Method and device for evaluating the adhesion of a coating on a water turbine blade

Abstract

The application discloses a kind of water turbine blade coating bonding force evaluation method and device, it is related to hydroelectric generation technical field.The method includes: to the coating area of the water turbine blade to be evaluated, the multimodal nondestructive testing data of synchronous acquisition;Multimodal nondestructive testing data are input to coating bonding force prediction model and are integrated to obtain coating bonding force spatial distribution diagram;Coating bonding force spatial distribution diagram is input to high fidelity digital twin and is coupled simulation and life analysis, and coating residual life distribution diagram is obtained;Coating bonding force spatial distribution diagram and coating residual life distribution diagram are input to intelligent maintenance decision model and are decided generation, and blade maintenance decision is obtained;Coating bonding force spatial distribution diagram and coating residual life distribution diagram are visualized and are shown, and blade coating bonding force evaluation report is generated.The application solves the problems of low evaluation accuracy, single information dimension and evaluation and decision disconnection existed in prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydroelectric power generation, and in particular to a method and device for evaluating the bonding force of a water turbine blade coating. BACKGROUND

[0002] A water turbine is the core energy conversion equipment of a hydropower station. Its blades are long-term operated in high-speed, sand-containing water flow, and the working conditions are harsh, which are prone to cavitation, abrasion and fatigue damage. In order to protect the blades and improve their efficiency, an anti-abrasion and anti-cavitation protective coating is usually coated on the surface of the blades. The performance of the coating, especially the bonding force between the coating and the blade substrate, directly determines the protection effect and the safe operation life of the blades. Once the coating peels off, not only will the substrate be exposed to the harsh environment to accelerate damage, but also the peeled fragments may cause secondary damage to other flow parts of the water turbine, leading to serious accidents.

[0003] Therefore, it is of great importance to accurately and efficiently evaluate the bonding force of the water turbine blade coating and develop a scientific maintenance strategy on this basis to ensure the safe, stable and economic operation of the hydropower station.

[0004] The existing coating bonding force evaluation methods mainly have the following shortcomings:

[0005] 1) Low evaluation accuracy: such as grid method, tensile method, etc., which need to sample or conduct destructive tests on the blades, and cannot comprehensively evaluate the in-service blades, and the representative of the sampling results is limited.

[0006] 2) Low evaluation accuracy: such as ultrasonic wave, eddy current, etc., which can usually only detect macroscopic defects such as interface debonding, and are not sensitive to early microscopic characterization of bonding force degradation (such as micro-crack initiation and interface performance degradation), and it is difficult to quantitatively evaluate the bonding force grade.

[0007] 3) Evaluation and decision-making are disconnected: the traditional evaluation results are mostly qualitative or discrete data points, which are difficult to intuitively reflect the spatial distribution state of the bonding force on the entire blade surface. At the same time, there is a lack of effective data-driven model connection between the evaluation results and the life prediction and maintenance decision-making, resulting in that the development of maintenance plan mainly depends on the experience of experts, which has subjectivity and lag, and cannot realize intelligent and refined management in the whole life cycle.

[0008] In summary, it is urgent to develop a comprehensive technical solution which can accurately, comprehensively and quantitatively evaluate the bonding force of the water turbine blade coating, intelligently predict the remaining life, and automatically generate the optimal maintenance decision. SUMMARY

[0009] The present application provides a method and device for evaluating the bonding force of a water turbine blade coating. The present application solves the problems of low evaluation accuracy, single information dimension and disconnection between evaluation and decision-making in the prior art.

[0010] In a first aspect, the embodiments of the present application provide a method for evaluating the coating adhesion of a water turbine blade, the method comprising:

[0011] Synchronously collecting multi-modal non-destructive testing data of a coating area of the water turbine blade to be evaluated, the multi-modal non-destructive testing data comprising macroscopic images, microscopic images, acoustic emission signals, and thermal imaging maps;

[0012] Inputting the multi-modal non-destructive testing data into a coating adhesion prediction model pre-trained based on a physical information multi-modal fusion algorithm to perform adhesion prediction and integration, and obtaining a coating adhesion spatial distribution map;

[0013] Inputting the coating adhesion spatial distribution map as an initial condition into a high-fidelity digital twin pre-constructed based on a digital twin technology to perform coupling simulation and life analysis, and obtaining a coating residual life distribution map;

[0014] Inputting the coating adhesion spatial distribution map and the coating residual life distribution map into an intelligent maintenance decision model pre-constructed based on a reinforcement learning algorithm to generate a decision, and obtaining a blade maintenance decision;

[0015] Using the high-fidelity digital twin to visually display the coating adhesion spatial distribution map and the coating residual life distribution map, and combining the blade maintenance decision to generate a blade coating adhesion evaluation report.

[0016] The technical solutions provided by the embodiments of the present application at least have the following beneficial effects:

[0017] By synchronously collecting multi-modal data such as macroscopic / microscopic images, acoustic emission, and thermal imaging, and using a physical information fusion model for prediction, the coating adhesion state can be comprehensively and deeply characterized from different physical dimensions, the limitations of single detection technology are overcome, and high-precision evaluation from qualitative to quantitative and from point to plane is realized; the high-precision coating adhesion spatial distribution map is used as an initial condition of the digital twin, the damage evolution process of the coating under real working conditions is deduced through high-fidelity coupling simulation, and dynamic and spatialized prediction of the coating residual life is realized, providing a scientific basis for preventive maintenance; the reinforcement learning model is introduced, multiple factors such as blade health status, power station operation, logistics support, and economic benefits are comprehensively considered, the optimal maintenance decision can be autonomously learned and generated, the blindness and subjectivity of traditional decision-making methods are avoided, and the maximization of operation and maintenance benefits is realized; a complete technical closed loop from data collection, state evaluation, life prediction to intelligent decision-making and report generation is constructed, and intuitive visual display is performed through the digital twin, greatly improving the transparency and efficiency of operation and maintenance work.

[0018] In an optional implementation, the multi-modal non-destructive testing data of the coating area of the water turbine blade to be evaluated is synchronously collected, including:

[0019] Based on the laser scanning technology, a three-dimensional point cloud model of the water turbine blade to be evaluated is obtained, and a global coordinate system of the water turbine blade is established;

[0020] On the three-dimensional point cloud model, a scanning path covering the coating area of the entire blade surface is planned, and a plurality of detection points on the blade surface of the water turbine blade are set according to the scanning path;

[0021] The multi-modal data collection probe is spatially registered in the global coordinate system to obtain the spatially registered multi-modal data collection probe;

[0022] Using the multi-modal data collection probe, multi-modal data is synchronously collected at each detection point to obtain multi-modal non-destructive testing data of the coating area of the water turbine blade to be evaluated.

[0023] In an optional implementation, the coating adhesion force prediction model includes an image feature extraction branch and a time sequence feature extraction branch connected in parallel, and an attention fusion module, a physical parameter output layer and an adhesion force prediction layer connected in sequence;

[0024] The image feature extraction branch includes a macro image feature extractor based on the ResNet-50 algorithm and a micro image feature extractor based on the EfficientNet-B0 algorithm connected in parallel, and a global average pooling layer;

[0025] The time sequence feature extraction branch includes an acoustic time sequence feature extractor based on the MLP algorithm and a thermal imaging time sequence feature extractor based on the 3D-CNN algorithm connected in parallel;

[0026] The coating adhesion force prediction model is provided with a total loss function combining physical information and data loss, the total loss function includes a total physical loss function and a data loss function, the total physical loss function includes a heat conduction physical loss function, a fatigue damage physical loss function and a physical consistency loss function, and the formula is:

[0027]

[0028] In the formula, is the total loss function; is the data loss function; is the total physical loss function; , is the total loss weight;

[0029]

[0030] In the formula, a heat conduction physics loss function; a fatigue damage physics loss function; a physics consistency loss function; a total physics loss weight;

[0031]

[0032] wherein, a MSE function; a predicted surface temperature response curve calculated according to the interfacial thermal resistance output by the physics parameter output layer; a measured surface temperature response curve;

[0033]

[0034] wherein, an acoustic emission signal energy calculation function; an acoustic emission signal energy; a mapping function of the stress factor mapped to the damage space; a stress intensity factor range;

[0035]

[0036] wherein, an interfacial thermal resistance; a microcrack density; a basic interfacial thermal resistance; a damage-thermal resistance coupling coefficient;

[0037] The intelligent maintenance decision model is constructed based on a PPO algorithm, and the intelligent maintenance decision model is provided with an agent and an experience replay pool.

[0038] In an optional implementation, a method for constructing a coating bond force prediction model includes:

[0039] In the laboratory, a plurality of coating test blocks of water turbine blades simulating different working conditions are prepared, for each coating test block, sample multi-modal non-destructive testing data is collected;

[0040] For each coating test block, destructive testing and surface temperature testing are performed to obtain corresponding bond force true value labels and measured surface temperature response curves, and the corresponding sample multi-modal non-destructive testing data is added to obtain a sample data set;

[0041] Based on a physical information multi-modal fusion algorithm, an initial coating bond force prediction model is constructed;

[0042] ​​Based on the total loss function, the hyperparameters of the initial coating adhesion force prediction model are optimized using a quantum-enhanced intelligent optimizer to obtain a final coating adhesion force prediction model.

[0043] In an optional implementation, based on the total loss function, the hyperparameters of the initial coating adhesion force prediction model are optimized using a quantum-enhanced intelligent optimizer to obtain a final coating adhesion force prediction model, comprising:

[0044] The combination of hyperparameters of the initial coating adhesion force prediction model is associated with the QAOA parameters of the quantum-enhanced intelligent optimizer, and the total loss function is used as the cost function of the QAOA parameters;

[0045] The quantum-enhanced intelligent optimizer is initialized using Logistic chaotic mapping to generate initial QAOA parameters and enter a hybrid iterative optimization loop;

[0046] In each hybrid iterative optimization loop, the QAOA circuit is run on a quantum computer using the current QAOA parameters to obtain the current quantum state, and the current hyperparameter combination is sampled;

[0047] On a classical computer, using a sample data set, a candidate coating adhesion force prediction model is quickly trained based on the current hyperparameter combination, and the cost function is used to calculate the current cost value of the candidate coating adhesion force prediction model;

[0048] A flight random number is generated, and if the flight random number is less than a preset threshold, Levy flight exploration is performed to update the current QAOA parameters to obtain updated QAOA parameters, otherwise, gradient descent is used to update the current QAOA parameters to obtain updated QAOA parameters, and the hyperparameter combination obtaining and cost value calculating step is returned;

[0049] If the current number of hybrid iterations reaches a hybrid iteration threshold or the cost value improvement is lower than an improvement threshold, the hybrid iterative optimization step is terminated, and the optimal hyperparameter combination corresponding to the optimal cost value is output;

[0050] According to the optimal hyperparameter combination, the hyperparameters of the initial coating adhesion force prediction model are set to obtain a final coating adhesion force prediction model.

[0051] In an optional implementation, the multi-modal non-destructive testing data is input into the coating adhesion force prediction model pre-trained based on the physical information multi-modal fusion algorithm for adhesion force prediction and integration to obtain a coating adhesion force spatial distribution map, comprising:

[0052] The multi-modal non-destructive testing data is pre-processed to obtain pre-processed multi-modal non-destructive testing data, and the pre-processed multi-modal non-destructive testing data is input into a coating adhesion prediction model pre-trained based on a physical information multi-modal fusion algorithm;

[0053] The macro image feature extractor of the coating adhesion prediction model is used to extract the macro image feature map of the macro image in the pre-processed multi-modal non-destructive testing data, and the global average pooling layer is input to perform global average pooling to obtain the macro image feature;

[0054] The micro image feature extractor of the coating adhesion prediction model is used to extract the micro image feature map of the micro image in the pre-processed multi-modal non-destructive testing data, and the global average pooling layer is input to perform global average pooling to obtain the micro image feature;

[0055] The acoustic time sequence feature extractor of the coating adhesion prediction model is used to extract the acoustic time sequence feature of the acoustic emission signal in the pre-processed multi-modal non-destructive testing data;

[0056] The thermal imaging time sequence feature extractor of the coating adhesion prediction model is used to extract the thermal imaging time sequence feature of the thermal imaging map in the pre-processed multi-modal non-destructive testing data;

[0057] The macro image feature, the micro image feature, the acoustic time sequence feature and the thermal imaging time sequence feature are spliced to obtain a fusion feature vector, which is input into an attention fusion module of the coating adhesion prediction model to perform attention fusion to obtain a weighted fusion feature;

[0058] According to the weighted fusion feature, a physical parameter output layer of the coating adhesion prediction model is used to generate a physical parameter; the physical parameter includes an interfacial thermal resistance, a micro crack density and an interfacial bonding strength;

[0059] According to the physical parameter, a bonding force prediction layer of the coating adhesion prediction model is used to call a Softmax function to generate a bonding force grade, and the corresponding physical parameter is combined to obtain a bonding force prediction result;

[0060] According to the spatial coordinate value of each detection point in the global coordinate system, the bonding force prediction results of all detection points are integrated to obtain a corresponding coating adhesion spatial distribution map.

[0061] In an optional implementation, the coating adhesion spatial distribution map is input into a high-fidelity digital twin pre-constructed based on digital twin technology as an initial condition to perform coupling simulation and life analysis to obtain a coating residual life distribution map, including:

[0062] The three-dimensional point cloud model of the water turbine blade to be evaluated is imported into a simulation software, and a high-quality hexahedral dominant mesh is divided to obtain a corresponding three-dimensional simulation model;

[0063] Based on digital twin technology, the elastoplastic material models of the matrix and coating are defined in the three-dimensional simulation model, and the damage evolution equation corresponding to the elastoplastic material model of the coating is set to obtain a pre-constructed high-fidelity digital twin.

[0064] The spatial distribution map of coating adhesion is used as the initial condition for the damage evolution equation, mapped to a high-fidelity digital twin, and the high-fidelity digital twin is then used for model calibration.

[0065] Coupled simulation and damage accumulation of a high-fidelity digital twin are performed under simulated specific working conditions to obtain the damage value of each detection point;

[0066] If the damage value exceeds the damage threshold for the first time, the detection point is taken as the critical detection point, and the coupled simulation time experienced by the critical detection point is taken as the corresponding predicted remaining lifetime.

[0067] If the damage value does not exceed the damage threshold, then continue the coupled simulation and damage accumulation based on that detection point;

[0068] Based on the spatial coordinates of each key detection point in the global coordinate system, the predicted remaining lifetime of all key detection points is integrated to obtain the coating remaining lifetime distribution map.

[0069] In one optional implementation, the spatial distribution map of coating adhesion and the distribution map of remaining coating lifetime are input into a pre-built intelligent maintenance decision model based on a reinforcement learning algorithm to generate a blade maintenance decision, including:

[0070] Global statistical features, local key area features, and spatial context features of the spatial distribution map of coating adhesion and the distribution map of coating remaining life are extracted and combined with hydropower station status data to obtain complete status observation.

[0071] Write complete state observations into the state space of the agent in a pre-constructed intelligent maintenance decision model based on reinforcement learning algorithms;

[0072] Based on the state space, an intelligent agent selects the optimal action in the preset action space to obtain the blade maintenance decision and collects real-time experience in the decision generation process.

[0073] Several historical experiences are randomly selected from the experience replay pool and combined with real-time experiences to continuously train the intelligent maintenance decision model, resulting in an updated intelligent maintenance decision model.

[0074] In one alternative implementation, a high-fidelity digital twin is used to visualize the spatial distribution map of coating adhesion and the distribution map of remaining coating lifetime. Combined with blade maintenance decisions, a blade coating adhesion assessment report is generated, including:

[0075] load a high-fidelity digital twin of the water turbine blade to be evaluated;

[0076] The coating adhesion spatial distribution map and the coating residual life distribution map are superimposed and rendered on the high-fidelity digital twin in the form of heat maps of different colors as independent layers, and are visually displayed.

[0077] The coating adhesion spatial distribution map, the coating residual life distribution map, and the corresponding blade maintenance decision are written into a preset report template to generate a blade coating adhesion evaluation report, and are visually displayed.

[0078] In a second aspect, an embodiment of the present application provides a water turbine blade coating adhesion evaluation device for implementing the water turbine blade coating adhesion evaluation method, and the device comprises:

[0079] A multi-modal data acquisition unit is configured to synchronously acquire multi-modal non-destructive testing data of a coating area of the water turbine blade to be evaluated.

[0080] A coating adhesion prediction unit is configured to input the multi-modal non-destructive testing data into a coating adhesion prediction model based on a physical information multi-modal fusion algorithm to perform adhesion prediction and integration, and obtain a coating adhesion spatial distribution map.

[0081] A coating residual life analysis unit is configured to input the coating adhesion spatial distribution map as an initial condition into a high-fidelity digital twin based on a digital twin technology to perform coupling simulation and life analysis, and obtain a coating residual life distribution map.

[0082] A blade maintenance decision generation unit is configured to input the coating adhesion spatial distribution map and the coating residual life distribution map into an intelligent maintenance decision model based on a reinforcement learning algorithm to generate a decision, and obtain a blade maintenance decision.

[0083] A visual display unit is configured to visually display the coating adhesion spatial distribution map and the coating residual life distribution map using the high-fidelity digital twin, and generate a blade coating adhesion evaluation report in combination with the blade maintenance decision.

[0084] An embodiment of the present application provides an electronic device, which comprises:

[0085] at least one processor; and a memory connected with the at least one processor; wherein

[0086] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the first aspect of the present application.

[0087] The fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the method according to the first aspect of the embodiment of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0088] Figure 1 The electronic device structure schematic diagram of the hardware running environment related to the embodiment of the present application.

[0089] Figure 2 The step flow chart of the water turbine blade coating bonding force evaluation method provided by the embodiment of the present application.

[0090] Figure 3 The functional unit schematic diagram of the water turbine blade coating bonding force evaluation device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0091] In order to make the above objectives, characteristics and advantages of the present application more apparent, comprehensible and easier to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0092] The scheme of the present application will be further described below with reference to the drawings.

[0093] REFERENCE Figure 1 , Figure 1 The electronic device structure schematic diagram of the hardware running environment related to the embodiment of the present application.

[0094] As Figure 1As shown, the electronic device can include: a processor 1001, for example, a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, a memory 1005. Among them, the communication bus 1002 is used to realize the connection communication between these components. The user interface 1003 can include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 can also include a standard wired interface, a wireless interface. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface). The memory 1005 can be a high-speed random access memory (RAM) memory, or a stable non-volatile memory (NVM), such as a disk memory. The memory 1005 can also be an independent storage device from the aforementioned processor 1001.

[0095] Those skilled in the art can understand that, Figure 1 The structure shown in the figure does not constitute a limitation on the electronic device, and can include more or fewer components than the figure, or combine certain components, or different component arrangements.

[0096] As Figure 1 As shown, the memory 1005 as a storage medium can include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program of the water turbine blade coating bonding force evaluation device.

[0097] In Figure 1 As shown in the electronic device, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present application can be arranged in the electronic device, and the electronic device calls the electronic program of the water turbine blade coating bonding force evaluation device stored in the memory 1005 through the processor 1001, and executes the water turbine blade coating bonding force evaluation method provided by the embodiment of the present application.

[0098] Referring to Figure 2 , the embodiment of the present application provides a water turbine blade coating bonding force evaluation method, the method comprising:

[0099] S201: synchronously collecting multi-modal non-destructive testing data on the coating area of the water turbine blade to be evaluated; the multi-modal non-destructive testing data includes macroscopic image, microscopic image, acoustic emission signal and thermal imaging spectrum;

[0100] S202: input the multi-modal non-destructive testing data into the coating adhesion prediction model pre-trained based on the physical information multi-modal fusion algorithm for adhesion prediction and integration, to obtain a coating adhesion spatial distribution map;

[0101] S203: input the coating adhesion spatial distribution map as an initial condition into the high-fidelity digital twin pre-constructed based on the digital twin technology for coupling simulation and life analysis, to obtain a coating residual life distribution map;

[0102] S204: input the coating adhesion spatial distribution map and the coating residual life distribution map into the intelligent maintenance decision model pre-constructed based on the reinforcement learning algorithm for decision generation, to obtain a blade maintenance decision;

[0103] S205: use the high-fidelity digital twin to visually display the coating adhesion spatial distribution map and the coating residual life distribution map, and combine the blade maintenance decision to generate a blade coating adhesion evaluation report.

[0104] The technical scheme provided by the embodiment of the application at least brings the following beneficial effects:

[0105] By synchronously collecting multi-modal data such as macro / micro images, acoustic emission, thermal imaging, and using a physical information fusion model for prediction, the coating adhesion state can be comprehensively and deeply represented from different physical dimensions, the limitations of single detection technology are overcome, and high-precision evaluation from qualitative to quantitative and from point to plane is realized; the high-precision coating adhesion spatial distribution map is used as an initial condition of the digital twin, the damage evolution process of the coating under real working conditions is deduced through high-fidelity coupling simulation, and dynamic and spatial prediction of the coating residual life is realized, providing a scientific basis for preventive maintenance; the reinforcement learning model is introduced, multiple factors such as blade health status, power station operation, logistics support and economic benefits are considered, the optimal maintenance decision can be autonomously learned and generated, the blindness and subjectivity of traditional decision-making methods are avoided, and the maximization of operation and maintenance benefits is realized; a complete technical closed loop from data acquisition, state evaluation, life prediction to intelligent decision-making and report generation is constructed, and intuitive visual display is realized through the digital twin, greatly improving the transparency and efficiency of operation and maintenance work.

[0106] In an optional implementation, multi-modal non-destructive testing data of a coating area of a water turbine blade to be evaluated is synchronously collected, including:

[0107] S2011: based on a laser scanning technology, a three-dimensional point cloud model of the water turbine blade to be evaluated is obtained, and a global coordinate system of the water turbine blade is established;

[0108] In the embodiment, the detailed steps include:

[0109] Using ground-based or handheld three-dimensional laser scanners, the water turbine blade in the stopped state is scanned comprehensively to obtain original point cloud data containing millions of data points;

[0110] The original point cloud data is denoised, registered (aligning the data of multiple scans), and packaged using point cloud processing software (such as Geomagic, PolyWorks) to generate a continuous, flawless three-dimensional point cloud model (STL or OBJ format) of the water turbine blade to be evaluated;

[0111] A global three-dimensional Cartesian coordinate system (O-XYZ), i.e., a global coordinate system, is defined on the three-dimensional point cloud model; generally, the origin O can be set at the center of the root flange of the water turbine blade to be evaluated, the X-axis can point radially to the outer edge of the blade, and the Z-axis can be along the direction of the main shaft of the water turbine; all subsequent spatial data will be aligned with this coordinate system;

[0112] S2012: On the three-dimensional point cloud model, a scanning path covering the entire coating area of the blade surface is planned, and a plurality of detection points are set on the blade surface of the water turbine blade according to the scanning path;

[0113] In this embodiment, the detailed steps include:

[0114] On the three-dimensional point cloud model, a path planning algorithm is run; the algorithm automatically generates one or more "S" or spiral-shaped scanning paths that can efficiently cover the entire surface based on the curvature distribution of the blade and the stress distribution history (if any);

[0115] In areas with high curvature and stress concentration (such as the water inlet edge and the water outlet edge), the path density is higher and the detection point spacing is smaller (such as 5 cm); in flat areas, the spacing can be appropriately increased (such as 10 cm);

[0116] A series of discrete detection points are generated along the planned path at fixed intervals, and each detection point has its precise coordinates in the global coordinate system and a surface normal vector;

[0117] S2013: Spatially registering the multi-modal data acquisition probe in the global coordinate system to obtain a spatially registered multi-modal data acquisition probe;

[0118] In this embodiment, the detailed steps include:

[0119] An industrial camera, a long working distance microscope, an acoustic emission sensor array, an infrared thermal imager, an electromagnetic exciter, etc. are integrated in a compact multi-modal data acquisition probe housing; a high-precision positioning target (such as a reflective marker ball) is installed at the end of the multi-modal data acquisition probe;

[0120] Through the "hand-eye calibration" process, the transformation relationship between the coordinate system of each sensor inside the probe and the coordinate system of the positioning target is accurately determined;

[0121] The multi-modal data acquisition probe is installed on a robot arm or a rail system; one or more tracking cameras (such as a laser tracker) are set up beside the water turbine blade to be evaluated; the tracking camera tracks the positioning target on the probe in real time, thereby calculating the accurate pose (position and attitude) of the probe end (i.e. the center of the sensor) in the global coordinate system O-XYZ in real time; this enables each frame of data collected to be accurately assigned a three-dimensional spatial coordinate;

[0122] S2014: Using the multi-modal data acquisition probe, a standardized multi-modal data acquisition process is performed at each detection point in sequence, and multi-modal data is synchronously collected to obtain multi-modal non-destructive testing data of the coating area of the water turbine blade to be evaluated;

[0123] In this embodiment, the detailed steps include:

[0124] The robot arm moves the multi-modal data acquisition probe to the detection point and adjusts the pose of the multi-modal data acquisition probe according to the surface normal vector so that it is perpendicular to the measured surface; the central controller sends a synchronous trigger signal;

[0125] Macro and micro image acquisition:

[0126] Macro image: The industrial camera captures an image covering an area of about 10 cm x 10 cm under the uniform illumination of the ring-shaped LED light source, with a resolution of no less than 24 million pixels;

[0127] Micro image: A high-power optical microscope (such as 200x) is aligned with the center area of the macro image to capture the micro morphology of the coating surface, such as pores, micro-cracks, and particle distribution, etc.

[0128] Acoustic emission signal acquisition:

[0129] Excitation: The electromagnetic exciter generates a pulsed mechanical wave with a frequency of 20 kHz and a duration of 5 ms to apply a small dynamic stress to the coating;

[0130] Acquisition: Four acoustic emission sensors (acoustic emission sensor array, sampling rate set to 5 MHz) arranged around the probe synchronously collect acoustic emission signals within 50 ms after excitation; through the time difference of arrival algorithm, the acoustic source can be preliminarily located;

[0131] Thermal imaging map acquisition:

[0132] Excitation: Two high-energy pulse xenon lamps (energy about 6 kJ) are aimed at the detection area for 2 ms of flash heating;

[0133] Acquisition: Infrared thermal imager (spectral range 3-5 μm, thermal sensitivity <20 mK, frame rate 200 Hz) records the complete temperature change sequence from the beginning of heating to the end of cooling (about 10 seconds);

[0134] The spatial coordinates, timestamps, and all collected raw data of the detection point are packaged to obtain multi-modal nondestructive testing data and stored in a database.

[0135] In an optional implementation, the coating adhesion force prediction model includes an image feature extraction branch and a time sequence feature extraction branch connected in parallel, and an attention fusion module, a physical parameter output layer, and an adhesion force prediction layer connected in sequence;

[0136] The image feature extraction branch includes a macroscopic image feature extractor constructed based on a 50-layer residual network (Residual Network 50, ResNet-50) algorithm and a microscopic image feature extractor based on an EfficientNet-B0 algorithm, and a global average pooling layer;

[0137] The time sequence feature extraction branch includes an acoustic time sequence feature extractor constructed based on a multi-layer perceptron (Multi-Layer Perceptron, MLP) algorithm and a thermal imaging time sequence feature extractor based on a 3-dimensional convolutional neural network (3-Dimensional Convolutional Neural Network, 3D-CNN) algorithm connected in parallel;

[0138] The coating adhesion force prediction model is provided with a total loss function combining physical information and data loss, the total loss function includes a total physical loss function and a data loss function, the total physical loss function includes a heat conduction physical loss function, a fatigue damage physical loss function, and a physical consistency loss function, and the formula is:

[0139]

[0140] In the formula, is the total loss function; is the data loss function; is the total physical loss function; , is the total loss weight;

[0141]

[0142] In the formula, is the heat conduction physical loss function; is the fatigue damage physical loss function; is the physical consistency loss function; , , is the total physical loss weight;

[0143] For thermal imaging data, a one-dimensional transient heat conduction model is established, and the formula is:

[0144]

[0145] In the formula, is the temperature; x is the spatial position; t is the time; pc is the volumetric heat capacity; k is the thermal conductivity; is the rate of change of temperature with respect to time; is the temperature gradient; is the divergence of heat flow; is the partial differential symbol;

[0146] Where, at the coating-substrate interface , there is a temperature jump caused by the interface thermal resistance , where, d is the interface position, and the formula is:

[0147]

[0148] In the formula, is the interface coating side temperature; is the interface substrate side temperature; is the temperature jump; is the heat flux density at the interface;

[0149] The interface thermal resistance predicted by the coating bond force prediction model is taken as a known parameter, and the partial differential equation is solved by finite difference method to obtain the predicted surface temperature response curve , and the mean square error between and the measured surface temperature response curve is calculated, and the formula is:

[0150]

[0151] In the formula, is the mean square error (MSE) function; is the predicted surface temperature response curve calculated according to the interface thermal resistance output by the physical parameter output layer; is the measured surface temperature response curve;

[0152] For acoustic emission signals, the energy can be regarded as the damage increment Correlation, meanwhile, according to the Paris law of fracture mechanics, the damage increment and the stress intensity factor range Correlation; estimate the stress amplitude of the detection point through a simplified blade stress model or finite element simulation, and then calculate the stress intensity factor range ;

[0153]

[0154] In the formula, is the acoustic emission signal energy calculation function; is the acoustic emission signal energy; is the mapping function of the stress factor mapped to the damage space; is the stress intensity factor range;

[0155]

[0156] In the formula, is the interfacial thermal resistance; is the microcrack density; is the base interfacial thermal resistance; is the damage-thermal resistance coupling coefficient;

[0157] The intelligent maintenance decision model is constructed based on a proximal policy optimization (PPO) algorithm, and the intelligent maintenance decision model is provided with an agent and an experience replay pool.

[0158] In an optional implementation, the method for constructing the coating bond force prediction model comprises:

[0159] A-1: A plurality of coating test blocks of water turbine blades simulating different working conditions are prepared in a laboratory, for each coating test block, sample multi-modal non-destructive testing data is collected;

[0160] A-2: For each coating test block, destructive testing and surface temperature testing are performed to obtain corresponding bond force true value labels and measured surface temperature response curves, and the labels and curves are added to the corresponding sample multi-modal non-destructive testing data to obtain a sample data set;

[0161] A-3: An initial coating bond force prediction model is constructed based on a physical information multi-modal fusion algorithm;

[0162] A-4: Based on a total loss function, a quantum-enhanced intelligent optimizer is used to optimize the hyperparameters of the initial coating bond force prediction model to obtain a final coating bond force prediction model.

[0163] In an optional implementation, based on the total loss function, the hyperparameters of the initial coating adhesion force prediction model are optimized using a quantum-enhanced intelligent optimizer to obtain a final coating adhesion force prediction model, including:

[0164] A-41: The hyperparameter combination of the initial coating adhesion force prediction model is associated to the quantum approximate optimization algorithm (QAOA) parameters of the quantum-enhanced intelligent optimizer, and the total loss function is taken as the cost function of the QAOA parameters;

[0165] In this embodiment, the hyperparameter combination includes network structure parameters, learning rate, total loss weight , total physical loss weight , Dropout ratio, etc.

[0166] The hyperparameter optimization problem is mapped to the cost function minimization problem of QAOA, and each hyperparameter in the hyperparameter combination is mapped to a set of computational ground states.

[0167] The formula of the cost function is:

[0168]

[0169] In the formula, is the cost function, that is, the cost Hamiltonian; is the hyperparameter combination; is the hyperparameter search space; is the hyperparameter combination The model constructed in the verification set uses the total loss value obtained by the total loss function; is the computational ground state; is the conjugate transpose of the computational ground state;

[0170] A-42: The quantum-enhanced intelligent optimizer is initialized using Logistic chaotic mapping to generate initial QAOA parameters, and enters a hybrid iterative optimization loop;

[0171] The formula is:

[0172]

[0173] In the formula, , is the first n+ 1, n chaotic variable; is a control parameter; compared with random initialization, chaotic initialization can ensure that the population is uniformly distributed in the solution space and enhance diversity; nfor the iteration index;

[0174] The generated chaotic sequence is linearly mapped to the typical range [0, 2π] of QAOA parameters:

[0175]

[0176] where, , is the initial QAOA parameter; is the initial cost parameter; is the initial mixing parameter; is the number of QAOA layers; is the total number of QAOA layers; , is the QAOA parameter of the - 1, + - 1 chaotic variable;

[0177] A-43: In each mixing iteration optimization cycle, run the QAOA circuit on the quantum computer using the current QAOA parameters to obtain the current quantum state, and sample the current hyperparameter combination;

[0178] Prepare the initial uniform superposition state on the quantum computer, and apply the parameterized quantum circuit , whose formula is:

[0179]

[0180] where, is the parameterized unitary operator, and is the overall evolution matrix of the QAOA circuit, which acts on the initial quantum state to generate a superposition state containing optimization problem information, whose behavior is completely controlled by the parameter ; is the QAOA parameter of the th mixing iteration; is the cost parameter of the th mixing iteration; is the mixing parameter of the th mixing iteration; is the number of mixing iterations;

[0181]

[0182] where, is the continuous product; is the problem unit evolution, which is used to encode the optimization problem information into the quantum state; For hybrid unit evolution, used for mixing and exploration in the solution space; , For the first The first mixed iteration The QAOA parameters of the layer; For the first The first mixed iteration Cost parameters of the layer; For the first The first mixed iteration Layer blending parameters; Operators for mixing Hamiltonians to achieve quantum tunneling and state transition; It is a quantum state; To achieve cost Hamiltonians, we need to optimize the quantum representation of the problem. i The imaginary unit; It is a natural constant;

[0183] For quantum states Multiple measurements are performed to obtain a series of measurement results. Each measurement result is decoded to obtain a sampled hyperparameter combination. ,in, j This is a hyperparameter combination indicator;

[0184] A-44: On a classic computer, using a sample dataset, quickly train alternative coating adhesion prediction models based on the current hyperparameter combination, and use a cost function to calculate the current cost value of the alternative coating adhesion prediction models.

[0185] The formula is:

[0186]

[0187] In the formula, For the first The expected cost of the next hybrid iteration, i.e., the cost value; Hyperparameter combination The constructed model on the validation set The total loss value obtained using the total loss function; For quantum states Number of measurements performed;

[0188] A-45: Generate a flight random number. If the flight random number is less than the preset threshold, perform Levy flight exploration, update the current QAOA parameters, and obtain the updated QAOA parameters. Otherwise, perform gradient descent to update the current QAOA parameters, obtain the updated QAOA parameters, and return the hyperparameter combination acquisition and cost value calculation steps.

[0189] Generate a uniformly distributed flight random number.r~ , if the flight random number r is less than a preset threshold , Levy flight exploration is performed, and the formula is:

[0190]

[0191] wherein, is a Levy distributed random number; b is a Levy step size, and b ∈[1, 2]; is the QAOA parameter of the m-th mixed iteration; +1; is the cost parameter of the m-th mixed iteration; +1; is the mixing parameter of the m-th mixed iteration; is an adaptive step size; is an element-wise multiplication;

[0192]

[0193] wherein, is an initial step size; is a maximum value and a minimum value of a convergence factor; is a maximum number of iterations; , is an adjustment parameter; is a hyperbolic tangent function;

[0194] Otherwise, gradient descent is performed, and the cost gradient is estimated on a quantum computer using a parameter shift rule, and the formula is:

[0195]

[0196] wherein, is the gradient of the expected cost with respect to the parameter , that is, ; is the expected cost after forward shifting; is the expected cost after negative shifting;

[0197]

[0198] wherein, is a classical learning rate, which is a step size of gradient descent; is the gradient of the expected cost, which is a vector pointing to the steepest rising direction of the cost function;

[0199] ​A-46: If the current number of hybrid iterations reaches the number of hybrid iteration threshold or the cost value improvement is lower than the improvement threshold, terminate the hybrid iteration optimization step, and output the optimal cost value corresponding to the optimal hyperparameter combination;

[0200] A-47: According to the optimal hyperparameter combination, set the hyperparameters of the initial coating adhesion prediction model to obtain the final coating adhesion prediction model.

[0201] In an optional implementation, the multi-modal non-destructive testing data is input into the coating adhesion prediction model pre-trained based on the physical information multi-modal fusion algorithm for adhesion prediction and integration to obtain a coating adhesion spatial distribution map, including:

[0202] S2021: Data preprocessing is performed on the multi-modal non-destructive testing data to obtain pre-processed multi-modal non-destructive testing data, and the pre-processed multi-modal non-destructive testing data is input into the coating adhesion prediction model pre-trained based on the physical information multi-modal fusion algorithm;

[0203] In this embodiment, the data preprocessing includes:

[0204] Image data: Distortion correction, white balance adjustment, contrast limited adaptive histogram equalization enhancement are performed on the macro and micro images, and the images are cropped to a uniform size (such as 224x224 pixels);

[0205] Acoustic emission signal: The original waveform is subjected to a 20kHz-2MHz bandpass filter, and then key features such as event count, ring count, energy, amplitude, duration, rise time are extracted, or the processed time domain waveform is directly input into the model;

[0206] Thermal imaging map: The original temperature sequence is subjected to non-uniformity correction and noise filtering; then, the temperature-time curve of each pixel point is fitted, and key thermal features such as peak temperature, maximum temperature difference, cooling rate constant are extracted; or the temperature sequence frames are directly stacked into a pseudo-three-dimensional data block to input into the model;

[0207] S2022: The macro image feature extractor of the coating adhesion prediction model is used to extract the macro feature map of the macro image in the pre-processed multi-modal non-destructive testing data, and the global average pooling layer is input to perform global average pooling to obtain the macro image feature (a 2048-dimensional feature vector);

[0208] S2023: The micro image feature extractor of the coating adhesion prediction model is used to extract the micro feature map of the micro image in the pre-processed multi-modal non-destructive testing data, and the global average pooling layer is input to perform global average pooling to obtain the micro image feature (a 1280-dimensional feature vector);

[0209] S2024: An acoustic time series feature extractor of the coating adhesion force prediction model composed of 3 fully connected layers, ReLU activation function and Dropout layer are used after each layer to prevent overfitting, acoustic time series features (256-dimensional feature vector) of the acoustic emission signal in the preprocessed multi-modal non-destructive testing data are extracted;

[0210] S2025: A thermal imaging time series feature extractor of the coating adhesion force prediction model composed of two 3D convolution layers, one 3D max pooling layer and one flattening layer, aiming to capture the dynamic change pattern of temperature in time and space, thermal imaging time series features (spatiotemporal dynamic features, 512-dimensional feature vector) of the thermal imaging map in the preprocessed multi-modal non-destructive testing data are extracted;

[0211] S2026: The macroscopic image features, microscopic image features, acoustic time series features and thermal imaging time series features are spliced to obtain a fusion feature vector, which is input into the attention fusion module of the coating adhesion force prediction model, the module calculates the importance weight of each feature dimension, generates a weighted fusion feature vector, so that the model can focus on the most critical information, and performs attention fusion to obtain a weighted fusion feature;

[0212] S2027: According to the weighted fusion feature, a physical parameter output layer of the coating adhesion force prediction model is used to generate physical parameters (parameters with clear physical meaning); the physical parameters include interfacial thermal resistance, microcrack density and interfacial adhesion strength;

[0213] In this embodiment, the physical parameter output layer includes a thermal resistance branch, a crack density branch and an adhesion strength branch;

[0214] Thermal resistance branch: a fully connected layer outputs a scalar value, which is the predicted interfacial thermal resistance (unit: m²K / W);

[0215] Crack density branch: a fully connected layer outputs a scalar value, which is the predicted microcrack density (unit: pieces / mm²);

[0216] Adhesion strength branch: a fully connected layer outputs a scalar value, which is the predicted interfacial adhesion strength (unit: MPa);

[0217] S2028: According to the physical parameters, a bonding force prediction layer of the coating adhesion force prediction model is used to call the Softmax function to generate the bonding force grade (output the probability distribution of belonging to the four grades of "excellent, good, medium and poor"), and combine the corresponding physical parameters to obtain the bonding force prediction result;

[0218] S2029: According to the spatial coordinate values of each detection point in the global coordinate system, the bonding force prediction results of all detection points are integrated to obtain the corresponding coating bonding force spatial distribution map;

[0219] In this embodiment, the spatial coordinates of each detection point are associated with the corresponding bonding force prediction results, and the data of these discrete points are expanded to the entire blade surface through an interpolation algorithm (such as Kriging interpolation) to generate a continuous and visual "blade coating bonding force spatial distribution map".

[0220] In an optional implementation, the coating bonding force spatial distribution map is input as an initial condition to the high-fidelity digital twin body pre-constructed based on the digital twin technology for coupling simulation and life analysis to obtain a coating residual life distribution map, including:

[0221] S2031: The three-dimensional point cloud model of the water turbine blade to be evaluated is imported into a simulation software (such as ANSYS or ABAQUS, etc.), and high-quality hexahedral dominant mesh division is performed thereon, especially in the coating area with a thickness of only microns or millimeters, the mesh is extremely refined to ensure that the physical field at the interface can be accurately captured to obtain a corresponding three-dimensional simulation model;

[0222] S2032: Based on the digital twin technology, an elastic-plastic material model of the substrate and the coating is defined in the three-dimensional simulation model, and a damage evolution equation corresponding to the elastic-plastic material model of the coating is set to obtain a pre-constructed high-fidelity digital twin body;

[0223] In this embodiment, the substrate: the elastic-plastic constitutive model of the blade substrate material (such as stainless steel) is defined;

[0224] Coating: the elastic-plastic constitutive model of the coating is defined, and a damage evolution equation is introduced; for example, a continuous medium damage mechanics model is used, and the relationship between the damage variable and the elastic modulus is When =0, the material is intact; when =1, the material is completely failed;

[0225] Digital twin body generation: the defined geometry, mesh, material model, and boundary condition setting jointly constitute the high-fidelity digital twin body of the blade;

[0226] S2033: The coating bonding force spatial distribution map is mapped to the high-fidelity digital twin body as an initial condition of the damage evolution equation, and the high-fidelity digital twin body is calibrated;

[0227] Using the "bonding force spatial distribution map" as an initial condition, it is mapped to the coating part of a high-fidelity digital twin through a script or API, and the initial damage variables are obtained. Establish a mapping relationship, the formula is:

[0228]

[0229] In the formula, The initial damage variable; For interface bonding strength; The ultimate tensile strength of the coating or substrate material; This is the shape factor, used to adjust the steepness of the curve; It is a natural constant;

[0230] The model is calibrated using known failure cases from the blade's historical operating data. The parameters in the damage evolution equation are adjusted in reverse to make the simulation results consistent with historical records, thereby improving the model's prediction accuracy.

[0231] S2034: Perform coupled simulation and damage accumulation on a high-fidelity digital twin under simulated specific working conditions to obtain the damage value at each detection point;

[0232] In this embodiment, future power generation plans (such as head, flow rate, and output curves for the next year) are obtained from the power plant monitoring system; these data are used as input conditions to perform fluid-structure interaction simulation on a high-fidelity digital twin; the simulation software calculates the water flow pressure field, which is applied to the structural finite element module to calculate the dynamic stress-strain field of the blades;

[0233] After each time step (or load cycle), based on the stress level and current damage value of each grid within the high-fidelity digital twin coating section. Calculate the damage increment in this step. ;

[0234] Using Miner's linear cumulative damage theory, the total damage value ,in, For the first l Damage value of the mesh, l For grid indication quantity, The number of cycles required to cause failure at the current stress level; update the damage value for each mesh. ;

[0235] S2035: If the damage value exceeds the damage threshold for the first time, the detection point is taken as the critical detection point, and the coupled simulation time experienced by the critical detection point is taken as the corresponding predicted remaining lifetime.

[0236] S2036: If the damage value does not exceed the damage threshold, continue coupling simulation and damage accumulation based on the detection point;

[0237] S2037: According to the spatial coordinate values of each key detection point in the global coordinate system, integrate the predicted residual life of all key detection points to obtain a coating residual life distribution map.

[0238] In an optional implementation, the method for constructing the intelligent maintenance decision model comprises:

[0239] B-1: Taking the hydropower station operation and maintenance environment as the simulation environment;

[0240] B-2: Defining the state space of the agent of the intelligent maintenance decision model, including global statistical features, local key area features, spatial context features, operation state vectors corresponding to hydropower station state data, logistics state vectors, and economic state vectors, etc.

[0241] The global statistical features of the bond force spatial distribution map include:

[0242] Deterioration area ratio (for example, calculate the percentage of the area with a bond force of "poor" (level 3) in the total blade area. This is a very intuitive risk indicator), average bond force / standard deviation (the average value reflects the overall level, and the standard deviation reflects the uniformity of damage, and a high standard deviation may mean that there is local severe damage), number of deterioration areas (through connected component analysis, calculate the number of all deterioration areas, and multiple small isolated damages may have higher risk than a single large damage);

[0243] The global statistical features of the residual life distribution map include:

[0244] Shortest residual life (this is the most critical feature, which directly determines the urgency of maintenance), average residual life / standard deviation (reflecting overall life expectancy and unevenness of life distribution), critical life area ratio (for example, calculate the area ratio of the area with a residual life of less than 3 months);

[0245] Key areas include: weakest area (find the point (or area) with the worst bond force in the bond force spatial distribution map), shortest life area (find the point (or area) with the shortest residual life in the residual life distribution map), high-risk area (define a comprehensive risk index and find the area with the highest comprehensive risk index);

[0246] The formula is:

[0247]

[0248] In the formula, is the comprehensive risk index; is a normalization function; Bonding force degradation value: the bonding force evaluation result for a certain region, which is the average value of the bonding force level for all the grids in this region (e.g., level 0-3, the larger the value, the worse the result); Shortest remaining life for this region; , Risk weight;

[0249] Local key region features include:

[0250] Location information: the 3D coordinates of the region centroid; this is important because damages at different locations (e.g., suction side vs. pressure side) have different maintenance strategies and risks;

[0251] Size information: the area or the equivalent diameter of the region;

[0252] Severity indicators: the average bonding force and the minimum remaining life in this region;

[0253] Spatial context features include:

[0254] Radial distribution features: divide the blade into multiple annular bands (e.g., 0-10%, 10%-20%, …, 90%-100%); calculate the average bonding force and the average remaining life in each annular band; this can reveal whether damages are more concentrated near the blade tip (where the stress is usually higher);

[0255] Spatial autocorrelation: calculate the Moran’s I index to measure whether the damages are clustered, random, or dispersed; a clustered damage pattern can indicate a common cause (e.g., cavitation);

[0256] Operational state vector includes: current time, current head, current flow, current output, predicted load curve for a certain period in the future, etc.;

[0257] Logistical state vector includes: spare parts inventory (e.g., amount of repair coating material), maintenance team availability (e.g., whether they are on-site, whether they have other tasks), planned downtime window, etc.;

[0258] Economic state vector includes: current electricity price, expected future electricity price, unit downtime loss cost, unit maintenance cost, etc.;

[0259] B-2: Define the action space of the agent for the intelligent maintenance decision model, including: {immediate major repair, immediate local repair, delay for 1 month, delay for 3 months, 10% load reduction, 20% load reduction, no operation};

[0260] B-3: Define a complex reward function to evaluate the long-term benefits of each action;

[0261]

[0262] wherein, is a reward value; is a repair cost; is a downtime loss per unit time; is a repair duration; is a power generation loss during downtime; is a potential loss penalty caused by a risk (e.g., in-flight shedding);

[0263] B-4: In a simulation environment, train the agent for millions of interactions, build an intelligent maintenance decision-making model, and set up an experience replay pool;

[0264] The agent learns an optimal strategy through continuous trial and error, which can select the action with the highest expected cumulative reward in any given state.

[0265] In an optional implementation, the coating adhesion force spatial distribution map and the coating residual life distribution map are input into an intelligent maintenance decision-making model pre-built based on a reinforcement learning algorithm to generate a decision, obtaining a blade maintenance decision, including:

[0266] S2041: Extract global statistical features, local key area features, and spatial context features of the coating adhesion force spatial distribution map and the coating residual life distribution map, and combine with hydropower station state data to obtain a complete state observation;

[0267] S2042: Write the complete state observation into the state space of the agent of the intelligent maintenance decision-making model pre-built based on the reinforcement learning algorithm;

[0268] S2043: Based on the state space, use the agent to select an optimal action in a preset action space, obtain a blade maintenance decision, and collect real-time experience in the decision-making process;

[0269] S2044: Randomly extract a number of historical experiences from the experience replay pool, and combine the real-time experience to continuously train the intelligent maintenance decision-making model, obtaining an updated intelligent maintenance decision-making model.

[0270] In an optional implementation, a high-fidelity digital twin is used to visually display the coating adhesion force spatial distribution map and the coating residual life distribution map, and a blade coating adhesion force evaluation report is generated in combination with the blade maintenance decision, including:

[0271] S2051: Load the high-fidelity digital twin of the water turbine blade to be evaluated;

[0272] S2052: The coating adhesion spatial distribution map and the coating residual life distribution map are superimposed and rendered on the high-fidelity digital twin in the form of heat maps of different colors as independent layers, and visualized display is performed;

[0273] S2053: The coating adhesion spatial distribution map, the coating residual life distribution map and the corresponding blade maintenance decision are written into a preset report template, a blade coating adhesion evaluation report is generated, and visualized display is performed;

[0274] In this embodiment, the blade coating adhesion evaluation report is a structured PDF or web page report, which can include:

[0275] Executive summary: blade overall health score, number and location of high-risk areas, predicted earliest failure time, blade maintenance decision;

[0276] Detailed analysis chart: adhesion distribution histogram, life prediction trend chart, cost-benefit comparison chart of different blade maintenance decisions;

[0277] Decision basis: explain why the intelligent decision model recommends this maintenance strategy, and list the risk analysis and economic calculation behind it;

[0278] Appendix: contains links to the original data of all detection points for experts to conduct in-depth review.

[0279] The embodiment of the application also provides a water turbine blade coating adhesion evaluation device, which refers to Figure 3 , shows a functional unit diagram of a water turbine blade coating adhesion evaluation device 300, which can include the following units:

[0280] The multi-modal data acquisition unit 301 is used for synchronously acquiring multi-modal non-destructive testing data of the coating area of the water turbine blade to be evaluated;

[0281] The coating adhesion prediction unit 302 is used for inputting the multi-modal non-destructive testing data into a coating adhesion prediction model based on a physical information multi-modal fusion algorithm to predict and integrate the adhesion, and obtain a coating adhesion spatial distribution map;

[0282] The coating residual life analysis unit 303 is used for inputting the coating adhesion spatial distribution map as an initial condition into a high-fidelity digital twin based on a digital twin technology to perform coupling simulation and life analysis, and obtain a coating residual life distribution map;

[0283] The blade maintenance decision generation unit 304 is used for inputting the coating adhesion spatial distribution map and the coating residual life distribution map into an intelligent maintenance decision model based on a reinforcement learning algorithm to generate a decision, and obtain a blade maintenance decision;

[0284] The visualization display unit 305 is configured to use the high-fidelity digital twin to visualize the coating adhesion force spatial distribution map and the coating residual life distribution map, and generate a blade coating adhesion force evaluation report in combination with the blade maintenance decision.

[0285] Based on the same inventive concept, another embodiment of the present application provides an electronic device comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus,

[0286] The memory is configured to store a computer program.

[0287] The processor is configured to execute the program stored in the memory to implement the water turbine blade coating adhesion force evaluation method.

[0288] The communication bus of the terminal mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EI) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus. The communication interface is used for communication between the terminal and other devices. The memory can include a Random Access Memory (RAM) and can also include a non-volatile memory, such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the aforementioned processor.

[0289] The processor mentioned above can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; can also be a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0290] In addition, to achieve the above object, the embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the water turbine blade coating bonding force evaluation method of the embodiment of the present application.

[0291] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, device, or computer program product. Therefore, the embodiments of the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present application can adopt a form of a computer program product implemented on one or more computer usable hardware (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer usable program codes.

[0292] The embodiments of the present application are described with reference to flowcharts and / or block diagrams according to the method, terminal device (apparatus), and computer program product of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be realized by computer program instructions. These computer program instructions can be provided to a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the computer or other programmable data processing terminal device produce a device for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more blocks.

[0293] These computer program instructions can also be stored in a computer readable memory capable of guiding a computer or other programmable data processing terminal device to work in a specific manner, so that the instructions stored in the computer readable memory produce a product including instruction apparatus, which realizes the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more blocks.

[0294] These computer program instructions can also be loaded into a computer or other programmable data processing terminal device, so that a series of operation steps are performed on the computer or other programmable terminal device to produce a computer implemented process, so that the instructions executed on the computer or other programmable terminal device provide a process for realizing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more blocks.

[0295] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements shall be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for evaluating the adhesion of a coating on a water turbine blade, characterized in that, The method includes: Multimodal nondestructive testing data were simultaneously acquired for the coating area of ​​the turbine blades to be evaluated; the multimodal nondestructive testing data included macroscopic images, microscopic images, acoustic emission signals, and thermal imaging spectra. Multimodal nondestructive testing data is input into a coating adhesion prediction model pre-trained based on a physical information multimodal fusion algorithm for adhesion prediction and integration, resulting in a spatial distribution map of coating adhesion. The spatial distribution map of coating adhesion is used as the initial condition and input into a high-fidelity digital twin pre-constructed based on digital twin technology for coupled simulation and lifetime analysis to obtain the distribution map of the remaining lifetime of the coating. The spatial distribution map of coating adhesion and the distribution map of coating remaining life are input into the intelligent maintenance decision model pre-built based on reinforcement learning algorithm to generate the blade maintenance decision. Using a high-fidelity digital twin, the spatial distribution map of coating adhesion and the distribution map of remaining coating lifetime are visualized, and combined with blade maintenance decisions, a blade coating adhesion assessment report is generated. The coating adhesion prediction model includes parallel image feature extraction branches and temporal feature extraction branches, as well as an attention fusion module, a physical parameter output layer and an adhesion prediction layer connected in sequence; The image feature extraction branch includes a macroscopic image feature extractor based on the ResNet-50 algorithm and a microscopic image feature extractor based on the EfficientNet-B0 algorithm, which are connected in parallel, as well as a global average pooling layer. The temporal feature extraction branch includes a parallel acoustic temporal feature extractor based on the MLP algorithm and a thermal imaging temporal feature extractor based on the 3D-CNN algorithm; The coating adhesion prediction model is configured with a total loss function that incorporates both physical information and data loss. This total loss function includes a total physical loss function and a data loss function. The total physical loss function comprises a heat conduction physical loss function, a fatigue damage physical loss function, and a physical consistency loss function, as shown in the following formula: In the formula, This is the total loss function; For data loss function; This is the total physical loss function; , Weighted by the total loss; In the formula, This is the physical loss function for heat conduction; The physical loss function is the fatigue damage function. The physical consistency loss function; , , This is the weight for the total physical loss; In the formula, For the MSE function; The predicted surface temperature response curve is obtained by calculating the interfacial thermal resistance output by the output layer based on the physical parameters. This is the measured surface temperature response curve; In the formula, This is a function for calculating the energy of acoustic emission signals. The energy of the acoustic emission signal; This is the mapping function from the stress factor to the damage space; The range of stress intensity factors; In the formula, For interfacial thermal resistance; Microcrack density; Based on the thermal resistance of the interface; The damage-thermal resistance coupling coefficient; The intelligent maintenance decision model is built based on the PPO algorithm, and the intelligent maintenance decision model is equipped with an intelligent agent and an experience playback pool. The method for constructing the coating adhesion prediction model includes: Several coated test blocks of turbine blades simulating different working conditions were prepared in the laboratory. For each coated test block, multimodal nondestructive testing data were collected. For each coated specimen, destructive testing and surface temperature testing were performed to obtain the corresponding true value label of adhesion force and the measured surface temperature response curve, which were then added to the corresponding sample multimodal nondestructive testing data to obtain the sample dataset. An initial coating adhesion prediction model was constructed based on a physical information multimodal fusion algorithm. Based on the total loss function, a quantum-enhanced intelligent optimizer is used to optimize the hyperparameters of the initial coating adhesion prediction model, resulting in the final coating adhesion prediction model.

2. The method for evaluating the adhesion of turbine blade coatings according to claim 1, characterized in that, Multimodal nondestructive testing data were simultaneously collected for the coated area of ​​the turbine blades to be evaluated, including: Based on laser scanning technology, a three-dimensional point cloud model of the turbine blade to be evaluated is obtained, and a global coordinate system of the turbine blade is established. On the 3D point cloud model, a scanning path covering the coating area of ​​the entire blade surface is planned, and according to the scanning path, several detection points are set on the blade surface of the turbine blade. The multimodal data acquisition probe is spatially registered in the global coordinate system to obtain the spatially registered multimodal data acquisition probe. Using a multimodal data acquisition probe, multimodal data is simultaneously acquired at each detection point to obtain multimodal nondestructive testing data of the coating area of ​​the turbine blade to be evaluated.

3. The method for evaluating the adhesion of turbine blade coatings according to claim 2, characterized in that, Based on the total loss function, a quantum-enhanced intelligent optimizer is used to optimize the hyperparameters of the initial coating adhesion prediction model, resulting in the final coating adhesion prediction model, including: The hyperparameter combination of the initial coating adhesion prediction model is associated with the QAOA parameters of the quantum-enhanced intelligent optimizer, and the total loss function is used as the cost function of the QAOA parameters. The quantum-enhanced intelligent optimizer is initialized using a Logistic chaotic map to generate initial QAOA parameters and then enters a hybrid iterative optimization loop. In each hybrid iterative optimization loop, on the quantum computer, the QAOA circuit is run with the current QAOA parameters to obtain the current quantum state and sample the current hyperparameter combination; On a classic computer, using a sample dataset, alternative coating adhesion prediction models are quickly trained based on the current hyperparameter combinations, and the current cost value of the alternative coating adhesion prediction models is calculated using a cost function. Generate a flight random number. If the flight random number is less than a preset threshold, perform Levy flight exploration, update the current QAOA parameters, and obtain the updated QAOA parameters. Otherwise, perform gradient descent to update the current QAOA parameters, obtain the updated QAOA parameters, and return the hyperparameter combination acquisition and cost value calculation steps. If the current number of hybrid iterations reaches the hybrid iteration threshold or the cost improvement is lower than the improvement threshold, terminate the hybrid iteration optimization step and output the optimal hyperparameter combination corresponding to the optimal cost value; Based on the optimal combination of hyperparameters, the hyperparameters of the initial coating adhesion prediction model are set to obtain the final coating adhesion prediction model.

4. The method for evaluating the adhesion of turbine blade coatings according to claim 3, characterized in that, Multimodal nondestructive testing data is input into a pre-trained coating adhesion prediction model based on a physical information-based multimodal fusion algorithm for adhesion prediction and integration, resulting in a spatial distribution map of coating adhesion, including: The multimodal nondestructive testing data is preprocessed to obtain preprocessed multimodal nondestructive testing data, which is then input into a coating adhesion prediction model pre-trained based on a physical information-based multimodal fusion algorithm. The macroscopic image feature extractor using the coating adhesion prediction model extracts the macroscopic feature map of the macroscopic image in the preprocessed multimodal nondestructive testing data, and inputs it into the global average pooling layer for global average pooling to obtain the macroscopic image features. The micro-image feature extractor using the coating adhesion prediction model extracts the micro-feature map of the micro-image in the preprocessed multimodal nondestructive testing data, and inputs it into the global average pooling layer for global average pooling to obtain the micro-image features. An acoustic temporal feature extractor based on a coating adhesion prediction model is used to extract the acoustic temporal features of acoustic emission signals from preprocessed multimodal nondestructive testing data. The thermal imaging time-series feature extractor of the coating adhesion prediction model is used to extract the thermal imaging time-series features of the thermal imaging spectrum in the preprocessed multimodal nondestructive testing data. Macroscopic image features, microscopic image features, acoustic temporal features, and thermal imaging temporal features are concatenated to obtain a fused feature vector, which is then input into the attention fusion module of the coating adhesion prediction model for attention fusion to obtain a weighted fused feature. Based on the weighted fusion characteristics, physical parameters are generated using the physical parameter output layer of the coating adhesion prediction model; the physical parameters include interfacial thermal resistance, microcrack density, and interfacial bonding strength. Based on the physical parameters, the bonding force prediction layer of the coating bonding force prediction model is used to call the Softmax function to generate the bonding force level, and combined with the corresponding physical parameters, the bonding force prediction result is obtained. Based on the spatial coordinates of each detection point in the global coordinate system, the predicted adhesion force of all detection points is integrated to obtain the corresponding spatial distribution map of coating adhesion force.

5. The method for evaluating the adhesion of turbine blade coatings according to claim 4, characterized in that, Using the spatial distribution map of coating adhesion as initial conditions, it is input into a high-fidelity digital twin pre-constructed based on digital twin technology for coupled simulation and lifetime analysis, resulting in a coating remaining lifetime distribution map, including: The three-dimensional point cloud model of the turbine blade to be evaluated is imported into the simulation software, and a high-quality hexahedral dominant mesh is generated to obtain the corresponding three-dimensional simulation model. Based on digital twin technology, the elastoplastic material models of the matrix and coating are defined in the three-dimensional simulation model, and the damage evolution equation corresponding to the elastoplastic material model of the coating is set to obtain a pre-constructed high-fidelity digital twin. The spatial distribution map of coating adhesion is used as the initial condition for the damage evolution equation, mapped to a high-fidelity digital twin, and the high-fidelity digital twin is then used for model calibration. Coupled simulation and damage accumulation of a high-fidelity digital twin are performed under simulated specific working conditions to obtain the damage value of each detection point; If the damage value exceeds the damage threshold for the first time, the detection point is taken as the critical detection point, and the coupled simulation time experienced by the critical detection point is taken as the corresponding predicted remaining lifetime. If the damage value does not exceed the damage threshold, then continue the coupled simulation and damage accumulation based on that detection point; Based on the spatial coordinates of each key detection point in the global coordinate system, the predicted remaining lifetime of all key detection points is integrated to obtain the coating remaining lifetime distribution map.

6. The method for evaluating the adhesion of turbine blade coatings according to claim 5, characterized in that, The spatial distribution map of coating adhesion and the distribution map of remaining coating lifetime are input into a pre-built intelligent maintenance decision model based on reinforcement learning algorithm to generate blade maintenance decisions, including: Global statistical features, local key area features, and spatial context features of the spatial distribution map of coating adhesion and the distribution map of coating remaining life are extracted and combined with hydropower station status data to obtain complete status observation. Write complete state observations into the state space of the agent in a pre-constructed intelligent maintenance decision model based on reinforcement learning algorithms; Based on the state space, an intelligent agent selects the optimal action in the preset action space to obtain the blade maintenance decision and collects real-time experience in the decision generation process. Several historical experiences are randomly selected from the experience replay pool and combined with real-time experiences to continuously train the intelligent maintenance decision model, resulting in an updated intelligent maintenance decision model.

7. The method for evaluating the adhesion of turbine blade coatings according to claim 6, characterized in that, Using a high-fidelity digital twin, the spatial distribution map of coating adhesion and the distribution map of remaining coating lifetime are visualized. Combined with blade maintenance decisions, a blade coating adhesion assessment report is generated, including: Loading a high-fidelity digital twin of the turbine blades to be evaluated; The spatial distribution map of coating adhesion and the distribution map of coating remaining life are rendered as independent layers on a high-fidelity digital twin in the form of heat maps of different colors, and then visualized. The spatial distribution map of coating adhesion, the distribution map of remaining coating life, and the corresponding blade maintenance decisions are written into a preset report template to generate a blade coating adhesion assessment report, which is then visualized.

8. A device for evaluating the adhesion of a turbine blade coating, used to implement the method for evaluating the adhesion of a turbine blade coating as described in any one of claims 1-7, characterized in that, The device includes: The multimodal data acquisition unit is used to simultaneously acquire multimodal nondestructive testing data for the coating area of ​​the turbine blades to be evaluated. The coating adhesion prediction unit is used to input multimodal nondestructive testing data into a pre-trained coating adhesion prediction model based on a physical information multimodal fusion algorithm to predict and integrate the adhesion force, thereby obtaining a spatial distribution map of the coating adhesion force. The coating remaining lifetime analysis unit is used to input the spatial distribution map of coating adhesion force as the initial condition into a high-fidelity digital twin pre-constructed based on digital twin technology for coupled simulation and lifetime analysis, so as to obtain the coating remaining lifetime distribution map. The blade maintenance decision generation unit is used to input the spatial distribution map of coating adhesion force and the distribution map of coating remaining life into the intelligent maintenance decision model pre-built based on reinforcement learning algorithm to generate a blade maintenance decision. The visualization unit is used to visualize the spatial distribution map of coating adhesion and the distribution map of remaining coating life using a high-fidelity digital twin, and generates a blade coating adhesion assessment report in conjunction with blade maintenance decisions.

Images