Method for constructing digital twin model of titanium alloy propeller based on SDAE model

Abstract

The invention discloses a titanium alloy propeller digital twin model construction method based on an SDAE model. The method comprises the steps that propeller multi-source sensing data are collected and obtained; preprocessing the multi-source sensing data of the propeller to obtain a spatio-temporal data set of the propeller; constructing a propeller digital twin model based on an SDAE algorithm model; performing model training and model evaluation on the propeller digital twin model according to the propeller spatio-temporal data set to obtain an optimal propeller digital twin model; and mapping the output of the optimal propeller digital twin model to a preset three-dimensional propeller grid model, and obtaining a real-time stress cloud picture and a fatigue damage hotspot distribution cloud picture of the propeller to realize health monitoring of the titanium alloy propeller. The problems that stress distribution and fatigue damage caused by high rigidity and electromagnetic shielding characteristics of the titanium alloy propeller are difficult to capture in real time through a traditional propeller health monitoring method under complex working conditions, and an existing digital twin model is insufficient in noise and nonlinear dynamic adaptability are solved.

Description

Technical Field

[0001] The present invention relates to the fields of digital twin technology and intelligent sensing technology, and particularly to a method for constructing a digital twin model of a titanium alloy propeller based on an SDAE model. Background Art

[0002] Traditional propeller health monitoring methods rely on finite element simulation or off-line detection, and it is difficult to dynamically capture the stress distribution and fatigue damage under complex working conditions. In the prior art, stress monitoring of composite propellers is achieved by embedding sensors. However, due to the high stiffness, complex stress transfer characteristics and electromagnetic shielding problems of titanium alloy propellers, it is difficult to directly adopt a similar scheme for stress monitoring by embedding sensors. In addition, existing digital twin models are mostly constructed based on physical equations, and there are problems of insufficient adaptability to noise data and non-linear dynamics. Summary of the Invention

[0003] The present invention provides a method for constructing a digital twin model of a titanium alloy propeller based on an SDAE model to overcome the above technical problems.

[0004] To achieve the above object, the technical solution of the present invention is as follows:

[0005] A method for constructing a digital twin model of a titanium alloy propeller based on an SDAE model specifically includes the following steps:

[0006] S1: Based on sensors embedded in the titanium alloy propeller, collect and obtain multi-source sensing data of the propeller;

[0007] And the multi-source sensing data of the propeller at least includes strain forces, vibration frequencies and blade surface pressures corresponding to the root, middle and tip of the propeller blade respectively;

[0008] S2: Preprocess the multi-source sensing data of the propeller to obtain a spatio-temporal data set of the propeller;

[0009] And the preprocessing includes: denoising the multi-source sensing data of the propeller by using a wavelet threshold denoising method to eliminate high-frequency electromagnetic interference, and obtaining denoised data of the propeller;

[0010] Based on the Kalman filtering algorithm, perform signal compensation on the signal attenuation caused by electromagnetic shielding of the titanium alloy material in the denoised data of the propeller to obtain spatio-temporal data of the propeller;

[0011] S3: Construct a digital twin model of the propeller based on the SDAE algorithm model;

[0012] S4: Randomly divide the spatio-temporal data set of the propeller into a training set and a test set;

[0013] Train and evaluate the propeller digital twin model based on the training set and the test set to obtain the optimal propeller digital twin model;

[0014] S5: Map the output of the optimal propeller digital twin model to a preset three-dimensional propeller grid model to obtain the real-time stress nephogram and fatigue damage hotspot distribution nephogram of the propeller, so as to realize the health monitoring of the titanium alloy propeller.

[0015] Furthermore, the propeller digital twin model constructed based on the SDAE algorithm model in S3 includes an input layer, a stacked denoising autoencoder layer, a feature dynamic fusion layer, and a decoding layer;

[0016] The input layer is used to input the propeller spatio-temporal data into the stacked denoising autoencoder layer;

[0017] The stacked denoising autoencoder layer is used to perform layer-by-layer encoding extraction and dimensionality reduction processing on the propeller spatio-temporal data to obtain the low-dimensional robustness features of the propeller;

[0018] The feature dynamic fusion layer is used to perform feature fusion according to the low-dimensional robustness features of the propeller combined with the finite element real-time simulation data to obtain the dynamic fusion features of the propeller; and the finite element real-time simulation data is the propeller structure data of the propeller three-dimensional physical model established based on the finite element method according to the physical structure of the titanium alloy propeller;

[0019] The decoding layer is used to decode and reconstruct the propeller spatio-temporal data according to the dynamic fusion features of the propeller, and then output the predicted propeller reconstruction data.

[0020] Furthermore, S4 specifically includes the following steps:

[0021] S41: Define the propeller spatio-temporal data set D and

[0022] where X t represents the propeller spatio-temporal data; T represents the sampling period of the propeller multi-source sensing data; t represents the sampling time point; N represents the number of sensor nodes; d represents the feature dimension, namely the strain force, vibration frequency, and blade surface pressure amplitude;

[0023] Use the sliding window mechanism to divide the propeller spatio-temporal data set into several local segment data according to the preset sliding window;

[0024] And randomly divide the local segment data into a training set and a test set;

[0025] S42: Train the propeller digital twin model constructed based on the SDAE algorithm model according to the training set to obtain the trained propeller digital twin model;

[0026] The specific model training is as follows:

[0027] S421: Input the spatio-temporal data of the propeller in the training set into the stacked denoising autoencoder layer through the input layer;

[0028] S422: Perform layer-by-layer encoding extraction and dimensionality reduction processing on the spatio-temporal data of the propeller through the stacked denoising autoencoder layer to obtain the low-dimensional robustness features of the propeller;

[0029] And the expression of the stacked denoising autoencoder layer is

[0030]

[0031] In the formula: h (k) represents the output of the k-th denoising autoencoder layer; σ represents the ReLU activation function; represents the data received by the input layer; W k , b k represent the trainable weight parameters; K represents the stacked number of denoising autoencoder layers; h (k-1) represents the output of the (k - 1)-th denoising autoencoder layer;

[0032] S423: Through the feature dynamic fusion layer, fuse the low-dimensional robustness features of the propeller with the finite element real-time simulation data to obtain the dynamic fusion features of the propeller;

[0033] S424: Decode the dynamic fusion features of the propeller through the decoding layer and reconstruct the spatio-temporal data of the propeller, and output the predicted reconstructed data of the propeller;

[0034] S43: Based on the hybrid loss function of the constructed propeller digital twin model, evaluate the trained propeller digital twin model according to the test set, and judge whether the output of the trained propeller digital twin model converges;

[0035] If so, confirm that the trained propeller digital twin model at this time is the optimal propeller digital twin model;

[0036] Otherwise, adaptively adjust the weight parameters of the trained propeller digital twin model based on the backpropagation method, and repeat step S42.

[0037] Furthermore, the expression of the hybrid loss function constructed in S43 is

[0038]

[0039] In the formula: L hybrid represents the hybrid loss function; α represents the weight coefficient; L SDAE represents the loss function of the predicted reconstructed data of the propeller and the input spatio-temporal data of the propeller; h(K) represents the output of the Kth noise reduction autoencoder layer; W p represents the mapping matrix of the physical structure characteristics of the titanium alloy propeller; F FEM represents the finite element real-time simulation data; represents the predicted propeller reconstruction data; λ represents the L2 regularization coefficient.

[0040] Furthermore, the embedding method of the titanium alloy propeller sensor in S1 is specifically as follows

[0041] Using laser micro-machining technology, node grooves for embedding micro fiber Bragg grating sensors are opened at the root, middle, and tip of the titanium alloy propeller;

[0042] And the groove depth of the node groove of the sensor is 2 / 3 of the sensor thickness, and a titanium alloy protective layer with a reserved thickness of 0.1 mm is provided at the bottom of the node groove to avoid fluid corrosion;

[0043] After the micro fiber Bragg grating sensor is encapsulated with silicon nitride ceramic glue and embedded in the node groove, titanium alloy powder is covered on the surface, and its surface flatness is repaired by selective laser melting technology SLM;

[0044] Based on a preset near-field communication NFC module, a wireless transmission communication is established for the integrated power supply system of the micro fiber Bragg grating sensor and the propeller rotation axis to collect and obtain multi-source sensing data of the propeller.

[0045] Beneficial effects: The present invention provides a method for constructing a digital twin model of a titanium alloy propeller based on the SDAE model. By analyzing the mechanical properties of the CFRP propeller structure, stress characteristic nodes of the CFRP propeller structure are obtained, and sensors are embedded in the titanium alloy propeller blade according to the stress characteristic nodes to collect and obtain multi-source sensing data of the propeller; by preprocessing the multi-source sensing data of the propeller, spatio-temporal data of the propeller after denoising and signal compensation are obtained, solving the problem of signal attenuation of the multi-source sensing data of the propeller due to high-frequency electromagnetic interference and electromagnetic shielding of titanium alloy materials; by constructing a digital twin model of the propeller based on the SDAE algorithm model and performing model training and model evaluation, an optimal digital twin model of the propeller is obtained, solving the problem of insufficient adaptability to noise data and non-linear dynamics in the existing digital twin models constructed based on physical equations, greatly improving the prediction of the spatio-temporal data of the titanium alloy propeller, and then mapping the output of the optimal digital twin model of the propeller to a preset three-dimensional propeller grid model to obtain a real-time stress cloud map and a fatigue damage hot spot distribution cloud map of the propeller, solving the problem that traditional propeller health monitoring relies on offline simulation or limited sensors and is difficult to capture in real time the stress distribution and fatigue damage caused by the high stiffness and electromagnetic shielding characteristics of the titanium alloy propeller, greatly improving the health monitoring accuracy and efficiency of the titanium alloy propeller. Brief Description of the Drawings

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0047] Figure 1 It is a flowchart of a method for constructing a digital twin model of a titanium alloy propeller based on the SDAE model of the present invention;

[0048] Figure 2 It is a schematic diagram of the sensor layout and groove machining of the titanium alloy propeller in this embodiment;

[0049] Figure 3 It is a schematic diagram of the digital twin visualization interface (stress nephogram and warning prompt) in this embodiment.

[0050] In the figure: 1, root node groove of the blade; 2, middle node groove of the blade; 3, tip node groove of the blade. Detailed Embodiments

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0052] This embodiment provides a method for constructing a digital twin model of a titanium alloy propeller based on the SDAE model, as Figure 1 shown, which specifically includes the following steps:

[0053] S1: Based on the sensors embedded in the titanium alloy propeller, collect and obtain multi-source sensing data of the propeller;

[0054] And the multi-source sensing data of the propeller includes strain forces, vibration frequencies, blade surface pressures, etc. corresponding to the root, middle, and tip of the propeller blade respectively;

[0055] Specifically, the embedding method of the titanium alloy propeller sensors in this embodiment is

[0056] By analyzing the mechanical properties of the CFRP propeller structure, stress characteristic nodes of the CFRP propeller structure are obtained, and sensors are embedded into the titanium alloy propeller blades according to the stress characteristic nodes, that is, by using laser micro-machining technology, node grooves for embedding miniature fiber Bragg grating sensors are opened at the root, middle, and tip of the titanium alloy propeller blade; as Figure 2 shown, the root node groove 1, the middle node groove 2, and the tip node groove 3;

[0057] And the depth of the node groove of the sensor is 2 / 3 of the thickness of the sensor, and a titanium alloy protective layer with a reserved thickness of 0.1 mm is provided at the bottom of the node groove to avoid fluid corrosion;

[0058] After the miniature fiber Bragg grating sensor is encapsulated with silicon nitride ceramic glue and embedded into the node groove, titanium alloy powder is covered on the surface, and its surface flatness is repaired by selective laser melting technology SLM;

[0059] Based on the preset near-field communication NFC module, a wireless transmission communication is established for the integrated power supply system of the miniature fiber Bragg grating sensor and the propeller rotation axis to solve the data backhaul problem under the high-speed rotation of the propeller, so as to collect and obtain multi-source sensing data of the propeller;

[0060] S2: Preprocess the multi-source sensing data of the propeller to obtain the spatio-temporal data set of the propeller;

[0061] And the preprocessing includes: denoising the multi-source sensing data of the propeller by the wavelet threshold denoising method to eliminate high-frequency electromagnetic interference, and obtaining the denoised data of the propeller;

[0062] Specifically, the wavelet threshold denoising method is a well-known technical means in the art. In this embodiment, it is only used for noise suppression of the multi-source sensing data of the propeller, and its core lies in using the multi-resolution characteristic of wavelet transform to separate noise from useful signals, which will not be elaborated here;

[0063] Based on the Kalman filtering algorithm, signal compensation is performed on the signal attenuation caused by the electromagnetic shielding of the titanium alloy material in the denoised data of the propeller to obtain the spatio-temporal data of the propeller;

[0064] Specifically, the Kalman filtering algorithm is a well-known technical means in the art. In this embodiment, it is only used for signal compensation of the signal attenuation caused by the electromagnetic shielding of the titanium alloy material in the denoised data of the propeller. Kalman filtering is an efficient recursive algorithm used to dynamically estimate the state of the system from the observed data containing noise. Its core principle is to achieve the optimal compensation of the signal by combining prediction (system model) and update (observed data), and the specific content will not be elaborated here;

[0065] S3: Construct a digital twin model of the propeller based on the SDAE algorithm model;

[0066] Specifically, the propeller digital twin model based on the SDAE algorithm model constructed in S3 includes an input layer, a stacked denoising autoencoder layer, a feature dynamic fusion layer, and a decoding layer;

[0067] The input layer is used to input the propeller spatio-temporal data into the stacked denoising autoencoder layer;

[0068] The stacked denoising autoencoder layer is used to perform layer-by-layer encoding extraction and dimensionality reduction processing on the propeller spatio-temporal data to obtain low-dimensional robust features of the propeller;

[0069] The feature dynamic fusion layer is used to perform feature fusion based on the low-dimensional robust features of the propeller combined with the finite element real-time simulation data to obtain the dynamic fusion features of the propeller; and the finite element real-time simulation data is the propeller structure data of the three-dimensional physical model of the propeller established based on the finite element method according to the physical structure of the titanium alloy propeller;

[0070] The decoding layer is used to decode and reconstruct the propeller spatio-temporal data according to the dynamic fusion features of the propeller, and then output the predicted propeller reconstruction data;

[0071] S4: Randomly divide the propeller spatio-temporal data set into a training set and a test set;

[0072] Train and evaluate the propeller digital twin model according to the training set and the test set to obtain the optimal propeller digital twin model;

[0073] Specifically, it includes the following steps:

[0074] S41: Define the propeller spatio-temporal data set D and

[0075] where X t represents the propeller spatio-temporal data; T represents the sampling period of the propeller multi-source sensing data; t represents the sampling time point; N represents the number of sensor nodes; d represents the feature dimension, i.e., the stress, vibration frequency, and blade surface pressure amplitude;

[0076] Adopt a sliding window mechanism to divide the propeller spatio-temporal data set into several local segment data according to a preset sliding window to adapt to the material property degradation during the service process of the propeller, and randomly divide the local segment data into a training set and a test set;

[0077] S42: Train the propeller digital twin model based on the SDAE algorithm model constructed according to the training set to obtain the trained propeller digital twin model;

[0078] The model training is specifically as follows:

[0079] S421: Input the spatio-temporal data of the propeller in the training set into the stacked denoising autoencoder layer through the input layer;

[0080] S422: Use the stacked denoising autoencoder layer (SDAE) to encode the spatio-temporal data of the propeller layer by layer to extract the low-dimensional robust features of the propeller;

[0081] And the expression of the stacked denoising autoencoder layer is

[0082]

[0083] In the formula: h (k) represents the output of the k-th denoising autoencoder layer; σ represents the ReLU activation function; represents the data received by the input layer; W k , b k represent the trainable weight parameters; K represents the number of stacked layers of the denoising autoencoder layer; h (k-1) represents the output of the (k - 1)-th denoising autoencoder layer;

[0084] S423: Use the feature dynamic fusion layer to fuse the low-dimensional robust features of the propeller and the finite element real-time simulation data to obtain the dynamic fusion features of the propeller;

[0085] S424: Use the decoding layer to decode the dynamic fusion features of the propeller and reconstruct the spatio-temporal data of the propeller, and output the predicted reconstructed data of the propeller;

[0086] S43: Based on the hybrid loss function of the constructed propeller digital twin model, evaluate the trained propeller digital twin model according to the test set, and judge whether the output of the trained propeller digital twin model converges;

[0087] In a specific embodiment, the expression of the constructed hybrid loss function is

[0088]

[0089] In the formula: L hybrid represents the hybrid loss function; α represents the weight coefficient; L SDAE represents the loss function between the predicted reconstructed data of the propeller and the input spatio-temporal data of the propeller, which is used to extract the low-dimensional features through layer-by-layer encoding and iterative dimensionality reduction. During the iteration process, this loss function is used as the evaluation criterion, and finally the low-dimensional features are obtained; h (K) represents the output of the K-th denoising autoencoder layer; W p represents the mapping matrix of the physical structure features of the titanium alloy propeller; F FEM represents the finite element real-time simulation data; represents the predicted reconstructed data of the propeller; λ represents the L2 regularization coefficient;

[0090] If so, confirm that the trained propeller digital twin model at this time is the optimal propeller digital twin model;

[0091] Otherwise, adaptively adjust the weight parameters of the trained propeller digital twin model based on the backpropagation method, and repeat step S42;

[0092] S5: Map the output of the optimal propeller digital twin model to a preset three-dimensional propeller grid model to obtain the real-time stress nephogram and fatigue damage hot spot distribution nephogram of the propeller, thereby realizing the health monitoring of the titanium alloy propeller. As Figure 3 shown, in this embodiment, a propeller strain threshold τ can also be set, and according to the output of the optimal propeller digital twin model, when the propeller strain threshold τ is not satisfied, an alarm is triggered through the preset monitoring system of the propeller, so as to facilitate the operator to adjust in time or stop for maintenance.

[0093] Beneficial effects: In this embodiment, by analyzing the mechanical properties of the CFRP propeller structure, the stress characteristic nodes of the CFRP propeller structure are obtained, and sensors are embedded into the titanium alloy propeller blades according to the stress characteristic nodes to collect and obtain multi-source sensing data of the propeller; by preprocessing the multi-source sensing data of the propeller, the spatio-temporal data of the propeller after denoising and signal compensation are obtained, solving the problem of signal attenuation of the multi-source sensing data of the propeller due to high-frequency electromagnetic interference and electromagnetic shielding of titanium alloy materials; by constructing a propeller digital twin model based on the SDAE algorithm model, and performing model training and model evaluation, the optimal propeller digital twin model is obtained, solving the problem of insufficient adaptability to noise data and nonlinear dynamics existing in the existing digital twin models constructed based on physical equations, greatly improving the prediction of the spatio-temporal data of the titanium alloy propeller, and then mapping the output of the optimal propeller digital twin model to a preset three-dimensional propeller grid model to obtain the real-time stress nephogram and fatigue damage hot spot distribution nephogram of the propeller, solving the problem that traditional propeller health monitoring depends on offline simulation or limited sensors and is difficult to capture the stress distribution and fatigue damage caused by the high stiffness and electromagnetic shielding characteristics of titanium alloy propellers in real time, greatly improving the health monitoring accuracy and efficiency of titanium alloy propellers.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements 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 method for constructing a digital twin model of a titanium alloy propeller based on an SDAE model, characterized in that, Specifically, it includes the following steps: S1: Based on the sensors embedded in the titanium alloy propeller, collect and obtain multi-source sensing data of the propeller; And the multi-source sensing data of the propeller at least includes the strain force, vibration frequency and blade surface pressure corresponding to the root, middle and tip of the propeller blade respectively; S2: Preprocess the multi-source sensing data of the propeller to obtain the spatio-temporal data set of the propeller; And the preprocessing includes: denoise the multi-source sensing data of the propeller by the wavelet threshold denoising method to eliminate high-frequency electromagnetic interference, and obtain the denoised data of the propeller; Based on the Kalman filtering algorithm, perform signal compensation on the signal attenuation caused by the electromagnetic shielding of the titanium alloy material in the denoised data of the propeller to obtain the spatio-temporal data of the propeller; S3: Construct a digital twin model of the propeller based on the SDAE algorithm model; S4: Randomly divide the spatio-temporal data set of the propeller into a training set and a test set; According to the training set and the test set, train and evaluate the digital twin model of the propeller to obtain the optimal digital twin model of the propeller; S5: Map the output of the optimal digital twin model of the propeller to a preset three-dimensional propeller grid model to obtain the real-time stress cloud map and fatigue damage hot spot distribution cloud map of the propeller, thereby realizing the health monitoring of the titanium alloy propeller.

2. A method for constructing a digital twin model of a titanium alloy propeller based on the SDAE model according to claim 1, characterized in that The digital twin model of the propeller based on the SDAE algorithm model constructed in S3 includes an input layer, a stacked denoising autoencoder layer, a feature dynamic fusion layer and a decoding layer; The input layer is used to input the spatio-temporal data of the propeller into the stacked denoising autoencoder layer; The stacked denoising autoencoder layer is used to perform layer-by-layer encoding extraction and dimensionality reduction processing on the spatio-temporal data of the propeller to obtain the low-dimensional robustness features of the propeller; The feature dynamic fusion layer is used to perform feature fusion on the low-dimensional robustness features of the propeller combined with the finite element real-time simulation data to obtain the dynamic fusion features of the propeller; and the finite element real-time simulation data is the propeller structure data of the three-dimensional physical model of the propeller established based on the finite element method according to the physical structure of the titanium alloy propeller; The decoding layer is used to decode and reconstruct the spatio-temporal data of the propeller according to the dynamic fusion features of the propeller, and then output the predicted reconstructed data of the propeller.

3. A method for constructing a digital twin model of a titanium alloy propeller based on the SDAE model according to claim 1, characterized in that, The specific steps of S4 are as follows: S41: Define the propeller spatio-temporal data set D and Among them, X t represents the spatio-temporal data of the propeller; T represents the sampling period of the multi-source sensing data of the propeller; t represents the sampling time point; N represents the number of sensor nodes; d represents the feature dimension, namely the strain force, vibration frequency, and blade surface pressure amplitude; Adopt the sliding window mechanism to divide the spatio-temporal data set of the propeller into several local segment data according to the preset sliding window; And randomly divide the local segment data into a training set and a test set; S42: Train the digital twin model of the propeller constructed based on the SDAE algorithm model according to the training set to obtain the trained digital twin model of the propeller; The specific model training is as follows: S421: Input the spatio-temporal data in the training set into the stacked denoising autoencoder layer through the input layer; S422: Perform layer-by-layer encoding extraction and dimensionality reduction processing on the spatio-temporal data of the propeller through the stacked denoising autoencoder layer to obtain the low-dimensional robustness features of the propeller; And the expression of the stacked denoising autoencoder layer is where: h (k) represents the output of the k-th denoising autoencoder layer; σ represents the ReLU activation function; represents the data received by the input layer; W k , b k represent trainable weight parameters; K represents the number of stacked denoising autoencoder layers; h (k-1) represents the output of the (k - 1)-th denoising autoencoder layer; S423: Through the feature dynamic fusion layer, perform feature fusion on the low-dimensional robustness features of the propeller and the finite element real-time simulation data to obtain the dynamic fusion features of the propeller; S424: Decode the dynamic fusion features of the propeller through the decoding layer, reconstruct the spatio-temporal data of the propeller, and output the predicted propeller reconstruction data; S43: Based on the hybrid loss function of the constructed propeller digital twin model, evaluate the trained propeller digital twin model according to the test set, and determine whether the output of the trained propeller digital twin model converges; If so, confirm that the trained propeller digital twin model at this time is the optimal propeller digital twin model; Otherwise, adaptively adjust the weight parameters of the trained propeller digital twin model based on the backpropagation method, and repeat step S42.

4. A method for constructing a digital twin model of a titanium alloy propeller based on the SDAE model according to claim 3, characterized in that, The hybrid loss function constructed in S43 has the following expression Where: L hybrid represents the mixed loss function; α represents the weight coefficient; L SDAE represents the loss function of the predicted propeller reconstruction data and the input propeller spatio-temporal data; h (K) represents the output of the K-th denoising autoencoder layer; W p represents the mapping matrix of the physical structure characteristics of the titanium alloy propeller; F FEM represents the finite element real-time simulation data; represents the predicted propeller reconstruction data; λ represents the L2 regularization coefficient.

5. A method for constructing a digital twin model of a titanium alloy propeller based on the SDAE model according to claim 1, characterized in that, The embedding method of the titanium alloy propeller sensor in S1 is specifically to use laser microfabrication technology to create node grooves for embedding miniature fiber Bragg grating sensors at the root, middle, and tip of the titanium alloy propeller; And the groove depth of the sensor node groove is 2 / 3 of the sensor thickness, and a titanium alloy protective layer with a reserved thickness of 0.1 mm is provided at the bottom of the node groove to avoid fluid corrosion; After the miniature fiber Bragg grating sensor is encapsulated with silicon nitride ceramic glue and embedded in the node groove, titanium alloy powder is covered on the surface, and its surface flatness is repaired by selective laser melting technology SLM; Based on the preset near-field communication NFC module, establish wireless transmission communication for the miniature fiber Bragg grating sensor and the integrated power supply system of the propeller rotation axis to collect and obtain multi-source sensing data of the propeller.

Images