An autonomous navigation control system suitable for a fan blade inner cavity inspection trolley

Abstract

This invention discloses an autonomous navigation and control system for a wind turbine blade internal cavity inspection vehicle, belonging to the field of autonomous navigation and control technology. After the inspection vehicle enters the blade internal cavity, the system reads the three-dimensional structural model and its current position and pose, determines the current axial section number and circumferential region number, collects test response data, and forms an active test observation group for the same section. The system performs sequence alignment and bidirectional excitation parity decomposition on the test responses, constructs an odd component response matrix, obtains the structural response position and direction through robust principal component decomposition, and filters the field correction grid domain. It generates a bias vector field and writes it into the original SDF field value, forming a reverse-proof corrected biasable traversable shell field. The current position of the inspection vehicle is mapped to the three-dimensional mesh of the shell field, and the navigation control quantities of the inspection vehicle are corrected by constructing shell control reference quantities and inspection constraint terms, enabling the inspection vehicle to adapt to changes in the wind turbine blade internal cavity structure and travel along the traversable shell region.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation and control technology, specifically to an autonomous navigation and control system suitable for a wind turbine blade internal cavity inspection vehicle. Background Technology

[0002] Autonomous navigation control typically involves the positioning, path planning, attitude adjustment, and motion control of mobile vehicles in unknown or semi-enclosed spaces. Wind turbine blades, as crucial components of wind power equipment, possess elongated, curved, and significantly varied cross-sectional cavity structures, featuring internal walls, webs, reinforcing ribs, adhesive areas, and connectors. When inspecting such spaces, the inspection vehicle needs to identify its axial cross-section and circumferential area within the wind turbine blade's internal cavity and generate navigation control parameters based on the vehicle's attitude, cavity boundaries, and passable areas. Therefore, autonomous navigation control of the wind turbine blade internal cavity inspection vehicle is a vital technical aspect of intelligent inspection of wind power equipment.

[0003] Existing autonomous navigation control methods for inspection vehicles mostly rely on preset 3D models, fixed paths, or single sensor feedback for driving control. Within the internal cavity of wind turbine blades, deviations may exist between the theoretical model and the actual cavity morphology. Inner wall boundaries, webs, reinforcing ribs, or bonded areas can cause changes in image boundaries, attitude disturbances, wheel speed fluctuations, and heading correction deviations. If the control system only relies on the original SDF field or a fixed passable space for navigation, it is difficult to accurately reflect the structural response position and direction within the current cross-section, easily leading to problems such as inaccurate target point selection, deviations in judging the space occupied by the vehicle's external outline, mismatched driving speed control, and insufficient advance yaw. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an autonomous navigation and control system for a wind turbine blade internal cavity inspection vehicle, solving the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: an autonomous navigation control system for a wind turbine blade internal cavity inspection vehicle, comprising a cross-sectional response acquisition module, an offset kernel generation module, a shell reconstruction module, and a constraint control module; The cross-section response acquisition module reads the three-dimensional structural model of the inner cavity of the wind turbine blade and determines the current axial cross-section number and circumferential region number. It extracts the theoretical structural position of the blade, the original SDF field value and the original SDF gradient direction, establishes a trial sampling window and collects trial response data. The bias kernel generation module performs sequence alignment of the left-biased and right-biased test responses based on the benchmark test response in the test response data, performs bidirectional excitation parity decomposition, constructs the odd component response matrix, and obtains the structural response location and structural response direction through robust principal component decomposition. The shell reconstruction module reads the three-dimensional mesh of the blade cavity and the solid mesh of the blade inner wall. It limits the candidate reconstruction region by the current axial section number, circumferential region number and structural response direction, and filters the field correction mesh domain. It generates the bias vector field and writes the original SDF field value to form the reverse proof correction bias pass-through shell field. The constraint control module maps the current position coordinates of the inspection vehicle to the three-dimensional mesh of the offset passage shell field, obtains the sampling point set of the vehicle body outline and determines the current target point, and constructs shell control reference quantities and inspection constraint terms to constrain and correct the inspection vehicle.

[0006] Preferably, the cross-section response acquisition module includes a cross-section recognition unit and a deflection execution unit; The cross-section recognition unit is used to read the three-dimensional structural model of the wind turbine blade cavity after the inspection trolley enters the inner cavity of the wind turbine blade, and extract the model coordinate system in the three-dimensional structural model as the blade coordinate system to obtain the current pose of the inspection trolley in the blade coordinate system. The current pose includes the position coordinates of the centroid of the inspection trolley and the vehicle attitude angle. The axial cross-section unit to which the inspection trolley belongs is determined according to the axial coordinate in the position coordinate, and the number of the axial cross-section unit is used as the current axial cross-section number. The circumferential region to which the inspection trolley belongs is determined according to the circumferential angle of the position coordinate relative to the blade axis in the current axial cross-section, and the number of the circumferential region is used as the current circumferential region number. Using the current axial section number as an index, the theoretical structural position of the blade within the corresponding axial section is read from the three-dimensional structural model, and the original SDF field value and original SDF gradient direction corresponding to the current position of the inspection trolley are read from the original SDF field value to form the cross-sectional reference data of the current axial section.

[0007] Preferably, the deflection execution unit is used to establish a trial sampling window corresponding to the current axial section unit using the current axial section number as an index, and to generate a basic driving command containing trial linear velocity, trial distance and sampling number using the current navigation direction when the inspection trolley enters the trial sampling window as the basic driving direction. Under the condition of keeping the basic driving command unchanged, zero deflection control quantity, left deflection control quantity and right deflection control quantity are superimposed on the basic driving command to generate a reference trial state, a left deflection trial state and a right deflection trial state. The inspection trolley is controlled to drive within the trial sampling window according to the order of the reference trial state, the left deflection trial state and the right deflection trial state, and the corresponding trial response data is collected in each trial driving state. The trial response data is collected into the corresponding trial sampling window according to the sampling time. The left yaw control quantity is opposite in direction to the right yaw control quantity and is executed through the steering control of the inspection trolley or the differential speed control of the left and right wheels. The test response data includes the baseline test response, the left yaw test response, and the right yaw test response. The baseline test response, the left yaw test response, and the right yaw test response all include the image boundary response, attitude response, wheel speed response, and heading correction quantity. The test response data at each sampling time is bound to the current axial section number, the current circumferential region number, the theoretical structure position of the blade, the original SDF field value, and the original SDF gradient direction.

[0008] Preferably, the bias kernel generation module includes a trial alignment unit, a parity decomposition unit, and a bias kernel generation unit; The trial alignment unit is used to align the left-skewed trial response and the right-skewed trial response with a reference using a dynamic time warping algorithm, so as to obtain the reference trial response sequence, the left-skewed trial response sequence and the right-skewed trial response sequence. The parity decomposition unit is used to perform response difference processing on the left-biased trial response and the right-biased trial response with the benchmark trial response sequence to obtain the left-biased relative response sequence and the right-biased relative response sequence, and to perform bidirectional excitation parity decomposition on the left-biased relative response sequence and the right-biased relative response sequence to obtain the even component response sequence and the odd component response sequence. The even component response sequence is composed of the same-direction components of the left-biased relative response sequence and the right-biased relative response sequence, while the odd component response sequence is composed of the opposite-direction components of the left-biased relative response sequence and the right-biased relative response sequence. The left yaw relative response sequence includes left yaw image boundary relative response, left yaw attitude relative response, left yaw wheel speed relative response, and left yaw heading correction relative response; The right yaw trial response sequence includes the right yaw image boundary relative response, the right yaw attitude relative response, the right yaw wheel speed relative response, and the right yaw heading correction relative response.

[0009] Preferably, the bias kernel unit is used to arrange the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components in the odd component response sequence according to the same sampling number to construct the odd component response matrix of the current axial section; each row of the odd component response matrix corresponds to a sampling number, and each column corresponds to a response type; the response types include image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components; Robust principal component decomposition is performed on the odd component response matrix, which decomposes the odd component response matrix into a low-rank structured response matrix and a sparse perturbation matrix; After normalizing each column of the low-rank structural response matrix L, the response intensity is calculated by weighted sum of squares and square root for each response type. The continuous trial sampling window with the largest response intensity is selected as the structural response window. The positions of each sampling time in the blade coordinate system within the current structural response window are weighted and summed to obtain the structural response position. Within the structural response window, the sum of squares of the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components are calculated respectively. The ratio of the sum of squares of each response type to the sum of squares of all response types is taken as the response contribution of the corresponding response type. Based on the sign direction of the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components within the structural response window, the structural response direction is determined. When there are multiple response types with inconsistent sign directions, the sign directions of each response type are weighted and synthesized using the response contribution as the weight to obtain the structural response direction of the current axial section.

[0010] Preferably, the shell reconstruction module includes a domain delimitation unit, a vector generation unit, and a shell generation unit; The domain delimiting unit is used to read the three-dimensional mesh of the blade cavity and the solid mesh of the blade inner wall in the three-dimensional structural model, and select the current axial section unit and its adjacent axial mesh with the current axial section number as the center to form the axial candidate range; select the current circumferential region and the corresponding adjacent circumferential region with the current circumferential region number as the center and according to the structural response direction to form the circumferential candidate range; within the axial candidate range and the circumferential candidate range, the mesh of the blade inner wall solid is removed according to the original SDF field value, and the mesh containing the corresponding and continuously connected structural response position is retained as the candidate reconstruction region; Within the candidate reconstruction region, the structural response location is used as the starting grid point. Connectivity filtering is performed along the adjacent direction of the starting grid point, and the grid connected to the structural response location is retained to obtain the field-corrected grid domain. The structural response location is mapped to the grid in the field-corrected grid domain to obtain the source point element group. The original SDF gradient direction and structural response direction in the source point element group are read. The structural response direction is projected onto the cross-sectional plane of the current axial section to obtain the cross-sectional response direction.

[0011] Preferably, the vector generation unit is used to compare the cross-sectional response direction with the structural reference position using the theoretical structural position of the blade as the structural reference position; When the cross-sectional response direction points to the structural reference position, the opposite direction of the cross-sectional response direction is taken as the bias direction of the passing shell. When the cross-sectional response direction deviates from the structural reference position, the cross-sectional response direction is taken as the bias direction of the passing shell. The odd-component response sequence and the even-component response sequence are subjected to L2 norm squared calculation to obtain the odd-component response energy and the even-component response energy respectively. Then, the odd-component response energy is multiplied by the matrix norm of the low-rank structural response matrix to obtain the structural response quantity. Then, the even-component response energy is multiplied by the matrix norm of the sparse perturbation component to obtain the proof-of-contrast response quantity. The bias vector field amplitude is obtained by calculating the proportion of the structural response quantity in the sum of the structural response quantity and the proof-of-contrast response quantity. When the sum of the structural response quantity and the proof-of-contrast response quantity is zero, the bias vector field amplitude is zero. The bias vector field is obtained by combining the source element group, the bias direction of the passing shell, and the magnitude of the bias vector field.

[0012] Preferably, the shell generation unit is used to perform weighted Poisson field reconstruction within the field correction grid domain using a bias vector field as gradient correction input, and generate a scalar correction field. Using the field correction grid domain as the writing range, the scalar correction field is written into the original SDF field value of the corresponding grid in the current field correction grid domain to form the bias correction SDF field value. After generating the bias correction SDF field value, Eikonal re-initialization processing is performed on the bias correction SDF field value. The SDF field value obtained by re-initialization is used as the shell field value. Based on the shell field values ​​of adjacent grid nodes, the field value difference of the current grid in the blade coordinate system is calculated. The field value difference is combined to form the spatial gradient vector of the current grid node to obtain the shell gradient direction of the current grid. Read the corresponding shell field value in each grid cell, and combine it with the outer contour occupancy size of the inspection vehicle to determine the shell-based passage area of ​​the current grid cell: If the shell field value corresponding to the mesh cell can accommodate the size occupied by the outer contour of the inspection vehicle, then the current mesh cell is written into the shelled passable area. If the shell field value corresponding to the grid cell cannot accommodate the size occupied by the outer contour of the inspection vehicle, then the current grid cell is written into the shelled inaccessible region. Based on the difference in shell field values ​​between adjacent grid nodes, the grids in the shelled passable area are divided into velocity zones. When the difference in shell field values ​​between adjacent grid nodes reaches the set velocity zone limit, the corresponding grid node is assigned to the velocity control zone. The shell gradient direction at the current position is synthesized with the structural response direction to generate an advance yaw vector; By combining the shell field values, shell gradient direction, shelled passable region, local velocity control region, advance yaw vector, and field reconstruction record, a reverse proof corrected bias passable shell field is constructed.

[0013] Preferably, the constraint control module includes a shell reference generation unit and a constraint control unit; The shell reference generation unit is used to map the current position coordinates of the inspection vehicle to the three-dimensional mesh of the reverse proof correction bias passage shell field, obtain the vehicle outline sampling point set and determine the mesh cell corresponding to the current position of the inspection vehicle, and set the current mesh cell as the current target point; The shelling determination for the current target point is performed as follows: Determine whether the current target point is located within the shelled passable area; If the current target point is located within the shelled passable area, then the current target point will be used as the shelled target point; If the current target point is outside the shelled passable region, then search for the grid position with the smallest Euclidean distance to the current target point and belonging to the shelled passable region in the proof by contradiction modified bias passable shell field as the shelled target point; The current pose of the inspection vehicle, the shell-shaped target point, the shell field value corresponding to the current position, the shell gradient direction, the velocity control zone marker, and the advance yaw vector are combined into a shell control reference quantity.

[0014] Preferably, the constraint control unit is used to construct inspection constraint items, including: Read the shell field value and shelled passable region corresponding to each vehicle outline sampling point, and construct the vehicle outline shell constraint term as follows; When each candidate vehicle outline sampling point is located within the shelled passable area, the current inspection car control quantity satisfies the vehicle outline shell constraint. When any candidate vehicle outline sampling point is located outside the shelled passable area, the current inspection car control quantity does not meet the vehicle outline shell constraint. After obtaining the outer shell constraint terms of the vehicle body, the target direction from the current position to the shell target point is determined based on the current position of the inspection vehicle and the shell target point. At the same time, the shell gradient direction corresponding to the current position is read. The target direction is used as the main direction of the candidate motion direction, and the shell gradient direction is used as the correction direction of the candidate motion direction. The candidate motion direction is simultaneously directed toward the shell target point and changes along the extension direction of the shell passable area to construct the target approach constraint terms. When the inspection vehicle is currently located within the speed control zone, the speed limit value corresponding to the current speed control zone is read, and the candidate driving speed is controlled not to exceed the speed limit value to construct a speed constraint term. When the current axial section number of the inspection trolley falls within the advance yaw trigger range, an advance yaw constraint term is constructed; wherein, the advance yaw trigger range consists of the axial section numbers before the axial section pointed to by the advance yaw vector; The navigation control quantity for the current control cycle is determined using the shell control reference quantity, and the inspection constraint term is used as a constraint condition to constrain and correct the navigation control quantity.

[0015] This invention provides an autonomous navigation and control system for a wind turbine blade internal cavity inspection vehicle. It has the following advantages: (1) The system's cross-sectional response acquisition module is used to read the three-dimensional structural model when the inspection trolley enters the inner cavity of the wind turbine blade and triggers cross-sectional recognition, determine the current axial cross-sectional number and circumferential region number, extract the theoretical structural position of the blade, the original SDF field value and the original SDF gradient direction, establish a trial sampling window, and acquire the benchmark trial response, left yaw trial response and right yaw trial response within the trial sampling window to form an active trial observation group for the same cross-section. Through this module, the system can incorporate the image boundary response, attitude response, wheel speed response and heading correction of the inspection trolley in the current axial cross-section into the navigation control judgment, so that the autonomous navigation control of the inspection trolley no longer relies solely on the preset three-dimensional structural model or fixed path, but combines the actual driving response of the inner cavity of the blade to carry out subsequent analysis.

[0016] (2) The bias kernel generation module of this system is used to perform bidirectional excitation parity decomposition on the left bias test response and the right bias test response after sequence alignment based on the benchmark test response in the same section active test observation group, construct the odd component response matrix, and obtain the structural response position and structural response direction through robust principal component decomposition. The shell reconstruction module is used to read the three-dimensional mesh of the blade cavity and the solid mesh of the blade inner wall, limit the candidate reconstruction area with the current axial section number, circumferential region number and structural response direction, and screen the field correction mesh domain, generate the bias vector field and write it into the original SDF field value to form the reverse proof correction bias passage shell field; compared with the navigation control method based only on the original SDF field value or the theoretical structural position of the blade, the above two modules can write the structural response obtained by active test into the passage shell field, so that the judgment of the passable area is closer to the actual structural state of the wind turbine blade cavity.

[0017] (3) The system's constraint control module is used to map the current position coordinates of the inspection trolley to the three-dimensional mesh of the reverse proof correction bias passage shell field, obtain the vehicle outline sampling point set and determine the current target point, construct the shell control reference quantity and inspection constraint terms, and perform constraint correction on the inspection trolley. Through the cooperation of the cross-section response acquisition module, the kernel generation module, the shell reconstruction module and the constraint control module, the system completes the identification of the current axial cross section and the current circumferential region, active trial response acquisition, structural response position and structural response direction extraction, reverse proof correction bias passage shell field construction, shell-based target point determination, vehicle outline shell constraint, target approach constraint, speed constraint and advance yaw constraint, etc. This scheme can control and correct navigation anomalies caused by changes in the cross section of the wind turbine blade cavity, deviation of the inner wall boundary, stiffeners, web, glued area or connectors, so that the inspection trolley can travel in the shell-based passage area, improve problems such as target point jump, wall-hugging driving, steering lag, speed control inaccuracy and vehicle outline approaching the impassable area. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the autonomous navigation control system for a wind turbine blade internal cavity inspection vehicle according to the present invention; Figure 2 This is a block diagram of the operation logic of an autonomous navigation control system for a wind turbine blade internal cavity inspection vehicle according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 Please see Figure 1 This invention provides an autonomous navigation control system for a wind turbine blade internal cavity inspection vehicle. To achieve the above objectives, this invention is implemented through the following technical solutions: including a cross-sectional response acquisition module, an offset kernel generation module, a shell reconstruction module, and a constraint control module. The cross-section response acquisition module reads the three-dimensional structural model of the inner cavity of the wind turbine blade and determines the current axial cross-section number and circumferential region number. It extracts the theoretical structural position of the blade, the original SDF field value and the original SDF gradient direction, establishes a trial sampling window and collects trial response data. The bias kernel generation module performs sequence alignment of the left-biased and right-biased test responses based on the benchmark test response in the test response data, performs bidirectional excitation parity decomposition, constructs the odd component response matrix, and obtains the structural response location and structural response direction through robust principal component decomposition. The shell reconstruction module reads the three-dimensional mesh of the blade cavity and the solid mesh of the blade inner wall. It limits the candidate reconstruction region by the current axial section number, circumferential region number and structural response direction, and filters the field correction mesh domain. It generates the bias vector field and writes the original SDF field value to form the reverse proof correction bias pass-through shell field. The constraint control module maps the current position coordinates of the inspection vehicle to the three-dimensional mesh of the offset passage shell field, obtains the sampling point set of the vehicle body outline and determines the current target point, and constructs shell control reference quantities and inspection constraint terms to constrain and correct the inspection vehicle.

[0021] In this embodiment, the cross-sectional response acquisition module first undertakes the basic perception and active exploration tasks after the wind turbine blade internal cavity inspection vehicle enters the inspection environment. It reads the three-dimensional structural model of the wind turbine blade internal cavity, determines the current axial cross-section number and circumferential region number based on the inspection vehicle's position in the blade coordinate system, and then extracts the theoretical structural position of the blade near that location, the original SDF field value, and the original SDF gradient direction. Subsequently, it establishes an exploration sampling window and collects exploration response data. For example, when the inspection vehicle travels from the blade root towards the blade tip and enters a section of internal cavity space near the web and bonding area, the system first determines which axial cross-section and circumferential region the vehicle is in. Then, it controls the vehicle to perform baseline exploration, left-side exploration, and right-side exploration within the exploration sampling window, collecting image boundary response, attitude response, wheel speed response, and heading correction, respectively. Through this module, the inspection vehicle no longer simply travels along a pre-set route but incorporates the actual driving feedback under the current blade internal cavity environment into subsequent judgments, enabling the system to understand the impact of the internal cavity structure of that cross-section on the vehicle's movement.

[0022] The bias kernel generation module and the shell reconstruction module are responsible for extracting structural responses and constructing the passing shell field based on the aforementioned trial data. The bias kernel generation module uses the baseline trial response in the trial response data as a reference, performs sequence alignment on the left-biased and right-biased trial responses, ensuring that the three types of trial responses fall on the same axial sampling position. It then performs bidirectional excitation parity decomposition to construct the odd-component response matrix, and obtains the structural response location and direction through robust principal component decomposition. For example, if the image boundary changes significantly when the vehicle veers left during a trial, and the attitude and wheel speed change in opposite directions when it veers right, the system will use this difference in left and right trials as a structural response clue, separating the structural components caused by the inner wall, web, stiffeners, or adhesive areas. Subsequently, the shell reconstruction module reads the three-dimensional mesh of the blade cavity and the solid mesh of the blade inner wall. It defines the candidate reconstruction region and filters the field correction mesh domain by using the current axial section number, circumferential region number and structural response direction. It generates an offset vector field and writes it into the original SDF field value to form a reverse proof correction offset passage shell field. Compared with the existing technical means that only rely on the original SDF field value, the theoretical structural position of the blade or the fixed passage area, this solution can write the actual structural response during the inspection process into the passage space judgment, so that the passage area is more consistent with the actual structural state of the wind turbine blade cavity.

[0023] After the formation of the reverse-proof correction bias passage shell field, the constraint control module undertakes the task of correcting the driving control of the inspection vehicle. The specific process is as follows: the current position coordinates of the inspection vehicle are mapped to the 3D mesh of the reverse-proof correction bias passage shell field, obtaining the set of sampling points on the vehicle's outer contour and determining the current target point. Then, shell control reference quantities and inspection constraint terms are constructed to correct the constraints on the inspection vehicle. For example, when the inspection vehicle approaches a bend in the inner wall, the system will determine whether the sampling points on the vehicle's outer contour are still within the shelled passable area; if the current target point falls outside the impassable area, a passable position is selected in the reverse-proof correction bias passage shell field as the shelled target point, and the navigation control quantity is corrected by combining the shell field value, shell gradient direction, speed control zone, and advance yaw vector. Through the coordinated efforts of the cross-sectional response acquisition module, offset kernel generation module, shell reconstruction module, and constraint control module, the system can undertake tasks such as identifying the current axial cross-section and circumferential region, acquiring trial response data, determining the structural response position and direction, constructing the offset passable shell field for verification correction, judging the passability of the vehicle's outer contour, determining the current target point, and correcting inspection constraints. This solution can improve upon existing inspection trolleys that are prone to abnormal situations in the inner cavity of wind turbine blades, such as target point deviation, wall-hugging driving, steering lag, speed control mismatch, and the vehicle's outer contour approaching impassable areas. It enables the inspection trolley to travel along the shelled passable area, achieving the goal of adapting to the complex structure of the inner cavity of wind turbine blades and carrying out continuous inspections.

[0024] Example 2 Please refer to Figure 2 Specifically: the cross-section response acquisition module includes a cross-section recognition unit and a deflection execution unit; The cross-section recognition unit is used to read the three-dimensional structural model of the wind turbine blade cavity after the inspection trolley enters the inner cavity of the wind turbine blade, and extract the model coordinate system in the three-dimensional structural model as the blade coordinate system to obtain the current pose of the inspection trolley in the blade coordinate system. The current pose includes the position coordinates of the centroid of the inspection trolley and the vehicle attitude angle. The axial cross-section unit to which the inspection trolley belongs is determined according to the axial coordinate in the position coordinate, and the number of the axial cross-section unit is used as the current axial cross-section number. The circumferential region to which the inspection trolley belongs is determined according to the circumferential angle of the position coordinate relative to the blade axis in the current axial cross-section, and the number of the circumferential region is used as the current circumferential region number. Using the current axial section number as an index, the theoretical structural position of the blade within the corresponding axial section is read from the three-dimensional structural model, and the original SDF field value and original SDF gradient direction corresponding to the current position of the inspection vehicle are read from the original SDF field value to form the cross-sectional reference data of the current axial section. Among them, the theoretical structural position of the blade is the set of positions of the inner wall boundary, stiffener, web, adhesive area or connector determined by the blade design model within the current axial section; the original SDF field value is the sign distance value from the current position to the inner wall or internal structural boundary of the blade; and the original SDF gradient direction is the spatial change direction of the original SDF field value at the current position.

[0025] The deflection execution unit is used to establish a test sampling window corresponding to the current axial section unit, using the current axial section number as an index. It uses the current navigation direction when the inspection trolley enters the test sampling window as the basic driving direction and generates a basic driving command containing the test linear velocity, test distance, and number of samplings. While keeping the basic driving command unchanged, it superimposes zero deflection control, left deflection control, and right deflection control to the basic driving command to generate a reference test state, a left deflection test state, and a right deflection test state. It controls the inspection trolley to drive within the test sampling window according to the order of the reference test state, the left deflection test state, and the right deflection test state, and collects the corresponding test response data in each test driving state. It also collects the test response data into the corresponding test sampling window according to the sampling time. The left yaw control quantity is opposite in direction to the right yaw control quantity and is executed through the steering control of the inspection trolley or the differential speed control of the left and right wheels. The test response data includes the baseline test response, the left yaw test response, and the right yaw test response. The baseline test response includes the baseline image boundary response, the baseline attitude response, the baseline wheel speed response, and the baseline heading correction. The left yaw test response includes the left yaw image boundary response, the left yaw attitude response, the left yaw wheel speed response, and the left yaw heading correction. The right yaw test response includes the right yaw image boundary response, the right yaw attitude response, the right yaw wheel speed response, and the right yaw heading correction. The test response data at each sampling time is bound to the current axial section number, the current circumferential region number, the theoretical structure position of the blade, the original SDF field value, and the original SDF gradient direction.

[0026] In this embodiment, after the inspection trolley enters the inner cavity of the wind turbine blade, the section recognition unit reads the three-dimensional structural model of the inner cavity of the wind turbine blade and extracts the model coordinate system in the three-dimensional structural model as the blade coordinate system to obtain the current pose of the inspection trolley in the blade coordinate system; it determines the axial section element to which the inspection trolley belongs based on the axial coordinates in the position coordinates of the inspection trolley's centroid point, and uses the number of the axial section element as the current axial section number; it determines the circumferential region to which the inspection trolley belongs based on the circumferential angle of the position coordinates relative to the blade axis in the current axial section, and uses the number of the circumferential region as the current circumferential region number. Subsequently, using the current axial section number as an index, it reads the theoretical structural position of the blade in the corresponding axial section from the three-dimensional structural model, and reads the original SDF field value and original SDF gradient direction corresponding to the current position of the inspection trolley from the original SDF field value to form the section reference data of the current axial section. The deflection execution unit uses the current axial section number as an index to establish a test sampling window corresponding to the current axial section unit. It uses the current navigation direction when the inspection trolley enters the test sampling window as the basic travel direction and generates a basic travel command containing the test linear velocity, test travel distance, and number of samplings. While keeping the basic travel command unchanged, it superimposes the zero deflection control quantity, left deflection control quantity, and right deflection control quantity to generate the reference test state, left deflection test state, and right deflection test state. The inspection trolley is controlled to travel within the test sampling window according to the above test states, and the reference test response, left deflection test response, and right deflection test response are collected. The test response data at each sampling time is bound to the current axial section number, the current circumferential region number, the theoretical structure position of the blade, the original SDF field value, and the original SDF gradient direction. Through this implementation method, the system can obtain various types of driving feedback during the actual inspection of the inner cavity of the wind turbine blade, such as image boundary response, attitude response, wheel speed response, and heading correction. This solves the problem that relying solely on static three-dimensional structural models, fixed paths, or single location information is insufficient to reflect the true passage status of the inner cavity of the blade. Compared with the existing technical means of inspection trolleys directly traveling in the inner cavity of wind turbine blades based on preset paths or original SDF field values, this solution can incorporate the current axial section, current circumferential region, theoretical structural position of the blade, original SDF field value, original SDF gradient direction, and active trial response into the navigation control process. This allows the inspection trolley to identify response differences caused by structural changes when passing through inner wall boundaries, reinforcing ribs, webs, adhesive areas, or near connectors. This improves abnormal situations such as target point deviation, wall-hugging driving, steering lag, inaccurate heading correction, and the vehicle's outer contour approaching impassable areas, thus achieving the goal of adapting to the complex structure of the inner cavity of wind turbine blades and supporting continuous inspection.

[0027] Example 3 Please refer to Figure 2 Specifically: the bias kernel generation module includes a trial alignment unit, a parity decomposition unit, and a bias kernel generation unit; The probe alignment unit is used to align the left-skewed probe response and the right-skewed probe response with a reference and a dynamic time warping algorithm, so that the sampled data in the reference probe response, the left-skewed probe response and the right-skewed probe response correspond to the same axial sampling position in the current axial section, and obtain the reference probe response sequence, the left-skewed probe response sequence and the right-skewed probe response sequence. The parity decomposition unit is used to perform response difference processing on the left-biased trial response and the right-biased trial response with the benchmark trial response sequence to obtain the left-biased relative response sequence and the right-biased relative response sequence, and to perform bidirectional excitation parity decomposition on the left-biased relative response sequence and the right-biased relative response sequence to obtain the even component response sequence and the odd component response sequence. The response difference processing includes: differentiating the left-leaning image boundary response with the reference image boundary response to obtain the left-leaning image boundary change; differentiating the right-leaning image boundary response with the reference image boundary response to obtain the right-leaning image boundary change; differentiating the left-leaning attitude response with the reference attitude response to obtain the left-leaning attitude change; differentiating the right-leaning attitude response with the reference attitude response to obtain the right-leaning attitude change; differentiating the left-leaning wheel speed response with the reference wheel speed response to obtain the left-leaning wheel speed change; differentiating the right-leaning wheel speed response with the reference wheel speed response to obtain the right-leaning wheel speed change; differentiating the left-leaning heading correction amount with the reference heading correction amount to obtain the left-leaning heading correction change; and differentiating the right-leaning heading correction amount with the reference heading correction amount to obtain the right-leaning heading correction change. The even-component response sequence consists of the same-direction components of the left-leaning relative response sequence and the right-leaning relative response sequence, while the odd-component response sequence consists of the opposite-direction components of the left-leaning relative response sequence and the right-leaning relative response sequence. The left yaw relative response sequence includes left yaw image boundary relative response, left yaw attitude relative response, left yaw wheel speed relative response, and left yaw heading correction relative response; The right yaw trial response sequence includes the right yaw image boundary relative response, the right yaw attitude relative response, the right yaw wheel speed relative response, and the right yaw heading correction relative response.

[0028] The biased kernel unit is used to arrange the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components in the odd component response sequence according to the same sampling number to construct the odd component response matrix of the current axial section; each row of the odd component response matrix corresponds to a sampling number, and each column corresponds to a response type; the response types include image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components; Robust principal component decomposition is performed on the odd component response matrix, which decomposes the odd component response matrix into a low-rank structured response matrix and a sparse perturbation matrix; The low-rank structure response matrix L and the sparse perturbation matrix S are obtained by solving for the following constraints: ; Where Q represents the odd component response matrix, L represents the low-rank structure response matrix, and S represents the sparse perturbation matrix. Represents the nuclear norm. Let λ denote the norm, λ be the decomposition weight parameter, and stQ=L+S represent the solution under the condition that the odd component response matrix is ​​decomposed into a low-rank structure response matrix and a sparse perturbation matrix. After normalizing each column of the low-rank structural response matrix L, the response intensity is calculated by weighted sum of squares and square root for each response type. The continuous trial sampling window with the largest response intensity is selected as the structural response window. The positions of each sampling time in the blade coordinate system within the current structural response window are weighted and summed to obtain the structural response position. Within the structural response window, the sum of squares of the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components are calculated respectively. The ratio of the sum of squares of each response type to the sum of squares of all response types is taken as the response contribution of the corresponding response type. Based on the sign direction of the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components within the structural response window, the structural response direction is determined. When there are multiple response types with inconsistent sign directions, the sign directions of each response type are weighted and synthesized using the response contribution as the weight to obtain the structural response direction of the current axial section.

[0029] In this embodiment, the trial alignment unit uses the reference trial response as a reference and employs a dynamic time warping algorithm to perform sequence alignment on the left-skewed trial response and the right-skewed trial response, respectively, so that the sampled data in the reference trial response, the left-skewed trial response, and the right-skewed trial response correspond to the same axial sampling position within the current axial section, thus obtaining the reference trial response sequence, the left-skewed trial response sequence, and the right-skewed trial response sequence. The parity decomposition unit uses the reference trial response sequence as a reference, performs response difference processing on the left-skewed trial response and the right-skewed trial response with the reference trial response, respectively, to obtain the left-skewed relative response sequence and the right-skewed relative response sequence. Then, it performs bidirectional excitation parity decomposition on the left-skewed relative response sequence and the right-skewed relative response sequence to obtain the even component response sequence and the odd component response sequence. The response difference processing specifically includes the difference processing between the left-leaning image boundary response and the reference image boundary response, the difference processing between the right-leaning image boundary response and the reference image boundary response, the difference processing between the left-leaning attitude response and the reference attitude response, the difference processing between the right-leaning attitude response and the reference attitude response, the difference processing between the left-leaning wheel speed response and the reference wheel speed response, the difference processing between the right-leaning wheel speed response and the reference wheel speed response, the difference processing between the left-leaning heading correction and the reference heading correction, and the difference processing between the right-leaning heading correction and the reference heading correction. The resulting left-leaning relative response sequence includes the left-leaning image boundary relative response, the left-leaning attitude relative response, the left-leaning wheel speed relative response, and the left-leaning heading correction relative response. The right-leaning trial response sequence includes the right-leaning image boundary relative response, the right-leaning attitude relative response, the right-leaning wheel speed relative response, and the right-leaning heading correction relative response. Even-component response sequences are used to characterize the same-direction components in the left-leaning and right-leaning relative response sequences, and odd-component response sequences are used to characterize the opposite-direction components in the left-leaning and right-leaning relative response sequences.The biased kernel unit arranges the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components in the odd component response sequence according to the same sampling number, constructing the odd component response matrix of the current axial section. Each row of the odd component response matrix corresponds to a sampling number, and each column corresponds to a response type. Subsequently, robust principal component decomposition is performed on the odd component response matrix, decomposing it into a low-rank structural response matrix and a sparse perturbation matrix. Then, each column of the low-rank structural response matrix is ​​normalized, and the response intensity is calculated by weighted sum of squares and square root for each response type. The continuous trial sampling window with the largest response intensity is selected as the structural response window. The system calculates the weighted sum of the positions of each sampling moment within the current structural response window in the blade coordinate system to obtain the structural response position. Simultaneously, it calculates the sum of squares of the odd components of the image boundary, attitude, wheel speed, and heading correction within the structural response window. The ratio of the sum of squares of each response type to the sum of squares of all response types is used as the response contribution of the corresponding response type. The structural response direction is determined based on the sign direction of each response type within the structural response window. When the sign directions of multiple response types are inconsistent, the response contribution is used as a weight to weight and synthesize the sign directions of each response type to obtain the structural response direction of the current axial section. Through the above implementation, the bias kernel generation module can unify the reference test response, left bias test response, and right bias test response to the same axial sampling position for analysis, distinguish between the reverse response caused by the structural boundary and the same-direction response caused by environmental disturbances, and extract the structural response position and structural response direction from the image boundary response, attitude response, wheel speed response, and heading correction. This provides structural basis for the shell reconstruction module to generate a bias vector field and form a reverse-proof corrected bias passage shell field. Compared with existing navigation control methods that directly rely on the original SDF field value, the theoretical structural position of the blade, or a fixed path, this implementation can transform the active trial feedback of the inspection trolley within the current axial section of the wind turbine blade cavity into a structural response result that can be used for field correction. This makes the subsequent candidate reconstruction region limitation, field correction grid domain screening, bias vector field generation, and inspection trolley constraint correction more closely match the actual driving state of the wind turbine blade cavity. It improves abnormal situations such as target point deviation, wall-hugging driving, steering lag, speed control mismatch, and the vehicle outline approaching impassable areas. This enables the inspection trolley to carry out continuous inspections along the traversable area in the biased traversable shell field of the reverse proof correction.

[0030] Example 4 Please refer to Figure 2 Specifically: the shell reconstruction module includes a domain delimitation unit, a vector generation unit, and a shell generation unit; The domain delimiting unit is used to read the three-dimensional mesh of the blade cavity and the solid mesh of the blade inner wall in the three-dimensional structural model, and select the current axial section unit and its adjacent axial mesh with the current axial section number as the center to form the axial candidate range; select the current circumferential region and the corresponding adjacent circumferential region with the current circumferential region number as the center and according to the structural response direction to form the circumferential candidate range; within the axial candidate range and the circumferential candidate range, the mesh of the blade inner wall solid is removed according to the original SDF field value, and the mesh containing the corresponding and continuously connected structural response position is retained as the candidate reconstruction region; Within the candidate reconstruction region, the structural response location is used as the starting grid point. Connectivity filtering is performed along the adjacent direction of the starting grid point, and the grids connected to the structural response location are retained to obtain the field correction grid domain. The field correction grid domain covers the axial section element and the corresponding circumferential region where the structural response location is located, and includes the grids adjacent to the current axial section element. The structural response location is mapped to the grid in the field-corrected grid domain to obtain the source element group. The original SDF gradient direction and structural response direction in the source element group are read. The structural response direction is projected onto the cross-sectional plane of the current axial section to obtain the cross-sectional response direction.

[0031] The vector generation unit is used to compare the cross-sectional response direction with the structural reference position, using the theoretical structural position of the blade as the structural reference position. When the cross-sectional response direction points to the structural reference position, the opposite direction of the cross-sectional response direction is taken as the bias direction of the passing shell. When the cross-sectional response direction deviates from the structural reference position, the cross-sectional response direction is taken as the bias direction of the passing shell. The odd-component response sequence and the even-component response sequence are subjected to L2 norm squared calculation to obtain the odd-component response energy and the even-component response energy respectively. Then, the odd-component response energy is multiplied by the matrix norm of the low-rank structural response matrix to obtain the structural response quantity. Then, the even-component response energy is multiplied by the matrix norm of the sparse perturbation component to obtain the proof-of-contrast response quantity. The bias vector field amplitude is obtained by calculating the proportion of the structural response quantity in the sum of the structural response quantity and the proof-of-contrast response quantity. When the sum of the structural response quantity and the proof-of-contrast response quantity is zero, the bias vector field amplitude is zero. The bias vector field is obtained by combining the source element group, the bias direction of the passing shell, and the magnitude of the bias vector field.

[0032] The shell generation unit is used to perform weighted Poisson field reconstruction within the field correction grid domain using the bias vector field as the gradient correction input, and generate a scalar correction field. The weighted Poisson field reconstruction includes: defining the reconstruction direction by the direction of the bias vector; not changing the original SDF gradient direction at the grid with the same sign direction; using the field correction grid domain boundary as the connection boundary; connecting with the original SDF field value at the connection boundary grid; and not participating in the scalar correction field solution for even component response energy and sparse perturbation components. Using the field correction grid domain as the writing range, the scalar correction field is written into the original SDF field value of the corresponding grid in the current field correction grid domain to form the bias correction SDF field value. After generating the bias correction SDF field value, Eikonal re-initialization processing is performed on the bias correction SDF field value. The SDF field value obtained by re-initialization is used as the shell field value. Based on the shell field values ​​of adjacent grid nodes, the field value difference of the current grid in the blade coordinate system is calculated. The field value difference is combined to form the spatial gradient vector of the current grid node to obtain the shell gradient direction of the current grid. Read the corresponding shell field value in each grid cell, and combine it with the outer contour occupancy size of the inspection vehicle to determine the shell-based passage area of ​​the current grid cell: If the shell field value corresponding to the mesh cell can accommodate the size occupied by the outer contour of the inspection vehicle, then the current mesh cell is written into the shelled passable area. If the shell field value corresponding to the grid cell cannot accommodate the size occupied by the outer contour of the inspection vehicle, then the current grid cell is written into the shelled inaccessible region. Based on the difference in shell field values ​​between adjacent grid nodes, the grids in the shelled passable area are divided into velocity zones. When the difference in shell field values ​​between adjacent grid nodes reaches the set velocity zone limit, the corresponding grid node is assigned to the velocity control zone. The shell gradient direction at the current position is synthesized with the structural response direction to generate an advance yaw vector; By combining the shell field values, shell gradient direction, shelled passable region, local velocity control region, advance yaw vector, and field reconstruction record, a reverse proof corrected bias passable shell field is constructed.

[0033] In this embodiment, the domain delimiting unit reads the three-dimensional mesh of the blade cavity and the solid mesh of the blade inner wall in the three-dimensional structural model. Taking the current axial section number as the center, it selects the current axial section element and its adjacent axial mesh to form an axial candidate range. Taking the current circumferential region number as the center, it selects the current circumferential region and its corresponding adjacent circumferential region to form a circumferential candidate range according to the structural response direction. Then, based on the original SDF field value, it removes the mesh of the blade inner wall solid and retains the mesh that corresponds to the structural response position and is continuously connected as the candidate reconstruction region. It then performs connectivity filtering with the structural response position as the starting mesh point to obtain the field correction mesh domain. The structural response position is then mapped to the mesh in the field correction mesh domain to obtain the source point element group. Finally, the structural response direction is projected onto the cross-sectional plane of the current axial section to obtain the cross-sectional response direction. The vector generation unit uses the theoretical structural position of the blade as the structural reference position. It compares the direction of the cross-sectional response with the structural reference position. When the direction of the cross-sectional response points to the structural reference position, the opposite direction of the cross-sectional response is taken as the bias direction of the passing shell. When the direction of the cross-sectional response deviates from the structural reference position, the cross-sectional response is taken as the bias direction of the passing shell. The bias vector field amplitude is obtained by combining the odd component response energy, even component response energy, matrix norm of the low-rank structural response matrix, and matrix norm of the sparse perturbation component. Finally, the bias vector field is obtained by combining the source point unit group, the passing shell bias direction, and the bias vector field amplitude. The shell generation unit uses the bias vector field as the gradient correction input in the field correction grid domain, performs weighted Poisson field reconstruction to generate a scalar correction field, and writes the scalar correction field into the original SDF field value to form the bias correction SDF field value. Then, it undergoes Eikonal re-initialization processing to obtain the shell field value and shell gradient direction. At the same time, it combines the outer contour occupancy size of the inspection vehicle to determine the shelled passable area, divides the shelled passable area and the shelled impassable area, and performs velocity partitioning based on the difference of shell field value between adjacent grid nodes to generate an advance yaw vector. Finally, it constructs the inverse proof correction biased passable shell field. Through the above implementation methods, the system can transform the structural response position and direction obtained by the bias kernel generation module into a passable shell field that can be used for navigation control. This solves the problem that it is difficult to reflect the actual structural changes in the internal cavity when relying solely on the original SDF field value or the theoretical structural position of the blade. This allows the constraint control module to closely follow the actual passable state of the wind turbine blade's internal cavity when determining the current target point, the set of sampling points on the vehicle's outer contour, the shell control reference quantity, and the inspection constraint terms. This achieves the purpose of correcting the inspection trolley's driving direction, passable area judgment, speed zone control, and advance yaw control. It also improves abnormal phenomena such as wall-hugging driving, target point deviation, steering lag, speed control mismatch, and the vehicle's outer contour approaching impassable areas that occur near the inner wall boundary, web, reinforcing ribs, adhesive areas, or connectors.

[0034] Example 5 Please refer to Figure 2 Specifically: the constraint control module includes a shell reference generation unit and a constraint control unit; The shell reference generation unit is used to map the current position coordinates of the inspection vehicle to the three-dimensional mesh of the reverse proof correction bias passage shell field, obtain the vehicle outline sampling point set and determine the mesh cell corresponding to the current position of the inspection vehicle, and set the current mesh cell as the current target point; The shelling determination for the current target point is performed as follows: Determine whether the current target point is located within the shelled passable area; If the current target point is located within the shelled passable area, then the current target point will be used as the shelled target point; If the current target point is outside the shelled passable region, then search for the grid position with the smallest Euclidean distance to the current target point and belonging to the shelled passable region in the proof by contradiction modified bias passable shell field as the shelled target point; The current pose of the inspection vehicle, the shell-shaped target point, the shell field value corresponding to the current position, the shell gradient direction, the velocity control zone marker, and the advance yaw vector are combined into a shell control reference quantity.

[0035] The constraint control unit is used to construct inspection constraint items, including: Read the shell field value and shelled passable region corresponding to each vehicle outline sampling point, and construct the vehicle outline shell constraint term as follows; When each candidate vehicle outline sampling point is located within the shelled passable area, the current inspection car control quantity satisfies the vehicle outline shell constraint. When any candidate vehicle outline sampling point is located outside the shelled passable area, the current inspection car control quantity does not meet the vehicle outline shell constraint. After obtaining the outer shell constraint terms of the vehicle body, the target direction from the current position to the shell target point is determined based on the current position of the inspection vehicle and the shell target point. At the same time, the shell gradient direction corresponding to the current position is read. The target direction is used as the main direction of the candidate motion direction, and the shell gradient direction is used as the correction direction of the candidate motion direction. The candidate motion direction is simultaneously directed toward the shell target point and changes along the extension direction of the shell passable area to construct the target approach constraint terms. When the inspection vehicle is currently located within the speed control zone, the speed limit value corresponding to the current speed control zone is read, and the candidate driving speed is controlled not to exceed the speed limit value to construct a speed constraint term. When the current axial section number of the inspection trolley falls within the advance yaw trigger range, an advance yaw constraint term is constructed; wherein, the advance yaw trigger range consists of the axial section numbers before the axial section pointed to by the advance yaw vector; The navigation control quantity for the current control cycle is determined using the shell control reference quantity, and the inspection constraint term is used as a constraint condition to constrain and correct the navigation control quantity.

[0036] In this embodiment, the shell reference generation unit maps the current position coordinates of the inspection vehicle to the three-dimensional mesh of the reverse proof correction bias travel shell field, obtains the vehicle outline sampling point set and determines the current target point. After shelling judgment, the shelled target point is obtained. The current pose of the inspection vehicle, the shelled target point, the shell field value corresponding to the current position, the shell gradient direction, the velocity control area marker, and the advance yaw vector are combined into the shell control reference quantity. The constraint control unit further constructs the vehicle outline shell constraint term, the target approach constraint term, the velocity constraint term, and the advance yaw constraint term, and uses the shell control reference quantity to determine the navigation control quantity of the current control cycle, and performs constraint correction on the navigation control quantity. Through the above implementation method, the system enables the inspection vehicle to identify the current axial section and circumferential region in the inner cavity of the wind turbine blade, extract the structural response position and structural response direction based on the active trial response, and write the structural response into the travel shell field, so as to achieve the purpose of navigation control according to the shelled travelable area. Compared to existing methods that rely solely on preset paths, original SDF field values, or static 3D structural models, this solution can control and correct changes in the passage space caused by inner wall boundaries, stiffeners, webs, adhesive areas, or connectors. This makes the selection of the current target point, the judgment of the vehicle's outer profile, the correction of the direction of movement, the speed control, and the advance yaw control more consistent with the actual structural state of the blade's inner cavity. It improves abnormal situations such as target point deviation, driving close to the wall, steering lag, speed mismatch, and the vehicle's outer profile approaching impassable areas. This allows the inspection trolley to maintain continuous driving and stable inspection in the narrow, curved, and cross-sectionally varied inner cavity of the wind turbine blade.

[0037] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended technical solutions and their equivalents.

Claims

1. An autonomous navigation control system for a wind turbine blade internal cavity inspection trolley, characterized in that: It includes a cross-sectional response acquisition module, an offset kernel generation module, a shell reconstruction module, and a constraint control module; The cross-section response acquisition module reads the three-dimensional structural model of the inner cavity of the wind turbine blade and determines the current axial cross-section number and circumferential region number. It extracts the theoretical structural position of the blade, the original SDF field value and the original SDF gradient direction, establishes a trial sampling window and collects trial response data. The bias kernel generation module performs sequence alignment of the left-biased and right-biased test responses based on the benchmark test response in the test response data, performs bidirectional excitation parity decomposition, constructs the odd component response matrix, and obtains the structural response location and structural response direction through robust principal component decomposition. The shell reconstruction module reads the three-dimensional mesh of the blade cavity and the solid mesh of the blade inner wall. It limits the candidate reconstruction region by the current axial section number, circumferential region number and structural response direction, and filters the field correction mesh domain. It generates the bias vector field and writes the original SDF field value to form the reverse proof correction bias pass-through shell field. The constraint control module maps the current position coordinates of the inspection vehicle to the three-dimensional mesh of the offset passage shell field, obtains the sampling point set of the vehicle body outline and determines the current target point, and constructs shell control reference quantities and inspection constraint terms to constrain and correct the inspection vehicle.

2. The autonomous navigation control system for a wind turbine blade internal cavity inspection trolley according to claim 1, characterized in that: The cross-section response acquisition module includes a cross-section recognition unit and a deflection execution unit; The cross-section recognition unit is used to read the three-dimensional structural model of the wind turbine blade cavity after the inspection trolley enters the inner cavity of the wind turbine blade, and extract the model coordinate system in the three-dimensional structural model as the blade coordinate system to obtain the current pose of the inspection trolley in the blade coordinate system. The current pose includes the position coordinates of the centroid of the inspection trolley and the vehicle attitude angle. The axial cross-section unit to which the inspection trolley belongs is determined according to the axial coordinate in the position coordinate, and the number of the axial cross-section unit is used as the current axial cross-section number. The circumferential region to which the inspection trolley belongs is determined according to the circumferential angle of the position coordinate relative to the blade axis in the current axial cross-section, and the number of the circumferential region is used as the current circumferential region number. Using the current axial section number as an index, the theoretical structural position of the blade within the corresponding axial section is read from the three-dimensional structural model, and the original SDF field value and original SDF gradient direction corresponding to the current position of the inspection trolley are read from the original SDF field value to form the cross-sectional reference data of the current axial section.

3. The autonomous navigation control system for a wind turbine blade internal cavity inspection trolley according to claim 2, characterized in that: The deflection execution unit is used to establish a test sampling window corresponding to the current axial section unit, using the current axial section number as an index, and the current navigation direction when the inspection trolley enters the test sampling window as the basic driving direction. It generates a basic driving command containing the test linear velocity, test distance, and number of samplings. While keeping the basic driving command unchanged, it superimposes zero deflection control, left deflection control, and right deflection control to the basic driving command to generate a reference test state, a left deflection test state, and a right deflection test state. It controls the inspection trolley to drive within the test sampling window according to the order of the reference test state, the left deflection test state, and the right deflection test state, and collects the corresponding test response data in each test driving state. It also collects the test response data into the corresponding test sampling window according to the sampling time. The left yaw control quantity is opposite in direction to the right yaw control quantity and is executed through the steering control of the inspection trolley or the differential speed control of the left and right wheels. The test response data includes the baseline test response, the left yaw test response, and the right yaw test response. The baseline test response, the left yaw test response, and the right yaw test response all include the image boundary response, attitude response, wheel speed response, and heading correction quantity. The test response data at each sampling time is bound to the current axial section number, the current circumferential region number, the theoretical structure position of the blade, the original SDF field value, and the original SDF gradient direction.

4. The autonomous navigation control system for a wind turbine blade internal cavity inspection trolley according to claim 3, characterized in that: The bias kernel generation module includes a trial alignment unit, a parity decomposition unit, and a bias kernel generation unit. The trial alignment unit is used to align the left-skewed trial response and the right-skewed trial response with a reference using a dynamic time warping algorithm, so as to obtain the reference trial response sequence, the left-skewed trial response sequence and the right-skewed trial response sequence. The parity decomposition unit is used to perform response difference processing on the left-biased trial response and the right-biased trial response with the benchmark trial response sequence to obtain the left-biased relative response sequence and the right-biased relative response sequence, and to perform bidirectional excitation parity decomposition on the left-biased relative response sequence and the right-biased relative response sequence to obtain the even component response sequence and the odd component response sequence. The even component response sequence is composed of the same-direction components of the left-biased relative response sequence and the right-biased relative response sequence, while the odd component response sequence is composed of the opposite-direction components of the left-biased relative response sequence and the right-biased relative response sequence. The left yaw relative response sequence includes left yaw image boundary relative response, left yaw attitude relative response, left yaw wheel speed relative response, and left yaw heading correction relative response; The right yaw trial response sequence includes the right yaw image boundary relative response, the right yaw attitude relative response, the right yaw wheel speed relative response, and the right yaw heading correction relative response.

5. The autonomous navigation control system for a wind turbine blade internal cavity inspection trolley according to claim 4, characterized in that: The biased kernel unit is used to arrange the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components in the odd component response sequence according to the same sampling number to construct the odd component response matrix of the current axial section; each row of the odd component response matrix corresponds to a sampling number, and each column corresponds to a response type; the response types include image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components; Robust principal component decomposition is performed on the odd component response matrix, which decomposes the odd component response matrix into a low-rank structured response matrix and a sparse perturbation matrix; After normalizing each column of the low-rank structural response matrix L, the response intensity is calculated by weighted sum of squares and square root for each response type. The continuous trial sampling window with the largest response intensity is selected as the structural response window. The positions of each sampling time in the blade coordinate system within the current structural response window are weighted and summed to obtain the structural response position. Within the structural response window, the sum of squares of the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components are calculated respectively. The ratio of the sum of squares of each response type to the sum of squares of all response types is taken as the response contribution of the corresponding response type. Based on the sign direction of the image boundary odd components, attitude odd components, wheel speed odd components, and heading correction odd components within the structural response window, the structural response direction is determined. When there are multiple response types with inconsistent sign directions, the sign directions of each response type are weighted and synthesized using the response contribution as the weight to obtain the structural response direction of the current axial section.

6. The autonomous navigation control system for a wind turbine blade internal cavity inspection trolley according to claim 5, characterized in that: The shell reconstruction module includes a domain delimitation unit, a vector generation unit, and a shell generation unit; The domain delimiting unit is used to read the three-dimensional mesh of the blade cavity and the solid mesh of the blade inner wall in the three-dimensional structural model, and select the current axial section unit and its adjacent axial mesh with the current axial section number as the center to form the axial candidate range; select the current circumferential region and the corresponding adjacent circumferential region with the current circumferential region number as the center and according to the structural response direction to form the circumferential candidate range; within the axial candidate range and the circumferential candidate range, the mesh of the blade inner wall solid is removed according to the original SDF field value, and the mesh containing the corresponding and continuously connected structural response position is retained as the candidate reconstruction region; Within the candidate reconstruction region, starting with the structural response location as the initial grid point, connectivity filtering is performed along the adjacent direction of the initial grid point, retaining the grids connected to the structural response location to obtain the field-corrected grid domain; The structural response location is mapped to the grid in the field-corrected grid domain to obtain the source element group. The original SDF gradient direction and structural response direction in the source element group are read. The structural response direction is projected onto the cross-sectional plane of the current axial section to obtain the cross-sectional response direction.

7. The autonomous navigation control system for a wind turbine blade internal cavity inspection trolley according to claim 6, characterized in that: The vector generation unit is used to compare the cross-sectional response direction with the structural reference position, using the theoretical structural position of the blade as the structural reference position. When the cross-sectional response direction points to the structural reference position, the opposite direction of the cross-sectional response direction is taken as the bias direction of the passing shell. When the cross-sectional response direction deviates from the structural reference position, the cross-sectional response direction is taken as the bias direction of the passing shell. The odd-component response sequence and the even-component response sequence are subjected to L2 norm squared calculation to obtain the odd-component response energy and the even-component response energy respectively. Then, the odd-component response energy is multiplied by the matrix norm of the low-rank structural response matrix to obtain the structural response quantity. Then, the even-component response energy is multiplied by the matrix norm of the sparse perturbation component to obtain the proof-of-contrast response quantity. The bias vector field amplitude is obtained by calculating the proportion of the structural response quantity in the sum of the structural response quantity and the proof-of-contrast response quantity. When the sum of the structural response quantity and the proof-of-contrast response quantity is zero, the bias vector field amplitude is zero. The bias vector field is obtained by combining the source element group, the bias direction of the passing shell, and the magnitude of the bias vector field.

8. The autonomous navigation control system for a wind turbine blade internal cavity inspection trolley according to claim 6, characterized in that: The shell generation unit is used to perform weighted Poisson field reconstruction within the field correction grid domain using the bias vector field as the gradient correction input, and generate a scalar correction field. Using the field correction grid domain as the writing range, the scalar correction field is written into the original SDF field value of the corresponding grid in the current field correction grid domain to form the bias correction SDF field value. After generating the bias correction SDF field value, Eikonal re-initialization processing is performed on the bias correction SDF field value. The SDF field value obtained by re-initialization is used as the shell field value. Based on the shell field values ​​of adjacent grid nodes, the field value difference of the current grid in the blade coordinate system is calculated. The field value difference is combined to form the spatial gradient vector of the current grid node to obtain the shell gradient direction of the current grid. Read the corresponding shell field value in each grid cell, and combine it with the outer contour occupancy size of the inspection vehicle to determine the shell-based passage area of ​​the current grid cell: If the shell field value corresponding to the mesh cell can accommodate the size occupied by the outer contour of the inspection vehicle, then the current mesh cell is written into the shelled passable area. If the shell field value corresponding to the grid cell cannot accommodate the size occupied by the outer contour of the inspection vehicle, then the current grid cell is written into the shelled inaccessible region. Based on the difference in shell field values ​​between adjacent grid nodes, the grids in the shelled passable area are divided into velocity zones. When the difference in shell field values ​​between adjacent grid nodes reaches the set velocity zone limit, the corresponding grid node is assigned to the velocity control zone. The shell gradient direction at the current position is synthesized with the structural response direction to generate an advance yaw vector; By combining the shell field values, shell gradient direction, shelled passable region, local velocity control region, advance yaw vector, and field reconstruction record, a reverse proof corrected bias passable shell field is constructed.

9. The autonomous navigation control system for a wind turbine blade internal cavity inspection trolley according to claim 8, characterized in that: The constraint control module includes a shell reference generation unit and a constraint control unit; The shell reference generation unit is used to map the current position coordinates of the inspection vehicle to the three-dimensional mesh of the reverse proof correction bias passage shell field, obtain the vehicle outline sampling point set and determine the mesh cell corresponding to the current position of the inspection vehicle, and set the current mesh cell as the current target point; The shelling determination for the current target point is performed as follows: Determine whether the current target point is located within the shelled passable area; If the current target point is located within the shelled passable area, then the current target point will be used as the shelled target point; If the current target point is outside the shelled passable region, then search for the grid position with the smallest Euclidean distance to the current target point and belonging to the shelled passable region in the proof by contradiction modified bias passable shell field as the shelled target point; The current pose of the inspection vehicle, the shell-shaped target point, the shell field value corresponding to the current position, the shell gradient direction, the velocity control zone marker, and the advance yaw vector are combined into a shell control reference quantity.

10. An autonomous navigation control system for a wind turbine blade internal cavity inspection trolley according to claim 9, characterized in that: The constraint control unit is used to construct inspection constraint items, including: Read the shell field value and shelled passable region corresponding to each vehicle outline sampling point, and construct the vehicle outline shell constraint term as follows; When each candidate vehicle outline sampling point is located within the shelled passable area, the current inspection car control quantity satisfies the vehicle outline shell constraint. When any candidate vehicle outline sampling point is located outside the shelled passable area, the current inspection car control quantity does not meet the vehicle outline shell constraint. After obtaining the outer shell constraint terms of the vehicle body, the target direction from the current position to the shell target point is determined based on the current position of the inspection vehicle and the shell target point. At the same time, the shell gradient direction corresponding to the current position is read. The target direction is used as the main direction of the candidate motion direction, and the shell gradient direction is used as the correction direction of the candidate motion direction. The candidate motion direction is simultaneously directed toward the shell target point and changes along the extension direction of the shell passable area to construct the target approach constraint terms. When the inspection vehicle is currently located within the speed control zone, read the speed limit value corresponding to the current speed control zone, and control the candidate driving speed to not exceed the speed limit value, thus constructing a speed constraint term; When the current axial section number of the inspection trolley falls within the advance yaw trigger range, an advance yaw constraint term is constructed; wherein, the advance yaw trigger range consists of the axial section numbers before the axial section pointed to by the advance yaw vector; The navigation control quantity for the current control cycle is determined using the shell control reference quantity, and the inspection constraint term is used as a constraint condition to constrain and correct the navigation control quantity.

Images