A method for attitude adjustment of a turbine blade based on six-point positioning

Abstract

The present invention relates to the field of turbine blade attitude adjustment, and specifically discloses a turbine blade attitude adjustment method based on six-point positioning: Based on the six-point positioning theory, the turbine blade is measured by measurement sensors at different positions, and the target point equation of the turbine blade is fitted to obtain the six-point positioning measurement data of the turbine blade; Based on the six-point positioning measurement data of the turbine blade, a translation matrix and a rotation matrix are constructed, and the registration points are determined from the turbine blade measurement data; The turbine blade to be adjusted is iteratively registered with the standard turbine blade data, a registration objective function is constructed, a Markov decision model of the registration objective is established, and the solution of the registration objective function is converted into formulating a registration action strategy under deep reinforcement learning to achieve the attitude adjustment of the turbine blade. It solves the problems that the prior art is difficult to be applied to the on-line measurement of turbine blades, the attitude adjustment method of turbine blades is not intelligent enough, and the efficiency is not high.

Description

Technical Field

[0001] The present invention relates to the field of turbine blade attitude adjustment, and particularly to a method for adjusting the attitude of a turbine blade based on six-point positioning. Background Art

[0002] Turbine blades are key components of aeroengines and operate in extreme environments. The reliability of turbine blades is crucial for the stability of the engine, which requires high dimensional accuracy of the turbine blades. Therefore, accurately measuring the attitude data of turbine blades and making timely adjustments is of great significance. Due to the large size, large surface area, and complex shape of turbine blades, it is difficult to perform non-destructive measurements even in manufacturing or maintenance facilities. During the measurement of turbine blades in the factory, the shear deformation recording method can be used to detect fiber waves in the spar caps and other areas of the blade, but this technique is slow and costly and is usually only carried out when there are suspected known problems. The adaptive adjustment of the attitude of the core in the inner cavity of the wax mold pressing die based on the six-point positioning principle can greatly improve the compliance of turbine blade castings, shorten the development cycle, reduce the die cost, make the positioning measurement process more convenient, have lower requirements for the measurement environment, stronger adaptability, and higher efficiency in attitude measurement and adjustment.

[0003] Chinese Patent Publication No. CN112344875A discloses an automatic measurement planning method for turbine blades. First, the curved surface contour of the turbine blade is evenly meshed, and uniformly distributed point clouds to be measured are extracted from the turbine blade design model, and then the local tangent plane is fitted to calculate the normal vectors of the points to be measured. Then, kinematic modeling of the measuring machine tool is carried out, and combined with the coordinates and normal vectors of the points to be measured, kinematic inverse solution is performed to obtain the normal measurement attitude of each point to be measured. The problem of planning the measurement order between each point to be measured is transformed into an optimization problem, and the optimization goal is to minimize the movement time of the machine tool during the measurement process, and the optimization is carried out through the ant colony algorithm. Finally, according to the normal measurement attitude and measurement order of each point to be measured, a measurement program is automatically generated to perform the contour measurement of the turbine blade, and the measurement result can be parsed into the measured point cloud according to the machine tool structure. This invention improves the measurement accuracy and shortens the measurement time.

[0004] However, the above technologies have at least the following technical problems: The prior art generally uses the equal-height section method for turbine blade contour measurement, with insufficient measurement accuracy, and the attitude adjustment method is not comprehensive enough. The measurement point cloud is unevenly distributed on the surface of the turbine blade, the measurement speed is slow, it is difficult to be applied to the on-line measurement of turbine blades, the attitude adjustment method of turbine blades is not intelligent enough, and the efficiency is not high. Summary of the Invention

[0005] The present invention provides a method for attitude adjustment of a turbine blade based on six-point positioning, which solves the problems in the prior art that the contour measurement of a turbine blade generally adopts the equal-height section method, with insufficient measurement accuracy, and the attitude adjustment method is not comprehensive enough, the measured point cloud is unevenly distributed on the surface of the turbine blade, the measurement speed is slow, it is difficult to be applied to the on-line measurement of the turbine blade, the attitude adjustment method of the turbine blade is not intelligent enough, and the efficiency is not high. It can finally obtain a relatively accurate target image, select evenly distributed registration points, and perform intelligent registration to realize the intelligentization and high efficiency of attitude adjustment.

[0006] The present invention specifically includes the following technical solutions:

[0007] A method for attitude adjustment of a turbine blade based on six-point positioning includes the following steps:

[0008] Step S1. Based on the six-point positioning theory, measure the turbine blade by measurement sensors at different positions, fit the target point equation of the turbine blade, and obtain the six-point positioning measurement data of the turbine blade.

[0009] Step S2. Based on the six-point positioning measurement data of the turbine blade, construct a translation matrix and a rotation matrix, and determine the registration points from the turbine blade measurement data.

[0010] Step S3. Iteratively register the turbine blade to be adjusted with the standard turbine blade data, construct a registration objective function, establish a Markov decision model of the registration target, and convert the solution of the registration objective function into formulating a registration action strategy under deep reinforcement learning to realize the attitude adjustment of the turbine blade.

[0011] Further, the specific steps of step S1 include:

[0012] Emitting X-rays to the turbine blade by measurement sensors at different positions, receiving the X-rays reflected from the turbine blade, pre-marking target points on the turbine blade, and representing the target points of the turbine blade according to the X-rays of the target points of the turbine blade received by the measurement sensors at different positions.

[0013] Further, step S1 also includes:

[0014] Fusing the data measured by the measurement sensors at different positions to form the target point equation of the turbine blade, and fusing the target point equation of the turbine blade with the translation coordinate system and the rotation coordinate system to obtain the coordinates of each target point, so as to locate the turbine blade and obtain the six-point positioning measurement data of the turbine blade.

[0015] Further, the specific steps of step S2 include:

[0016] Determine the registration points from the turbine blade measurement data, form a turbine blade model from the turbine blade measurement data, perform uniform two-dimensional meshing on the turbine blade model, extract the vertices of each grid cell to form a set of contour feature points, select a point Q from the set of contour feature points, and construct a translation matrix and a rotation matrix for point Q.

[0017] Further, step S2 further includes:

[0018] After obtaining the normal vectors of all the contour feature points, calculate the difference in the direction angles of the normal vectors of adjacent two points, set an angle threshold, and select the points where the angle difference is greater than the angle threshold; calculate the distances between the selected points pairwise, and select the final registration points according to the distance threshold.

[0019] Further, step S3 specifically includes:

[0020] Iteratively register the turbine blade to be adjusted with the standard turbine blade data until the error between the registration points of the turbine blade to be adjusted and the registration points at the same position of the standard turbine blade data meets the preset range, and construct a registration objective function based on the change in the position coordinates of the registration points before and after registration.

[0021] Further, step S3 further includes:

[0022] Establish a Markov decision model for the registration objective, define a registration vector, define a reward function, and convert the registration objective where the maximum distance difference between the registration points of the turbine blade to be adjusted and the standard image at the same position is within the accuracy threshold range into a registration action policy under deep reinforcement learning.

[0023] Further, step S3 further includes:

[0024] The goal of the registration action policy under deep reinforcement learning is to form an optimal policy C through deep learning of the registration process, maximize the registration action function, so convert the solution of the registration objective function into the solution of the optimal policy C, and select the optimal policy C by constructing a policy neural network and an evaluation neural network, and then select the optimal registration action to achieve the purpose of successful registration.

[0025] Further, step S3 further includes:

[0026] Obtain the registration action of the current registration point by constructing a policy neural network. The policy neural network includes an input layer, a hidden layer Ⅰ, a hidden layer Ⅱ, and an output layer; the input data of the input layer is the registration vector of the registration point r v The number of neurons is the dimension of the registration vector; the hidden layer Ⅰ uses the ReLu activation function, the hidden layer Ⅱ uses the tanh activation function, and the output layer outputs the registration point r vThe registration action; obtaining the function value of the registration action by constructing an evaluation neural network, where the evaluation neural network includes an input layer, a hidden layer, and an output layer; the input of the input layer is the registration point r v The registration vector and the registration action of, and the function value of the registration action is output by the output layer.

[0027] The present invention has at least the following technical effects or advantages:

[0028] 1. The turbine blade is measured by measurement sensors at different positions, the target point equation of the turbine blade is fitted, and the six-point positioning measurement data of the turbine blade is obtained, quickly and effectively solving the problem of large measurement error of the shape accuracy of the turbine blade. By restricting the degrees of freedom in different directions through six-point positioning, the turbine blade is positioned, and the positioning measurement process is more convenient, the measurement environment requirements are low, and the adaptability is stronger.

[0029] 2. Based on the measurement data of the turbine blade, a turbine blade model is constructed. By uniformly dividing the two-dimensional grid of the model, the vertices of each grid unit are extracted to form a set of contour feature points. The translation matrix and rotation matrix of each feature point are measured to obtain the normal vector, and the registration points are selected according to the angle difference in the direction of the normal vector. The selected registration points are evenly distributed on the contour line, so that the obtained registration result is more accurate.

[0030] 3. The measurement data of the turbine blade to be adjusted is iteratively registered with the standard turbine blade data, a registration objective function is constructed, a Markov decision model of the registration objective is established, and the solution of the registration objective function is converted into a strategy for formulating a registration action under deep reinforcement learning. A strategy-evaluation neural network structure is adopted, and the deep reinforcement learning method is used, which has good autonomy, improves the algorithm stability, performs registration intelligently, and realizes the intelligentization and high efficiency of attitude adjustment.

[0031] 4. The technical solution of the present invention can effectively solve the problems in the prior art that generally use the equal-height cross-section method for turbine blade contour measurement, with insufficient measurement accuracy, and the attitude adjustment method is not comprehensive enough, the measurement point cloud is unevenly distributed on the surface of the turbine blade, the measurement speed is slow, it is difficult to be applied to the on-line measurement of the turbine blade, the attitude adjustment method of the turbine blade is not intelligent enough, and the efficiency is not high. It can quickly and effectively solve the problem of large measurement error of the shape accuracy of the turbine blade, the positioning measurement process is more convenient, the attitude adjustment process is more accurate, and the intelligentization and high efficiency of attitude adjustment are realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flowchart of a method for attitude adjustment of a turbine blade based on six-point positioning according to the present invention;

[0033] Figure 2 is a schematic diagram of six-point positioning according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0034] The technical solutions in the embodiments of the present application to solve the above problems are generally as follows:

[0035] The present invention measures the turbine blade with measurement sensors at different positions, fits the target point equation of the turbine blade, and obtains the six-point positioning measurement data of the turbine blade, quickly and effectively solving the problem of large measurement error of the shape accuracy of the turbine blade. By restricting the degrees of freedom in different directions through six-point positioning, the turbine blade is positioned, and the positioning measurement process is more convenient, the measurement environment requirements are low, and the adaptability is stronger; based on the measurement data of the turbine blade, a turbine blade model is constructed. By uniformly dividing the two-dimensional grid of the model, the vertexes of each grid unit are extracted to form a set of contour feature points, and the translation matrix and rotation matrix of each feature point are measured to obtain the normal vector. The registration points are selected according to the angular difference in the direction of the normal vector, and the selected registration points are evenly distributed on the contour line, so that the obtained registration result is more accurate; the measurement data of the turbine blade to be adjusted is iteratively registered with the standard turbine blade data, a registration objective function is constructed, a Markov decision model of the registration objective is established, and the solution of the registration objective function is converted into formulating a registration action strategy under deep reinforcement learning. A strategy-evaluation neural network structure is adopted, and the deep reinforcement learning method is used, which has good autonomy, improves the algorithm stability, and intelligently performs registration, realizing the intelligence and high efficiency of attitude adjustment.

[0036] To better understand the above technical solutions, the above technical solutions will be described in detail below in conjunction with the accompanying drawings of the specification and specific embodiments.

[0037] Refer to the attached Figure 1 , a method for attitude adjustment of a turbine blade based on six-point positioning according to the present invention includes the following steps:

[0038] S1. Based on the six-point positioning theory, measure the turbine blade with measurement sensors at different positions, fit the target point equation of the turbine blade, and obtain the six-point positioning measurement data of the turbine blade.

[0039] To solve the problem of spatial positioning of the turbine blade, the free rigid body motion theory is introduced. The free rigid body motion theory means that a free rigid body has six degrees of freedom in three-dimensional space, and any motion state of the rigid body can be decomposed into translation and rotation of the rigid body relative to the base point.

[0040] Construct a spatial rectangular coordinate system of the turbine blade, with the center point of the turbine blade as the base point, and use oxyz to represent the spatial rectangular coordinate system. Taking the base point o as the reference point, construct a function of the base point position changing with time t:

[0041] o′ = f t (o)

[0042] where, o′ represents the new position point to which the base point o moves over time, and f t represents a time function. As the position of the base point changes, a coordinate system o′x′y′z′ that is translated relative to the spatial rectangular coordinate system oxyz with the base point o′ as the origin is defined, which is called the translation coordinate system; the rotation of the blade relative to the base point o′ forms a new rotation coordinate system where, x θ represents the rotation angle of the blade in the x′-axis direction, y θ represents the rotation angle of the blade in the y-axis direction, represents the rotation angle of the blade in the z′-axis direction.

[0043] The turbine blade has 6 degrees of freedom in three-dimensional space, namely the translational degrees of freedom along 3 mutually perpendicular coordinate axes and the rotational degrees of freedom about 3 coordinate axes. Keeping the machine tool spindle fixed, the turbine blade is clamped on the adjustment device, a measuring sensor is installed on the machine tool spindle, and the degrees of freedom in different directions are restricted by six-point positioning. The distance between the measuring sensor and the turbine blade is adjusted, and the point cloud coordinate set of the target point of the turbine blade is measured based on the six-point positioning method to form a target point equation.

[0044] X-rays are emitted from measuring sensors at different positions to the turbine blade, and the X-rays reflected from the turbine blade are received. Target points are pre-marked on the turbine blade, and the target points of the turbine blade are represented based on the X-rays received by the measuring sensors at different positions, and the target point equation of the turbine blade is fitted. The specific fitting process is as follows:

[0045]

[0046] where, y is the representation of the target point of the turbine blade, represents the emission distance at which the measuring sensor at the i-th position emits X-rays to the turbine blade, represents the reflection distance at which the measuring sensor at the i-th position receives the X-rays reflected from the turbine blade, A 0 represents the initial emission distance, B 0 represents the initial reflection distance.

[0047] The data measured by the measuring sensors at different positions are fused to form the target point equation of the turbine blade, that is,

[0048]

[0049] where, Y represents the target point equation of the turbine blade, and are respectively the emission distance and the reflection distance of the measuring sensor at the n-th position, n represents the number of different positions of the measurement sensors. By integrating the target point equation of the turbine blade with the translation coordinate system and the rotation coordinate system, the coordinates of each target point are obtained, thereby positioning the turbine blade and obtaining the six-point positioning measurement data of the turbine blade.

[0050] S2. Based on the six-point positioning measurement data of the turbine blade, construct a translation matrix and a rotation matrix, and determine the registration points from the turbine blade measurement data.

[0051] Using industrial CT, measure and scan the standard turbine blade core model under non-destructive conditions, and extract the blade tissue contour from the scanned grayscale image. Import the measurement data of the turbine blade to be adjusted and the standard turbine blade data into three-dimensional modeling software for iterative registration, thereby obtaining the data required for attitude adjustment of the turbine blade to be adjusted and forming a three-dimensional model of the turbine blade, which clearly, accurately, and intuitively displays the internal structure, composition, material, and defect status of the turbine blade in the form of a three-dimensional stereoscopic image.

[0052] Before data registration, it is first necessary to determine the registration points from the turbine blade measurement data. Form a turbine blade model from the turbine blade measurement data, perform uniform two-dimensional mesh division on the turbine blade model, extract the vertexes of each mesh unit to form a set of contour feature points, select a point Q in the set of contour feature points, and the coordinates of point Q in the translation coordinate system and the rotation coordinate system are (x′(Q), y′(Q), z′(Q)) and Then the translation matrix and rotation matrix of point Q are respectively:

[0053]

[0054]

[0055] Among them, G M (Q) is the translation matrix of point Q, and G R (Q) is the rotation matrix of point Q.

[0056] The unit vector of the normal direction of point Q is (μ x , μ y , μ z ), then the normal vector of point Q is

[0057] Select the registration points according to the angular difference in the direction of the normal vector. After obtaining the normal vectors of all contour feature points, calculate the angular difference in the direction of the normal vectors of adjacent two points, set an angular threshold, and select the points with the angular difference greater than the angular threshold; calculate the distances between the selected points in pairs, and select the final registration points according to the distance threshold. The selection method of the registration points is:

[0058] ε 1≤ω(NV(Q)) - ω(NV(Q′)) ≤ ε 2

[0059] ∈ 1 ≤d|Q - Q′| ≤ ∈ 2

[0060] where ε 1 and ε 2 are the upper and lower limits of the angle threshold, ω represents the angle, NV(Q) represents the normal vector of point Q, ∈ 1 and ∈ 2 are the upper and lower limits of the distance threshold, d represents the distance function. Thus, the registration points are obtained.

[0061] S3. Iteratively register the turbine blade to be adjusted with the standard turbine blade data, construct a registration objective function, establish a Markov decision model for the registration objective, and convert the solution of the registration objective function into formulating a registration action strategy under deep reinforcement learning to achieve the attitude adjustment of the turbine blade.

[0062] Iteratively register the turbine blade to be adjusted with the standard turbine blade data until the error between the registration points of the turbine blade to be adjusted and the corresponding registration points of the standard turbine blade data at the same position meets the preset range, that is: the maximum distance difference between the registration points of the turbine blade to be adjusted and the standard turbine blade at the same position after registration is within the accuracy threshold range.

[0063] The change in the position coordinates of the registration points before and after registration is:

[0064]

[0065] where represents the coordinates of the v-th registration point of the blade profile after registration, n represents the number of iterations, α u and β u are the translational degree of freedom and rotational degree of freedom respectively in the u-th iterative registration process, r v represents the coordinates of the v-th registration point of the blade profile before registration.

[0066] Construct the registration objective function:

[0067]

[0068] where F is the registration objective function, β is the rotation factor, α is the translation factor, ||·|| is the norm. The registration objective is: the maximum distance difference between the registration points of the turbine blade to be adjusted and the standard turbine blade at the same position after registration is within the accuracy threshold range, that is, minimizing the objective function.

[0069] The registration points at the same position of the turbine blade to be adjusted and the standard turbine blade data are called registration point pairs. A Markov decision model of the registration target is established, and a registration vector is defined. Define the reward function γ, and the reward function represents the registration point r v After registration, a registration point pair is formed. The obtained reward. The registration target with the difference between the maximum distances of the registration points at the same position of the turbine blade to be adjusted and the standard turbine blade constructed within the accuracy threshold range is converted into a registration action strategy under deep reinforcement learning. Specifically, the strategy is defined as the probability of the success of the registration action, and the registration action function is defined as:

[0070]

[0071] Among them, represents the registration action function, and E c represents the expected value of the policy c for the reward function and the registration action.

[0072] The goal of the registration action strategy under deep reinforcement learning is to form an optimal strategy C through deep learning of the registration process, so that is maximized. Therefore, the solution of the registration target function is converted into the solution of the optimal strategy C.

[0073] An optimal strategy C is selected by constructing a policy neural network and an evaluation neural network, and then an optimal registration action is selected to achieve the purpose of successful registration.

[0074] The registration action of the current registration point is obtained by constructing a policy neural network. The policy neural network includes an input layer, a hidden layer I, a hidden layer II, and an output layer; the input data of the input layer is the registration vector of the registration point r v , and the number of neurons is the dimension of the registration vector; the hidden layer I uses the ReLu activation function, the hidden layer II uses the tanh activation function, and the output layer outputs the registration action of the registration point r v .

[0075] The function value of the registration action is obtained by constructing an evaluation neural network The evaluation neural network includes an input layer, a hidden layer, and an output layer; the input of the input layer is the registration vector and the registration action of the registration point r v .

[0076] The calculation process of the hidden layer is:

[0077]

[0078] Among them, represents the output of the hidden layer, γ v represents the reward function, σ represents the error factor, and y(s v) represents the output of the decision neural network, s v represents the registration point r v The registration vector of. The function value of the registration action is output by the output layer Among them, w represents the connection weight, and b represents the bias of the output layer.

[0079] After successful registration, the pose offset of the turbine blade to be adjusted is obtained according to the difference between the registration point pairs of the turbine blade to be adjusted and the standard turbine blade, and the turbine blade is reversely biased based on the pose offset, thereby completing the attitude adjustment of the turbine blade.

[0080] In summary, the attitude adjustment method of a turbine blade based on six-point positioning described in the present invention is completed.

[0081] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks the device with the specified functions.

[0082] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing in the process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks the steps with the specified functions.

[0083] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0084] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for adjusting the attitude of a turbine blade based on six-point positioning, characterized in that: The following steps are involved: Step S1. Based on the six-point positioning theory, the turbine blades are measured by measuring sensors at different positions, and the turbine blade target point equation is fitted to obtain the six-point positioning measurement data of the turbine blades; Step S2. Based on the six-point positioning measurement data of the turbine blade, a translation matrix and a rotation matrix are constructed, the registration point is determined from the turbine blade measurement data, the turbine blade measurement data is formed into a turbine blade model, the turbine blade model is uniformly meshed, the vertices of each mesh unit are extracted to form a contour feature point set, and a point Q in the contour feature point set is selected, and the coordinates of point Q in the translation coordinate system and the rotation coordinate system are (x′(Q), y′(Q), z′(Q)) and Then the translation matrix and rotation matrix of point Q are: Among them, G M (Q) is the translation matrix of point Q, G R (Q) is the rotation matrix of point Q; The unit vector in the normal direction of point Q is (μ x , μ y , μ z ), then the normal vector of point Q is Select the registration point according to the angle difference of the normal vector direction; after obtaining the normal vectors of all contour feature points, calculate the direction angle difference of the normal vectors of two adjacent points, set an angle threshold, and select the point whose angle difference is greater than the angle threshold; Calculate the distance between the selected points and select the final registration point according to the distance threshold; Step S3. Iteratively align the turbine blade to be adjusted with the standard turbine blade data, construct a registration objective function, establish a Markov decision model of the registration target, convert the solution of the registration objective function into a registration action strategy under deep reinforcement learning, and realize the posture adjustment of the turbine blade.

2. A method for adjusting the attitude of a turbine blade based on six-point positioning according to claim 1, characterized in that: The step S1 specifically includes: Measurement sensors at different positions emit X-rays to the turbine blades, receive X-rays reflected from the turbine blades, pre-mark target points on the turbine blades, and represent the turbine blade target points based on the X-rays of the turbine blade target points received by the measurement sensors at different positions.

3. A method for adjusting the attitude of a turbine blade based on six-point positioning as claimed in claim 2, characterized in that: The step S1 further comprises: The data measured by measurement sensors at different positions are fused to form the turbine blade target point equation, and the turbine blade target point equation is fused with the translation coordinate system and the rotation coordinate system to obtain the coordinates of each target point, thereby locating the turbine blade and obtaining the six-point positioning measurement data of the turbine blade.

4. A method for adjusting the attitude of a turbine blade based on six-point positioning according to claim 1, characterized in that: The step S3 specifically includes: The turbine blade to be adjusted is iteratively registered with the standard turbine blade data until the error between the registration point of the turbine blade to be adjusted and the registration point at the same position of the standard turbine blade data meets the preset range, and the registration objective function is constructed according to the change of the position coordinates of the registration points before and after the registration.

5. A method for adjusting the attitude of a turbine blade based on six-point positioning as claimed in claim 4, characterized in that: The step S3 further comprises: A Markov decision model of the registration target is established, the registration vector is defined, the reward function is defined, and the registration target whose maximum distance difference between the registration point at the same position of the turbine blade to be adjusted and the standard image is within the accuracy threshold is converted into a registration action strategy under deep reinforcement learning.

6. A method for adjusting the attitude of a turbine blade based on six-point positioning as claimed in claim 5, characterized in that: The step S3 further comprises: The goal of the registration action strategy under deep reinforcement learning is to form an optimal strategy C through deep learning of the registration process to maximize the registration action function. Therefore, the solution of the registration objective function is converted into the solution of the optimal strategy C. The optimal strategy C is selected by constructing a policy neural network and an evaluation neural network, and then the optimal registration action is selected to achieve the purpose of successful registration.

7. A method for adjusting the attitude of a turbine blade based on six-point positioning as claimed in claim 6, characterized in that: The step S3 further comprises: The registration action of the current registration point is obtained by constructing a strategy neural network. The strategy neural network includes an input layer, a hidden layer I, a hidden layer II and an output layer. The input data of the input layer is the registration point r v The registration vector is the number of neurons, and the dimension of the registration vector is the number of neurons. The hidden layer I uses the ReLu activation function, the hidden layer II uses the tanh activation function, and the output layer outputs the registration point r v The function value of the registration action is obtained by constructing an evaluation neural network, wherein the evaluation neural network includes an input layer, a hidden layer and an output layer; the input layer is the registration point r v The registration vector and registration action are output by the output layer, and the function value of the registration action is output by the output layer.

Images