A multi-source information fusion wind turbine blade damage mode recognition method

Abstract

The application discloses a wind turbine blade damage mode recognition method based on multi-source information fusion, and belongs to the field of wind turbine blade damage recognition. The wind turbine blade damage mode recognition method comprises the following steps: determining a recognition framework of the wind turbine blade damage mode, and acquiring three single-source signals used for recognizing a typical damage mode of the blade; calculating a basic probability assignment result of each single-source signal corresponding to the damage mode; solving the conflictivity between the evidences corresponding to each single-source signal and the information amount of each evidence; solving the credibility and the robustness of the evidence corresponding to each single-source signal; calculating the comprehensive weight of the evidence corresponding to each single-source signal; weighting to obtain a new evidence, and obtaining the basic probability assignment result of the final fault mode after fusing the new evidence. The application can assign more reasonable weights to the fused evidence and obtain more accurate fusion recognition results by constructing a new conflictivity solving formula and proposing an evidence robustness and propagation iteration mechanism.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine blade damage identification, and more specifically, it relates to a method for wind turbine blade damage pattern identification by multi-source information fusion. Background Technology

[0002] Wind energy, as a clean energy source, occupies an important position in the global energy structure. Wind turbine blades, as the core components for capturing wind energy, are constantly exposed to complex natural environments, enduring multiple loads such as airflow impact, temperature changes, humidity erosion, dust abrasion, and lightning strikes. This makes them prone to various forms of damage, including cracks, skin peeling, and leading-edge erosion. If blade damage is not identified and addressed promptly, it can lead to decreased wind energy capture efficiency and even safety accidents.

[0003] Traditional blade damage identification methods primarily rely on periodic manual inspections or monitoring based on single sensors. Manual inspections are time-consuming, labor-intensive, and highly susceptible to weather conditions; while single sensors (such as those relying solely on vibration signals) often struggle to extract effective fault characteristics in environments with wind and sand interference, easily leading to missed detections or false alarms. For example, vibration signals generated by wind and sand impacting the blade can be confused with vibration signals generated by blade damage, making it difficult for single vibration analysis to distinguish them. Furthermore, the complex operating environment of wind turbines means that single information sources often contain uncertainties and conflicts, making it difficult to comprehensively reflect the true health status of the blades. Therefore, there is an urgent need for a blade damage pattern recognition method that can integrate information from multiple sensors to improve recognition accuracy and robustness. Summary of the Invention

[0004] To address the problems of missed detections, false alarms, and insufficient robustness under environmental interference inherent in existing wind turbine blade damage identification methods that rely on single sensors, this invention provides a multi-source information fusion-based wind turbine blade damage pattern recognition method. This method first identifies typical blade damage patterns and collects multi-source signals. Then, it calculates the basic probability allocation of each single-source signal to the corresponding evidence for the damage pattern using a single-source probabilistic neural network model. Next, it calculates the conflict between the evidence corresponding to each single-source signal and the information content of the evidence itself. Based on this, it evaluates the credibility and robustness of each piece of evidence and determines its comprehensive weight in the fusion process. Finally, it performs weighted synthesis of the evidence based on the weights and iterative fusion using DS evidence theory to output the final damage pattern recognition result. This invention can effectively fuse multi-source heterogeneous information, significantly improving the accuracy and environmental adaptability of damage identification.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] This invention provides a method for identifying damage patterns in wind turbine blades based on multi-source information fusion, characterized by comprising the following steps:

[0007] Step 1: Determine the identification framework for wind turbine blade damage modes and acquire multiple single-source signals for identifying blade damage modes, including blade vibration signals, acoustic emission signals, and strain signals;

[0008] Step 2: Calculate the basic probability allocation results of the evidence corresponding to each single-source signal for each damage mode using a single-source probabilistic neural network model;

[0009] Step 3: Solve for the conflict between the evidence corresponding to each of the single-source signals and the information content of the evidence corresponding to each single-source signal;

[0010] Step 4: Based on the conflict between the evidence corresponding to each single-source signal calculated in Step 3, determine the credibility of the evidence corresponding to each single-source signal;

[0011] Step 5: Based on the information content of the evidence corresponding to each single-source signal calculated in Step 3, solve for the robustness of the evidence corresponding to each single-source signal;

[0012] Step Six: Based on the credibility of the evidence corresponding to each single-source signal determined in Step Four, and the robustness of the evidence corresponding to each single-source signal determined in Step Five, calculate the comprehensive weight of the evidence corresponding to each single-source signal.

[0013] Step 7: Based on the comprehensive weight of the evidence corresponding to each single-source signal calculated in Step 6, the basic probability allocation result of the evidence corresponding to each single-source signal calculated in Step 2 is weighted to obtain the basic probability allocation result of the new evidence to the damage mode. Through multiple fusions, the final basic probability allocation result of the typical blade failure mode is obtained.

[0014] Step 8: The damage mode corresponding to the maximum value in the basic probability allocation result of the final typical blade failure modes in Step 7 is the finally identified blade damage mode.

[0015] Furthermore, in step one, the framework for identifying wind turbine blade damage modes is represented as follows: F1 represents crack damage, F2 represents sand hole damage, F3 represents skin detachment, and F4 represents leading edge erosion. Multiple single-source signals were acquired to identify typical blade damage modes, including representative vibration signals, representative acoustic emission signals, and representative strain signals.

[0016] Furthermore, step two includes:

[0017] Step 2.1: Construct three single-source probabilistic neural network models respectively. The blade vibration signal, acoustic emission signal and strain signal are used as the inputs of the three single-source probabilistic neural network models respectively, and the damage mode of the wind turbine blade is used as the output of the three probabilistic neural network models.

[0018] Step 2.2: Train the corresponding single-source probabilistic neural network model constructed in Step 2.1 using the historical blade vibration signal, historical acoustic emission signal and historical strain signal of the wind turbine blade damage mode respectively;

[0019] Step 2.3: Input the blade vibration signal, acoustic emission signal, and strain signal obtained in Step 1 into the corresponding single-source probabilistic neural network model trained in Step 2.2. The output value of each single-source probabilistic neural network model is the basic probability allocation result m of the evidence corresponding to each single-source signal to the damage mode. i ( ), where m i Let F represent the basic probability allocation function of the evidence corresponding to the i-th single-source signal, i=1, 2, 3, m1 represent the basic probability allocation function of the evidence corresponding to the blade vibration signal, m2 represent the basic probability allocation function of the evidence corresponding to the acoustic emission signal, and m3 represent the basic probability allocation function of the evidence corresponding to the strain signal; j Represents the j-th damage mode, j=1, 2, 3, 4.

[0020] Furthermore, step three includes:

[0021] Step 3.1: Solve for the confidence function and plausibility function of the evidence corresponding to each single-source signal, expressed as:

[0022] (1)

[0023] In equation (1), Let represent the confidence function of the evidence corresponding to the i-th single-source signal for the j-th damage mode, i=1, 2, 3, j=1, 2, 3, 4; F represents the similarity function of the evidence corresponding to the i-th single-source signal to the j-th damage mode; j B represents the j-th damage mode; B belongs to the recognition framework. A subset; m i The basic probability allocation function represents the evidence corresponding to the i-th single-source signal;

[0024] Step 3.2: Solve for the conflict between the evidence corresponding to each of the single-source signals:

[0025] First, we need to resolve the conflict between evidence m1 corresponding to the blade vibration signal and evidence m2 corresponding to the acoustic emission signal. , is represented as:

[0026] (2)

[0027] In equation (2), α is the adjustment coefficient, α∈(0, 1)∪(1, ∞); Let i represent the belief similarity function of the evidence corresponding to the i-th single-source signal for the j-th damage mode, i=1,2,3,j=1,2,3,4; The belief likelihood function representing the evidence corresponding to the i-th single-source signal for the j-th damage mode. of Power; The belief similarity function of the evidence corresponding to the i-th single-source signal for the j-th damage mode. of Power;

[0028] The belief similarity function of the evidence corresponding to the i-th single-source signal for the j-th damage mode. The solution method is as follows:

[0029] (3)

[0030] In equation (3), Represents the damage pattern cardinality, with a value of 1; This represents the cardinality of the recognition framework, with a value of 4;

[0031] Further solving yields the following: the conflict between evidence m1 corresponding to the blade vibration signal and evidence m3 corresponding to the strain signal. The conflict between evidence m2 corresponding to the acoustic emission signal and evidence m1 corresponding to the blade vibration signal. The conflict between evidence m2 corresponding to the acoustic emission signal and evidence m3 corresponding to the strain signal. The conflict between evidence m3 corresponding to the strain signal and evidence m1 corresponding to the blade vibration signal. The conflict between evidence m3 corresponding to the strain signal and evidence m2 corresponding to the acoustic emission signal. ;

[0032] Step 3.3: Calculate the information content of the evidence corresponding to each single-source signal, expressed as:

[0033] (4)

[0034] In equation (4), Inf(m) i ) represents the information content of the evidence corresponding to the i-th single-source signal; m i ( ) represents the basic probability allocation result of the evidence corresponding to the i-th single-source signal to the j-th damage mode; mi F represents the basic probability allocation function for the evidence corresponding to the i-th single-source signal, where i = 1, 2, 3; j Represents the j-th damage mode, j=1, 2, 3, 4.

[0035] Furthermore, step four includes:

[0036] Step 4.1: Based on the conflict between the evidence corresponding to each pair of single-source signals obtained in Step 3.2, construct the conflict matrix D between the evidence corresponding to all single-source signals, expressed as:

[0037] (5)

[0038] Step 4.2: Calculate the difference between the evidence corresponding to each single-source signal and the evidence corresponding to other single-source signals, expressed as:

[0039] (6)

[0040] In equation (6), Diff(m) i ) represents the difference between the evidence corresponding to the i-th single-source signal and the evidence corresponding to the other two single-source signals, i=1, 2, 3;

[0041] Step 4.3: Obtain the support level (Sup) of the evidence corresponding to each single-source signal, expressed as:

[0042] (7)

[0043] In equation (7), Sup(m) i ) represents the support level of the evidence corresponding to the i-th single-source signal, i=1, 2, 3;

[0044] Step 4.4: Calculate the credibility Cre of the evidence corresponding to each single-source signal, expressed as:

[0045] (8)

[0046] In equation (8), Cre(m) i ) represents the credibility of the evidence corresponding to the i-th single-source signal, i=1, 2, 3.

[0047] Furthermore, step five includes:

[0048] Step 5.1: Based on the information content of the evidence corresponding to each single-source signal calculated in Step 3.3, construct an initial robustness graphical model. ,in, E represents the initial robustness of the evidence node corresponding to the i-th single-source signal. iqThe edge weights represent the initial robustness graph model;

[0049] First, the initial robustness of the evidence node corresponding to each single-source signal is calculated, expressed as:

[0050] (9)

[0051] In equation (9), Let Inf(m) represent the initial robustness of the evidence node corresponding to the i-th single-source signal, where i = 1, 2, 3; i ) represents the information content of the evidence corresponding to the i-th single-source signal, i=1, 2, 3;

[0052] Then, the edge weights E of the initial robust graphical model are calculated. iq , is represented as:

[0053] (10)

[0054] In equation (10), JD represents the Jousselme distance; m i Let m be the basic probability assignment function representing the evidence corresponding to the i-th single-source signal, where i = 1, 2, 3, m. q The basic probability assignment function represents the evidence corresponding to the adjacent single-source signals of the i-th single-source signal, where q = 1, 2, 3 and i ≠ q;

[0055] Step 5.2: Define the iterative propagation mechanism for robustness. The iterative propagation process for the robustness of the evidence node corresponding to the i-th single-source signal is expressed as:

[0056] (11)

[0057] In equation (11), Represents the damping factor; This represents the robustness of the evidence node corresponding to the i-th single-source signal after the (k+1)-th iteration; Let l represent the robustness of the evidence node corresponding to the neighboring single-source signal q of the i-th single-source signal after the k-th iteration; q represents the neighboring single-source signal of the i-th single-source signal; l represents the neighboring single-source signal of single-source signal q, E ql This represents the edge weight between the evidence node corresponding to the single-source signal q and the evidence node corresponding to the single-source signal l.

[0058] Step 5.3: Calculate the robustness of the evidence node corresponding to the i-th single-source signal after the iteration terminates. The iteration termination condition is expressed as:

[0059] (12)

[0060] At this point, the stability of the evidence node corresponding to the i-th single-source signal is represented by R. i ;R i (k) This represents the robustness of the evidence node corresponding to the i-th single-source signal after the k-th iteration.

[0061] Furthermore, step six includes:

[0062] Based on the credibility determined in step 4.4 and the robustness determined in step 5.3, the comprehensive weight of the evidence corresponding to each single-source signal in the fusion process is determined, expressed as:

[0063] (13)

[0064] In equation (13), w(m) i ) represents the comprehensive weight of the evidence corresponding to the i-th single-source signal in the fusion process, i=1,2,3; Cre(m i R represents the credibility of the evidence corresponding to the i-th single-source signal, where i = 1, 2, 3; i This represents the stability of the evidence node corresponding to the i-th single-source signal, where i = 1, 2, 3.

[0065] Furthermore, step seven includes:

[0066] Step 7.1: Weight the basic probability allocation result of the evidence corresponding to the damage mode calculated in Step 2 with the corresponding comprehensive weight calculated in Step 6 to obtain the basic probability allocation result of the new evidence to the damage mode, expressed as:

[0067] (14)

[0068] In equation (14), The basic probability assignment of the new evidence to the j-th damage mode is given, where j = 1, 2, 3, 4; w(m i ) represents the comprehensive weight of the evidence corresponding to the i-th single-source signal in the fusion process, i=1, 2, 3;

[0069] Step 7.2: Using the DS evidence theory, the new evidence obtained in Step 7.1 is fused with the basic probability allocation results of the damage modes twice to obtain the final basic probability allocation results of the typical blade failure modes.

[0070] The present invention has the following beneficial effects:

[0071] 1. This invention provides a method for identifying damage patterns of wind turbine blades based on multi-source information fusion. It abandons the simple conflict measurement method based solely on intersection in the traditional DS evidence theory and constructs a new formula to solve for the conflict between evidence. The new conflict solution formula can quantify the nonlinear and high-order degree of conflict, providing a more accurate data foundation for subsequent evidence processing and significantly improving the system's ability to identify real damage patterns.

[0072] 2. This invention constructs a robustness graph model and an iterative propagation mechanism. Through the graph model, this invention believes that the robustness of a piece of evidence depends not only on itself, but also on the influence of other neighboring node evidence. By introducing an iterative propagation mechanism, the robustness flows and updates continuously among the evidence nodes. This mechanism can automatically weaken the influence of evidence that is severely interfered with in the wind turbine service environment and enhance the robustness of the system.

[0073] 3. This invention proposes a dual decision-making mechanism that integrates credibility and robustness. Credibility focuses on assessing the credibility of evidence from the perspective of conflict, while robustness focuses on assessing the quality of evidence from the perspective of information content and group consensus. The combination of the two constitutes a more comprehensive evaluation system, making the determined comprehensive weight allocation more reasonable and increasing the recognition accuracy of the final fusion result. Attached Figure Description

[0074] Figure 1 This is a flowchart of a wind turbine blade damage pattern recognition method based on multi-source information fusion as described in this invention;

[0075] Figure 2 This is a schematic diagram of the iterative propagation process of evidence robustness;

[0076] Figure 3 This is a graph showing the changes in the robustness of the evidence before and after the iteration. Detailed Implementation

[0077] To better illustrate the purpose and advantages of the present invention, the invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0078] This invention provides a method for identifying damage patterns in wind turbine blades based on multi-source information fusion, such as... Figure 1 As shown, it includes the following steps:

[0079] Step 1: Determine the identification framework for wind turbine blade damage modes and acquire multi-source signals for identifying blade damage modes:

[0080] A framework for identifying typical damage modes of wind turbine blades is defined, and this framework is represented as follows: F1 represents crack damage, F2 represents pinhole damage, F3 represents skin detachment, and F4 represents leading edge erosion.

[0081] A multi-source signal S is acquired to identify typical damage modes of the blade. The multi-source signal S is composed of multiple single-source signals, i.e., S={S1,S2,S3}, where S1 represents the vibration signal, S2 represents the acoustic emission signal, and S3 represents the strain signal.

[0082] Step 2: Based on the wind turbine blade damage modes determined in Step 1 and the acquired multi-source signals, calculate the basic probability allocation results of each single-source signal for each damage mode using a single-source probabilistic neural network model.

[0083] Step 2.1: Construct three single-source probabilistic neural network models respectively. The blade vibration signal S1, acoustic emission signal S2 and strain signal S3 are used as inputs to the three single-source probabilistic neural network models respectively. The damage mode of the wind turbine blade is used as the output of the three probabilistic neural network models, namely crack damage F1, sand hole damage F2, skin peeling F3 and leading edge erosion F4.

[0084] Step 2.2: Train the corresponding single-source probabilistic neural network model constructed in Step 2.1 using the historical blade vibration signal, historical acoustic emission signal and historical strain signal of the wind turbine blade damage mode respectively;

[0085] Step 2.3: Input the online (i.e., the blade vibration signal, acoustic emission signal, and strain signal obtained in Step 1) into the corresponding single-source probabilistic neural network model trained in Step 2.2. The output value of each single-source probabilistic neural network model is the basic probability allocation result m of the evidence corresponding to each single-source signal to the damage mode. i ( ), where m i Let m1 represent the basic probability allocation function of the evidence corresponding to the i-th single-source signal, i=1, 2, 3. Specifically, m1 represents the basic probability allocation function of the evidence corresponding to the vibration signal, m2 represents the basic probability allocation function of the evidence corresponding to the acoustic emission signal, and m3 represents the basic probability allocation function of the evidence corresponding to the strain signal; F jLet F1 represent the j-th damage mode, where j = 1, 2, 3, 4. Specifically, F1 represents crack damage, F2 represents pinhole damage, F3 represents skin detachment, and F4 represents leading-edge erosion. Specifically, m1(F1) represents the basic probability allocation of evidence corresponding to blade vibration signals for crack damage; m1(F2) represents the basic probability allocation of evidence corresponding to blade vibration signals for pinhole damage; m1(F3) represents the basic probability allocation of evidence corresponding to blade vibration signals for skin detachment; m1(F4) represents the basic probability allocation of evidence corresponding to blade vibration signals for leading-edge erosion; m2(F1) represents the basic probability allocation of evidence corresponding to acoustic emission signals for crack damage; m2(F2) represents the basic probability allocation of evidence corresponding to acoustic emission signals for pinhole damage. The probability distribution results are as follows: m2(F3) represents the basic probability distribution result of the evidence corresponding to the acoustic emission signal for skin detachment; m2(F4) represents the basic probability distribution result of the evidence corresponding to the acoustic emission signal for leading edge erosion; m3(F1) represents the basic probability distribution result of the evidence corresponding to the strain signal for crack damage; m3(F2) represents the basic probability distribution result of the evidence corresponding to the strain signal for pinhole damage; m3(F3) represents the basic probability distribution result of the evidence corresponding to the strain signal for skin detachment; m3(F4) represents the basic probability distribution result of the evidence corresponding to the strain signal for leading edge erosion.

[0086] Step 3: Based on the basic probability allocation results of the evidence corresponding to each single source signal in Step 2, solve for the conflict between the evidence corresponding to each single source signal and the information content of the evidence corresponding to each single source signal.

[0087] Step 3.1: Solve for the confidence function of the evidence corresponding to each single-source signal. and similarity function The framework for identifying damage patterns in wind turbine blades is as follows: Then the trust function and similarity function Represented as:

[0088] (1)

[0089] In equation (1), Let represent the confidence function of the evidence corresponding to the i-th single-source signal for the j-th damage mode, i=1, 2, 3, j=1, 2, 3, 4; F represents the similarity function of the evidence corresponding to the i-th single-source signal to the j-th damage mode; j B represents the j-th damage mode; B belongs to the recognition framework. A subset; m iF1 represents the basic probability assignment function of the evidence corresponding to the i-th single-source signal; specifically, F1 represents crack damage, F2 represents sand hole damage, F3 represents skin detachment, and F4 represents leading edge erosion.

[0090] Step 3.2: Solve the conflict between the evidence corresponding to each of the single-source signals within the framework of wind turbine blade damage mode identification. Next, the conflict between evidence m1 corresponding to the blade vibration signal and evidence m2 corresponding to the acoustic emission signal is resolved. , is represented as:

[0091] (2)

[0092] In equation (2), α is the adjustment coefficient, α∈(0, 1)∪(1, ∞); Let i represent the belief similarity function of the evidence corresponding to the i-th single-source signal for the j-th damage mode, i=1,2,3,j=1,2,3,4; The belief likelihood function representing the evidence corresponding to the i-th single-source signal for the j-th damage mode. of Power; The belief similarity function of the evidence corresponding to the i-th single-source signal for the j-th damage mode. of Power;

[0093] BPF stands for Belief Plausibility Function. BPF transforms basic probability assignments into probabilistic forms by integrating the belief function Bel and the plausibility function Pl. The solution method for BPF is as follows:

[0094] (3)

[0095] In equation (3), Represents the damage pattern cardinality, with a value of 1; This represents the cardinality of the recognition framework, with a value of 4;

[0096] Following the same approach described in step 3.2, we can continue to solve for the results. , , , , The value of .

[0097] Step 3.3: In the recognition framework Next, the information content of the evidence corresponding to each single-source signal is calculated and expressed as follows:

[0098] (4)

[0099] In equation (4), Inf(m) i ) represents the information content of the evidence corresponding to the i-th single-source signal; m i ( ) represents the basic probability allocation result of the evidence corresponding to the i-th single-source signal to the j-th damage mode; m i F represents the basic probability allocation function for the evidence corresponding to the i-th single-source signal, where i = 1, 2, 3; j Let m1 represent the j-th damage mode, where j = 1, 2, 3, 4. Specifically, m1 represents the basic probability allocation function of the evidence corresponding to the vibration signal, m2 represents the basic probability allocation function of the evidence corresponding to the acoustic emission signal, and m3 represents the basic probability allocation function of the evidence corresponding to the strain signal; F1 represents crack damage, F2 represents pinhole damage, F3 represents skin detachment, and F4 represents leading edge erosion.

[0100] Step 4: Based on the conflict between the evidence corresponding to each single-source signal calculated in Step 3, determine the credibility of the evidence corresponding to each single-source signal.

[0101] Step 4.1: Based on the conflict between the evidence corresponding to each pair of single-source signals obtained in Step 3.2, construct the conflict matrix D between the evidence corresponding to all single-source signals, expressed as:

[0102] (5)

[0103] Step 4.2: Based on the conflict matrix D from Step 4.1, calculate the difference (Diff) between the evidence corresponding to each single-source signal and the evidence corresponding to other single-source signals, as follows:

[0104] (6)

[0105] In equation (6), Diff(m1) represents the difference between the evidence corresponding to the vibration signal and the evidence corresponding to the other two single-source signals; Diff(m2) represents the difference between the evidence corresponding to the acoustic emission signal and the evidence corresponding to the other two single-source signals; Diff(m3) represents the difference between the evidence corresponding to the strain signal and the evidence corresponding to the other two single-source signals.

[0106] Step 4.3: Based on the differences in Step 4.2, obtain the support level of the evidence corresponding to each single-source signal, expressed as follows:

[0107] (7)

[0108] In equation (7), Sup(m1) represents the support level of the evidence corresponding to the vibration signal; Sup(m2) represents the support level of the evidence corresponding to the acoustic emission signal; and Sup(m3) represents the support level of the evidence corresponding to the strain signal.

[0109] Step 4.4: Based on the support in Step 4.3, calculate the credibility (Credibility, Cre) of the evidence corresponding to each single-source signal, expressed as follows:

[0110] (8)

[0111] In equation (8), Cre(m1) represents the credibility of the evidence corresponding to the vibration signal; Cre(m2) represents the credibility of the evidence corresponding to the acoustic emission signal; and Cre(m3) represents the credibility of the evidence corresponding to the strain signal.

[0112] Step 5: Based on the information content of the evidence corresponding to each single-source signal calculated in Step 3, solve for the robustness of the evidence corresponding to each single-source signal;

[0113] Step 5.1: Based on the information content of the evidence corresponding to each single-source signal calculated in Step 3.3, construct an initial robustness graphical model. ;in, E represents the initial robustness of the evidence node corresponding to the i-th single-source signal. iq The edge weights represent the initial robustness graph model;

[0114] First, the initial robustness of the evidence node corresponding to each single-source signal is calculated, expressed as:

[0115] (9)

[0116] In equation (9), This represents the initial robustness of the evidence node corresponding to the vibration signal; This represents the initial robustness of the evidence node corresponding to the acoustic emission signal; Inf(m1) represents the initial robustness of the evidence node corresponding to the strain signal; Inf(m2) represents the information content of the evidence corresponding to the vibration signal; Inf(m3) represents the information content of the evidence corresponding to the acoustic emission signal; Inf(m4) represents the information content of the evidence corresponding to the strain signal; Inf(m5) represents the information content of the evidence corresponding to the strain signal; Inf(m6) represents the initial robustness of the evidence node corresponding to the strain signal; Inf(m7) represents the initial robustness of the evidence node corresponding to the strain signal; Inf(m8) represents the initial robustness of the evidence node corresponding to the strain signal; Inf(m9) represents the initial robustness of the evidence node corresponding to the strain signal; Inf(m1) represents the initial robustness of the evidence node corresponding to the vibration signal; Inf(m1) represents the initial robustness of the evidence node corresponding to the strain signal; Inf(m2) represents the initial robustness of the evidence node corresponding to the acoustic emission signal; Inf(m1) represents the initial i ) represents the information content of the evidence corresponding to the i-th single-source signal, i=1, 2, 3;

[0117] Then, the edge weights E of the initial robust graphical model are calculated. iq , is represented as:

[0118] (10)

[0119] In equation (10), JD represents the Jousselme distance, which is usually used to measure the distance between pieces of evidence. i Let m be the basic probability assignment function representing the evidence corresponding to the i-th single-source signal, where i = 1, 2, 3, m. q The basic probability assignment function represents the evidence corresponding to the adjacent single-source signals of the i-th single-source signal, where q = 1, 2, 3 and i ≠ q;

[0120] Step 5.2: Define the iterative propagation mechanism for robustness. The robustness of an evidence node consists of two parts: its initial robustness and the robustness inherited from other evidence nodes. The iterative propagation process can be represented as:

[0121] (11)

[0122] In equation (11), This represents the damping factor, reflecting the tendency of the evidence node to maintain initial robustness. This represents the initial robustness of the evidence node corresponding to the i-th single-source signal; This represents the robustness of the i-th single-source signal evidence node after the (k+1)-th iteration. The robustness of the evidence node corresponding to the adjacent single-source signal q of the i-th single-source signal after the k-th iteration;

[0123] Summation symbol In the formula, the subscript q represents the adjacent single-source signals of the i-th single-source signal, indicating that all evidence nodes except the i-th single-source signal evidence node itself are traversed. The denominator of the formula is " The subscript 'l' in the text represents the adjacent single-source signals of the single-source signal 'q', indicating that the traversal is performed except for evidence m. q All evidence nodes other than the node itself, E ql Representative evidence m q Node and all other evidence m l The edge weights of nodes;

[0124] Step 5.3: Calculate the robustness of the evidence node corresponding to the i-th single-source signal after iteration termination. The iterative propagation process stops when robustness propagation and redistribution reach a stable state, i.e., when the robustness inflow obtained by each evidence node from its neighboring nodes reaches equilibrium with its own state. The termination condition is expressed as:

[0125] (12)

[0126] R i (k) This represents the robustness of the evidence node corresponding to the i-th single-source signal after the k-th iteration; when evidence mi The iteration terminates when the maximum absolute value of the difference between the data from the k-th iteration and the data from the (k+1)-th iteration of a node is less than 1e-6, and the evidence m... i Once the robustness propagation and allocation of nodes reach a stable state, the stability of the evidence node corresponding to the i-th single-source signal is represented by R. i .

[0127] Step Six: Based on the credibility of the evidence corresponding to each single-source signal determined in Step Four, and the robustness of the evidence corresponding to each single-source signal determined in Step Five, calculate the comprehensive weight of the evidence corresponding to each single-source signal.

[0128] Based on the credibility determined in step 4.4 and the robustness determined in step 5.3, the comprehensive weight w of the evidence corresponding to each single-source signal in the fusion process is determined, expressed as:

[0129] (13)

[0130] In equation (13), w(m1) represents the comprehensive weight of the evidence corresponding to the vibration signal in the fusion process; w(m2) represents the comprehensive weight of the evidence corresponding to the acoustic emission signal in the fusion process; w(m3) represents the comprehensive weight of the evidence corresponding to the strain signal in the fusion process; R i This represents the stability of the evidence node corresponding to the i-th single-source signal, where i = 1, 2, 3;

[0131] Step 7: Based on the comprehensive weight of the evidence corresponding to each single-source signal calculated in Step 6, the basic probability allocation result of the evidence corresponding to each single-source signal calculated in Step 2 is weighted to obtain the basic probability allocation result of the new evidence to the damage mode. Through multiple fusions, the final basic probability allocation result of the typical blade failure mode is obtained.

[0132] Step 7.1: Multiply the evidence corresponding to each single-source signal by the corresponding comprehensive weight obtained in Step 6 to obtain the basic probability allocation result of the new evidence for the damage mode, expressed as:

[0133] (14)

[0134] In equation (14), The basic probability assignment of F1 for crack damage based on new evidence; The basic probability assignment results of the new evidence for F2 trachoma injury; The basic probability assignment results for skin detachment F3 based on new evidence; The basic probability assignment result of the new evidence for the leading edge erosion F4 is represented by w(m1); w(m2) represents the comprehensive weight of the evidence corresponding to the vibration signal in the fusion process; w(m3) represents the comprehensive weight of the evidence corresponding to the acoustic emission signal in the fusion process; w(m3) represents the comprehensive weight of the evidence corresponding to the strain signal in the fusion process.

[0135] Step 7.2: Using the DS evidence theory, the new evidence obtained in Step 7.1 is fused with the basic probability allocation results of the damage modes twice to obtain the final basic probability allocation results of the typical blade failure modes.

[0136] Step 8: The damage mode corresponding to the maximum value in the basic probability allocation result of the final typical blade failure modes in Step 7 is the finally identified blade damage mode.

[0137] Example:

[0138] This embodiment presents a method for identifying damage patterns in wind turbine blades through multi-source information fusion, comprising the following steps:

[0139] Step 1: Determine the typical damage modes of wind turbine blades to form an identification framework. The typical damage modes are crack damage (F1), pinhole damage (F2), skin detachment (F3), and leading edge erosion (F4). The identification framework for blade damage modes is then represented as follows: Three multi-source signals were acquired to identify typical damage modes of the blade: vibration signal S1, acoustic emission signal S2, and strain signal S3.

[0140] Step 2: Based on the blade vibration signal, acoustic emission signal and strain signal from Step 1, calculate the basic probability allocation result of the evidence corresponding to each single source signal for each damage mode;

[0141] Step 2.1: Based on the blade vibration signal S1, acoustic emission signal S2, and strain signal S3 from Step 1, construct three single-source probabilistic neural network models respectively. The blade vibration signal S1, acoustic emission signal S2, and strain signal S3 are used as the inputs of the three single-source probabilistic neural network models, and the damage mode of the wind turbine blade is used as the output of the three single-source probabilistic neural network models, namely, crack damage F1, sand hole damage F2, skin peeling F3, and leading edge erosion F4 are the outputs.

[0142] Step 2.2: Train the three single-source probabilistic neural network models constructed in Step 2.1 using historical blade vibration signals, acoustic emission signals, and strain signals of wind turbine blade damage patterns;

[0143] Step 2.3: Input the online blade vibration signal, acoustic emission signal, and strain signal into the single-source probabilistic neural network model trained in Step 2.2. The output value of each single-source probabilistic neural network model is the basic probability allocation result of the evidence corresponding to that single-source signal with respect to the damage mode; the basic probability allocation result of the evidence corresponding to each single-source signal with respect to the damage mode is expressed as m. i ( ), where i represents the type of single-source signal, i=1, 2, 3; m1 represents the basic probability allocation function of the evidence corresponding to the vibration signal; m2 represents the basic probability allocation function of the evidence corresponding to the acoustic emission signal; m3 represents the basic probability allocation function of the evidence corresponding to the strain signal; j represents the type of damage mode, j=1, 2, 3, 4, F j The damage modes are represented as follows: F1 represents crack damage, F2 represents pinhole damage, F3 represents skin detachment, and F4 represents leading edge erosion. In this embodiment, the basic probability allocation results of the evidence corresponding to each single-source signal for the damage mode are shown in Table 1.

[0144] Table 1. Basic probability allocation results of evidence corresponding to single-source signals

[0145]

[0146] As can be seen from the data in Table 1, m1 and m3 mainly support the occurrence of F1 (crack damage) and are the actual damage modes, while m2 supports the occurrence of F2 (sand hole damage). There is a clear conflict between m2 and other evidence.

[0147] Step 3: Based on the basic probability allocation results of the evidence corresponding to each single source signal in Step 2, solve for the conflict between the evidence corresponding to each single source signal and the information content of the evidence corresponding to each single source signal.

[0148] Step 3.1: Solve for the confidence function and plausibility function of the evidence corresponding to each single-source signal. The identification framework for blade damage patterns is then written as follows: In this embodiment, the trust function Bel and the plausibility function Pl of the evidence corresponding to each single-source signal are respectively:

[0149] m1:Bel(F1) = 0.5, Bel(F2) = 0.2, Bel(F3) = 0.2, Bel(F4) = 0.1;

[0150] Pl(F1) = 0.5, Pl(F2) = 0.2, Pl(F3) = 0.2, Pl(F4) = 0.1;

[0151] m2:Bel(F1) = 0.0, Bel(F2) = 0.8, Bel(F3) = 0.1, Bel(F4) = 0.1;

[0152] Pl(F1) = 0.0, Pl(F2) = 0.8, Pl(F3) = 0.1, Pl(F4) = 0.1;

[0153] m3: Bel(F1) = 0.55, Bel(F2) = 0.15, Bel(F3) = 0.1, Bel(F4) = 0.2;

[0154] Pl(F1) = 0.55, Pl(F2) = 0.15, Pl(F3) = 0.1, Pl(F4) = 0.2;

[0155] Step 3.2: Solve the conflict of evidence corresponding to each single-source signal within the framework of blade damage pattern identification. Next, the conflict of evidence corresponding to each of the two single-source signals is calculated. Preferably, the adjustment coefficient α in the conflict calculation formula is set to 2. Therefore, in this embodiment, the conflict of evidence corresponding to each of the two single-source signals is as follows:

[0156] BPRD2(m1, m2) = 6.8191, BPRD2(m1, m3) = 0.1565,

[0157] BPRD2(m2, m1) = 6.8191, BPRD2(m2, m3) = 7.0526,

[0158] BPRD2(m3, m1) = 0.1565, BPRD2(m3, m2) = 7.0526;

[0159] Step 3.3: Calculate the information (Inf) of the evidence corresponding to each single-source signal. In this embodiment, the information is as follows:

[0160] Inf(m1) = 5.8180, Inf(m2) = 2.5141, Inf(m3) = 5.3736;

[0161] Step 4: Based on the conflict of evidence corresponding to each single-source signal in Step 3, determine the credibility of the evidence corresponding to each single-source signal.

[0162] Step 4.1: Based on the conflict of evidence corresponding to each pair of single-source signals obtained in Step 3.2, construct the conflict matrix D between all the evidence corresponding to single-source signals. In this embodiment, the conflict matrix D is:

[0163] ;

[0164] Step 4.2: Based on the conflict matrix in Step 4.1, calculate the difference (Diff) between each piece of evidence and the other evidence. In this embodiment, these are:

[0165] Diff(m1) = 6.9755, Diff(m2) = 13.8717, Diff(m1) = 7.0291;

[0166] Step 4.3: Based on the differences in Step 4.2, obtain the support level for each piece of evidence. In this embodiment, these are:

[0167] Sup(m1) = 0.1434, Sup(m2) = 0.0721, Sup(m1) = 0.1387;

[0168] Step 4.4: Based on the support level in Step 4.3, calculate the credibility (Credibility, Cre) of each piece of evidence. In this embodiment, the credibility of each piece of evidence is as follows:

[0169] Cre(m1) = 0.4048, Cre(m2) = 0.2035, Cre(m1) = 0.3917;

[0170] Step 5: Based on the information content of the evidence corresponding to each single-source signal in Step 3, calculate the robustness of the evidence corresponding to the single-source signal;

[0171] Step 5.1: Construct an initial robustness graph model based on the information content of each piece of evidence calculated in Step 3. First, the initial robustness of each evidence node is calculated. In this embodiment, the initial robustness of each evidence node is:

[0172] , , ;

[0173] Then, the edge weights of the initial robustness graph model are calculated. In this embodiment, the edge weights are:

[0174] E 12 = 0.0816, E 13 = 0.8156,

[0175] E 21 = 0.0816, E 23 = 0.0000,

[0176] E31 = 0.8156, E 32 = 0.0000;

[0177] Step 5.2: Define the iterative propagation mechanism of robustness. The robustness of an evidence node consists of two parts: its own initial robustness and the robustness passed down from other evidence nodes.

[0178] Step 5.3: Calculate the robustness after the iteration terminates. When the robustness propagation and redistribution of the robustness graph model reaches a stable state, that is, when the robustness inflow obtained by each evidence node from its neighboring nodes reaches a balance with its own state, the iterative propagation process stops.

[0179] As a preferred option, damping factor Taking 0.85, the schematic diagram of the evidence robustness iterative propagation process is shown in the attached figure. Figure 2 As shown in the attached figure, the changes in the robustness of the evidence before and after the iteration are as follows. Figure 3 As shown.

[0180] In this embodiment, after the iteration terminates, the robustness of each piece of evidence is expressed as follows:

[0181] R1 = 0.4442, R2 = 0.1620, R3 = 0.3938;

[0182] Step Six: Based on the credibility of the evidence corresponding to each single-source signal determined in Step Four, and the robustness of the evidence corresponding to each single-source signal determined in Step Five, calculate the comprehensive weight of the evidence corresponding to each single-source signal.

[0183] Based on the evidence credibility determined in step 4.4 and the evidence robustness determined in step 5.3, the comprehensive weight of each piece of evidence in the fusion process is determined. In this embodiment, the comprehensive weight is:

[0184] w(m1) = 0.4899, w(m2) = 0.0898, w(m1) = 0.4203;

[0185] Step 7: Based on the comprehensive weight of the evidence corresponding to each single source signal calculated in Step 6, the basic probability allocation results of the evidence corresponding to each original single source signal for the damage mode are weighted to obtain new evidence. The new evidence is then fused to obtain the final basic probability allocation results of the typical blade failure modes.

[0186] Step 7.1: Multiply the evidence corresponding to each single-source piece of information by the comprehensive weight from Step 6, and then weight them to obtain new evidence, represented as:

[0187] , , , ;

[0188] Step 7.2: Using the DS evidence theory, fuse the new evidence obtained in Step 7.1 twice to obtain the final fusion result:

[0189] , , , ;

[0190] Step 8: The damage mode corresponding to the maximum value in the basic probability allocation result of the typical blade damage modes in Step 7 is the finally identified blade damage mode, and the final identification result is crack loss F1.

[0191] To further highlight the recognition effect of this method, the results of recognition using the DS evidence theory are compared with the results of recognition in this embodiment. The comparison results are shown in Table 2.

[0192] Table 2. Recognition performance of the comparison methods

[0193]

[0194] The results in Table 2 show that the DS method identifies the lesion as F2 (trachoma), which is inconsistent with the identification result of this embodiment, resulting in a situation contrary to reality. Therefore, this further demonstrates the identification effectiveness of the method provided by the present invention.

[0195] The above detailed description further elaborates on the purpose, technical solution, and effective effects of the invention. However, the embodiments of the present invention are not limited thereto. Any modifications made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying damage patterns in wind turbine blades based on multi-source information fusion, characterized in that, Includes the following steps: Step 1: Determine the identification framework for wind turbine blade damage modes and acquire multiple single-source signals for identifying blade damage modes. These single-source signals include blade vibration signals, acoustic emission signals, and strain signals. The identification framework for wind turbine blade damage modes is represented as follows: F1 represents crack damage, F2 represents sand hole damage, F3 represents skin detachment, and F4 represents leading edge erosion. Multiple single-source signals, including representative vibration signals, representative acoustic emission signals, and representative strain signals, were acquired to identify typical blade damage modes. Step 2: Calculate the basic probability allocation results of the evidence corresponding to each single-source signal for each damage mode using a single-source probabilistic neural network model; Step 3: Solve for the conflict between the evidence corresponding to each of the single-source signals and the information content of the evidence corresponding to each single-source signal; Step 4: Based on the conflict between the evidence corresponding to each single-source signal calculated in Step 3, determine the credibility of the evidence corresponding to each single-source signal; Step 5: Based on the information content of the evidence corresponding to each single-source signal calculated in Step 3, solve for the robustness of the evidence corresponding to each single-source signal; Step Six: Based on the credibility of the evidence corresponding to each single-source signal determined in Step Four, and the robustness of the evidence corresponding to each single-source signal determined in Step Five, calculate the comprehensive weight of the evidence corresponding to each single-source signal. Step 7: Based on the comprehensive weight of the evidence corresponding to each single-source signal calculated in Step 6, the basic probability allocation result of the evidence corresponding to each single-source signal calculated in Step 2 is weighted to obtain the basic probability allocation result of the new evidence to the damage mode. Through multiple fusions, the final basic probability allocation result of the typical blade failure mode is obtained. Step 8: The damage mode corresponding to the maximum value in the basic probability allocation result of the final typical blade failure modes in Step 7 is the finally identified blade damage mode. Step three includes: Step 3.1: Solve for the confidence function and plausibility function of the evidence corresponding to each single-source signal, expressed as: （1） In equation (1), Let represent the confidence function of the evidence corresponding to the i-th single-source signal for the j-th damage mode, i=1,2,3, j=1,2,3,4; F represents the similarity function of the evidence corresponding to the i-th single-source signal to the j-th damage mode; j B represents the j-th damage mode; B belongs to the recognition framework. A subset; m i The basic probability assignment function represents the evidence corresponding to the i-th single-source signal; Step 3.2: Solve for the conflict between the evidence corresponding to each of the single-source signals: First, we need to resolve the conflict between evidence m1 corresponding to the blade vibration signal and evidence m2 corresponding to the acoustic emission signal. , represented as: （2） In equation (2), α is the adjustment coefficient, α∈(0, 1)∪(1, ∞); Let i represent the belief similarity function of the evidence corresponding to the i-th single-source signal for the j-th damage mode, i=1,2,3,j=1,2,3,4; The belief likelihood function representing the evidence corresponding to the i-th single-source signal for the j-th damage mode. of Power; The belief similarity function of the evidence corresponding to the i-th single-source signal for the j-th damage mode. of Power; The belief similarity function of the evidence corresponding to the i-th single-source signal for the j-th damage mode. The solution method is as follows: （3） In equation (3), This represents the damage pattern cardinality, with a value of 1. This represents the cardinality of the recognition framework, with a value of 4. Further solving yields the following: the conflict between evidence m1 corresponding to the blade vibration signal and evidence m3 corresponding to the strain signal. The conflict between evidence m2 corresponding to the acoustic emission signal and evidence m1 corresponding to the blade vibration signal. The conflict between evidence m2 corresponding to the acoustic emission signal and evidence m3 corresponding to the strain signal. The conflict between evidence m3 corresponding to the strain signal and evidence m1 corresponding to the blade vibration signal. The conflict between evidence m3 corresponding to the strain signal and evidence m2 corresponding to the acoustic emission signal. ; Step 3.3: Calculate the information content of the evidence corresponding to each single-source signal, expressed as: （4） In equation (4), Inf(m) i ) represents the information content of the evidence corresponding to the i-th single-source signal; m i ( ) represents the basic probability allocation result of the evidence corresponding to the i-th single-source signal to the j-th damage mode.

2. The method for identifying wind turbine blade damage patterns by multi-source information fusion as described in claim 1, characterized in that, Step two includes: Step 2.1: Construct three single-source probabilistic neural network models respectively. The blade vibration signal, acoustic emission signal and strain signal are used as the inputs of the three single-source probabilistic neural network models respectively, and the damage mode of the wind turbine blade is used as the output of the three probabilistic neural network models. Step 2.2: Train the corresponding single-source probabilistic neural network model constructed in Step 2.1 using the historical blade vibration signal, historical acoustic emission signal and historical strain signal of the wind turbine blade damage mode respectively; Step 2.3: Input the blade vibration signal, acoustic emission signal, and strain signal obtained in Step 1 into the corresponding single-source probabilistic neural network model trained in Step 2.

2. The output value of each single-source probabilistic neural network model is the basic probability allocation result m of the evidence corresponding to each single-source signal to the damage mode. i ( ), where m1 represents the basic probability allocation function of the evidence corresponding to the blade vibration signal, m2 represents the basic probability allocation function of the evidence corresponding to the acoustic emission signal, and m3 represents the basic probability allocation function of the evidence corresponding to the strain signal.

3. The method for identifying wind turbine blade damage patterns through multi-source information fusion as described in claim 1, characterized in that, Step four includes: Step 4.1: Based on the conflict between the evidence corresponding to each pair of single-source signals obtained in Step 3.2, construct the conflict matrix D between the evidence corresponding to all single-source signals, expressed as: （5） Step 4.2: Calculate the difference between the evidence corresponding to each single-source signal and the evidence corresponding to other single-source signals, expressed as: （6） In equation (6), Diff(m) i ) represents the difference between the evidence corresponding to the i-th single-source signal and the evidence corresponding to the other two single-source signals, i=1, 2, 3; Step 4.3: Obtain the support level (Sup) of the evidence corresponding to each single-source signal, expressed as: （7） In equation (7), Sup(m) i ) represents the support level of the evidence corresponding to the i-th single-source signal, i=1, 2, 3; Step 4.4: Calculate the credibility Cre of the evidence corresponding to each single-source signal, expressed as: （8） In equation (8), Cre(m) i ) represents the credibility of the evidence corresponding to the i-th single-source signal, i=1, 2, 3.

4. The wind turbine blade damage pattern recognition method based on multi-source information fusion as described in claim 1, characterized in that, Step five includes: Step 5.1: Based on the information content of the evidence corresponding to each single-source signal calculated in Step 3.3, construct an initial robustness graphical model. ,in, E represents the initial robustness of the evidence node corresponding to the i-th single-source signal. iq The edge weights represent the initial robustness graph model; First, the initial robustness of the evidence node corresponding to each single-source signal is calculated, expressed as: （9） In equation (9), Let Inf(m) represent the initial robustness of the evidence node corresponding to the i-th single-source signal, where i = 1, 2, 3; i ) represents the information content of the evidence corresponding to the i-th single-source signal, i=1, 2, 3; Then, the edge weights E of the initial robust graphical model are calculated. iq , represented as: （10） In equation (10), JD represents the Jousselme distance; m q The basic probability assignment function represents the evidence corresponding to the adjacent single-source signals of the i-th single-source signal, where q = 1, 2, 3 and i ≠ q; Step 5.2: Define the iterative propagation mechanism for robustness. The iterative propagation process for the robustness of the evidence node corresponding to the i-th single-source signal is expressed as: （11） In equation (11), Represents the damping factor; This represents the robustness of the evidence node corresponding to the i-th single-source signal after the (k+1)-th iteration; Let l represent the robustness of the evidence node corresponding to the neighboring single-source signal q of the i-th single-source signal after the k-th iteration; q represents the neighboring single-source signal of the i-th single-source signal; l represents the neighboring single-source signal of single-source signal q, E ql This represents the edge weight between the evidence node corresponding to the single-source signal q and the evidence node corresponding to the single-source signal l. Step 5.3: Calculate the robustness of the evidence node corresponding to the i-th single-source signal after the iteration terminates. The iteration termination condition is expressed as: （12） At this point, the stability of the evidence node corresponding to the i-th single-source signal is represented by R. i ;R i (k) This represents the robustness of the evidence node corresponding to the i-th single-source signal after the k-th iteration.

5. The wind turbine blade damage pattern recognition method based on multi-source information fusion as described in claim 4, characterized in that, Step six includes: Based on the credibility determined in step 4.4 and the robustness determined in step 5.3, the comprehensive weight of the evidence corresponding to each single-source signal in the fusion process is determined, expressed as: （13） In equation (13), w(m) i ) represents the comprehensive weight of the evidence corresponding to the i-th single-source signal in the fusion process, i=1, 2,3; Cre(m i R represents the credibility of the evidence corresponding to the i-th single-source signal, where i = 1, 2, 3; i This represents the stability of the evidence node corresponding to the i-th single-source signal, where i = 1, 2, 3.

6. The wind turbine blade damage pattern recognition method based on multi-source information fusion as described in claim 5, characterized in that, Step seven includes: Step 7.1: Weight the basic probability allocation result of the evidence corresponding to the damage mode calculated in Step 2 with the corresponding comprehensive weight calculated in Step 6 to obtain the basic probability allocation result of the new evidence to the damage mode, expressed as: （14） In equation (14), The basic probability assignment of the new evidence to the j-th damage mode is given, where j = 1, 2, 3, 4; w(m i ) represents the comprehensive weight of the evidence corresponding to the i-th single-source signal in the fusion process, i=1, 2, 3; Step 7.2: Using the DS evidence theory, the new evidence obtained in Step 7.1 is fused with the basic probability allocation results of the damage modes twice to obtain the final basic probability allocation results of the typical blade failure modes.

Images