Water-turbine generator set crack prediction method based on dynamic programming and application

Abstract

The invention discloses a hydroelectric generating set crack prediction method based on dynamic programming and application. The hydroelectric generating set crack prediction method comprises the steps of collecting stress data and temperature data of a hydroelectric generating set and sequentially preprocessing the data; a three-dimensional model of the water-turbine generator set is established, the processed data set is input into the three-dimensional model for mechanical simulation, and structural stress field distribution of the water-turbine generator set is obtained through calculation; calculating to obtain a theoretical minimum damage value of each crack propagation path individual through a genetic algorithm; calculating to obtain a fitness value of each crack propagation path individual; performing path probability optimization and verification or evaluation, and outputting specific data of the predicted crack propagation path of the water-turbine generator set; the defect and the problem that the accuracy and the efficiency of crack prediction cannot be considered due to the fact that a prediction model needs a large amount of resources and time for calculation in the prior art are solved, a large amount of resources and time are not needed for calculation, and the accuracy and the efficiency of crack prediction can be considered.

Description

Technical Field

[0001] The present invention belongs to the field of hydro-generators, and in particular relates to a hydro-generator set crack prediction method based on dynamic programming and its application. Background Art

[0002] As hydro-turbine generator sets are iteratively upgraded toward larger capacities and sizes, their relative mechanical strength has significantly decreased. This makes key components like the runner and main shaft susceptible to cracks and crack propagation risks under high torque, high-speed rotation, and complex vibration loads. If not discovered promptly, cracks can rapidly propagate, potentially leading to cascading failures such as rotor imbalance and shaft fracture, resulting in unplanned downtime or catastrophic accidents. To address this, crack prediction technology is often used to derive a predicted crack propagation path for a hydro-turbine generator set. Based on this predicted crack propagation path and path-related data (such as crack depth, crack direction, and crack propagation rate), maintenance and repairs can be performed on the hydro-turbine generator set in advance. Currently, commonly used prediction methods include finite element analysis, data-driven prediction methods, and physical model-based prediction methods. However, these methods all have certain limitations in practical applications. Although finite element analysis can provide relatively accurate prediction results based on high-precision meshing and material constitutive models, it requires a large amount of computing resources and time when processing complex geometric structures, resulting in poor prediction efficiency and real-time performance. Data-driven prediction methods mainly use historical data and real-time monitoring data to establish a crack propagation prediction model through statistical analysis, machine learning, and other means. However, this method has poor model generalization and adaptability, resulting in poor prediction accuracy. In addition, the processing of new data and model updates require time, which affects its real-time performance. Physical model-based prediction methods predict crack propagation paths and rates by establishing physical equations for crack propagation and combining theories such as material mechanics and fracture mechanics. However, the prediction results of this method model deviate from the actual situation, making it difficult to accurately and real-timely predict cracks. In summary, it is difficult for currently commonly used prediction methods to strike a balance between accuracy, real-time performance, and efficiency.

[0003] A Chinese patent application with application number 202410213653.1 and filing date February 27, 2024, discloses a method for establishing a crack prediction model, a crack prediction method, and a computing device. The method includes: obtaining dimensional parameters of multiple cracks on an axle at the initial and different expansion stages, the multiple cracks being located at target locations on the axle, the dimensional parameters including at least two of a depth value, a length value, and an aspect ratio; determining multiple fitting curves corresponding to the multiple cracks based on the dimensional parameters, each fitting curve representing the relationship between the depth value of the corresponding crack and the length value; dividing the fitting curves into different expansion stages at a critical point, wherein the changing patterns of the fitting curves of adjacent expansion stages at the critical point are different, while the changing patterns of the fitting curves of multiple cracks in the same expansion stage are consistent; and establishing a crack prediction model capable of predicting the depth value of the crack based on the crack length value based on the changing patterns of the multiple fitting curves at different expansion stages. Although the patent combines finite element analysis and fitting curves to establish a crack prediction model, which can achieve real-time prediction of axle cracks, it still has the following defects: The prediction model of this design requires a lot of resources and time for calculation, which makes it impossible to take into account both the accuracy and efficiency of crack prediction; therefore, it is necessary to design a crack prediction method for hydro-generator sets based on dynamic programming to solve the above problems. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for predicting cracks in a hydro-turbine generator set based on dynamic programming, aiming to solve the defects and problems of the existing technology that the prediction model requires a large amount of resources and time for calculation, resulting in the inability to take into account both the accuracy and efficiency of crack prediction. A method for predicting cracks in a hydro-turbine generator set based on dynamic programming and its application are provided, which does not require a large amount of resources and time for calculation and can take into account both the accuracy and efficiency of crack prediction.

[0005] In order to achieve the above technical effects, the technical solution adopted by the present invention is: A method for predicting cracks in a hydro-generator set based on dynamic programming, comprising: S1, real-time monitoring and collection of stress data and temperature data of a hydro-generator set, sequentially performing denoising, normalization, and standardization on the stress data to obtain a stress characteristic data set, and sequentially performing denoising, normalization, and standardization on the temperature data to obtain a temperature characteristic data set; S2, establishing a three-dimensional model of the hydro-generator set, inputting the stress characteristic data set into the three-dimensional model for mechanical simulation, and calculating the structural stress field distribution of the hydro-generator set; The above temperature characteristic data set is input into the three-dimensional model for thermodynamic simulation to calculate the thermal field distribution of the hydro-generator set; Based on the structural stress field distribution, a number of potential crack propagation paths are determined through fracture mechanics analysis, the potential crack propagation paths are then divided into a number of nodes, and then stress state data of each node is extracted from the structural stress field distribution, and temperature state data of each node is extracted from the thermal field distribution; Input the stress state data and temperature state data of each node into the state transition equation of the dynamic programming algorithm, calculate the minimum damage value of each node, and obtain and store the minimum damage value data set composed of the minimum damage values ​​of each node; S3, initializing a population using a genetic algorithm, wherein the population consists of a number of randomly generated crack propagation path individuals, each of which consists of a number of nodes; extracting the minimum damage value of the corresponding node in each crack propagation path individual from the above minimum damage value dataset, and calculating the theoretical minimum damage value of each crack propagation path individual; The actual damage value of each crack propagation path individual is calculated based on the stress characteristic data set and the temperature characteristic data set, and the fitness value of each crack propagation path individual is obtained by comparing the theoretical minimum damage value with the actual damage value through the fitness function; According to the fitness values ​​of the crack propagation path individuals, the population is subjected to selection, crossover, mutation and iterative updates until convergence to generate a global optimal crack propagation path candidate set; S4, optimizing the path probability of each candidate path in the above-mentioned global optimal crack propagation path candidate set by using an ant colony algorithm, and selecting the candidate path with a high path probability as the predicted crack propagation path of the hydro-generator set; S5, verifying or evaluating the predicted crack propagation path of the hydro-generator set, and outputting specific data of the predicted crack propagation path of the hydro-generator set.

[0006] Preferably, in step S1, real-time monitoring and collection of stress data and temperature data of the hydro-generator set includes: using optical fiber sensing technology to real-time monitor and collect stress data and temperature data of the turbine main shaft, runner, and frame in the hydro-generator set.

[0007] Preferably, in step S2, the state transition equation is: ; Where, represents the minimum cumulative damage value from node i to node j; represents the minimum damage value of node i; represents the crack extension cost from node i to node j; described Calculated by the following formula: ; Among them, w1, w2, w3, w4, and w5 are the weight factors of the node damage value, which respectively represent the weight ratio of stress intensity factor, crack depth, crack length, propagation distance, and temperature influence in the calculation of node damage value; the calculation formula of stress intensity factor (i) is: ; Where K(i) represents the stress intensity factor, σ represents the stress acting on the crack tip, which comes from the stress characteristic data set, and r represents the minimum distance between the crack tip and node i, which is obtained through the mechanical simulation calculation in the second step. The crack depth (i), crack length (i), and propagation distance (i) are also obtained through the mechanical simulation calculation in the second step. Indicates the effect of temperature on cracks; described Calculated by the following formula: ; in, represents the change in crack depth from node i to node j; represents the change in crack length from node i to node j; represents the change of stress intensity factor from node i to node j; 、 、 is a weight factor used to regulate the impact of different factors on the crack growth cost and is adjusted according to the needs of actual problems.

[0008] Preferably, in step S3, the fitness function is: ; in, represents the fitness value of the crack growth path; represents the actual damage value of node i, which is calculated from the stress characteristic data set; Indicates the minimum damage value of node i calculated by the dynamic programming algorithm.

[0009] Preferably, in step S3, a population is initialized by a genetic algorithm, and the population is composed of several randomly generated crack propagation path individuals, and each crack propagation path individual is composed of several nodes, including: randomly generating an initial population composed of crack propagation path individuals by a genetic algorithm, and each crack propagation path individual is composed of several nodes, and the nodes correspond to positions in the distribution of structural stress field and thermal field; in the minimum damage value data set, the minimum damage value of the corresponding node in each crack propagation path individual is extracted, and the theoretical minimum damage value of each crack propagation path individual is calculated, including: when calculating the theoretical minimum damage value of a crack propagation path individual, the minimum damage value of the node corresponding to the crack propagation path individual is extracted from the above-mentioned minimum damage value data set, and the theoretical minimum damage value of the crack propagation path individual is obtained by recursion, thereby calculating the theoretical minimum damage value of each crack propagation path individual.

[0010] Preferably, in step S3, the fitness value of each crack propagation path individual is obtained by comparing the theoretical minimum damage value with the actual damage value through the fitness function, which includes: inputting the theoretical minimum damage value and the actual damage value into the fitness function to calculate the fitness value of each crack propagation path individual, wherein the crack propagation path individual with a lower fitness value is better.

[0011] Preferably, in step S3, the selection operation includes: A roulette wheel selection method or a tournament selection method is used to select the crack propagation path individuals, and a number of crack propagation path individuals with low fitness values ​​are selected from the population to form a parent population; The crossover operation includes: randomly selecting two crack propagation path individuals from the parent population, and selecting a portion of nodes from the two crack propagation path individuals for crossover to generate a child path; The mutation operation includes: randomly adjusting one or several nodes of the offspring path to obtain a mutated offspring path, and adding the mutated offspring path to form a new population, or mixing the mutated offspring path with the parent population to form a new population.

[0012] Preferably, in step S3, iterative updating until convergence to generate a global optimal crack propagation path candidate set includes: calculating the fitness values ​​of all crack propagation path individuals in the new population, and repeating the selection operation, crossover operation, and mutation operation on the new population according to the fitness values ​​until the maximum number of iterations is reached or the fitness value of the new population no longer changes in several consecutive generations, thereby generating a global optimal crack propagation path candidate set.

[0013] Preferably, in step S4, performing path probability optimization on each candidate path in the global optimal crack propagation path candidate set by using an ant colony algorithm, and selecting a candidate path with a high path probability as the predicted crack propagation path of the hydro-generator set includes: First, the above-mentioned global optimal crack propagation path candidate set is used as the initial path set, and the pheromone concentration of each candidate path in the above-mentioned global optimal crack propagation path candidate set is initialized so that the pheromone concentration of each candidate path is the same; then, the path probability of each candidate path is calculated using the path selection probability formula; then, based on the calculated path probability, path selection and transfer are performed in the global optimal crack propagation path candidate set, and candidate paths with high path probabilities are preferentially selected for transfer. During the transfer process, the pheromone concentration of the candidate path is updated to enhance the attractiveness of high-quality paths; finally, path selection and pheromone update are iteratively performed until the convergence condition is met, and the candidate path with high path probability is output as the predicted crack propagation path of the hydro-generator set; The path selection probability formula is: ; in, represents the pheromone concentration on the path (i, j), which is updated by the ant colony algorithm; represents the crack extension cost from node i to node j, which is calculated from experimental data or physical model; It represents the minimum cumulative damage value from node i to node j calculated by the dynamic programming algorithm, reflecting the optimal damage cost from node i to node j; represents the minimum damage value from i to k; α is the weight factor, which indicates the influence of pheromone concentration; β is the weight factor, which indicates the influence of crack growth cost; γ is the weight factor, which indicates the influence of damage value difference; N represents the set of all adjacent path points reachable from the current node i; The pheromone concentration of the candidate path is updated according to the following formula: ; in, represents the pheromone concentration on the path (i, j) at time or iteration number t+1; Indicates the current moment or iteration number, used to mark the time step of the algorithm; represents the pheromone concentration on the path (i, j) at time or iteration number t; Indicates the pheromone volatility coefficient, which controls the volatility of pheromones; It represents the amount of pheromone added by the k-th contributing source on the path (i, j) at time or iteration number t; Indicates the total number of contributing sources; It represents the sum of pheromones released by all contributing sources on path (i, j).

[0014] Preferably, the method further includes application of the dynamic programming-based crack prediction method for a hydro-generator set to crack prediction of a turbine main shaft, a runner, and a frame in a hydro-generator set.

[0015] The beneficial effects of the present invention are as follows: 1. The present invention provides a method for predicting cracks in a hydro-generator set based on dynamic programming and its application. The method divides the potential crack propagation path into a plurality of nodes, calculates the minimum damage value of each node by using the state transition equation of the dynamic programming algorithm, obtains a minimum damage value data set, then performs the operation of the genetic algorithm, uses the minimum damage value data set as the fitness function input of the genetic algorithm, obtains a globally optimal crack propagation path candidate set, and then performs the operation of the ant colony algorithm on the above-mentioned globally optimal crack propagation path candidate set to obtain the predicted crack propagation path of the hydro-generator set. When applied, the method first divides the potential crack propagation path into a plurality of nodes according to the stress distribution of the hydro-generator set, calculates the minimum damage value of each node by using the state transition equation in the dynamic programming algorithm, and can quickly calculate the minimum damage value of each node through the optimization of dynamic programming, thereby significantly reducing the amount of calculation. On this basis, the genetic algorithm can quickly find the globally optimal crack propagation path candidate set by initializing the population, selecting, crossover and mutation operations, and avoid falling into the local optimal solution. Improve the diversity and global optimality of the solution. Then, the ant colony algorithm further optimizes the candidate set obtained by the genetic algorithm. Through pheromone updating and path selection probability formula, it can quickly screen out the optimal crack propagation path, improving the accuracy of the prediction. This method that combines dynamic programming, genetic algorithm and ant colony algorithm fully utilizes the advantages of each algorithm. Dynamic programming quickly calculates the minimum damage value through the state transition equation, providing efficient basic data and clear optimization direction for the genetic algorithm and ant colony algorithm. The genetic algorithm explores potential optimal paths in a global scope based on the fitness function, avoiding local optimal solutions. The ant colony algorithm uses the pheromone mechanism to conduct a refined evaluation of the global optimal crack propagation path candidate set, improving the accuracy of local optimization. The three form a hierarchical optimization strategy from coarse screening to wide search and then to fine refinement, effectively balancing computational efficiency and result accuracy, and improving the efficiency and accuracy of the entire prediction process. Compared with the existing technology, the present invention avoids complex calculations and model limitations through the combination of multiple algorithms, achieving a balance between high efficiency and accuracy. Therefore, the present invention can not only predict cracks in hydro-turbine generator sets, but also achieve accurate and efficient prediction results.

[0016] 2. In a method for predicting cracks in a hydro-generator set based on dynamic programming and its application, the present invention calculates the minimum damage value of each node, obtains and stores a minimum damage value data set consisting of the minimum damage values ​​of each node. When the state transition equation in the dynamic programming algorithm is calculated to a certain node, the minimum damage value of the node is calculated and stored. When the path node is encountered again in subsequent genetic algorithm and ant colony algorithm calculations, the result is directly read from the stored data, avoiding repeated calculations and improving computational efficiency. Therefore, the present invention not only predicts results efficiently and accurately, but also improves computational efficiency.

[0017] 3. The present invention discloses a method for predicting cracks in a hydro-turbine generator set based on dynamic programming and its application. Fiber optic sensing technology is used to monitor and collect stress and temperature data from the turbine main shaft, runner, and frame of the hydro-turbine generator set in real time. During application, fiber optic sensing technology is first used to monitor and collect stress and temperature data from key components of the hydro-turbine generator set, such as the main shaft, runner, and frame. Due to its high sensitivity, high precision, high anti-interference performance, high real-time performance, high transmission efficiency, and high layout flexibility, fiber optic sensing technology can provide highly accurate and real-time hydro-turbine generator set status data. This status data is then pre-processed and transmitted to a dynamic programming algorithm. Combined with a genetic ant colony algorithm for calculation, this method automatically analyzes the real-time data, immediately detecting the onset and development of cracks and providing timely warning information on crack expansion, thus avoiding the lag and errors associated with manual intervention. Compared to existing technologies, the present invention can provide warning information at the early stages of a crack, avoiding the lag and errors associated with traditional methods and significantly improving the real-time and accuracy of crack prediction. Therefore, the present invention not only improves computational efficiency but also exhibits excellent real-time performance.

[0018] 4. The present invention provides a dynamic programming-based crack prediction method for a hydro-turbine generator set and its application. The predicted crack propagation path of the hydro-turbine generator set is verified or evaluated, and specific data of the predicted crack propagation path of the hydro-turbine generator set is output. During application, the error between the calculated predicted crack propagation path of the hydro-turbine generator set and the actual crack propagation path is verified using actual experimental data. Alternatively, a numerical simulation is performed to simulate crack propagation along the calculated predicted crack propagation path of the hydro-turbine generator set, and the error between the simulation result and the predicted crack propagation path of the hydro-turbine generator set is evaluated. When the error meets a preset value (e.g., less than or equal to 5%), it is confirmed that the predicted crack propagation path of the hydro-turbine generator set accurately reflects the actual crack behavior. By verifying or evaluating the predicted crack propagation path, the accuracy and reliability of the prediction results can be ensured, providing a basis for subsequent predictions. Therefore, the present invention not only has good real-time performance but also improves the reliability of the prediction results.

[0019] 5. The present invention provides a method for predicting cracks in a hydro-turbine generator set based on dynamic programming and its application, including the application of the method for predicting cracks in a turbine main shaft, runner, and frame in a hydro-turbine generator set. During application, optical fiber sensing technology is used to monitor and collect status data of the turbine main shaft, runner, and frame in the hydro-turbine generator set in real time, and combined with a dynamic programming algorithm, a genetic algorithm, and an ant colony algorithm, the crack propagation paths of key parts such as the turbine main shaft, runner, and frame in the hydro-turbine generator set are predicted in real time, accurately, and efficiently. This method has a wide range of applications in crack prediction and is also applicable to other parts of a hydro-turbine generator set, with good versatility. Compared with traditional crack prediction models in the prior art, which are often targeted at specific equipment or structures and difficult to adapt to different working conditions and complex crack propagation paths, the present invention achieves efficient prediction of crack propagation paths in key components of a hydro-turbine generator set by integrating optical fiber sensing technology with a dynamic programming algorithm, a genetic algorithm, and an ant colony algorithm for multi-algorithm collaborative optimization, and can be flexibly applied to crack prediction in a variety of complex structures and equipment. Therefore, the present invention not only improves the reliability of the prediction results, but also has wide applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0021] Example 1: A method for predicting cracks in a hydro-generator set based on dynamic programming, comprising: S1, real-time monitoring and collection of stress data and temperature data of a hydro-generator set, sequentially performing denoising, normalization, and standardization on the stress data to obtain a stress characteristic data set, and sequentially performing denoising, normalization, and standardization on the temperature data to obtain a temperature characteristic data set; S2, establishing a three-dimensional model of the hydro-generator set, inputting the stress characteristic data set into the three-dimensional model for mechanical simulation, and calculating the structural stress field distribution of the hydro-generator set; The above temperature characteristic data set is input into the three-dimensional model for thermodynamic simulation to calculate the thermal field distribution of the hydro-generator set; Based on the structural stress field distribution, a number of potential crack propagation paths are determined through fracture mechanics analysis, the potential crack propagation paths are then divided into a number of nodes, and then stress state data of each node is extracted from the structural stress field distribution, and temperature state data of each node is extracted from the thermal field distribution; Input the stress state data and temperature state data of each node into the state transition equation of the dynamic programming algorithm, calculate the minimum damage value of each node, and obtain and store the minimum damage value data set composed of the minimum damage values ​​of each node; S3, initializing a population using a genetic algorithm, wherein the population consists of a number of randomly generated crack propagation path individuals, each of which consists of a number of nodes; extracting the minimum damage value of the corresponding node in each crack propagation path individual from the above minimum damage value dataset, and calculating the theoretical minimum damage value of each crack propagation path individual; The actual damage value of each crack propagation path individual is calculated based on the stress characteristic data set and the temperature characteristic data set, and the fitness value of each crack propagation path individual is obtained by comparing the theoretical minimum damage value with the actual damage value through the fitness function; According to the fitness values ​​of the crack propagation path individuals, the population is subjected to selection, crossover, mutation and iterative updates until convergence to generate a global optimal crack propagation path candidate set; S4, optimizing the path probability of each candidate path in the above-mentioned global optimal crack propagation path candidate set by using an ant colony algorithm, and selecting the candidate path with a high path probability as the predicted crack propagation path of the hydro-generator set; S5, verifying or evaluating the predicted crack propagation path of the hydro-generator set, and outputting specific data of the predicted crack propagation path of the hydro-generator set.

[0022] Preferably, in step S1, real-time monitoring and collection of stress data and temperature data of the hydro-generator set includes: using optical fiber sensing technology to real-time monitor and collect stress data and temperature data of the turbine main shaft, runner, and frame in the hydro-generator set.

[0023] Preferably, in step S2, the state transition equation is: ; Where, represents the minimum cumulative damage value from node i to node j; represents the minimum damage value of node i; represents the crack extension cost from node i to node j; described Calculated by the following formula: ; Among them, w1, w2, w3, w4, and w5 are the weight factors of the node damage value, which respectively represent the weight ratio of stress intensity factor, crack depth, crack length, propagation distance, and temperature influence in the calculation of node damage value; the calculation formula of stress intensity factor (i) is: ; Where K(i) represents the stress intensity factor, σ represents the stress acting on the crack tip, which comes from the stress characteristic data set, and r represents the minimum distance between the crack tip and node i, which is obtained through the mechanical simulation calculation in the second step. The crack depth (i), crack length (i), and propagation distance (i) are also obtained through the mechanical simulation calculation in the second step. Indicates the effect of temperature on cracks; described Calculated by the following formula: ; in, represents the change in crack depth from node i to node j; represents the change in crack length from node i to node j; represents the change of stress intensity factor from node i to node j; 、 、 is a weight factor used to regulate the impact of different factors on the crack growth cost and is adjusted according to the needs of actual problems.

[0024] Preferably, in step S3, the fitness function is: ; in, represents the fitness value of the crack growth path; represents the actual damage value of node i, which is calculated from the stress characteristic data set; Indicates the minimum damage value of node i calculated by the dynamic programming algorithm.

[0025] Preferably, in step S3, a population is initialized by a genetic algorithm, and the population is composed of several randomly generated crack propagation path individuals, and each crack propagation path individual is composed of several nodes, including: randomly generating an initial population composed of crack propagation path individuals by a genetic algorithm, and each crack propagation path individual is composed of several nodes, and the nodes correspond to positions in the distribution of structural stress field and thermal field; in the minimum damage value data set, the minimum damage value of the corresponding node in each crack propagation path individual is extracted, and the theoretical minimum damage value of each crack propagation path individual is calculated, including: when calculating the theoretical minimum damage value of a crack propagation path individual, the minimum damage value of the node corresponding to the crack propagation path individual is extracted from the above-mentioned minimum damage value data set, and the theoretical minimum damage value of the crack propagation path individual is obtained by recursion, thereby calculating the theoretical minimum damage value of each crack propagation path individual.

[0026] Preferably, in step S3, the fitness value of each crack propagation path individual is obtained by comparing the theoretical minimum damage value with the actual damage value through the fitness function, which includes: inputting the theoretical minimum damage value and the actual damage value into the fitness function to calculate the fitness value of each crack propagation path individual, wherein the crack propagation path individual with a lower fitness value is better.

[0027] Preferably, in step S3, the selection operation includes: A roulette wheel selection method or a tournament selection method is used to select the crack propagation path individuals, and a number of crack propagation path individuals with low fitness values ​​are selected from the population to form a parent population; The crossover operation includes: randomly selecting two crack propagation path individuals from the parent population, and selecting a portion of nodes from the two crack propagation path individuals for crossover to generate a child path; The mutation operation includes: randomly adjusting one or several nodes of the offspring path to obtain a mutated offspring path, and adding the mutated offspring path to form a new population, or mixing the mutated offspring path with the parent population to form a new population.

[0028] Preferably, in step S3, iterative updating until convergence to generate a global optimal crack propagation path candidate set includes: calculating the fitness values ​​of all crack propagation path individuals in the new population, and repeating the selection operation, crossover operation, and mutation operation on the new population according to the fitness values ​​until the maximum number of iterations is reached or the fitness value of the new population no longer changes in several consecutive generations, thereby generating a global optimal crack propagation path candidate set.

[0029] Preferably, in step S4, performing path probability optimization on each candidate path in the global optimal crack propagation path candidate set by using an ant colony algorithm, and selecting a candidate path with a high path probability as the predicted crack propagation path of the hydro-generator set includes: First, the above-mentioned global optimal crack propagation path candidate set is used as the initial path set, and the pheromone concentration of each candidate path in the above-mentioned global optimal crack propagation path candidate set is initialized so that the pheromone concentration of each candidate path is the same; then, the path probability of each candidate path is calculated using the path selection probability formula; then, based on the calculated path probability, path selection and transfer are performed in the global optimal crack propagation path candidate set, and candidate paths with high path probabilities are preferentially selected for transfer. During the transfer process, the pheromone concentration of the candidate path is updated to enhance the attractiveness of high-quality paths; finally, path selection and pheromone update are iteratively performed until the convergence condition is met, and the candidate path with high path probability is output as the predicted crack propagation path of the hydro-generator set; The path selection probability formula is: ; in, represents the pheromone concentration on the path (i, j), which is updated by the ant colony algorithm; represents the crack extension cost from node i to node j, which is calculated from experimental data or physical model; It represents the minimum cumulative damage value from node i to node j calculated by the dynamic programming algorithm, reflecting the optimal damage cost from node i to node j; represents the minimum damage value from i to k; α is the weight factor, which indicates the influence of pheromone concentration; β is the weight factor, which indicates the influence of crack growth cost; γ is the weight factor, which indicates the influence of damage value difference; N represents the set of all adjacent path points reachable from the current node i; The pheromone concentration of the candidate path is updated according to the following formula: ; in, represents the pheromone concentration on the path (i, j) at time or iteration number t+1; Indicates the current moment or iteration number, used to mark the time step of the algorithm; represents the pheromone concentration on the path (i, j) at time or iteration number t; Indicates the pheromone volatility coefficient, which controls the volatility of pheromones; It represents the amount of pheromone added by the k-th contributing source on the path (i, j) at time or iteration number t; Indicates the total number of contributing sources; It represents the sum of pheromones released by all contributing sources on path (i, j).

[0030] Preferably, the method further includes application of the dynamic programming-based crack prediction method for a hydro-generator set to crack prediction of a turbine main shaft, a runner, and a frame in a hydro-generator set.

[0031] Example 2: In the preferred first step of the present invention, the denoising process refers to removing out-of-frequency noise from the collected state data using a filtering method. Commonly used filtering methods include low-pass filtering and high-pass filtering. The specific formula is as follows: ; Among them, x(t) represents state data (such as stress, strain, and temperature data), h(i) represents the impulse response of the filter, and y(t) represents the filtered data.

[0032] In the preferred first step of the present invention, the normalization process refers to scaling the filtered data to the range of [0, 1] to eliminate the dimensional differences between the data. The specific formula is as follows: ; Among them, x(t) represents state data (such as stress, strain, temperature data), min(x) and max(x) represent the minimum and maximum values ​​of the state data respectively. is the normalized data.

[0033] In the preferred first step of the present invention, the normalization process refers to converting the data into a form with zero mean and unit variance, which is suitable for situations where the influence of data scale needs to be eliminated. The specific formula is as follows: ; in, Represents the mean of state data (such as stress, strain, and temperature data), represents the standard deviation of the state data, Represents the normalized data.

[0034] Preferably, in the second step of the present invention, the step of establishing a three-dimensional model of the hydro-generator set, inputting the above-mentioned stress characteristic data set into the three-dimensional model for mechanical simulation, and calculating the structural stress field distribution of the hydro-generator set comprises: creating an accurate three-dimensional geometric model based on the actual size and shape of the hydro-generator set, assigning material properties (such as elastic modulus, Poisson's ratio, yield strength, etc.) to each component; simulating the loads and constraints (such as water pressure, torque, rotational speed, etc.) that may be encountered in actual operation, and combining the actually measured stress data with finite element analysis software to calculate and obtain the structural stress field distribution of the hydro-generator set.

[0035] In the preferred second step of the present invention, the determining of several potential crack propagation paths by fracture mechanics analysis based on the structural stress field distribution refers to: finding stress concentration areas in the structural stress field distribution of the hydro-turbine unit, where the stress concentration areas are usually places where cracks are prone to initiation and propagation; and determining possible crack propagation directions and paths by fracture mechanics analysis based on the structural characteristics and stress field distribution of the hydro-turbine generator unit (these paths can be determined based on the mechanical properties of the material, the geometric shape of the structure, and historical crack propagation data), thereby obtaining several potential crack propagation paths.

[0036] In the preferred second step of the present invention, the dividing of the potential crack propagation path into a plurality of nodes means: first, determining the basis for path division based on the geometric characteristics of the potential crack propagation path (such as straight segments, curved segments), stress distribution (locations where stress is concentrated or changes significantly), and different stages of crack propagation (such as rapid propagation stage and stable propagation stage); then, sampling the crack propagation path data obtained through simulation, and representing the path as a series of point sets, each point containing position information and other relevant attributes; then, using equal distance division, equal step length division, or feature point extraction methods to divide the path into multiple nodes; finally, adjusting and optimizing the preliminarily divided nodes according to actual needs and computing resource limitations, and verifying the rationality of the node division through visualization tools or mathematical methods to ensure that the divided nodes can accurately reflect the characteristics of the crack propagation path, thereby providing a basis for subsequent dynamic programming algorithms and other analyses.

[0037] The present invention preferably adopts the second step in which the minimum damage value data set includes the minimum damage value of the potential crack propagation path from the starting point to each node, and these values ​​are obtained by recursive calculation through the state transfer equation; the optimal path damage value refers to the damage value of each node on an optimal crack propagation path calculated by the dynamic programming algorithm; the dynamic programming calculates the minimum damage value of each path node by recursion, and finally selects a crack propagation path with the least damage; these minimum damage value data sets are used as the fitness function input of the genetic algorithm to evaluate the pros and cons of each potential crack propagation path; the genetic algorithm further optimizes the path selection through selection, crossover and mutation operations, combined with the pheromone update and path selection probability formula of the ant colony algorithm; finally, the path with the highest pheromone concentration and the lowest fitness value is screened out from the global optimal crack propagation path candidate set as the predicted crack propagation path of the hydro-turbine generator set, and the minimum damage value at this time is the optimal path damage value; therefore, the minimum damage value data set is the basis for determining the optimal path damage value, and the optimal path damage value is obtained through further optimization algorithm based on these minimum damage value data sets.

[0038] In the preferred embodiment of the present invention, in the second step, the method for storing the minimum damage value data set composed of the minimum damage value of each node includes a memo array (or matrix), a hash table / dictionary (variable storage form), which is a key strategy for optimizing computational efficiency of the dynamic programming algorithm; wherein, the memo array can be a one-dimensional array or a two-dimensional matrix, and the specific implementation method depends on the scale and complexity of the problem; when the crack propagation path is linear, a one-dimensional array is used to store the minimum damage value of each node, for example, for the path nodes A→B→C→D→E, a one-dimensional array S[] is created, wherein S[0] stores the minimum damage value of A, S[1] stores the minimum damage value of B, and so on; and when the crack propagation path is complex and there are multiple branches, a two-dimensional matrix is ​​used to store the minimum damage value between each node pair, such as the matrix S[i][j] represents the minimum damage value from node i to node j; in addition to the minimum damage value of each node, the stored data can also include node-related state data, such as crack depth, length, stress intensity factor, etc., so as to more comprehensively describe the state of the crack at the node; in the calculation process, the state transition equation is recursively calculated for each node. The minimum damage value of each node is calculated and the result is stored in a memo array. When a node that has been calculated is encountered again, the stored minimum damage value is directly read from the memo to avoid repeated calculations, thereby significantly improving the efficiency of the algorithm. The specific operation is as follows: when calculating the minimum damage value of each node, first check whether the node already exists in the memo array (or matrix). If so, the corresponding minimum damage value is directly read from the memo using the node's index or identifier; if not, the state transition equation is used for calculation and the result is stored in the memo for subsequent direct reading; this method of storing and reading intermediate results not only reduces the amount of calculation but also optimizes the use of computing resources by avoiding redundant calculations; thereby obtaining a dataset of minimum damage values ​​for each node on the potential crack propagation path; if there is no order relationship between nodes and states (for example, path points have no fixed order or node names are discontinuous), a hash table or dictionary can be used to store the damage values ​​of the path points. The hash table provides constant time complexity for search and insertion operations; when searching, the hash table can be used to find and read the result by the path point identifier (such as node name or number).

[0039] In the preferred genetic algorithm of the present invention, the factors for evaluating the fitness function include the following: stress intensity factor SIF (reflecting the potential and extensibility of crack growth), node damage value (minimum damage value of each path segment calculated by dynamic programming), crack propagation distance (the distance the crack propagates from the starting point to the path point), material properties (including tensile strength, elastic modulus, etc.), crack morphology and initial crack conditions (crack shape and starting state), environmental factors (external factors such as temperature and humidity may affect crack growth), and crack propagation rate (the rate at which the crack propagates along the path). These factors can be used individually or in combination to evaluate the fitness function. The specific selection of which factor or combination of factors is mainly determined based on actual needs. The purpose of this is to comprehensively evaluate the advantages and disadvantages of each potential crack propagation path, thereby effectively guiding the genetic algorithm to search for the global optimal solution.

[0040] In the third step of the present invention, the crossover operation method preferably includes single-point crossover, multi-point crossover and uniform crossover.

[0041] In the preferred fourth step of the present invention, the maximum number of iterations of the ant colony algorithm is determined according to the complexity of the problem. If the problem is relatively simple, it is set to 50-200; if the problem is relatively difficult, it is set to more than 1000; thereby balancing the solution quality and computational efficiency.

[0042] Example 3: See also Figure 1 A method for predicting cracks in a hydro-generator set based on dynamic programming and its application, the prediction method comprising the following steps: Step 1: First, real-time monitoring and collection of stress data and temperature data of the hydro-generator set. Then, the stress data is subjected to denoising, normalization, and standardization in sequence to obtain a stress feature dataset. The temperature data is also subjected to denoising, normalization, and standardization in sequence to obtain a temperature feature dataset. Step 2: First, a three-dimensional model of the hydro-generator set is established, and the above-mentioned stress characteristic data set is input into the three-dimensional model for mechanical simulation to calculate the structural stress field distribution of the hydro-generator set, and the above-mentioned temperature characteristic data set is input into the three-dimensional model for thermodynamic simulation to calculate the thermal field distribution of the hydro-generator set; then, based on the structural stress field distribution, a number of potential crack propagation paths are determined through fracture mechanics analysis, and the potential crack propagation paths are divided into a number of nodes, and then the stress state data of each node is extracted from the structural stress field distribution, and the temperature state data of each node is extracted from the thermal field distribution; finally, the stress state data and the temperature state data of each node are input into the state transition equation of the dynamic programming algorithm, and the minimum damage value of each node is calculated, and a minimum damage value data set consisting of the minimum damage values ​​of each node is obtained and stored; Step 3: First, a population is initialized using a genetic algorithm. The population consists of several randomly generated crack propagation path individuals, each of which consists of several nodes. Then, in the above-mentioned minimum damage value dataset, the minimum damage value of the corresponding node in each crack propagation path individual is extracted, and the theoretical minimum damage value of each crack propagation path individual is calculated. Then, the actual damage value of each crack propagation path individual is calculated based on the above-mentioned stress characteristic dataset and temperature characteristic dataset. The theoretical minimum damage value is compared with the actual damage value through a fitness function to obtain the fitness value of each crack propagation path individual. Finally, selection operations, crossover operations, mutation operations and iterative updates are performed on the population according to the fitness values ​​of the crack propagation path individuals until convergence to generate a global optimal crack propagation path candidate set. Step 4: Optimize the path probability of each candidate path in the above global optimal crack propagation path candidate set through the ant colony algorithm, and select the candidate path with high path probability as the predicted crack propagation path of the hydro-generator set.

[0043] When applied, dynamic programming algorithm, genetic algorithm and ant colony algorithm work together to ensure the quality of crack prediction in terms of accuracy and efficiency. In terms of accuracy, the dynamic programming algorithm decomposes the crack propagation path prediction problem into multiple interrelated sub-problems, and uses the optimal substructure and state transition characteristics to gradually calculate the optimal solution of each sub-problem. In the crack propagation path prediction of hydro-generator sets, the crack propagation path is complex and changeable. Dynamic programming calculates the damage value of each node (such as stress intensity factor, crack depth, etc.), comprehensively considers environmental conditions and load changes, and ensures that each step of the decision is optimal, thereby improving the crack path prediction accuracy and avoiding the possible neglect of details by traditional methods. The genetic algorithm is based on natural selection and genetic mechanism, has a strong global search capability, can avoid falling into the local optimal solution, in the crack propagation path prediction, the crack may be along The crack propagation path expands in multiple directions, forming numerous potential paths. The genetic algorithm comprehensively explores all possible paths through selection, crossover, and mutation operations, helping to find more accurate crack propagation paths and forming a global optimal crack propagation path candidate set, laying the foundation for local optimization. It is particularly suitable for processing complex structures such as hydro-turbine generator sets. The ant colony algorithm simulates the behavior of ants in finding the shortest path, guides path selection through the pheromone update mechanism, and excels at local optimization. In crack propagation prediction, crack propagation involves complex stress changes in local areas. The ant colony algorithm simulates the selection behavior of ants on local paths and optimizes the crack propagation path in local areas. Combined with the global search capability of the genetic algorithm, after finding the potential crack propagation path globally, the ant colony algorithm quickly and finely optimizes, eliminates invalid paths, retains and strengthens valid paths, and further improves the overall prediction accuracy. In terms of efficiency, the dynamic programming algorithm stores intermediate calculation results (the minimum damage value data set of each node) through "memorized recursion" or "memorandum method", avoiding a large amount of repeated calculations in traditional methods. This can significantly reduce the calculation time when processing the crack propagation path prediction of complex structures of hydro-turbine generator sets, greatly improving the calculation efficiency. This efficiency advantage is even more obvious when facing large-scale data and complex models. The genetic algorithm quickly explores potential crack propagation paths on a global scale and determines the approximate direction and range, which saves time for subsequent local optimization. After finding the potential crack propagation paths globally, the ant colony algorithm can quickly and finely optimize these paths, further reducing the calculation time. The pheromone update mechanism of the ant colony algorithm makes the path selection more centralized and efficient, avoiding excessive calculations on invalid paths. In summary, in the crack prediction of hydro-turbine generator sets, the dynamic programming algorithm provides basic path prediction, the genetic algorithm performs global optimization, and the ant colony algorithm performs local optimization. The three work together to ensure the accuracy, comprehensiveness and efficiency of crack propagation path prediction.

[0044] Example 4: The basic content is the same as that of Example 3, except that: in the first step, the real-time monitoring and collection of stress data and temperature data of the hydro-generator set refers to: using optical fiber sensing technology to monitor and collect stress data and temperature data of the turbine main shaft, runner, and frame in the hydro-generator set in real time.

[0045] When applied, the stress and temperature data of key parts such as the turbine main shaft, runner, and frame in the hydro-turbine generator set are first monitored and collected in real time through fiber optic sensing technology. Fiber optic sensing technology has many advantages. First, it has high sensitivity and high precision, can sense tiny stress changes, detect physical quantities based on optical properties, provide high-precision and stable data, and has high response sensitivity to tiny stress changes, which is particularly suitable for crack detection and stress monitoring. Secondly, the distributed monitoring capability of fiber optic sensors can realize global monitoring, and the high spatial resolution helps to capture tiny stress concentration areas and crack initiation. In addition, fiber optic sensors are not affected by electromagnetic interference, adapt to complex environments, and can maintain high stability under harsh conditions such as high temperature, high humidity, and strong vibration. The fiber optic sensor has high accuracy and high precision. Moreover, the signal transmission loss of the fiber optic sensor is small, and data can be transmitted over long distances without significant attenuation, ensuring the accuracy and real-time performance of data from various parts of the structure to the data center. Finally, the fiber optic sensor is small in size and light in weight, which is convenient for embedded environmental monitoring and can be flexibly arranged in key parts of the hydro-turbine generator set. These pre-processed high-precision and high-real-time status data are transmitted to the dynamic programming algorithm and combined with the genetic ant colony algorithm for calculation, which can significantly improve the computing efficiency. The combination of fiber optic sensing technology, dynamic programming algorithm and genetic ant colony algorithm ensures the high efficiency of crack propagation path prediction, can quickly respond to crack changes, provide real-time early warning information, and provide a strong guarantee for the safe operation of the hydro-turbine generator set.

[0046] Embodiment 5: The basic content is the same as that of the third embodiment, except that in the second step, the state transition equation is: ; Where, represents the minimum cumulative damage value from node i to node j; represents the minimum damage value of node i; represents the crack extension cost from node i to node j; described Calculated by the following formula: ; Among them, w1, w2, w3, w4, and w5 are the weight factors of the node damage value, which respectively represent the weight ratio of stress intensity factor, crack depth, crack length, propagation distance, and temperature influence in the calculation of node damage value; the calculation formula of stress intensity factor (i) is: ; Where K(i) represents the stress intensity factor, σ represents the stress acting on the crack tip, which comes from the stress characteristic data set, and r represents the minimum distance between the crack tip and node i, which is obtained through the mechanical simulation calculation in the second step. The crack depth (i), crack length (i), and propagation distance (i) are also obtained through the mechanical simulation calculation in the second step. Indicates the effect of temperature on cracks; described Calculated by the following formula: ; in, represents the change in crack depth from node i to node j; represents the change in crack length from node i to node j; represents the change of stress intensity factor from node i to node j; 、 、 is a weight factor used to regulate the impact of different factors on the crack growth cost and is adjusted according to the needs of actual problems.

[0047] When applied, in the calculation During the process, w i The weight of is generally set manually or the optimal value is found through experiments. r is related to the specific characteristics of the crack propagation path and reflects the specific geometric size and development characteristics of the crack.

[0048] Example 6: The basic content is the same as that of Example 3, except that in the third step, the fitness function is: ; in, represents the fitness value of the crack growth path; represents the actual damage value of node i, which is calculated from the stress characteristic data set; Indicates the minimum damage value of node i calculated by the dynamic programming algorithm.

[0049] In the third step, the population is initialized by the genetic algorithm, and the population is composed of a number of randomly generated crack propagation path individuals, and each crack propagation path individual is composed of a number of nodes. This means that: an initial population composed of crack propagation path individuals is randomly generated by the genetic algorithm, and each crack propagation path individual is composed of a number of nodes, and the nodes correspond to positions in the distribution of the structural stress field and the thermal field; in the third step, the minimum damage value of the corresponding node in each crack propagation path individual is extracted from the above minimum damage value data set, and the theoretical minimum damage value of each crack propagation path individual is calculated to be It means that when calculating the theoretical minimum damage value of an individual crack propagation path, the minimum damage value of the node corresponding to the individual crack propagation path is extracted from the above minimum damage value data set, and the theoretical minimum damage value of the individual crack propagation path is obtained by recursion, thereby calculating the theoretical minimum damage value of each individual crack propagation path; in the third step, the fitness value of each individual crack propagation path is obtained by comparing the theoretical minimum damage value with the actual damage value through the fitness function. It means that the fitness value of each individual crack propagation path is calculated by inputting the theoretical minimum damage value and the actual damage value into the fitness function. fitness value, wherein the crack propagation path individual with a low fitness value is better; in the third step, the selection operation refers to: using the roulette selection method or the tournament selection method to select the crack propagation path individuals, and selecting a number of crack propagation path individuals with low fitness values ​​from the population to form a parent population; in the third step, the crossover operation refers to: randomly selecting two crack propagation path individuals from the parent population, and selecting a part of nodes from the two crack propagation path individuals for crossover to generate a child path; in the third step, the mutation operation refers to: performing a mutation operation on the child path One or several nodes are randomly adjusted to obtain a mutated offspring path, and the mutated offspring path is added to form a new population, or the mutated offspring path is mixed with the parent population to form a new population; in the third step, the iterative update until convergence to generate a global optimal crack propagation path candidate set means: calculating the fitness value of all crack propagation path individuals in the new population, and repeating the selection operation, crossover operation, and mutation operation on the new population according to the fitness value until the maximum number of iterations is reached or the fitness value of the new population does not change in several consecutive generations, thereby generating a global optimal crack propagation path candidate set.

[0050] When applied, first, the crack propagation path individual is composed of multiple nodes, and the crack propagation path individual is represented by P, that is, , represents a crack propagation path individual P from the crack starting point P1 to the end point P n A sequence of which Represents the i-th node of the potential crack propagation path; with the help of the dynamic programming algorithm, the minimum damage value of each node in the crack propagation path individual is pre-calculated and stored; for one of the crack propagation path individuals P, starting from the starting point P1, by comparing and extracting the minimum damage value stored in the dynamic programming algorithm, and then calculating the minimum damage value from the starting point P1 to the end point P by recursion, n The cumulative minimum damage value, that is, S in the formula DP (i) represents the minimum damage accumulation under ideal conditions; at the same time, in this crack propagation path individual P, from the starting point P1 to the end point P n Each segment has a stress intensity factor or other traditional damage value S(i) calculated based on actual conditions. These S(i) values ​​are calculated from the stress and temperature data of the turbine generator set monitored and collected in real time, reflecting the damage of crack propagation under actual conditions. The fitness function comprehensively compares the ideal and actual damage values, and converts S(i) into DP (i) and S(i) are combined to calculate a fitness value; if the fitness value is small, it indicates that the actual damage condition of the crack propagation path individual P is closer to the ideal state, that is, the crack propagation path is better; if the fitness value is large, it indicates that the actual damage condition of the crack propagation path individual P is greatly different from the ideal state, that is, the crack propagation path is worse; by combining the ideal and actual damage values, a scientific and effective basis is provided for the optimization of the crack propagation path; according to the calculated fitness value of each crack propagation path individual, a roulette mechanism or tournament selection is used to select the path with low fitness value to enter the next generation, and a parent population composed of paths with low fitness value is obtained, and then two parent crack propagation path individuals are randomly selected from the parent population, and a part of the path nodes are selected from the two parent crack propagation path individuals to cross and generate a new path; if two paths with lower fitness values ​​are selected as parent 1 and parent 2, the child path P is generated by the formula, and the formula is Then randomly change some nodes of the offspring path P to perform mutation operation, simulate natural mutation, and obtain the offspring path after mutation ; A new offspring path group is generated through crossover and mutation, and then fitness calculation, selection, crossover, mutation and other operations are performed to continuously iterate and optimize the path. When the fitness value converges, that is, there is no significant improvement for multiple generations or no obvious improvement for several consecutive generations, the algorithm terminates early.

[0051] Embodiment seven: The basic content is the same as that of Example 3, except that: in the fourth step, the path probability optimization of each candidate path in the above-mentioned global optimal crack propagation path candidate set is performed by using the ant colony algorithm, and the candidate path with high path probability is selected as the predicted crack propagation path of the hydro-turbine generator set, which means: first, the above-mentioned global optimal crack propagation path candidate set is used as the initial path set, and the pheromone concentration of each candidate path in the above-mentioned global optimal crack propagation path candidate set is initialized so that the pheromone concentration of each candidate path is the same; then, the path probability of each candidate path is calculated by using the path selection probability formula; then, based on the calculated path probability, path selection and transfer are performed in the global optimal crack propagation path candidate set, and candidate paths with high path probability are preferentially selected for transfer. During the transfer process, the pheromone concentration of the candidate path is updated to enhance the attractiveness of the high-quality path; finally, the path selection and pheromone update are iteratively performed until the convergence condition is met, and the candidate path with high path probability is output as the predicted crack propagation path of the hydro-turbine generator set; The path selection probability formula is: ; in, represents the pheromone concentration on the path (i, j), which is updated by the ant colony algorithm; represents the crack extension cost from node i to node j, which is calculated from experimental data or physical model; It represents the minimum cumulative damage value from node i to node j calculated by the dynamic programming algorithm, reflecting the optimal damage cost from node i to node j; represents the minimum damage value from i to k; α is the weight factor, which indicates the influence of pheromone concentration; β is the weight factor, which indicates the influence of crack extension cost; γ is the weight factor, which indicates the influence of damage value difference; N represents the set of all adjacent path points reachable from the current node i.

[0052] When applied, α represents the degree of influence of pheromone concentration, which controls the influence of pheromone concentration on path selection. Paths with higher pheromone concentrations are more likely to be selected because ants tend to choose paths they have traveled before. α is usually set to 1; β represents the degree of influence of crack extension cost, that is, the crack extension cost from node i to j. This cost reflects the difficulty of crack extension in the path. A higher cost means that the path extension is more difficult; γ represents the degree of influence of damage value difference. If the difference in crack damage value has a greater impact on path selection, γ can be increased; the path selection and pheromone update process of the ant colony algorithm, combined with the optimal damage value of dynamic programming, ensures the accuracy of local path optimization. The main task of the genetic algorithm is to explore crack extension paths through global search and select a better path through fitness evaluation. Dynamic programming can optimize the fitness evaluation of the genetic algorithm by providing the optimal damage value for each path.

[0053] Embodiment 8: The basic content is the same as that of Example 7, except that the pheromone concentration of the candidate path is updated according to the following formula: ; in, represents the pheromone concentration on the path (i, j) at time or iteration number t+1; Indicates the current moment or iteration number, used to mark the time step of the algorithm; represents the pheromone concentration on the path (i, j) at time or iteration number t; Indicates the pheromone volatility coefficient, which controls the volatility of pheromones; It represents the amount of pheromone added by the k-th contributing source on the path (i, j) at time or iteration number t; Indicates the total number of contributing sources; It represents the sum of pheromones released by all contributing sources on path (i, j).

[0054] When applied, the pheromone concentrations of all paths are first initialized so that the pheromone concentrations of all paths are set to the same initial value, which is usually a fixed constant, such as 0 or other appropriate numbers, indicating that the attractiveness of each path in the initial stage is the same. This setting ensures that the algorithm will not favor any specific path at the beginning, but gradually adjust the priority of the path according to subsequent pheromone updates; in each round of iteration, the path is selected according to the current pheromone concentration and heuristic information. After the path selection is completed, the pheromone concentrations of all paths are reduced according to the pheromone volatility coefficient ρ, and then the pheromone concentrations of the corresponding paths are increased according to the contribution of each path. This process is repeated until the preset number of iterations is reached or the optimal solution is found. In this way, the pheromone concentration update can help the ant colony algorithm find a better path in the crack propagation path prediction and improve the accuracy and reliability of the prediction results; in addition, ρ represents the pheromone volatility coefficient, which controls the volatility of the pheromone. It is generally [0,1] and is set to 0.5 in the present invention; is the amount of pheromone added from path node i to j at time t, Where Q is a constant, which represents the total amount of pheromone or the amount of pheromone carried by the ant after each path selection, L ij represents the length of the path from node i to node j, which is usually a cost function of the path, such as the damage value of crack extension or other factors; the pheromone concentration update can dynamically adjust the path probability, thereby gradually strengthening high-quality paths and eliminating low-quality paths during the iteration process; specifically, the pheromone volatilization mechanism can prevent the algorithm from converging to the local optimal solution too early, and the addition of new pheromones can enhance the attractiveness of high-quality paths, making the algorithm more inclined to select these paths. This dynamic adjustment mechanism enables the ant colony algorithm to continuously optimize path selection during the search process, improving global search capabilities and convergence speed.

[0055] Embodiment 9: The basic content is the same as that of the third embodiment, except that the prediction method further includes: Step 5: Verify or evaluate the predicted crack propagation path of the hydro-generator set, and output specific data of the predicted crack propagation path of the hydro-generator set; The specific operation is: verifying the error between the predicted crack propagation path of the above-mentioned hydro-generator set obtained by calculation and the actual crack propagation path through actual experimental data; when the error is less than or equal to 5%, confirming that the predicted crack propagation path of the above-mentioned hydro-generator set obtained by calculation is consistent with the actual crack propagation path; or simulating the crack propagation along the predicted crack propagation path of the above-mentioned hydro-generator set obtained by calculation through numerical simulation, evaluating the error between the simulation result and the predicted crack propagation path of the above-mentioned hydro-generator set; when the error is less than or equal to 5%, confirming that the predicted crack propagation path of the above-mentioned hydro-generator set accurately reflects the actual crack behavior.

[0056] When applying, the consistency between the predicted crack propagation path and the actual path is first verified through experiments. In specific operations, sensors are set at key parts of the hydro-turbine generator set to collect stress and temperature data during the crack propagation process in real time, and compared with the predicted results for analysis; if the error is within an acceptable range (for example, less than or equal to 5%), the accuracy of the prediction model can be confirmed; in addition, the crack is simulated along the calculated predicted crack propagation path through numerical simulation, and the error between the simulation result and the predicted crack propagation path is evaluated. If the error is also less than or equal to 5%, the accuracy of the predicted crack propagation path is confirmed; the reliability of the predicted path is ensured through verification or evaluation of these two methods; finally, the specific data of the predicted crack propagation path of the hydro-turbine generator set is output, including crack depth, crack direction, crack propagation speed and stress intensity factor; these specific data are of great significance to the safe operation of the hydro-turbine generator set. It has important guiding significance; among them, crack depth: deeper cracks usually affect the stability and safety of the structure. Therefore, the crack depth is crucial to assessing the health status of the unit; crack direction: the direction of crack expansion is very important for judging the fatigue life of the equipment and selecting repair methods. For example, if the crack expands along the stress direction of the key component, it may cause more serious damage; crack expansion rate: the crack expansion rate can help evaluate the growth rate of the crack, and then predict the remaining service life of the equipment, and help formulate maintenance plans; stress intensity factor (SIF): by predicting the SIF at the crack tip, the risk of crack expansion can be effectively assessed, helping to provide early warning of possible structural failure; these data not only verify the accuracy of the prediction model, but also provide maintenance personnel with detailed crack expansion information, thereby helping to formulate more accurate maintenance plans, reduce equipment downtime, and improve operating efficiency and safety.

[0057] Embodiment 10: The basic content is the same as that of the third embodiment, except that the method for predicting cracks in a hydro-generator set based on dynamic programming is applied to the crack prediction of a main shaft, a runner, and a frame of a hydro-generator set.

[0058] During operation, the hydro-turbine generator set is subjected to complex stress, vibration and temperature changes for a long time, which can easily lead to structural fatigue and cracks. If the cracks are not discovered in time and continue to expand, it may cause serious malfunction or even catastrophic failure of the equipment. The present invention uses crack prediction to identify potential crack problems in advance and take appropriate measures (such as shutdown inspection or local repair) to prevent accidents and ensure the safe operation of the equipment. Accurate and efficient crack prediction is of great significance in practical work. First, it can help operation and maintenance personnel to detect problems in time before the cracks expand to a dangerous level, realize preventive maintenance, avoid large-scale maintenance work and long shutdowns, reduce downtime and reduce maintenance costs. At the same time, by timely detection and treatment of cracks, further expansion of cracks can be avoided, structural damage can be reduced, and the service life of the hydro-turbine generator set can be extended. In addition, crack prediction enables operation and maintenance personnel to have a clearer understanding of the health status of the equipment, assist in scientific decision-making, and prevent premature scrapping or excessive maintenance. Accurate prediction and treatment in the early stage of crack expansion can avoid more extensive damage and reduce high maintenance costs. Accurate crack prediction also provides data support for equipment management, optimizes resource allocation, improves management efficiency, and provides The method provides a reference for future technological upgrades or structural optimizations and supports decision-making. For example, operation and maintenance personnel can arrange spare parts procurement and personnel deployment based on the prediction results, thereby improving overall management efficiency. The present invention predicts the crack paths of key components of the turbine main shaft, runner, and frame in the hydro-turbine generator set to obtain prediction results of the crack propagation path, including the direction, speed and development trend of the crack, as well as key parameters related to crack propagation, such as crack depth, width, and propagation location. These results help to accurately judge the development status of the crack and provide a basis for taking appropriate preventive and maintenance measures. In addition, the method is also applicable to other components of the hydro-turbine generator set (such as runner blades, hydro-turbine generator rotors, turbine blades, etc.) and has good versatility. Compared with existing technologies, traditional crack prediction models are often targeted at specific equipment or structures and are difficult to adapt to different working conditions and complex crack propagation paths. The present invention achieves efficient prediction of crack propagation paths in key components of the hydro-turbine generator set by integrating fiber optic sensing technology with multi-algorithm collaborative optimization of dynamic programming algorithms, genetic algorithms, and ant colony algorithms. It can be flexibly applied to crack prediction in a variety of complex structures and equipment.

[0059] The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. Any equivalent modifications or changes made by ordinary technicians in this field based on the contents disclosed in the present invention should be included in the protection scope recorded in the claims.

Claims

1. A method for predicting cracks in a hydro-generator set based on dynamic programming, characterized in that: include: S1, real-time monitoring and collection of stress data and temperature data of a hydro-generator set, sequentially performing denoising, normalization, and standardization on the stress data to obtain a stress characteristic data set, and sequentially performing denoising, normalization, and standardization on the temperature data to obtain a temperature characteristic data set; S2, establishing a three-dimensional model of the hydro-generator set, inputting the stress characteristic data set into the three-dimensional model for mechanical simulation, and calculating the structural stress field distribution of the hydro-generator set; The above temperature characteristic data set is input into the three-dimensional model for thermodynamic simulation to calculate the thermal field distribution of the hydro-generator set; Based on the structural stress field distribution, a number of potential crack propagation paths are determined through fracture mechanics analysis, the potential crack propagation paths are then divided into a number of nodes, and then stress state data of each node is extracted from the structural stress field distribution, and temperature state data of each node is extracted from the thermal field distribution; Input the stress state data and temperature state data of each node into the state transition equation of the dynamic programming algorithm, calculate the minimum damage value of each node, and obtain and store the minimum damage value data set composed of the minimum damage values ​​of each node; S3, initializing a population using a genetic algorithm, wherein the population consists of a number of randomly generated crack propagation path individuals, each of which consists of a number of nodes; extracting the minimum damage value of the corresponding node in each crack propagation path individual from the above minimum damage value dataset, and calculating the theoretical minimum damage value of each crack propagation path individual; The actual damage value of each crack propagation path individual is calculated based on the stress characteristic data set and the temperature characteristic data set, and the fitness value of each crack propagation path individual is obtained by comparing the theoretical minimum damage value with the actual damage value through the fitness function; According to the fitness values ​​of the crack propagation path individuals, the population is subjected to selection, crossover, mutation and iterative updates until convergence to generate a global optimal crack propagation path candidate set; S4, optimizing the path probability of each candidate path in the above-mentioned global optimal crack propagation path candidate set by using an ant colony algorithm, and selecting the candidate path with a high path probability as the predicted crack propagation path of the hydro-generator set; S5, verifying or evaluating the predicted crack propagation path of the hydro-generator set, and outputting specific data of the predicted crack propagation path of the hydro-generator set.

2. The method for predicting cracks in a hydro-generator set based on dynamic programming according to claim 1, characterized in that: In step S1, real-time monitoring and collection of stress data and temperature data of the hydro-generator set includes: using optical fiber sensing technology to real-time monitor and collect stress data and temperature data of the turbine main shaft, runner, and frame of the hydro-generator set.

3. The method for predicting cracks in a hydro-generator set based on dynamic programming according to claim 1, characterized in that: In step S2, the state transition equation is: ； Where, represents the minimum cumulative damage value from node i to node j; represents the minimum damage value of node i; represents the crack extension cost from node i to node j; described Calculated by the following formula: ； Among them, w1, w2, w3, w4, and w5 are the weight factors of the node damage value, which respectively represent the weight ratio of stress intensity factor, crack depth, crack length, propagation distance, and temperature influence in the calculation of node damage value; the calculation formula of stress intensity factor (i) is: ； Where K(i) represents the stress intensity factor, σ represents the stress acting on the crack tip, which comes from the stress characteristic data set, and r represents the minimum distance between the crack tip and node i, which is obtained through the mechanical simulation calculation in the second step. The crack depth (i), crack length (i), and propagation distance (i) are also obtained through the mechanical simulation calculation in the second step. Indicates the effect of temperature on cracks; described Calculated by the following formula: ； in, represents the change in crack depth from node i to node j; represents the change in crack length from node i to node j; represents the change of stress intensity factor from node i to node j; 、 、 is a weight factor used to regulate the impact of different factors on the crack growth cost and is adjusted according to the needs of actual problems.

4. The method for predicting cracks in a hydro-generator set based on dynamic programming according to claim 1, characterized in that: In step S3, the fitness function is: ； in, represents the fitness value of the crack growth path; represents the actual damage value of node i, which is calculated from the stress characteristic data set; Indicates the minimum damage value of node i calculated by the dynamic programming algorithm.

5. The method for predicting cracks in a hydro-generator set based on dynamic programming according to claim 4, characterized in that: In step S3, a population is initialized by a genetic algorithm, where the population consists of several randomly generated crack propagation path individuals, and each crack propagation path individual consists of several nodes, including: randomly generating an initial population consisting of crack propagation path individuals by a genetic algorithm, where each crack propagation path individual consists of several nodes, and the nodes correspond to positions in the distribution of the structural stress field and the thermal field; extracting the minimum damage value of the corresponding node in each crack propagation path individual from the minimum damage value data set, and calculating the theoretical minimum damage value of each crack propagation path individual, including: when calculating the theoretical minimum damage value of a crack propagation path individual, extracting the minimum damage value of the node corresponding to the crack propagation path individual from the above minimum damage value data set, obtaining the theoretical minimum damage value of the crack propagation path individual by recursion, and thus calculating the theoretical minimum damage value of each crack propagation path individual.

6. The method for predicting cracks in a hydro-generator set based on dynamic programming according to claim 5, characterized in that: In step S3, the fitness value of each crack propagation path individual is obtained by comparing the theoretical minimum damage value with the actual damage value through the fitness function, which includes: inputting the theoretical minimum damage value and the actual damage value into the fitness function to calculate the fitness value of each crack propagation path individual, wherein the crack propagation path individual with a lower fitness value is better.

7. The method for predicting cracks in a hydro-generator set based on dynamic programming according to claim 6, characterized in that: In step S3, the selection operation includes: A roulette wheel selection method or a tournament selection method is used to select the crack propagation path individuals, and a number of crack propagation path individuals with low fitness values ​​are selected from the population to form a parent population; The crossover operation includes: randomly selecting two crack propagation path individuals from the parent population, and selecting a portion of nodes from the two crack propagation path individuals for crossover to generate a child path; The mutation operation includes: randomly adjusting one or several nodes of the offspring path to obtain a mutated offspring path, and adding the mutated offspring path to form a new population, or mixing the mutated offspring path with the parent population to form a new population.

8. The method for predicting cracks in a hydro-generator set based on dynamic programming according to claim 7, characterized in that: In step S3, iterative updating until convergence to generate a global optimal crack propagation path candidate set includes: calculating the fitness values ​​of all crack propagation path individuals in the new population, and repeatedly performing selection operations, crossover operations, and mutation operations on the new population according to the fitness values ​​until a maximum number of iterations is reached or the fitness value of the new population does not change in several consecutive generations, thereby generating a global optimal crack propagation path candidate set.

9. The method for predicting cracks in a hydro-generator set based on dynamic programming according to claim 1, characterized in that: In step S4, the path probability of each candidate path in the global optimal crack propagation path candidate set is optimized by using an ant colony algorithm, and a candidate path with a high path probability is selected as the predicted crack propagation path of the hydro-generator set, which includes: First, the above-mentioned global optimal crack propagation path candidate set is used as the initial path set, and the pheromone concentration of each candidate path in the above-mentioned global optimal crack propagation path candidate set is initialized so that the pheromone concentration of each candidate path is the same; then, the path probability of each candidate path is calculated using the path selection probability formula; then, based on the calculated path probability, path selection and transfer are performed in the global optimal crack propagation path candidate set, and candidate paths with high path probabilities are preferentially selected for transfer. During the transfer process, the pheromone concentration of the candidate path is updated to enhance the attractiveness of high-quality paths; finally, path selection and pheromone update are iteratively performed until the convergence condition is met, and the candidate path with high path probability is output as the predicted crack propagation path of the hydro-generator set; The path selection probability formula is: ； in, represents the pheromone concentration on the path (i, j), which is updated by the ant colony algorithm; represents the crack extension cost from node i to node j, which is calculated from experimental data or physical model; It represents the minimum cumulative damage value from node i to node j calculated by the dynamic programming algorithm, reflecting the optimal damage cost from node i to node j; represents the minimum damage value from i to k; α is the weight factor, which indicates the influence of pheromone concentration; β is the weight factor, which indicates the influence of crack growth cost; γ is the weight factor, which indicates the influence of damage value difference; N represents the set of all adjacent path points reachable from the current node i; The pheromone concentration of the candidate path is updated according to the following formula: ； in, represents the pheromone concentration on the path (i, j) at time or iteration number t+1; Indicates the current moment or iteration number, used to mark the time step of the algorithm; represents the pheromone concentration on the path (i, j) at time or iteration number t; Indicates the pheromone volatility coefficient, which controls the volatility of pheromones; It represents the amount of pheromone added by the k-th contributing source on the path (i, j) at time or iteration number t; Indicates the total number of contributing sources; It represents the sum of pheromones released by all contributing sources on path (i, j).

10. Application of the method for crack prediction of a hydro-generator set based on dynamic programming according to claims 1 to 9 in crack prediction of a turbine main shaft, runner and frame in a hydro-generator set.

Images