-order mode displacement mode and curvature mode as parameters to compare and analyze the accuracy of pipeline damage identification based on the normalized interval curvature difference parameter.

[0083] Combining all the above technical solutions, the beneficial effects of this invention are as follows:

[0084] (1) Based on the safety requirement that more than 50% of aero-engine in-flight shutdown incidents are caused by core engine external pipeline failures, and in view of the actual need for low accuracy of traditional aero-engine pipeline structure damage identification, this invention proposes an improved aero-engine pipeline structure damage identification method based on the identification of pipeline structure vibration mode displacement parameters, from first-order vibration mode displacement parameters, curvature mode parameters to interval curvature difference mode parameters.

[0085] (2) This invention solves the problem of inconsistent curvature magnitude at various measuring points by proposing new identification parameters and processing measures. The parameter values ​​are affected by different operating conditions and damage conditions of the aero-engine, as well as the curvature abrupt change caused by damage. The sample damage identification error is comprehensively evaluated by three new evaluation criteria.

[0086] (3) As a high-value product with extremely high safety requirements, the reliability and operational stability of the system components of aero engines are related to the reliability and safety of the engine, as well as the life and property safety of passengers and aircraft, and must be reliably guaranteed. The annual delivery of commercial aircraft will reach as many as 250 aircraft and 500 engines, not including the annual deployment of 1,000 military engines to meet market demand. This product will provide accurate pipeline damage detection and timely maintenance for the above engine products, which will be a market demand with great commercial value.

[0087] (4) Traditional structural damage identification based on modal parameters such as natural frequency, first-order mode displacement, and curvature can only identify the location or extent of damage and has a high identification error. This invention can accurately identify both single and double damage types in pipeline systems. The identification accuracy is far higher than that of other parameters, which effectively ensures the reliability of the pipeline system operation.

[0088] (5) Using the normalized interval curvature difference as a parameter to carry out structural damage identification can fully characterize different situations of structural damage, so that the damage status of pipeline structures at different locations or degrees can be well distinguished and characterized by this parameter. This avoids the fact that modal parameters such as natural frequency, 1-mode displacement, and curvature can only identify some damage features, greatly reducing the complexity of damage identification and improving the accuracy of damage identification. Attached Figure Description

[0089] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure;

[0090] Figure 1 This is a flowchart of the aero-engine pipeline damage mode identification method provided in the embodiments of the present invention;

[0091] Figure 2 This is a schematic diagram of the aircraft engine pipeline damage mode identification method provided in the embodiments of the present invention;

[0092] Figure 3 This is a diagram showing the structural damage occurring at a location with high curvature in the normalized curvature change before and after the occurrence of pipeline structural damage, provided in an embodiment of the present invention.

[0093] Figure 4 This is a diagram showing the structural damage occurring at a location with relatively low curvature in the normalized curvature change before and after the occurrence of pipeline structural damage, provided in an embodiment of the present invention.

[0094] Figure 5 This is a schematic diagram of the BP neural network provided in an embodiment of the present invention;

[0095] Figure 6 This is a simplified diagram of the straight pipe system provided in an embodiment of the present invention;

[0096] Figure 7 This is a simplified diagram of the fully constrained pipeline location provided in an embodiment of the present invention;

[0097] Figure 8 M is provided in the embodiments of the present invention. 1j Curvature values ​​at each measurement point in the model (j=1,2,3,4,5);

[0098] Figure 9 M is provided in the embodiments of the present invention. 1j (j=1,2,3,4,5) Normalized interval curvature difference diagram of each measurement point in the model;

[0099] Figure 10 M is provided in the embodiments of the present invention. 2j Curvature values ​​at each measurement point in the model (j=1,2,3,4,5);

[0100] Figure 11 M is provided in the embodiments of the present invention. 2j (j=1,2,3,4,5) Normalized interval curvature difference diagram of each measurement point in the model;

[0101] Figure 12 M is provided in the embodiments of the present invention. 3j Curvature values ​​at each measurement point in the model (j=1,2,3,4,5);

[0102] Figure 13 M is provided in the embodiments of the present invention. 3j (j=1,2,3,4,5) Normalized interval curvature difference diagram of each measurement point in the model;

[0103] Figure 14 M is provided in the embodiments of the present invention. 1n (n=0,1,2,3,4,5) Curvature values ​​at each measurement point in the model;

[0104] Figure 15 M is provided in the embodiments of the present invention. 1n (n=0,1,2,3,4,5) Normalized interval curvature difference map of each measurement point in the model;

[0105] Figure 16 M is provided in the embodiments of the present invention. 2n (n=0,1,2,3,4,5) Curvature values ​​at each measurement point in the model;

[0106] Figure 17 M is provided in the embodiments of the present invention. 2n (n=0,1,2,3,4,5) Normalized interval curvature difference map of each measurement point in the model;

[0107] Figure 18 This is an error map of each single damage sample based on different identification parameters provided in the embodiments of the present invention;

[0108] Figure 19 This is an error map of the identification of each double-damage sample based on different identification parameters provided in the embodiments of the present invention. Detailed Implementation

[0109] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0110] The innovation of this invention lies in:

[0111] 1. Since the characteristic function of pipeline damage curvature is not a linear function of the damage location but rather discrete data, the "central difference method" is used to transform the first-order mode displacement value of the measuring point into a curvature value, thus solving the problem that the curvature value of the measuring point cannot be directly calculated using the curvature function;

[0112] 2. A parameter normalization formula with a balance factor is proposed to eliminate the influence of the difference in the magnitude of curvature values ​​at different speeds and times, and to ensure that the maximum value of the parameter is within the range of (0-1) and not equal to 0. This solves the problem that the normalized parameter value may be 0, which may lead to the inability to calculate it later. At the same time, it eliminates the differential influence caused by vibration and noise excitation at different speeds during the operation of aero-engines.

[0113] 3. Based on multiple improvements to the curvature modal parameters, a curvature difference parameter was proposed to reduce and eliminate the influence of curvature abrupt error caused by structural damage. It also solved the problem that the curve formed by connecting the curvature differences of various measuring points in the pipeline structure has multiple abrupt change points when the pipeline structure damage occurs but is unknown.

[0114] 4. Normalized interval curvature difference is proposed as an identification parameter to replace first-order mode displacement, curvature mode parameters, etc., which significantly enhances the ability to characterize pipeline structure damage.

[0115] 5. A damage identification network is proposed to be constructed using a BP neural network. The data on possible or actual structural damage to the pipeline is used as input. As the data sample continues to increase, the structural damage identification capability will become stronger and the identification accuracy will become higher.

[0116] This invention proposes a secondary improvement to the damage identification parameters, eliminating the differential effects of vibration and noise excitation at different speeds during aero-engine operation. The method further improves the damage identification parameters to eliminate the differential effects of vibration and noise excitation at different speeds during aero-engine operation, and constructs a neural network using pipeline damage samples for aero-engine pipeline damage mode identification.

[0117] Example 1, such as Figure 1 As shown, the aero-engine pipeline damage mode identification method provided in this embodiment of the invention includes:

[0118] S1. Utilizing the structural characteristic that the first-order principal vibration mode of the pipeline is a bend, the displacement of the first-order bending mode is improved. After improvement by measuring the curvature mode value, the normalized curvature mode value, and the normalized curvature difference of the measuring points before and after damage, the normalized interval curvature difference of the measuring points of the pipeline sample is obtained as a parameter for identifying pipeline structural damage.

[0119] Significant changes in the normalized curvature of the measuring points before and after pipeline damage are visible. Figure 3In the normalized curvature changes before and after pipeline structural damage, the structural damage occurs at locations with higher curvature. Figure 4 The normalized curvature changes before and after pipeline structural damage show that the structural damage occurs at a location with lower curvature.

[0120] Where: M 00 For non-destructive piping model, M 15 For a pipeline model where the damage is located at one end of the pipeline, M 35 This is a pipeline model where the damage is located in the middle of the pipeline.

[0121] For example, improving the first-order bending mode displacement involves three steps: measuring the curvature mode value, the normalized curvature mode value, and the normalized curvature difference before and after damage. The final normalized interval curvature difference of the pipeline sample measurement points specifically includes:

[0122] Normalized interval curvature difference is used as the identification parameter for structural damage. The curvature values ​​of the pipeline measuring points are normalized according to the following formula. The identification error, average identification error, and identification error variance of structural damage are used as new evaluation standards for identification accuracy. To solve the problem of inconsistent curvature magnitudes at various measuring points under different operating conditions and vibration excitations, the curvature of each measuring point is normalized:

[0123]

[0124] In the formula: (D r ) * α is the normalized curvature value; α is the balance factor, which addresses the calculation bias caused by a parameter value of 0 during parameter value normalization. α is set to 0.9; D r Let D be the curvature of the pipeline structure damage characteristic function at measurement point r. r ) min For the minimum curvature of different damage models at the measurement point, (D) r ) max This represents the maximum curvature of different damage models at the measurement point.

[0125] It is understandable that the curvature normalization formula for each measuring point is used to normalize different parameters such as cost-type and benefit-type parameters, mainly to solve the problem that the value may be 0 after parameter normalization; it is also used here for parameter normalization, mainly to eliminate the problem of data magnitude difference caused by noise excitation caused by different working states of pipeline.

[0126] To address the issue of the inability to significantly characterize the changes in features of different types of pipeline damage, a curvature difference identification parameter is proposed:

[0127]

[0128] In the formula, (ΔD) r ) *The difference. Let be the curvature-normalized value of the characteristic function of the vibration response of the damaged pipeline at node r. The curvature normalized value of the characteristic function of the vibration response of the undamaged pipeline at node r;

[0129] It is understood that the innovative curvature difference identification formula proposed in this invention analyzes the changes in the structure before and after damage using normalized curvature values, and is mainly used to display the differences in curvature changes after structural damage.

[0130] To reduce and eliminate errors caused by sudden changes in the curvature of measuring points after pipeline damage, the difference in curvature of the normalized characteristic function before and after damage at adjacent measuring points is subtracted:

[0131] MI r =Δ(D) r+1 ) * -Δ(D r ) *

[0132] In the formula, MI r The normalized characteristic function interval curvature value before and after damage at adjacent measuring points represents the value of this value. The larger this value is, the more significant the abrupt change in the difference between adjacent measuring points r and r+1 before and after damage, indicating a higher probability of pipeline damage; Δ(D r+1 ) * Let Δ(D) be the normalized curvature difference of the structural model before and after damage at measurement point r+1 along the axial direction. r ) * This represents the normalized curvature difference of the structural model before and after damage at the r-point along the axial direction.

[0133] It is understood that the present invention innovatively proposes a formula for subtracting the curvature difference of the normalized characteristic function before and after damage at adjacent measuring points, which is used to propose that the structural damage morphology is characterized by the difference in the interval change value between measuring points instead of the curvature change value of the points.

[0134] S2, using the normalized characteristic function interval curvature value MI before and after damage at adjacent measuring points. r As parameters for identifying pipeline structural damage, a BP neural network identification model for pipeline damage is constructed.

[0135] The principle of BP neural network is as follows: Figure 5 As shown, the intermediate layer uses the function Transig(), and both the input and output layers use the sigmoid transfer function logsig(), with output values ​​in the range (0,1). The number of nodes in the intermediate layer is determined empirically, using the following formula:

[0136]

[0137] In the formula, l is the number of nodes in the intermediate layer or hidden layer, n is the number of nodes in the input layer, m is the number of nodes in the output layer, and a is an empirical value;

[0138] The network is set to a maximum training iteration count of 1000, a sum of squared errors of 0.0001, and a learning rate of 0.5.

[0139] In the pipeline damage BP neural network identification model, let the number of input nodes be n, and the input value of the nth node be x. n The number of output nodes is m, and the expected output value of the m-th node is t. m The calculation method of the pipeline damage BP neural network identification model is as follows:

[0140] The output of the hidden layer node is:

[0141]

[0142] In the formula, Z l For hidden layer nodes, ω ln Let f() be the weight between the input node and the hidden node, and θ be the output function. l The threshold value is the output parameter, and n is the number of input nodes in the neural network.

[0143] The output node outputs:

[0144]

[0145] In the formula, y m For the output node, ω ml is the weight between the hidden layer node and the output node, and l is the number of intermediate or hidden layer nodes in the neural network.

[0146] S3, based on the constructed pipeline damage BP neural network identification model, uses the identification error, average identification error, and error variance of the damage sample as error accuracy evaluation indicators to determine the pipeline damage sample identification indicators;

[0147] To address the issue of incomplete accuracy assessment in pipe damage identification, three major sample damage identification error evaluation indicators are used, namely, the identification indicators for pipe damage samples. The identification error of the damaged sample, the average identification error, and the error variance are used as the error accuracy evaluation indicators.

[0148] The expression for the identification error of damaged samples is:

[0149]

[0150] In the formula, E u For the identification error of damaged samples, This corresponds to the actual network output of the u-th damage model sample. This represents the network's expected output for the u-th damage model sample, where T is the transpose sign and p is the vector dimension.

[0151] The expression for the average recognition error is:

[0152]

[0153] In the formula, The average identification error is represented by c, where c is the total number of pipeline surface damage model samples and u is the u-th pipeline surface damage model sample.

[0154] The expression for the error variance is:

[0155]

[0156] In the formula, S represents the error variance.

[0157] It is understood that the innovative use of sample recognition error, average recognition error, and error variance as a combined measure of recognition accuracy in this invention demonstrates its precise recognition capability.

[0158] S4. Comparative analysis of different identification parameters: The accuracy of pipeline damage identification is compared and analyzed using the first-order mode displacement mode and curvature mode as parameters and the normalized interval curvature difference parameter.

[0159] The accuracy of pipeline damage identification was compared and analyzed using the first-order mode displacement mode and curvature mode as parameters and the normalized interval curvature difference parameter.

[0160] Example 2: This embodiment of the invention provides an aero-engine piping damage mode identification system, comprising:

[0161] The pipeline structure damage identification parameter acquisition module utilizes the structural characteristic of the pipeline vibration first-order principal mode being a bend to improve the first-order bending mode displacement. After improvement by measuring the curvature mode values, normalized curvature mode values, and normalized curvature difference values ​​of measuring points before and after damage, the normalized interval curvature difference value of the pipeline sample measuring points is obtained as the pipeline structure damage identification parameter. The measuring point curvature mode values ​​are obtained through vibration test or ANSYS simulation analysis.

[0162] The pipeline damage BP neural network identification model construction module is used to construct the normalized feature function interval curvature value MI before and after damage at adjacent measuring points. r A BP neural network identification model for pipeline damage was constructed using parameters for pipeline structure damage identification.

[0163] The pipeline damage sample identification index determination module is used to determine the pipeline damage sample identification index based on the constructed pipeline damage BP neural network identification model, using the identification error, average identification error, and error variance of the damage sample as error accuracy evaluation indexes.

[0164] The comparative analysis module for different identification parameters is used to compare and analyze the accuracy of pipeline damage identification based on the normalized interval curvature difference parameter using the first-order mode displacement mode and curvature mode as parameters.

[0165] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0166] The information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0167] To further illustrate the effects of the embodiments of the present invention, the following experiments were conducted.

[0168] like Figure 2 As shown, the aero-engine pipeline damage mode identification method provided in this embodiment of the invention includes:

[0169] Step 1: Model the pipeline to be tested, and use a hammer to excite the measuring points to complete the vibration modal test of the pipeline, or use ANSYS simulation calculation to continuously obtain the vibration modal parameter values ​​of the pipeline throughout its use, such as the natural frequency of the pipeline structure and the mode displacement of the pipeline (the first-order bending mode is the best).

[0170] Step 2: Using the central difference method, calculate the curvature mode value of each measuring point based on the obtained first-order mode displacement data of each measuring point in the pipeline;

[0171] Step 3: Introduce a balance factor a to normalize the curvature mode values ​​of each measuring point, eliminate the data magnitude difference caused by noise excitation from different working states of the pipeline, and make the curvature values ​​of the structural damage characteristic function comparable under different working conditions.

[0172] Step 4: To fully characterize the changes in features before and after pipeline damage, normalized interval curvature identification parameters are used instead of curvature parameters;

[0173] Step 5: In view of the problem that there may be abrupt changes in the curvature curves of various measuring points in the pipeline after the pipeline structure is damaged, in order to eliminate the influence of curvature abrupt change error caused by structural damage, the interval curvature parameter is improved to the difference of normalized interval curvature difference.

[0174] Step 6: By comparing the changes in the pipe modal parameter curvature and normalized interval curvature difference before and after damage at different locations and to different degrees in the pipe, it is confirmed that the normalized interval curvature difference modal parameter significantly characterizes the pipe damage status.

[0175] Step 7: Set the input and ideal output matrices for pipeline damage based on the damage location and degree of damage in the pipeline damage model;

[0176] Step 8: Determine the input layer and output layer data based on the input and ideal output matrix of pipeline damage, and estimate the hidden layer (intermediate layer) data based on empirical formulas;

[0177] Step 9: Construct an input-output network model for pipeline damage identification using a BP neural network, and perform simulation training to obtain the damage output of different damage samples, i.e., the actual output matrix of pipeline damage.

[0178] Step 10: Calculate the identification error, average identification error, and error variance of the damaged sample based on the formulas for identification error, average identification error, and identification error variance.

[0179] Step 11: Using curvature mode and first-order mode displacement as identification parameters, construct a BP neural network damage identification model according to the same idea, and obtain the identification error, average identification error, and error variance of the damage sample under different identification parameters;

[0180] Step 12: Analyze the identification capabilities of normalized interval curvature difference, curvature mode, and first-order mode displacement identification parameters for single and double damage in pipeline structures from three perspectives: identification error, average identification error, and error variance. Confirm that the normalized interval curvature difference parameter has the best damage identification capability and the highest accuracy.

[0181] For example, the aero-engine pipeline damage mode identification method provided in this embodiment of the invention includes the following:

[0182] (1) Simplified diagram of the system straight pipe, such as Figure 6 As shown.

[0183] A pipeline model M is established using a straight section of a certain aero-engine system. 00 The two ends of the pipeline, segments 12 and 1314, are fixed constraint ends, each 37mm in length. The total length of the pipeline is 860mm, symmetrically arranged around segment 78 (36mm long). The remaining sections of the pipeline, segments 23, 34, 45, 56, 910, 1011, 1112, and 1213, are all 75mm long. The outer diameter d1 of the pipeline is 19mm, the inner diameter d2 is 17mm, and the thickness Δ is 1mm. The coordinates of each measuring point in the pipeline, in mm, are as follows: 1 (0, 9.5, 0), 2 (0, 9.5, 37), 3 (0, 9.5, 112), 4 (0, 9.5, 187), 5 (0, 9.5, 262), 6 (0, 9.5, 337), 7 (0, 9.5, 412), 8 (0, 9.5, 448), 9 (0, 9.5, 523), 10 (0, 9.5, 598), 10 (0, 9.5, 673), 11 (0, 9.5, 748), 13 (0, 9.5, 823), 14 (0, 9.5, 860).

[0184] (2) Simplified diagram of the fully constrained location of the pipeline, as shown Figure 7 As shown.

[0185] The pipeline is constrained at sections 12 and 1314, which are shown visually; the constrained surface is the outer surface of the pipeline, with angles of (0, 24.9°), (155.1°, 180°), (180°, 204.9°), and (335.1°, 360°).

[0186] (3) Different damage models of pipelines are shown in Table 1.

[0187] Table 1 Pipeline Structure Model

[0188]

[0189]

[0190] (4) Characterization diagram of different single damage features of pipelines based on normalized interval curvature difference, such as Figure 8 For M 1j (j=1,2,3,4,5) Curvature values ​​at each measurement point in the model. Figure 9 For M 1j (j=1,2,3,4,5) Normalized interval curvature difference of each measurement point in the model; Figure 10 For M 2j (j=1,2,3,4,5) Curvature values ​​at each measurement point in the model. Figure 11 For M 2j (j=1,2,3,4,5) Normalized interval curvature difference values ​​for each measurement point in the model. Figure 12 For M 3j (j=1,2,3,4,5) Curvature values ​​at each measurement point in the model. Figure 13 For M 3j (j=1,2,3,4,5) Normalized interval curvature difference of each measurement point in the model.

[0191] The curvature values ​​of the measurement points and the normalized interval curvature differences between the measurement points for different damage levels in single-damage pipeline models at sections 23, 45, and 78 of the pipeline are shown in the figure. Figures 8-13 As shown.

[0192] analyze Figure 8 , Figure 10 , Figure 12 The calculated first-order vibration mode of each single-damage pipeline model is "Bend 1". It should be noted that the curvature calculation is based on the following formula:

[0193]

[0194] In the formula, D rLet be the curvature of the pipeline structure damage characteristic function at measurement point r, and d be the distance between two measurement points before and after the damage interval. Q” r Let Q be the curvature characteristic function of pipeline structural damage. r-1 Q represents the first-order mode displacement value of the pipeline damage model at the r-1 measuring point along the x-axis. r+1 Q represents the first-order mode displacement value of the pipeline damage model at point r+1 along the x-axis. r This represents the first-order mode displacement value of the pipeline damage model at point r along the x-axis.

[0195] Since the curvature values ​​at the two selected measurement points 1 and 14 for each model could not be calculated, the curvature values ​​of all measurement points were connected by a line starting from measurement point 2 and ending at measurement point 13, totaling 12 measurement points. Looking at the three figures, the damage differences between the various damage models were not significantly reflected in the curvature values ​​at each measurement point; the curvature values ​​at each measurement point remained essentially unchanged for models with different damage degrees. However… Figure 9 , Figure 11 , Figure 13 The opposite is true; the difference in normalized interval curvature at each measurement point in the model varies significantly for different damage levels. Figure 9 Measurement points 2, 3, 6, 7, 8, and 12 account for 54.55% of the total number of measurement points. Figure 11 All measurement points accounted for 100%; Figure 13 Measurement points 1, 6, 7, 8, 9, 10, and 12 account for 63.64% of all measurement points. It should be noted that the interval curvature difference calculation method proposed above results in the inability to calculate the corresponding interval curvature difference values ​​for measurement points 1, 13, and 14; only the interval curvature difference values ​​between measurement points 2 and 12 can be calculated. Therefore, the normalized interval curvature difference values ​​obtained better reflect the damage characteristics of the pipeline structure and are reasonable for identifying single damages in pipeline structures.

[0196] (5) Characterization diagram of different dual-damage features of pipelines based on normalized interval curvature difference ( Figures 14-17 ).

[0197] Dual-damage pipeline model M 1n One damage location is at pipe segment 78, where the thickness is reduced by 0.2 mm compared to the straight pipe radius. Other damage locations are at segments 12, 23, 34, 45, 56, and 67, where the thickness is reduced by 0.2 mm compared to the straight pipe radius. Therefore, a double-damaged pipe model M was established. 1n For dual-damage structure models with the same degree of damage, the curvature values ​​at the measurement points and the normalized interval curvature differences between relevant measurement points for each model are shown in [reference needed]. Figure 14 M 1n (n=0,1,2,3,4,5) Curvature values ​​at each measurement point in the model. Figure 15 M 1n(n=0,1,2,3,4,5) Normalized interval curvature difference of each measurement point in the model;

[0198] Dual-damage pipeline model M 2n One damage location is at pipe segment 78, where the thickness is reduced by 0.1 mm compared to the straight pipe radius. Other damage locations are at segments 12, 23, 34, 45, 56, and 67, where the thickness is reduced by 0.2 mm compared to the straight pipe radius. Therefore, a double-damaged pipe model M was established. 2n For dual-damage structure models with different damage levels, the curvature values ​​at measurement points and the normalized interval curvature differences between relevant measurement points in the pipeline models with different damage levels are shown below. Figure 16 M 2n (n=0,1,2,3,4,5) Curvature values ​​at each measurement point in the model. Figure 17 M 2n (n=0,1,2,3,4,5) Normalized interval curvature difference of each measurement point in the model.

[0199] analyze Figures 14-16 It can be seen that the first mode shape of each double-damaged pipeline model is still "Bend 1". From Figure 14 The curvature values ​​of the models before and after pipeline damage, as well as at different damage levels, did not change significantly; only eight measurement points showed some difference. Figure 16 As can be seen, the curvature values ​​at various measuring points changed significantly before and after the pipeline damage, particularly at measuring points 2, 3, 5, 6, 7, 8, 9, 10, and 12, accounting for 75% of all 12 measuring points. However, Figure 15 , Figure 17 The opposite is true; the normalized interval curvature difference values ​​at each measurement point in the model before and after pipeline damage, as well as for different degrees of damage, change significantly. Figure 15 Measurement points 2, 7, 8, 9, 10, 11, and 12 account for 63.64% of the total number of measurement points. Figure 17 Measurement points 2, 4, 5, 6, 7, 8, 9, 10, 11, and 12 account for 90.91% of all measurement points. Therefore, it can be concluded that the normalized interval curvature difference value reflects the damage characteristics of the pipeline structure well.

[0200] (6) Identification accuracy diagram of different identification parameters for pipeline damage samples ( Figure 18 (Identification error of each single damage sample based on different identification parameters).

[0201] Using the non-destructive piping model M 00 And single-damage pipeline model M ij (i = 1, 2, 3; j = 1, 2, 3, 4, 5) Construct a damage identification network model and use the normalized interval curvature difference as the identification parameter to obtain the input matrix p1 of the pipeline single damage identification BP neural network.

[0202] p1 = [MI MI]11 …MI 15 MI 21 …MI 25 MI 31 …MI 35 ] 11×16

[0203] In the formula, MI is the non-destructive pipeline structure model. 11 Based on the outer ring radius of the pipeline, the damage location is at the fan-shaped structure of the pipeline segment 23 with angles of (24.9°, 155.1°), and the radius thickness is reduced by 0.1 mm. MI 15 Based on the outer ring radius of the pipeline, the damage location is at the fan-shaped structure of the pipeline segment 23 with angles of (24.9°, 155.1°), and the radius thickness is reduced by 0.5 mm. MI 21 Based on the outer ring radius of the pipeline, the damage location is at a sector-shaped structure with angles of (24.9°, 155.1°) in the 45-degree segment of the pipeline, and the radius thickness is reduced by 0.1 mm. MI 25 Based on the outer ring radius of the pipeline, the damage location is at a sector-shaped structure with angles of (24.9°, 155.1°) in the 45-degree segment of the pipeline, and the radius thickness is reduced by 0.5 mm. MI 31 Based on the outer ring radius of the pipeline, the damage location is at the fan-shaped structure of the pipeline segment with angles of (24.9°, 155.1°) in segment 78, with the radius and thickness reduced by 0.1, 0.2, 0.3, 0.4, and 0.5 mm, respectively. MI 35 This is a single-damage pipeline model with the outer ring radius of the pipeline as the reference, the damage location being at the pipeline sector structure with angles of (24.9°, 155.1°) in segment 78 of the pipeline, and the radius and thickness reduced by 0.5 mm.

[0204] The output vector of the recognition network is defined as [xy]. T x is the length from the midpoint of the damaged location to the fixed end of the pipeline, and y is the ratio of the damaged thickness to the undamaged pipeline thickness.

[0205] M 11 The output vector corresponding to the model is [0.0745 0.1]. T M 35 The output vector corresponding to the model is [0.43000.5]. T Set M 00 The corresponding output vector is [0 0] T Define the ideal output matrix t1 of the single-damage recognition BP neural network as follows:

[0206]

[0207] From the input matrix p1 and the ideal output matrix t1, it can be seen that the input matrix is ​​an 11-dimensional column vector and the output matrix is ​​a 2-dimensional column vector. The input and output layer nodes of the BP neural network can be set to 11 and 2, respectively. The BP pipeline damage identification network is constructed using the MATLAB R2016b Neural Network Toolbox. The hidden layer nodes are set to 12 according to the trial and error method in Equation (9). The 3-layer network structure is 11×12×2. The network is trained using the Levenbrg-Marquardt algorithm. The maximum number of training steps is set to 1000. The sum of squared errors is set to 0.0001. The learning rate is set to 0.5. The input and output layers use the sigmoid transfer function logsin(), and the hidden layer uses the function Transig(). Training can be stopped when the sum of squared errors of the network reaches the expected value.

[0208] In addition, the neural network input matrix p2 and output matrix t1 based on the curvature mode structural damage are established, and a 12×13×2 network structure is constructed, while other network training parameters remain unchanged.

[0209] p2 = [MI MI] 11 …MI 15 MI 21 …MI 25 MI 31 …MI 35 ] 12×16

[0210] The neural network for a single-damaged pipeline, with the first-order mode displacement change rate as the identification parameter, uses p3 as the input and t1 as the output matrix and establishes a 10×12×2 network structure, while keeping other network training parameters unchanged.

[0211] p3 = [MI MI] 11 …MI 15 MI 21 …MI 25 MI 31 …MI 35 ] 10×16

[0212] The output for both of the above parameters is t1.

[0213] in, Figure 18Based on the identification errors of single-damage samples with different identification parameters, the analysis of sample identification errors shows that, for samples using curvature mode as the parameter, the identification errors of 9 samples are significantly greater than those using the normalized interval curvature difference parameter, the identification errors of 4 samples are close to those using the normalized interval curvature difference parameter, and only the identification errors of 2 samples are less than those using the normalized interval curvature difference parameter. Conversely, for samples using the first-order mode displacement mode as the parameter, the identification errors of 6 samples are significantly greater than those using the normalized interval curvature difference parameter, the identification errors of 5 samples are close to those using the normalized interval curvature difference parameter, and only the identification errors of 4 samples are less than those using the normalized interval curvature difference parameter. Therefore, the identification accuracy of the three parameters is in the order of "normalized interval curvature difference > first-order mode displacement mode > curvature mode".

[0214] Based on the analysis of the average identification error of the samples, the average identification error of the normalized interval curvature difference parameter is 1.7876% < the average identification error of the first-order mode displacement is 2.6353% < the average identification error of the curvature mode parameter is 7.4644%.

[0215] Based on the variance analysis of sample identification errors, the normalized interval curvature difference parameter is 3.3070 < 1st order mode displacement mode 3.3381 < curvature mode 8.7448. Therefore, the fluctuation of the damage identification error of the three parameters is in the order of curvature mode > 1st order mode displacement mode > normalized interval curvature difference, that is, the fluctuation of the damage sample identification error of the normalized interval curvature difference parameter is the smallest. From the analysis of the maximum identification error of the damaged sample, the curvature mode is 28.6565% > 1st order mode displacement mode 13.0835% > normalized interval curvature difference parameter is 11.3628%. Therefore, the identification error of the three parameters is in the order of "curvature mode > 1st order mode displacement mode > normalized interval curvature difference", that is, the damage identification error of the normalized interval curvature difference parameter is the smallest overall.

[0216] In summary: compared to curvature mode and first-order mode displacement modal parameters, the normalized interval curvature difference parameter has the highest accuracy in identifying single pipeline damage, followed by the first-order mode displacement and curvature mode parameters.

[0217] (7) Obtain the input matrix p4 of the dual-damage recognition BP neural network using the normalized interval curvature difference as a parameter;

[0218] p4 = [MI MI] 10 …MI 15 MI 20 …MI 25 ] 11×13

[0219] In the formula, MI 10 This is a pipeline damage model, with the damage location based on the outer ring radius of the pipeline. The damage thickness is reduced by 0.2 mm in both the 78th and 12th pipeline segments. (MI) 20This is a pipeline damage model. The damage location is based on the outer ring radius of the pipeline, and the damage thickness decreases by 0.2 mm in the 78th segment and by 0.1 mm in the 12th segment.

[0220] Define the ideal output matrix t2 of the damage identification BP neural network for the dual-damage pipeline model.

[0221]

[0222] From the input matrix p4 and the ideal output matrix t2, we know that the input matrix is ​​an 11-dimensional column vector and the output matrix is ​​a 4-dimensional column vector. Therefore, the input and output layers of the BP neural network can be set to 11 and 4 nodes, respectively, as shown in the following formula:

[0223]

[0224] In the formula, l is the number of nodes in the intermediate layer or hidden layer, n is the number of nodes in the input layer, m is the number of nodes in the output layer, and a is an empirical formula;

[0225] The trial-and-error method is used to set the hidden layer nodes to 12, and the 3-layer network structure is 11×12×4.

[0226] The BP neural network for structural dual-damage recognition based on curvature mode is used to input matrices p5 and t2, and a 12×13×4 network structure is established, while other network training parameters remain unchanged.

[0227] p5 = [MI MI] 10 …MI 15 MI 20 …MI 25 ] 12×13

[0228] The neural network input matrices p6 and t2 for a single-damage pipeline with the first-order mode displacement change rate as the identification parameter are established, and a 10×12×4 network structure is built, while other network training parameters are required to remain unchanged.

[0229] p6 = [MI MI] 10 …MI 15 MI 20 …MI 25 ] 10×13

[0230] in, Figure 19Based on the identification error maps of various dual-damage samples with different identification parameters, and based on the sample identification error analysis, for samples using curvature mode as the parameter, the identification error of 6 samples is much larger than that of the normalized interval curvature difference parameter, the identification error of 3 samples is close to that of the normalized interval curvature difference parameter, and only 3 samples have an identification error smaller than that of the normalized interval curvature difference parameter. Conversely, for samples using the first-order mode displacement mode as the parameter, the identification error of 8 samples is much larger than that of the normalized interval curvature difference parameter, the identification error of 2 samples is close to that of the normalized interval curvature difference parameter, and only 2 samples have an identification error smaller than that of the normalized interval curvature difference parameter. Overall, the damage identification error for all three parameters is characterized by "the normalized interval curvature difference being comparable to that of the curvature mode, and both being smaller than that of the first-order mode displacement mode."

[0231] Based on the analysis of the average identification error of the samples, the average identification error of the normalized interval curvature difference parameter is 3.1045% < the average identification error of the first-order mode displacement is 3.8629% < the average identification error of the curvature mode parameter is 4.2766%.

[0232] Analysis of the variance of sample identification errors shows that the first-order mode displacement mode (2.9822) < the curvature mode (4.6228) < the normalized interval curvature difference parameter (6.6043). Therefore, the fluctuation of the damage identification error for the three parameters is in the order of normalized interval curvature difference > curvature mode > first-order mode displacement mode, meaning that the normalized interval curvature difference parameter has the largest fluctuation in identification error. Analysis of the maximum identification error of the damage samples shows that the normalized interval curvature difference parameter (22.4646%) > the first-order mode displacement mode (12.6009%) > the curvature mode (9.1005%). Therefore, the identification error of individual samples for the three parameters is in the order of normalized interval curvature difference > first-order mode displacement mode > curvature mode, meaning that the overall identification error for the normalized interval curvature difference parameter is relatively large.

[0233] Therefore, the accuracy of dual-damage identification of pipelines with normalized interval curvature difference is comparable to that of curvature mode, and both are higher than that of first-order mode displacement. Since the sample identification variance is greater than that of curvature mode and first-order mode displacement mode, this leads to large fluctuations in identification error for different samples, and the maximum sample identification error becomes larger.

[0234] (8) The calculation results of this invention are shown in Tables 2-7.

[0235] Table 2. Simulation results of single-damage network for normalized interval curvature difference modal parameters.

[0236]

[0237] Table 3. Simulation output results of single-damage network for curvature modal parameters.

[0238]

[0239] Table 4. Single-damage network simulation output results of first-order mode displacement parameters.

[0240]

[0241] Table 5. Simulation output results of the dual-damage network with normalized interval curvature difference parameters.

[0242]

[0243]

[0244] Table 6. Network simulation output results for curvature modal parameters and dual-damage.

[0245]

[0246] Table 71 shows the simulation output results of the dual-damage network for the displacement modal parameters of the 11th order vibration mode.

[0247]

[0248]

[0249] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention and within the spirit and principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for identifying damage modes in aero-engine piping, characterized in that, This method involves a secondary improvement of the damage identification parameters to eliminate the differential effects of vibration and noise excitation at different speeds during aero-engine operation. It also constructs a neural network using pipeline damage samples to perform aero-engine pipeline damage mode identification, including the following steps: S1. Utilizing the structural characteristic that the first-order principal vibration mode of the pipeline is a bend, the displacement of the first-order bending mode is improved. After improvement by measuring the curvature mode value, the normalized curvature mode value, and the normalized curvature difference of the measuring points before and after damage, the normalized interval curvature difference of the measuring points of the pipeline sample is obtained as a parameter for identifying pipeline structural damage. S2, the normalized characteristic function interval curvature value before and after damage at adjacent measuring points. As parameters for identifying pipeline structural damage, a BP neural network identification model for pipeline damage is constructed. S3, based on the constructed pipeline damage BP neural network identification model, uses the identification error, average identification error, and error variance of the damage sample as error accuracy evaluation indicators to determine the pipeline damage sample identification indicators; S4, Comparative analysis of different identification parameters: The accuracy of pipeline damage identification is compared and analyzed using the first-order mode displacement mode and curvature mode as parameters and the normalized interval curvature difference parameter. In step S1, the improvement of the curvature mode normalization value includes: Normalized interval curvature difference is used as the identification parameter for structural damage. The curvature values ​​of the pipeline measuring points are normalized according to the following formula. The identification error, average identification error, and identification error variance of structural damage are used as new evaluation standards for identification accuracy. This addresses the inconsistency in the curvature magnitude of measuring points under different operating conditions and vibration excitations. The normalization of the curvature of each measuring point is expressed as follows: ； In the formula: The normalized curvature value; To balance the calculations, a parameter value of 0 is used to address the bias caused by parameter value normalization. The value is 0.9; For the characteristic function of pipeline structure damage in Curvature at the measuring point The minimum curvature of different damage models at the measurement point. The maximum curvature of different damage models at the measurement point; In step S1, the normalized curvature difference of the measuring points before and after damage is improved to obtain the normalized interval curvature difference of the measuring points in the pipeline sample, which serves as a parameter for identifying pipeline structural damage, including: The curvature difference identification parameter is used to characterize the feature changes of different types of damage in pipelines. The curvature difference identification parameter is expressed as follows: ； In the formula, The difference. The characteristic function of vibration response of damaged pipeline at the node The normalized value of curvature, The characteristic function of vibration response of undamaged pipeline at the node The normalized value of curvature; Subtracting the difference in curvature of the normalized characteristic function before and after damage at adjacent measuring points reduces and eliminates the error caused by the sudden change in curvature at measuring points after pipeline damage. The expression for subtracting the difference in curvature of the normalized characteristic function before and after damage at adjacent measuring points is as follows: ； In the formula, This represents the normalized characteristic function interval curvature value before and after damage at adjacent measuring points. The larger this value, the better the curvature of adjacent measuring points. The more significant the abrupt change in the difference before and after the damage, the higher the probability of pipeline damage. For the structural model along the axial direction Normalized curvature difference before and after damage to the structural model at the measuring point. For the structural model along the axial direction Normalized curvature difference before and after damage to the structural model at the measuring point.

2. The method for identifying damage modes in aero-engine piping according to claim 1, characterized in that, In step S2, a pipeline damage BP neural network identification model is constructed, including: The intermediate layer uses functions Both the input and output layers use sigmoid transfer functions. The output value is in Within the range; the number of intermediate layer nodes is determined empirically, using the following empirical formula: ； In the formula, This represents the number of nodes in the intermediate or hidden layers. The number of nodes in the input layer. This represents the number of nodes in the output layer. This is an empirical formula; The network is set to a maximum training iteration count of 1000, a sum of squared errors of 0.0001, and a learning rate of 0.

5.

3. The method for identifying damage modes in aero-engine piping according to claim 2, characterized in that, In the BP neural network identification model for pipeline damage Let the number of input nodes be , No. The input value of each node is The number of output nodes is , No. The expected output value of each node is The calculation method for the pipeline damage BP neural network identification model is as follows: The output of the hidden layer node is: ； In the formula, For hidden layer nodes, The weights between the input node and the hidden layer node are given. For the output function, For the output parameter threshold, The number of input nodes for the neural network; The output node outputs: ； In the formula, For output nodes, The weights between the hidden layer nodes and the output nodes. This represents the number of nodes in the intermediate or hidden layers of a neural network.

4. The method for identifying damage modes in aero-engine piping according to claim 1, characterized in that, In step S3, the expression for the identification error of the damaged sample is: ； In the formula, For the identification error of damaged samples, For the corresponding number The actual network output of a damage model sample For the corresponding number The expected network output for a sample of the damage model. It is the transpose symbol. Let the dimension be the vector. The expression for the average recognition error is: ； In the formula, For the average recognition error, This represents the total number of samples in the pipeline surface damage model. For the first A sample of a pipeline surface damage model; The expression for the error variance is: ； In the formula, This represents the error variance.

5. The method for identifying damage modes in aero-engine piping according to claim 1, characterized in that, In step S4, the accuracy of pipeline damage identification is compared and analyzed using the first-order mode displacement mode and curvature mode as parameters, and the normalized interval curvature difference parameter is analyzed. This includes: Step 1: Model the pipeline to be tested, and use a hammer to excite the measuring points to complete the vibration modal test of the pipeline, or use ANSYS simulation calculation to continuously obtain the vibration modal parameter values ​​of the pipeline throughout its use process; among which, the vibration modal parameter values ​​include: the natural frequency of the pipeline structure and the mode displacement of the pipeline. Step 2: Using the central difference method, calculate the curvature mode value of each measuring point based on the obtained first-order mode displacement data of each measuring point in the pipeline; Step 3: Introduce a balance factor The curvature mode values ​​of each measuring point are normalized to eliminate the data magnitude difference caused by noise excitation from different working states of the pipeline, so that the curvature values ​​of the structural damage characteristic function under different working conditions are comparable. Step 4: To fully characterize the changes in features before and after pipeline damage, normalized interval curvature identification parameters are used instead of curvature parameters; Step 5: In view of the problem that there may be abrupt changes in the curvature curves of various measuring points in the pipeline after the pipeline structure is damaged, in order to eliminate the influence of curvature abrupt change error caused by structural damage, the interval curvature parameter is improved to the difference of normalized interval curvature difference. Step 6: By comparing the changes in the pipe modal parameter curvature and normalized interval curvature difference before and after damage at different locations and to different degrees in the pipe, it is confirmed that the normalized interval curvature difference modal parameter significantly characterizes the pipe damage status. Step 7: Set the input and ideal output matrices for pipeline damage based on the damage location and degree of damage in the pipeline damage model; Step 8: Determine the input layer and output layer data based on the input and ideal output matrices of pipeline damage, and estimate the hidden layer data based on empirical formulas; Step 9: Construct an input-output network model for pipeline damage identification using a BP neural network, and perform simulation training to obtain the damage output of different damage samples, i.e., the actual output matrix of pipeline damage. Step 10: Calculate the identification error, average identification error, and error variance of the damaged sample based on the formulas for identification error, average identification error, and identification error variance. Step 11: Using curvature mode and first-order mode displacement as identification parameters, construct a BP neural network damage identification model according to the same idea, and obtain the identification error, average identification error, and error variance of the damage sample under different identification parameters; Step 12: Analyze the identification capabilities of the normalized interval curvature difference, curvature mode, and first-order mode displacement identification parameters for single and double damage in pipeline structures from three perspectives: identification error, average identification error, and error variance. Confirm the damage identification capability of the normalized interval curvature difference parameter.

6. The method for identifying damage modes in aero-engine piping according to claim 5, characterized in that, In step 12, the damage identification capability of the normalized interval curvature difference parameter is confirmed, including: Using a non-destructive pipeline model Single-damage pipeline model A damage identification network model was constructed, and the input matrix of the BP neural network for single-damage identification of pipelines was obtained by using the normalized interval curvature difference as the identification parameter. The expression is: ； In the formula, This is a non-destructive pipeline structure model. Based on the outer ring radius of the pipeline, the damage location is at the angle of segment 23 of the pipeline. A single-damage pipeline model with a reduced radius and thickness of 0.1 mm at the fan-shaped structure of the pipeline. Based on the outer ring radius of the pipeline, the damage location is at the angle of segment 23 of the pipeline. A single-damage pipeline model with a reduced radius and thickness of 0.5 mm at the fan-shaped structure of the pipeline. Based on the outer ring radius of the pipeline, the damage location is at an angle of 45 degrees on the pipeline. A single-damage pipeline model with a reduced radius and thickness of 0.1 mm at the fan-shaped structure of the pipeline. Based on the outer ring radius of the pipeline, the damage location is at an angle of 45 degrees on the pipeline. A single-damage pipeline model with a reduced radius and thickness of 0.5 mm at the fan-shaped structure of the pipeline. Based on the outer ring radius of the pipeline, the damage location is at the 78-degree angle of the pipeline segment. Single-damage pipeline models with reduced radius and thickness of 0.1, 0.2, 0.3, 0.4, and 0.5 mm at the fan-shaped structure of the pipeline. Based on the outer ring radius of the pipeline, the damage location is at the 78-degree angle of the pipeline segment. A single-damage pipeline model with a reduced radius and thickness of 0.5 mm at the fan-shaped structure of the pipeline; The recognition network output vector is defined as The distance from the midpoint of the damage location to the fixed starting end of the pipeline. This is the ratio of the damaged thickness to the undamaged pipe thickness. The output vector corresponding to the model is , The output vector corresponding to the model is ,set up The corresponding output vector is Define the ideal output matrix of a single-damage recognition BP neural network. The expression is: ； Neural network input matrix based on curvature mode structural damage and output matrix and establish The network structure is expressed as: ； The neural network input for a single damaged pipeline, using the first-order mode displacement change rate as the identification parameter. The output matrix is and establish The network structure is expressed as: ； The recognition output of the above two parameters is .

7. The method for identifying damage modes of aero-engine piping according to claim 6, characterized in that, The method also includes: The input matrix of the BP neural network for dual-damage recognition is obtained by using the normalized interval curvature difference as a parameter. The expression is: ； In the formula, This is a pipeline damage model, with the damage location based on the outer ring radius of the pipeline. The damage thickness is reduced by 0.2 mm in both the 78th and 12th pipeline segments. This is a pipeline damage model. The damage location is based on the outer ring radius of the pipeline, and the damage thickness decreases by 0.2 mm in the 78th segment and by 0.1 mm in the 12th segment. Define the ideal output matrix of the BP neural network for damage identification in a dual-damage pipeline model. The expression is: ； Input matrix of a BP neural network for structural dual-damage recognition based on curvature modes and input matrix and establish The network structure is expressed as: ； The neural network input matrix for a single damaged pipeline, using the first-order mode displacement change rate as the identification parameter. and input matrix and establish The network structure is expressed as: 。 8. A damage mode identification system for aero-engine piping, characterized in that, The system implements the aero-engine piping damage mode identification method according to any one of claims 1-7, and the system includes: The pipeline structure damage identification parameter acquisition module utilizes the structural characteristic that the first-order principal vibration mode of the pipeline is a bend, improves the displacement of the first-order bending vibration mode, and obtains the normalized interval curvature difference of the pipeline sample measurement points through the measurement point curvature mode value, the normalized curvature mode value, and the normalized curvature difference of the measurement points before and after damage, as the pipeline structure damage identification parameter. The pipeline damage BP neural network identification model construction module uses the normalized feature function interval curvature values ​​before and after damage at adjacent measuring points. As parameters for identifying pipeline structural damage, a BP neural network identification model for pipeline damage is constructed. The pipeline damage sample identification index determination module, based on the constructed pipeline damage BP neural network identification model, uses the identification error, average identification error, and error variance of the damage sample as error accuracy evaluation indicators to determine the pipeline damage sample identification index. The module compares and analyzes different identification parameters, using the first-order mode displacement mode and curvature mode as parameters to compare and analyze the accuracy of pipeline damage identification based on the normalized interval curvature difference parameter.

Images