A propeller damage prediction method based on vibration characteristic parameters and neural network
By measuring the difference in vibration characteristic parameters of propeller blades and using a BP neural network to establish a prediction model, the problem of accuracy in propeller damage prediction was solved, and rapid and accurate damage assessment was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- DALIAN MARITIME UNIVERSITY
- Filing Date
- 2023-04-14
- Publication Date
- 2026-05-15
AI Technical Summary
Existing technologies cannot accurately predict propeller damage based on the vibration characteristics of the propeller during its service life, resulting in an inability to quickly assess its damage status.
By measuring the difference in vibration characteristic parameters between undamaged and damaged propeller blades, a damage prediction model is established using a BP neural network. Combined with the material and operating parameters of the propeller blades, damage prediction is performed.
It enables rapid and accurate propeller damage prediction, improves detection efficiency and accuracy, and reduces personnel costs.
Smart Images
Figure CN116306314B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mechanical equipment damage prediction and assessment technology, and in particular to a propeller damage prediction method based on vibration characteristic parameters and neural networks. Background Technology
[0002] The propeller is a crucial output component of a ship's propulsion system, and its integrity directly impacts its service life and, consequently, the ship's navigation capabilities. During service, propellers are subjected to seawater corrosion and alternating loads, and also collide with floating debris, making their surfaces highly susceptible to various types of damage, including corrosion pits, fatigue cracks, and impact breakage. Once damage occurs, the propeller's efficiency is significantly reduced, and substantial eccentric vibrations can occur, potentially damaging the shafting and main engine, thus threatening the ship's safe operation. Current propeller damage prediction technologies cannot predict damage based on characteristic signals generated during propeller service, and without assessing the vibration characteristics of the propeller during service, rapid prediction of propeller damage is impossible.
[0003] The actual working conditions of ship propellers are very complex, which makes it extremely difficult to accurately predict their damage. Qian Xiaonan ([1] Qian Xiaonan. Cavitation damage of marine propeller surface [J]. Journal of Shanghai Jiaotong University, 1985(01):72-79.) analyzed the formation mechanism and development process of surface damage of marine propellers under cavitation erosion and verified it based on experiments. Liu Zongkai et al. ([1] Liu Zongkai, Liu Manhong, Guo Zhengyang, Li Zhanjiang. Damage identification mechanism of propeller blades of fully appended underwater vehicles [J]. Acta Ordnance et al., 2021, 42(12):2710-2721.) studied the damage identification mechanism of propeller blades of fully appended underwater vehicles and proposed a damage identification method based on the coupling relationship between flow field fluctuation and propeller dynamic characteristics. Deep learning, as a new research direction in the field of machine learning, has been widely applied to the state prediction research of materials and equipment. Among them, BP neural network has shown great potential in identifying the hidden patterns behind complex industrial data and determining various mechanical and fatigue parameters of materials. Zhan Xinpeng et al. ([1] Zhan Xinpeng, Wang Zixian, Ren Tongwei. Propeller comprehensive performance prediction algorithm based on neural network and random forest [J]. Computer and Information Technology, 2023, 31(01): 1-3+14) applied neural network and random forest model to predict the comprehensive performance of propeller and verified its effectiveness. Qiang Yiming et al. ([1] Qiang Yiming, Chen Shinan, Chen Yihong, Chu Xuesen. Ship propeller open water performance prediction proxy model based on machine learning [J]. China Shipbuilding, 2022, 63(05): 181-188.) based on propeller characteristic parameters and applied classic machine learning model to establish a ship propeller open water performance prediction proxy model.
[0004] According to existing research, there is currently a lack of effective means to predict propeller damage. In particular, if the relationship between damage and propeller vibration characteristics cannot be scientifically and rationally analyzed, accurate prediction of propeller damage cannot be achieved. Summary of the Invention
[0005] This invention provides a propeller damage prediction method based on vibration characteristic parameters and neural networks, which overcomes the current propeller damage prediction technology, which cannot predict damage based on the characteristic signals of the propeller during service, and cannot quickly predict the damage of the propeller if the vibration characteristics of the propeller during service cannot be evaluated.
[0006] To achieve the above objectives, the technical solution of the present invention is as follows:
[0007] A propeller damage prediction method based on vibration characteristic parameters and neural networks includes the following steps:
[0008] Step S1: Obtain propeller blades with random damage and undamaged propeller blades;
[0009] Step S2: Measure the damage geometry of the propeller blade with random damage;
[0010] Step S3: Conduct an open-water performance test on the undamaged propeller blade to obtain the first vibration characteristic parameters of the undamaged propeller blade;
[0011] The first vibration characteristic parameters include the first vibration frequency f1 and the first vibration amplitude A1;
[0012] Step S4: Conduct an open-water performance test on the damaged propeller blades to obtain the second vibration characteristic parameters under the same working conditions as the undamaged propeller blades;
[0013] The second vibration characteristic parameters include the second vibration frequency f2 and the second vibration amplitude A2;
[0014] Step S5: Based on the vibration characteristic parameters of the undamaged propeller blade and the damaged propeller blade, obtain the vibration frequency difference Δf and vibration amplitude difference ΔA of the undamaged propeller blade and the damaged propeller blade after stable operation under open water performance test.
[0015] Step S6: Based on the damage geometry of the propeller blades, the difference in vibration frequency Δf, and the difference in vibration amplitude ΔA, a propeller blade damage prediction model based on vibration characteristics is established using a BP neural network.
[0016] Step S7: Predict the propeller blade to be tested according to the propeller blade damage prediction model based on vibration characteristics.
[0017] Furthermore, in step S1, obtaining the propeller blades with random damage specifically involves...
[0018] Step S1.1: Obtain the spatial geometric parameters and material parameters of the propeller blade;
[0019] The spatial geometric parameters of the propeller blades include at least the propeller diameter, blade tilt angle, maximum thickness at the blade root position, and minimum thickness at the blade tip position.
[0020] The material parameters of the propeller blade include at least the material grade, density, tensile strength, bending strength, and yield strength;
[0021] Step S1.2: Based on the spatial geometric parameters and material parameters of the propeller blade, random damage is machined on the surface of the non-destructive propeller blade using methods such as milling, grinding, drilling, and abrasive waterjet machining, to obtain a propeller blade with random damage.
[0022] Furthermore, the maximum depth of random damage during the processing does not exceed 50% of the depth at the location.
[0023] Furthermore, in step S2, the damage geometry of the propeller blade with random damage is measured, specifically...
[0024] Step S2.1: Use a blue light industrial-grade 3D scanner to scan the three-dimensional dimensions of the propeller blade with random damage and obtain the propeller blade and damage point cloud data.
[0025] Step S2.2: Import the point cloud data into CAD software to obtain a three-dimensional model of the propeller blade with random damage;
[0026] Step S2.3: Extract damage data points from the three-dimensional model of the propeller blade with random damage to obtain damage data information;
[0027] The damage data information includes spatial points representing the maximum width, maximum depth, and maximum length of damage on the propeller blade surface.
[0028] Step S2.4: Use a three-dimensional semi-ellipse to spatially wrap the surface damage of the propeller blade in three dimensions, and wrap the spatial points of the maximum width, maximum depth and maximum length of the surface damage of the propeller blade within the three-dimensional ellipse to simplify the three-dimensional morphology of the damage.
[0029] The simplified three-dimensional morphology of the damage is equivalent to a semi-elliptical pit, and the key dimensions of the pit are obtained based on the semi-elliptical pit.
[0030] Furthermore, the key dimensions of the pit include the width w, depth d, and length h of the semi-elliptical pit, and these key dimensions are used as the three-dimensional dimensions of the damage.
[0031] Furthermore, the step S6, which involves establishing a propeller blade damage prediction model based on vibration characteristics using a BP neural network, specifically involves...
[0032] Step S6.1: Determine the training samples, which include the propeller blade material parameters, operating condition parameters and environmental parameters, three-dimensional dimensions of propeller blade damage, vibration frequency difference Δf and vibration amplitude difference ΔA;
[0033] The material parameters of the propeller blades include at least the material grade, density, tensile strength, bending strength, and yield strength;
[0034] The operating parameters and environmental parameters include at least the propeller speed, the ambient temperature under actual operating conditions, and the chemical composition of the actual operating environment.
[0035] Step S6.2: Divide the training samples into a training set and a test set by random partitioning; and input the training set into the BP neural network for training;
[0036] The BP neural network includes an input layer, an output layer, and several hidden layers. The input layer parameters are the propeller blade material parameters, operating condition parameters, environmental parameters, vibration frequency difference Δf, and vibration amplitude difference ΔA. The output layer parameters are the three-dimensional dimensions of the propeller blade damage.
[0037] Based on the quantity and dimension of the actual input propeller blade damage data, the number of hidden layers, the data weights of each hidden layer, and the threshold for judging the output data of the hidden layers are defined.
[0038] Step S6.3: Normalize the samples of the input layer and the output layer; the training set generates an output signal through nonlinear transformation, and passes through the input layer, several hidden layers and the output layer in sequence;
[0039] Step S6.4: Use the sigmoid function to normalize the output data of each hidden layer and input it into the next hidden layer until the initial output value y is obtained after passing through the last hidden layer;
[0040] The last hidden layer inputs the initial output value y to the output layer;
[0041] Step S6.5: The output layer compares the initial output value y with the expected output value to obtain the prediction error δ;
[0042] The output layer backpropagates the prediction error δ back to the initial hidden layer. Each hidden layer modifies the data weight w defined in each layer according to the backpropagated prediction error δ to obtain the optimized data weight w1 and the optimized output original value y1.
[0043] Step S6.6: Repeat step S6.5 until the mean square error of the prediction error between the original output value and the expected output value is minimized, then end the training and obtain the propeller blade damage prediction model based on vibration characteristics.
[0044] Step S6.7: The propeller blade damage prediction model based on vibration characteristics is used to predict the test set to achieve the prediction of propeller blade damage.
[0045] Beneficial Effects: This invention provides a propeller damage prediction method based on vibration characteristic parameters and neural networks. It involves pre-preparing propeller blades with random damage and measuring their damage geometry. Based on the actual operating parameters and environmental parameters of the propeller, open-water performance tests are conducted on undamaged and undamaged propeller blades to obtain their vibration frequency and amplitude during stable operation. The vibration frequency difference Δf and vibration amplitude difference ΔA are also obtained. Based on the propeller blade damage geometry, vibration frequency difference Δf, and vibration amplitude difference ΔA, a propeller blade damage prediction model based on vibration characteristics is established using a BP neural network. This solves the problem that current propeller damage prediction methods cannot assess the vibration characteristics during propeller service, enabling rapid prediction of propeller damage, reducing personnel costs in propeller damage detection, and significantly improving detection efficiency and accuracy. Attached Figure Description
[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0047] Figure 1 This is a flowchart of a propeller damage prediction method based on vibration characteristic parameters and neural networks according to the present invention.
[0048] Figure 2 This is a framework diagram of a propeller damage prediction method based on vibration characteristic parameters and neural networks according to the present invention.
[0049] Figure 3 This is a technical roadmap for a propeller damage prediction method based on vibration characteristic parameters and neural networks according to the present invention.
[0050] Figure 4 This is a schematic diagram of the BP neural network structure of a propeller damage prediction method based on vibration characteristic parameters and neural networks according to the present invention. Detailed Implementation
[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0052] This embodiment provides a propeller damage prediction method based on vibration characteristic parameters and neural networks, such as... Figure 1 As shown, it includes the following steps:
[0053] Step S1: Obtain propeller blades with random damage and undamaged propeller blades;
[0054] Step S2: Measure the damage geometry of the propeller blade with random damage;
[0055] Step S3: Conduct an open-water performance test on the undamaged propeller blade to obtain the first vibration characteristic parameters of the undamaged propeller blade;
[0056] The first vibration characteristic parameters include the first vibration frequency f1 and the first vibration amplitude A1;
[0057] Step S4: Conduct an open-water performance test on the damaged propeller blades to obtain the second vibration characteristic parameters under the same working conditions as the undamaged propeller blades;
[0058] The second vibration characteristic parameters include the second vibration frequency f2 and the second vibration amplitude A2;
[0059] Step S5: Based on the vibration characteristic parameters of the undamaged propeller blade and the damaged propeller blade, obtain the vibration frequency difference Δf and vibration amplitude difference ΔA of the undamaged propeller blade and the damaged propeller blade after stable operation under open water performance test.
[0060] Step S6: Based on the damage geometry of the propeller blades, the difference in vibration frequency Δf, and the difference in vibration amplitude ΔA, a propeller blade damage prediction model based on vibration characteristics is established using a BP neural network.
[0061] Step S7: Predict the propeller blade to be tested according to the propeller blade damage prediction model based on vibration characteristics.
[0062] By pre-fabricating propeller blades with random damage, the damage geometry of the propeller blades with random damage is measured, such as... Figure 2As shown, based on the actual operating parameters and environmental parameters of the propeller, open-water performance tests were conducted on propeller blades with and without damage. The open-water performance test is a known prior art and not the inventive point of this application, so it will not be elaborated upon here. The vibration frequency and amplitude of each blade during stable operation were obtained. Based on the vibration characteristic parameters of the propeller blades with random damage and those of the undamaged propeller blades, the vibration frequency difference Δf and vibration amplitude difference ΔA were obtained. Based on the propeller blade damage geometry, the vibration frequency difference Δf, and the vibration amplitude difference ΔA, and combined with the propeller blade material parameters, operating parameters, and environmental parameters, a propeller blade damage prediction model based on vibration characteristics was established using a BP neural network. This solves the problem that current propeller damage prediction cannot be evaluated based on the vibration characteristics during propeller service, enabling rapid prediction of propeller damage and greatly improving the prediction accuracy of propeller blade damage.
[0063] In a specific embodiment, obtaining the propeller blade with random damage in step S1 specifically involves step S1.1: obtaining the spatial geometric parameters and material parameters of the propeller blade;
[0064] The spatial geometric parameters of the propeller blades include at least the propeller diameter, blade tilt angle, maximum thickness at the blade root position, and minimum thickness at the blade tip position.
[0065] The material parameters of the propeller blade include at least the material grade, density, tensile strength, bending strength, and yield strength;
[0066] Step S1.2: Based on the spatial geometric parameters and material parameters of the propeller blade, random damage is machined on the surface of the non-destructive propeller blade using methods such as milling, grinding, drilling, and abrasive waterjet machining, to obtain a propeller blade with random damage.
[0067] In a specific embodiment, the maximum depth of the random damage caused by the processing does not exceed 50% of the depth at the location.
[0068] In a specific embodiment, step S2 involves measuring the damage geometry of the propeller blade with random damage, specifically...
[0069] Step S2.1: Use a blue light industrial-grade 3D scanner to scan the three-dimensional dimensions of the propeller blade with random damage and obtain the propeller blade and damage point cloud data.
[0070] Step S2.2: Import the point cloud data into CAD software to obtain a three-dimensional model of the propeller blade with random damage;
[0071] Step S2.3: Extract damage data points from the three-dimensional model of the propeller blade with random damage to obtain damage data information;
[0072] The damage data information includes spatial points representing the maximum width, maximum depth, and maximum length of damage on the propeller blade surface.
[0073] Step S2.4: Use a set minimum three-dimensional semi-ellipse to spatially encapsulate the surface damage of the propeller blade in three dimensions. Encapsulate the spatial points of the maximum width, maximum depth, and maximum length of the surface damage of the propeller blade within the three-dimensional ellipse to simplify the three-dimensional morphology of the damage. The method of using a set minimum three-dimensional semi-ellipse to spatially encapsulate the surface damage of the propeller blade in three dimensions is a known prior art in three-dimensional modeling and is not the inventive point of this application. It will not be described in detail here.
[0074] like Figure 3 As shown, the simplified three-dimensional morphology of the damage is equivalent to a semi-elliptical pit, and the key dimensions of the pit are obtained based on the semi-elliptical pit.
[0075] Furthermore, the key dimensions of the pit include the width w, depth d, and length h of the semi-elliptical pit, and these key dimensions are used as the three-dimensional dimensions of the damage.
[0076] In a specific embodiment, step S6, which involves establishing a propeller blade damage prediction model based on vibration characteristics using a BP neural network, specifically involves...
[0077] Step S6.1: Determine the training samples, which include the propeller blade material parameters, operating condition parameters and environmental parameters, three-dimensional dimensions of propeller blade damage, vibration frequency difference Δf and vibration amplitude difference ΔA;
[0078] The material parameters of the propeller blades include at least the material grade, density, tensile strength, bending strength, and yield strength;
[0079] The operating parameters and environmental parameters include at least the propeller speed, the ambient temperature under actual operating conditions, and the chemical composition of the actual operating environment, wherein the chemical composition of the actual operating environment is specifically the mass percentage of sodium chloride.
[0080] Step S6.2: Divide the training samples into a training set and a test set by random partitioning; the training set is used to train the BP neural network model, and the test set is used to verify whether the output of the trained BP neural network model meets the expected requirements; and input the training set into the BP neural network for training.
[0081] like Figure 4As shown, the BP neural network includes an input layer, an output layer, and several hidden layers. The input layer parameters are the propeller blade material parameters, operating condition parameters, environmental parameters, vibration frequency difference Δf, and vibration amplitude difference ΔA. The output signal is generated through nonlinear transformation, and the BP neural network model uses the three-dimensional dimensions of propeller blade damage as the output layer parameters.
[0082] Based on experience and the quantity and dimensions of actual input propeller blade damage data, the number of hidden layers, the data weights of each hidden layer, and the threshold for judging the output data of the hidden layers are defined.
[0083] Step S6.3: Normalize the samples of the input layer and the output layer; the training set generates an output signal through nonlinear transformation, and passes through the input layer, several hidden layers and the output layer in sequence;
[0084] Step S6.4: Use the sigmoid activation function to normalize the output data of each hidden layer and input it into the next hidden layer until the initial output value y is obtained after passing through the last hidden layer;
[0085] The last hidden layer inputs the initial output value y to the output layer;
[0086] Step S6.5: The output layer compares the initial output value y with the expected output value to obtain the prediction error δ;
[0087] The output layer backpropagates the prediction error δ back to the initial hidden layer. Each hidden layer modifies the data weight w defined in each layer according to the backpropagated prediction error δ to obtain the optimized data weight w1 and the optimized output original value y1.
[0088] Step S6.6: Repeat step S6.5 until the mean square error of the prediction error between the original output value and the expected output value is minimized, then end the training and obtain the propeller blade damage prediction model based on vibration characteristics.
[0089] Step S6.7: The propeller blade damage prediction model based on vibration characteristics is used to predict the test set to achieve the prediction of propeller blade damage.
[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A propeller damage prediction method based on vibration characteristic parameters and neural networks, characterized in that, Includes the following steps: Step S1: Obtain propeller blades with random damage and undamaged propeller blades; Step S2: Measure the damage geometry of the propeller blades with random damage, specifically: Step S2.1: Use a blue light industrial-grade 3D scanner to scan the three-dimensional dimensions of the propeller blade with random damage and obtain the propeller blade and damage point cloud data. Step S2.2: Import the point cloud data into CAD software to obtain a three-dimensional model of the propeller blade with random damage; Step S2.3: Extract damage data points from the three-dimensional model of the propeller blade with random damage to obtain damage data information; The damage data information includes spatial points representing the maximum width, maximum depth, and maximum length of damage on the propeller blade surface. Step S2.4: Use a three-dimensional semi-ellipse to spatially wrap the surface damage of the propeller blade in three dimensions, and wrap the spatial points of the maximum width, maximum depth and maximum length of the surface damage of the propeller blade within the three-dimensional ellipse to simplify the three-dimensional morphology of the damage. The simplified three-dimensional morphology of the damage is equivalent to a semi-elliptical pit, and the key dimensions of the pit are obtained based on the semi-elliptical pit. Furthermore, the key dimensions of the recess include the width of the semi-elliptical recess. w ,depth d and length h The key dimensions of the pit are used as the three-dimensional dimensions of the damage. Step S3: Conduct an open-water performance test on the undamaged propeller blade to obtain the first vibration characteristic parameters of the undamaged propeller blade; The first vibration characteristic parameter includes the first vibration frequency. f 1 and the first vibration amplitude A 1; Step S4: Conduct an open-water performance test on the damaged propeller blades to obtain the second vibration characteristic parameters under the same working conditions as the undamaged propeller blades; The second vibration characteristic parameter includes the second vibration frequency. f 2 and the second vibration amplitude A 2; Step S5: Based on the vibration characteristic parameters of the undamaged propeller blade and the damaged propeller blade, obtain the difference in vibration frequency between the undamaged propeller blade and the damaged propeller blade after stable operation under open water performance test. f Difference from vibration amplitude A ; Step S6: Based on the difference in the damaged geometry and vibration frequency of the propeller blades... f and vibration amplitude difference A A propeller blade damage prediction model based on vibration characteristics was established using a BP neural network, taking into account propeller blade material parameters, operating condition parameters, environmental parameters, and vibration frequency differences. f and vibration amplitude difference A The three-dimensional dimensions of propeller blade damage are used as input layer parameters and output layer parameters. Step S7: Predict the propeller blade to be tested according to the propeller blade damage prediction model based on vibration characteristics.
2. The propeller damage prediction method based on vibration characteristic parameters and neural networks according to claim 1, characterized in that, In step S1, the specific steps for obtaining propeller blades with random damage are as follows: Step S1.1: Obtain the spatial geometric parameters and material parameters of the propeller blade; The spatial geometric parameters of the propeller blades include at least the propeller diameter, blade tilt angle, maximum thickness at the blade root position, and minimum thickness at the blade tip position. The material parameters of the propeller blade include at least the material grade, density, tensile strength, bending strength, and yield strength; Step S1.2: Based on the spatial geometric parameters and material parameters of the propeller blade, random damage is machined on the surface of the non-destructive propeller blade using methods such as milling, grinding, drilling, and abrasive waterjet machining, to obtain a propeller blade with random damage.
3. The propeller damage prediction method based on vibration characteristic parameters and neural networks according to claim 2, characterized in that, The maximum depth of random damage during the processing does not exceed 50% of the depth at the location.
4. The propeller damage prediction method based on vibration characteristic parameters and neural networks according to claim 1, characterized in that, Step S6, which involves establishing a propeller blade damage prediction model based on vibration characteristics using a BP neural network, specifically involves... Step S6.1: Determine the training samples, which include the propeller blade material parameters, operating condition parameters and environmental parameters, three-dimensional dimensions of propeller blade damage, and vibration frequency differences. f and vibration amplitude difference A ; The material parameters of the propeller blades include at least the material grade, density, tensile strength, bending strength, and yield strength; The operating parameters and environmental parameters include at least the propeller speed, the ambient temperature under actual operating conditions, and the chemical composition of the actual operating environment. Step S6.2: Divide the training samples into a training set and a test set by random partitioning; and input the training set into the BP neural network for training; The BP neural network includes an input layer, an output layer, and several hidden layers, based on propeller blade material parameters, operating condition parameters, environmental parameters, and vibration frequency differences. f and vibration amplitude difference A The three-dimensional dimensions of propeller blade damage are used as input layer parameters and output layer parameters. Based on the quantity and dimension of the actual input propeller blade damage data, the number of hidden layers, the data weights of each hidden layer, and the threshold for judging the output data of the hidden layers are defined. Step S6.3: Normalize the samples of the input layer and the output layer; the training set generates an output signal through nonlinear transformation, and passes through the input layer, several hidden layers and the output layer in sequence; Step S6.4: Use the sigmoid function to normalize the output data of each hidden layer and input it into the next hidden layer until the initial output value y is obtained after passing through the last hidden layer; The last hidden layer inputs the initial output value y to the output layer; Step S6.5: The output layer compares the initial output value y with the expected output value to obtain the prediction error δ; The output layer backpropagates the prediction error δ back to the initial hidden layer. Each hidden layer modifies the data weight w defined in each layer according to the backpropagated prediction error δ to obtain the optimized data weight w1 and the optimized output original value y1. Step S6.6: Repeat step S6.5 until the mean square error of the prediction error between the original output value and the expected output value is minimized, then end the training and obtain the propeller blade damage prediction model based on vibration characteristics. Step S6.7: The propeller blade damage prediction model based on vibration characteristics is used to predict the test set to achieve the prediction of propeller blade damage.