An online error detection and correction system and method for a mechanical assembly process

Abstract

The application discloses an error online detection and correction system and method for a mechanical accessory assembly process, relates to the field of automatic assembly of heavy load hydraulic cylinders, and aims at the problem of visual profile distortion caused by eccentric oil film in high-reflective plunger rod blind insertion. The system calculates three-dimensional coordinate drift distance by combining a boundary splitting module with a depth constraint and a mapping matrix, splits out a real rod profile exposure point set and a meniscus edge covering point set, a direction inversion module reduces the dimension of the point set, constructs a first contact inversion direction vector according to the spatial opposite direction of the two centers, a directional correction module translates along the opposite direction of the vector and slightly tilts the posture based on the extracted net exposure amount and topological inclusion relationship, and a blind insertion execution module uses absolute depth mapping to guide space oblique interpolation under the condition of frozen visual feedback. The application excludes optical outward expansion illusion from the bottom algorithm, converts visual features into anti-collision vectors, and realizes high-precision soft assembly in a deep limited blind area.

Description

Technical Field

[0001] This invention relates to the field of automated assembly of heavy-duty hydraulic cylinders, and specifically to an online error detection and correction system and method for the assembly process of mechanical parts. Background Technology

[0002] In the field of automated assembly of heavy-duty hydraulic cylinders, the plunger rod enters the cylinder barrel by passing sequentially through the guide sleeve inlet and the rod sealing lip with preload, which is a typical blind insertion assembly in a confined space. To ensure service life and sealing performance, the outer cylindrical surface of the plunger rod is treated with chrome plating and polishing to achieve a highly reflective mirror finish. Simultaneously, a certain thickness of lubricating oil film must be maintained throughout the assembly process. Existing automated production lines generally rely on external machine vision systems to guide the plunger rod alignment. When an external light source illuminates the highly reflective cylindrical surface with a thin fluid film, the outer contour captured by the optical lens contains strong refracted and scattered light spots, resulting in an inherent deviation between the image edge and the actual spatial boundary of the metal entity.

[0003] Conventional vision servo alignment techniques typically extract the geometric contour center of the aforementioned imaging edge as the control reference and align it with the target hole center. However, in actual operation, due to the combined effects of gravity, initial alignment deviation, and sealing lip compression, the fluid oil film on the plunger rod surface cannot maintain a strictly axially symmetrical distribution. It easily converges and adheres to the side with the larger gap, forming a locally bulging eccentric oil film coating. Conventional algorithms directly identify the bright eccentric oil film coating as the metal rod boundary, resulting in a significant shift in the calculated apparent axis towards the oil film convergence side. In the subsequent penetration of the sealing lip without external visual monitoring, the side with the eccentric oil film provides fluid lubrication buffer, while the exposed metal area on the opposite side becomes a truly dangerous area prone to rigid impact. Existing vision solutions, based on distorted alignment results, forcibly drive the actuator downwards, inevitably causing the metal to directly impact the sealing lip or the inner wall of the guide sleeve, leading to assembly jamming and the scrapping of high-value seals. How to eliminate the optical interference of eccentric oil film from the mixed visual appearance and accurately deduce the safe collision avoidance approach direction has become a process bottleneck that urgently needs to be overcome in this field. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes an online error detection and correction system for the assembly process of mechanical parts. This system solves the problem of visual distortion of the appearance caused by the eccentric oil film meniscus locking edge during the blind insertion of high-reflectivity plunger rod segments, which leads to the failure of traditional alignment schemes and hard interference in assembly.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The boundary splitting module is used to combine depth constraints and mapping matrices to map multi-directional illumination optical data to three-dimensional space, calculate coordinate drift distance, split the real rod contour exposed area point set and the crescent lock edge covered area point set according to rigid tolerance threshold, and extract the guide sleeve entrance contour point set. The orientation inversion module is used to project the point set of the exposed area of ​​the real rod profile and the point set of the crescent-locked edge-covered area onto the two-dimensional observation plane, and construct the first contact inversion orientation vector based on the spatial opposing lateral directions of the two-dimensional geometric centers of the two point sets. The orientation correction module is used to calculate the minimum two-dimensional distance between the actual rod profile exposed area point set and the crescent edge covered area point set and the guide sleeve inlet profile point set in the two-dimensional observation plane, respectively. The absolute difference between the two minimum two-dimensional distances is taken as the net exposure amount. When the net exposure amount is greater than the natural uniform oil film thickness threshold, or when the two-dimensional projection of the actual rod profile exposed area point set is not completely contained within the closed polygonal area formed by the guide sleeve inlet profile point set, the module translates in the opposite direction of the first contact inversion direction vector and tilts slightly around the attitude axis perpendicular to the first contact inversion direction vector, and outputs the reconstructed actual mechanical entry boundary and the orientation-corrected plunger rod assembly attitude. The blind insertion execution module is used to interpolate and advance along the feed direction axis of the tool rectangular coordinate system bound to the plunger rod based on the reconstructed real mechanical entry boundary and the oriented correction of the plunger rod assembly posture. It combines the absolute encoder position data with the three-dimensional prior geometric dimension parameters to map the real-time physical insertion depth and complete the segmented blind insertion.

[0006] Furthermore, the boundary splitting module is equipped with at least one industrial camera and multiple sets of narrowband illumination units with different incident directions that are discretely distributed around the plunger rod in the circumferential direction; Multiple narrowband illumination units with different incident directions maintain an adjacent azimuth angle difference distribution of more than 60 degrees; The mapping matrix is ​​a pre-calibrated mapping matrix that is defined before implementation; Depth constraints include the known nominal radius constraint of the piston rod and the feed axis depth feedback data of the actuator; The boundary splitting module controls multiple narrowband illumination units with different incident directions to be triggered sequentially. It uses an industrial camera to acquire original images, extracts discrete edge abrupt change pixels from the original images, and combines depth constraints and calibration mapping matrices to transform the discrete edge abrupt change pixels into three-dimensional space, generating candidate boundary point sets for the plunger rod and candidate inlet point sets for the guide sleeve, respectively.

[0007] Furthermore, the boundary splitting module uses the same feed direction axis coordinate section or spatial nearest neighbor search as the matching benchmark to calculate the three-dimensional spatial physical Euclidean distance range of the same local edge under different lighting direction image mappings, and uses the three-dimensional spatial physical Euclidean distance range as the coordinate drift distance. The edge points whose coordinate drift distance is less than the rigid tolerance threshold and which extend continuously along the feed direction axis are converged to generate a set of points in the actual rod profile exposed area; The edge points that are continuously attached circumferentially to the outer side of the point set whose coordinate drift distance is greater than or equal to the rigid tolerance threshold and are located in the exposed area of ​​the real rod profile are gathered to generate the crescent-shaped edge-wrapping area point set.

[0008] Furthermore, the direction inversion module extracts the two-dimensional orthogonal coordinates of each point in the exposed area point set and the crescent edge covered area point set of the real rod profile, corresponding to the lateral correction direction axis and the side correction direction axis, and performs numerical accumulation. The accumulated result of the two-dimensional orthogonal coordinates is divided by the total number of points to calculate the two-dimensional geometric center of the exposed area and the two-dimensional geometric center of the crescent edge covered area of ​​the real rod profile. When the spatial distance between the two-dimensional geometric center of the exposed area of ​​the real rod profile and the two-dimensional geometric center of the crescent-locked edge covering area is less than the preset system zero drift constant, it is determined that there is no significant asymmetric eccentricity in the current transverse section, the first contact inversion direction vector is directly assigned the value of zero vector and the current direction inversion cycle is skipped.

[0009] Furthermore, when constructing the first contact inversion direction vector, the direction inversion module uses the spatial direction pointing from the two-dimensional geometric center of the crescent-shaped edge-covered area to the two-dimensional geometric center of the real rod contour exposed area as the biased opposing direction. The spatial vector of the biased opposing direction is normalized by dividing it by the spatial distance between the two-dimensional geometric center of the real rod contour exposed area and the two-dimensional geometric center of the crescent-shaped edge-covered area to obtain the first contact inversion direction vector.

[0010] Furthermore, when the orientation correction module translates in the opposite direction to the first contact inversion direction vector and tilts slightly around the attitude axis perpendicular to the first contact inversion direction vector, it outputs translation servo commands and tilt servo commands based on the preset servo step size. The preset servo step size is a fixed displacement increment, or a dynamically adjustable step size based on the ratio set by the difference between the net exposure amount and the uniform assembly tolerance range.

[0011] Furthermore, after each translation and micro-tilt combination adjustment, the orientation correction module triggers a closed-loop retest process to reacquire images and reconstruct boundary and distance data until the recalculated net exposure amount is less than or equal to the natural uniform oil film thickness threshold, and the two-dimensional orthogonal projection of the real rod profile exposed area point set is completely contained within the closed polygonal area formed by the guide sleeve inlet profile point set. Then, the reconstructed real mechanical entry boundary and the orientation-corrected plunger rod assembly posture are output.

[0012] Furthermore, the blind insertion execution module sequentially controls the plunger rod to complete the actual boundary entry of the guide sleeve inlet, the directional penetration entry of the rod sealing lip, and the segmented blind insertion entry into the cylinder. Furthermore, the blind insertion execution module forcibly freezes the visual closed-loop feedback throughout the entire blind insertion process, and determines the completion conditions of each entry stage based on the real-time physical insertion depth.

[0013] Furthermore, during the advancement process, the blind insertion execution module uses the absolute spatial position when the orientation correction module completes closed-loop correction as the depth zero point reference, and drives the plunger rod to perform spatial oblique interpolation advancement along the feed direction axis of the tool rectangular coordinate system tilted after orientation correction.

[0014] Furthermore, an online error detection and correction method for the assembly process of mechanical parts is proposed, which is applied to the above-mentioned online error detection and correction system for the assembly process of mechanical parts, including: By combining depth constraints and mapping matrices, multi-directional illumination optical data is mapped to three-dimensional space, coordinate drift distance is calculated, and the point set of the exposed area of ​​the real rod profile and the point set of the meniscus edge covering area are split according to the rigid tolerance threshold. The point set of the guide sleeve entrance profile is also extracted. Project the point set of the exposed area of ​​the real rod profile and the point set of the crescent-locked edge-enclosed area onto the two-dimensional observation plane, and construct the first contact inversion direction vector based on the spatial opposing lateral directions of the two-dimensional geometric centers of the two point sets. In the two-dimensional observation plane, the minimum two-dimensional distance between the point set of the exposed area of ​​the real rod profile, the point set of the crescent-shaped edge covering area, and the point set of the guide sleeve inlet profile is calculated respectively. The absolute difference between the two minimum two-dimensional distances is taken as the net exposure amount. When the net exposure amount is greater than the threshold of the natural uniform oil film thickness, or when the two-dimensional projection of the point set of the exposed area of ​​the real rod profile is not completely contained within the closed polygonal area formed by the point set of the guide sleeve inlet profile, the rod is translated in the opposite direction of the first contact inversion direction vector and slightly tilted around the attitude axis perpendicular to the first contact inversion direction vector. The reconstructed real mechanical entry boundary and the oriented correction of the plunger rod assembly attitude are output. Based on the reconstructed real mechanical entry boundary and the oriented correction of the plunger rod assembly posture, the tool is interpolated and advanced along the feed direction axis of the tool rectangular coordinate system bound to the plunger rod. The real-time physical insertion depth is mapped by combining the absolute encoder position data and the three-dimensional prior geometric dimension parameters to complete the segmented blind insertion.

[0015] Compared with existing technologies, it has the following advantages: This solution proposes an online error detection and correction system and method for the assembly process of mechanical parts. Addressing the optical parallax bias problem caused by residual lubricating oil layers on the surface of highly reflective cylindrical parts, it constructs a boundary splitting and direction inversion linkage calculation mechanism. Unlike traditional visual servo systems that directly treat the globally bright outer edge as a rigid body boundary, this system introduces multi-directional optical constraints and spatial drift distance calculations to accurately segment the mixed and overlapping apparent contours into a set of points representing the real rod contour exposed area of ​​the mechanical entity and a set of points representing the illusory meniscus-enclosed area of ​​the fluid. This disassembly action eliminates the optical expansion artifacts caused by local convergence of eccentric oil layers from the underlying algorithm, completely severing the chain of position misjudgments caused by visual artifacts in conventional detection, and establishing a highly reliable spatial position benchmark for the centering stage, which is a high-risk pre-blind insertion phase.

[0016] In the correction and blind insertion execution stages, this system cleverly utilizes the spatial coupling relationship between the uneven distribution of the lubrication layer and the mechanical assembly gaps. The system independently extracts the two-dimensional geometric centers of two types of point sets, and constructs the first contact inversion direction vector through their opposing lateral pointing relationship in space. This successfully transforms the previously extracted optical boundary classification data into an obstacle avoidance and anti-collision vector that directly reflects the actual assembly interference risk. Subsequently, based on the calculated net exposure amount, the system drives the actuator to perform translation and attitude micro-tilt adjustment along the opposite side of the inversion direction. When the plunger rod penetrates the obstruction area of ​​the sealing component, the external visual feedback is forcibly frozen, and the system fully enters the oblique interpolation propulsion state based on absolute depth data mapping. The full-process collaborative control strategy breaks through the limitation of conventional single visual guidance, which is prone to falling into the apparent centering trap. It realizes a logical closed loop from surface visual desiccation to smooth obstacle avoidance in deep blind areas, greatly reducing the risk of wedge jamming and cutting damage when heavy-duty rods penetrate the sealing system. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the system framework of the present invention.

[0018] Figure 2 This is a schematic diagram of the method flow of 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] Please see Figure 1This application provides an online error detection and correction system for the assembly process of mechanical parts, including a boundary splitting module, a direction inversion module, an orientation correction module, and a blind insertion execution module; In the specific implementation process, the boundary splitting module is executed first to obtain and split the optical contour data of the assembly front end. In the segmented blind insertion assembly scenario of heavy-duty hydraulic cylinders, the surface of the highly reflective chrome-plated plunger rod is prone to strong specular reflection. When the plunger rod approaches the rod sealing lip with preload, under the action of slight eccentricity and gravity, the residual lubricating oil film will form a continuously attached eccentric oil film meniscus in the local circumference of the rod body. Conventional optical alignment systems often mistake the bright and bulging oil film coverage boundary for the real mechanical geometric entity boundary, resulting in a serious deviation between the measured apparent center line and the actual physical access axis, ultimately causing hard interference or seal damage during the lip penetration stage.

[0021] Specifically, the boundary splitting module in this embodiment breaks away from the conventional image processing logic of directly treating all highlighted boundaries as outer contours. By utilizing the difference in optical response between the real metal entity and the oil film coating layer under multi-directional incident illumination, the mixed apparent contour is forcibly split into an unclosed region representing the real geometry and a closed region representing the oil film coating. Simultaneously, the guide sleeve inlet reference is extracted, thereby providing geometric input support after eliminating fluid optical interference for the subsequent inversion of the real mechanical entry axis.

[0022] To achieve the separation and extraction of the aforementioned boundaries, at least one industrial camera and multiple sets of narrowband illumination units with different incident directions are configured. The total number of illumination directions is set to J, the index number of the current illumination direction is defined as i, and the pixel coordinates of the two-dimensional image are defined as u and v. Multiple sets of narrowband illumination units are discretely distributed circumferentially around the axis of the plunger rod at a preset angle to ensure that the azimuth difference between adjacent beams can excite the fluid coating layer to produce an optical refraction drift difference greater than the system rigidity tolerance threshold. At the position of the plunger rod leading section near the guide sleeve inlet but not yet entering the sealing lip, the plunger rod is controlled to maintain a stationary hovering posture or, combined with high-speed displacement compensation of the actuator, the light sources in different directions are triggered sequentially to acquire the original image under the i-th illumination direction. For each frame of the original captured image Perform preprocessing operations, capture the effective assembly field of view including the plunger rod leading section and the guide sleeve inlet, filter out background frame information irrelevant to assembly, perform global brightness normalization, and output the preprocessed target area image. .

[0023] Specifically, J is preferably an integer from two to four, and i takes the value from one to J. The narrowband illumination unit uses a narrowband blue light source that matches the reflective properties of the chrome-plated surface, and maintains a spatial topological distribution with adjacent azimuth angle differences greater than sixty degrees. This ensures, at the hardware physical level, the significant deterioration of the optical response of the fluid interface and rigid body interface under multi-directional illumination. A static hovering or displacement compensation mechanism is introduced during the acquisition phase to avoid misinterpreting the actual mechanical motion displacement as fluid optical drift due to the continuous downward movement of the plunger rod, ensuring tight alignment of multiple illumination sequences in spatial physical positions. By capturing the initial response state of the same physical region under different optical excitations through multi-directional illumination with spatial topological constraints, unnecessary computational redundancy in subsequent algorithms is eliminated.

[0024] The pre-calibrated calibration mapping matrix M is invoked, and the sub-pixel edge detection operator is used to traverse the target region image. (i.e., multi-directional illumination optical data) to extract discrete edge abrupt change pixels. Then, combining depth constraints (specifically including the known nominal radius constraint of the plunger rod and the depth feedback data of the actuator's Z-axis, i.e., the feed axis), the calibration mapping matrix M is used to uniformly transform the aforementioned abrupt change pixels into a three-dimensional stationary Cartesian coordinate system X, Y, Z, generating candidate boundary point sets for the plunger rod. and the set of candidate entrance points for the guide loop .

[0025] Specifically, the calibration mapping matrix M is used to transform pixels in a two-dimensional image plane to coordinates in three-dimensional physical space. Depth and nominal radius constraints are introduced to overcome the limitations of inverse projection due to the lack of depth information in monocular vision. Because of the inclusion of Z-axis depth feedback data from the actuator, the calibration mapping matrix M contains scale interpolation coefficients that are dynamically updated with the Z-axis depth, or the system hardware directly uses a telecentric optical lens, thereby ensuring absolute physical consistency in the resolution of transverse two-dimensional orthogonal coordinates at different assembly depths. This is the set of discrete three-dimensional coordinate points extracted from the i-th preprocessed image, corresponding to the apparent physical positions of the plunger rod and the attached material. This represents the set of discrete three-dimensional coordinate points corresponding to the physical position of the guide sleeve rigid body. It accurately transforms the optical features of the two-dimensional pixel domain into geometric entity occupancy features in a three-dimensional physical space coordinate system, avoiding the spatial measurement limitations caused by optical lens distortion and the two-dimensional pixel scale in high-precision assembly alignment calculations.

[0026] Receive the above candidate point set Cross-frame coordinate comparison is performed within the families of sets extracted for all J illumination directions. For any spatial edge region, the maximum 3D coordinate drift distance of the same local edge under J illumination changes is calculated using the coordinate cross-section of the same feed direction axis (i.e., Z-axis) or spatial nearest neighbor search as the matching benchmark. If the drift distance is less than the set rigid tolerance threshold and the edge extends continuously along the feed direction axis of the piston rod, then the point within the matching region is defined as a single true boundary point. The points are converged to generate a set of exposed points in the true pole profile. The total number of points is recorded as If the drift distance is greater than or equal to the set rigid tolerance threshold and is located within the point set... If the outermost part is continuously attached circumferentially, then the points within the matching area are defined as individual wrapping boundary points. The points of the crescent-shaped edge-enclosing region are converged to generate the point set. The total number of points is recorded as .

[0027] Specifically, sets and included points This represents the boundary point cloud of a purely mechanical entity with high rigidity and stability in physical space. (Set) and included points The point cloud representing the boundary of the fluid coating layer exhibits optical drift characteristics and significant shifts in its expansion morphology that vary with illumination. Euclidean distance variation is calculated using the coordinate consistency of rigid entities and the optical refractive drift of the fluid interface to perform forced classification. Mixed apparent boundaries are decomposed using quantization rules, eliminating the misleading effect of eccentric oil film artifacts on visual edge extraction.

[0028] It should be noted that the criterion-based splitting of the actual rod profile and the meniscus-covered area is based on rigorous coordinate distance variation calculations, eliminating the interference of subjective visual perception. The maximum 3D coordinate drift distance refers to the extreme difference in the 3D spatial physical Euclidean distance for the same local matching feature, such as edge points at the same Z-axis advance depth, under different lighting directions. The set rigid tolerance threshold is based on the sum of the extreme value of the micro-roughness of the plunger rod surface and the system camera calibration error. When the drift distance is less than the rigid tolerance threshold, the optical reflecting surface is determined to be an undeformable, unaffected, bottom-layer chrome-plated metal surface, i.e., a real mechanical entity. When the drift distance is greater than the rigid tolerance threshold, the optical reflecting surface is determined to be a transparent oil film layer whose highlight position is refracted and shifted with the incident light angle, i.e., an eccentric oil film meniscus.

[0029] For the obtained guide entrance candidate point set Geometric morphology screening is performed. Based on the nominal aperture size of the guide sleeve, a Hough circle transformation algorithm or a template matching algorithm with a preset standard contour is introduced to eliminate scattered noise points and retain the continuous contour points that constitute the inner edge of the rigid inlet of the guide sleeve. The retained points are defined as individual inlet boundary points. This generates the guide sleeve entrance contour point set. .

[0030] Specifically, explicit mathematical morphological matching operators effectively filter out interference from pseudo-edges such as threads, chamfers, and tooling fixtures in complex industrial environments. Parameters and set This represents the set of three-dimensional coordinates of the clean edge of the target orifice after interference removal. The absolute position of the assembled orifice is anchored in the same physical coordinate system, establishing a reference anchor point for subsequent calculation of the spatial relative offset distance between the plunger rod entity boundary and the assembly inlet.

[0031] At this point, the processed external image is split into three independent three-dimensional geometric sets: the set of exposed points representing the true rod contours of the pure entity boundary. The set of points representing the influence range of the local eccentric oil film in the meniscus edge-sealing area. and the guide sleeve inlet profile point set characterizing the assembly target hole . The above point set with point set It is called as a basic input variable in subsequent execution, and the system extracts the point set. and The characteristic coordinates are used to calculate the spatial geometric center of each part, and the first contact inversion direction vector that determines the actual assembly direction is then derived from these coordinates.

[0032] The orientation inversion module, after completing the normalization and boundary splitting of the external optical image, executes the orientation inversion module to invert the true first contact orientation based on the spatial lateral relationship between the closed and unclosed regions. It receives the set of exposed points of the true rod contour output from the preceding sequence. crescent-shaped edge-locking area point set Extracting point sets All true boundary points The two-dimensional orthogonal coordinates corresponding to the lateral correction direction axis X and the side correction direction axis Y are calculated and accumulated. The accumulated result of the two-dimensional coordinates is then divided by the total number of points. The two-dimensional geometric center of the exposed area of ​​the true rod profile was calculated. Extracting point sets All enclosing boundary points The two-dimensional orthogonal coordinates corresponding to the lateral correction direction axis X and the side correction direction axis Y are calculated and accumulated. The accumulated result of the two-dimensional coordinates is then divided by the total number of points. The two-dimensional geometric center of the crescent-locked cladding region was calculated. Based on the calculated two-dimensional geometric center and two-dimensional geometric center Construct the first contact inversion direction vector The construction formula is as follows: Specifically, by stripping the coordinate components of the feed direction axis Z, the 3D point cloud is reduced to a 2D observation plane perpendicular to the feed axis, avoiding vector yaw interference caused by feature points at different depths during planar orientation inversion. 2D geometric center The topological center of mass and two-dimensional geometric center of the arc segment representing a real metallic entity within the observation plane. This represents the topological center of mass of the eccentric oil film boundary within the observation plane. The vector in the formula... This characterizes the spatial offset of the oil film-covered area relative to the exposed metal area within the two-dimensional correction plane. The negative sign in the formula represents a reverse interpretation mechanism, that is, setting the physically opposite side of the side where the eccentric oil film edge is located as the preferred contact side. The denominator... This represents the length of the offset direction vector, used for normalization. During the actual calculation, the system uses the spatial direction pointing from the two-dimensional geometric center of the crescent-shaped edge-covered area to the two-dimensional geometric center of the actual rod profile exposed area as the offset opposing direction. The spatial vector of this offset opposing direction is then normalized by dividing it by the spatial distance between the two-dimensional geometric centers of the actual rod profile exposed area and the crescent-shaped edge-covered area, thus obtaining the first contact inversion direction vector. This constitutes the prerequisite directional input data for subsequent directional corrections in the lateral and transverse directions.

[0033] It should be noted that in constructing the first contact inversion direction vector At that time, the system has a built-in singularity protection mechanism. When the denominator When the value of (i.e., the spatial distance between the two-dimensional geometric center of the exposed area of ​​the actual rod profile and the two-dimensional geometric center of the meniscus-covered area) is less than the preset system zero drift constant, it is determined that there is no significant asymmetric eccentricity within the current transverse section, and the system directly... By assigning a value of zero to the vector and skipping the current direction inversion cycle, the program exception caused by the division by zero operation is avoided from the bottom layer.

[0034] It should be noted that the first contact inversion direction vector is constructed. The underlying computational logic abandons the conventional fitting algorithm with poor stability that performs circle center fitting on short, local contours, and also abandons the conventional visual rule that equates the side with the highlighted boundary with the side with the assembly deviation. In the actual physical scenario of a heavy-duty plunger rod passing through the sealing lip, the side with the meniscus closed area is the side where the lubricating oil film preferentially adheres and produces optical expansion; the side with the unclosed area of ​​the real rod contour, which is topologically opposite to the meniscus closed area, is closer to the side of the real metal entity directly facing the guide sleeve inlet. According to the laws of fluid encapsulation buffering and mechanical contact mechanics, the side not fully encapsulated by the oil film is more likely to form rigid mechanical contact first during assembly and advancement. Based on the reverse derivation mechanism established by the topological mass centers of the two types of boundary sets, the purely optical measurement boundary data is transformed into a mechanical direction guide reflecting the risk of interference from real physical contact with extremely high noise resistance.

[0035] The orientation correction module, after inverting the initial contact direction vector, reconstructs the real mechanical entry boundary based on the misalignment relationship between the real boundary, the covering boundary, and the entry boundary, and completes the orientation correction. It receives the set of exposed points of the real rod profile output from the preceding sequence. crescent-shaped edge-locking coverage area point set , guide sleeve entrance contour point set and the first contact inversion direction vector After removing the Z-axis coordinate components from the feed direction of the aforementioned point set, all boundary points are uniformly reduced to a two-dimensional observation plane perpendicular to the feed axis. Within this observation plane, the reduced two-dimensional point set is calculated. Edge points and point sets The minimum two-dimensional Euclidean distance between each point on the edge is extracted as the true radial gap of the metal. Calculate the point set after dimensionality reduction in the observation plane. Edge points and point sets The minimum two-dimensional Euclidean distance between each point on the edge is extracted as the optical gap of the cladding boundary. Based on the extracted true metal radial gap and the optical gap at the coating boundary The net exposure E is constructed using the following formula: Specifically, the conventional visual control logic of directly aligning the center of the apparent contour with the center of the entrance is abandoned. Instead, the interference of three-dimensional height differences is eliminated, and independent boundary point sets are placed within the same two-dimensional assembly section for geometric calculation. (Real metal radial clearance) The physical characterization defines the minimum mechanical safety clearance between the actual metal solid boundary and the guide sleeve inlet boundary within the transverse assembly section. (Optical clearance of the cladding boundary) The physical characterization measures the visual proximity between the edge of the eccentric oil film coating and the guide sleeve inlet boundary within the transverse assembly section. The net exposure amount E is introduced into an absolute value formula, physically characterizing the asymmetric spatial difference between the actual radial clearance of the metal and the optical clearance of the coating boundary caused by eccentricity within the transverse assembly section. This constitutes a geometric misalignment index specifically used to quantify the degree of spatial eccentricity between the actual physical boundary and the optically apparent pseudo-boundary.

[0036] When the net exposure E is greater than the set threshold for natural uniform oil film thickness, or the point set The two-dimensional orthogonal projection is not completely contained within the point set. Within the closed polygonal region, if the optically visible apparent boundary deviates significantly from the actual mechanical entry boundary, the system suspends control commands using the apparent contour center as the servo propulsion reference. The system then calls the first contact inversion direction vector. The plunger rod undergoes directional translation and micro-tilt adjustment in sequence. The adjustment actions include, in order of timing: along the inverted direction vector in the Cartesian coordinate system of the workstation, relative to the initial contact direction. In the opposite direction (i.e.) The path (vector direction) is based on a preset servo step size to output translation servo commands, thus adjusting the radial clearance of the actual metal. It shows an increasing trend until the actual metal radial gap... Achieving uniform assembly tolerance range; subsequently, in a two-dimensional plane within the Cartesian coordinate system of the workstation, the vector perpendicular to the first contact inversion direction is retrieved. The horizontal attitude axis outputs a micro-tilt servo command based on a preset servo step size, so that the axis of the real metal entity boundary approaches the normal perpendicular state of the guide sleeve center line.

[0037] The preset servo step size can be set to a fixed displacement increment, or the step size can be dynamically adjusted proportionally based on the difference between the net exposure amount E and the uniform assembly tolerance range, so as to achieve adaptive anti-vibration control that allows large deviations to be quickly approximated and small deviations to be accurately converged.

[0038] Specifically, the control timing of translation correction precedes that of tilt correction, prioritizing the elimination of lateral geometric interference between assembled components, and subsequently compensating for attitude tilt caused by initial minor eccentricity. First contact inversion direction vector. Pointing towards the dangerous impact side where the gap is extremely compressed, along The system performs translational movements and, physically, aims to increase the radial safety clearance as a closed-loop objective, constituting a compliant escape strategy consistent with rigid body kinematics collision avoidance principles. After each translational and slight tilting adjustment, the system triggers a closed-loop retest process to reacquire images and reconstruct boundary and distance data. When the recalculated net exposure E is less than or equal to the natural uniform oil film thickness threshold, and the point set... The two-dimensional orthogonal projection is completely contained within the point set. Within the closed polygonal region formed, the reconstructed true mechanical entry boundary and the oriented corrected plunger rod assembly posture are output. The reconstructed assembly posture eliminates the misleading effects caused by the optical distortion of the eccentric oil film meniscus, forming a non-interference assembly posture reference that can be directly called upon in the subsequent segmented blind insertion process.

[0039] It should be noted that the underlying computational logic for constructing the net exposure amount E as the control threshold does not incorporate an uninterpretable deep learning model. Since the oil film on the piston rod surface always possesses an objectively non-zero physical thickness, using a naturally uniform oil film thickness threshold instead of the theoretical zero deviation value as the exit condition eliminates the logical paradox of the control system falling into an infinite loop. The exit condition introduces a point set. The projection is completely contained within the orifice point set. The geometric determination within the system concretizes the abstract assembly entry state into a rigorous mathematical internal and external topological inclusion relationship. Addressing the optical distortion caused by unilateral locking of the eccentric oil film in specific scenarios, the net exposure amount E based on the absolute value difference provides a quantitative defense line for distinguishing between optical artifacts and real eccentricity. Those skilled in the art can directly program and implement the above control rules using conventional multi-axis robotic arms and point cloud distance calculation algorithms, ensuring the high robustness and reproducibility of the correction servo actions in industrial settings.

[0040] The blind insertion execution module, after completing the lateral and attitude orientation correction, performs segmented blind insertion assembly based on the reconstructed real mechanical entry boundary. It receives the reconstructed real mechanical entry boundary and the oriented correction plunger rod assembly posture from the orientation correction module. The system preloads the three-dimensional prior geometric parameters of the plunger rod leading section, guide sleeve inlet, and rod sealing lip. The system executes a coordinate system transformation command, switching the servo propulsion reference from the station global coordinate system to the tool Cartesian coordinate system bound to the plunger rod. The system stops intervening in the lateral and lateral degrees of freedom in the station global coordinate system and drives the plunger rod to perform spatial interpolation propulsion along the Z-axis of the tool Cartesian coordinate system tilted after orientation correction. During propulsion, using the spatial absolute position at the completion of the orientation correction module's closed-loop correction as the depth zero-point reference, the system dynamically maps the real-time absolute encoder position feedback data of the servo actuator with the three-dimensional prior geometric parameters to calculate the real-time physical insertion depth. Based on the real-time physical insertion depth, the plunger rod is sequentially controlled to complete the actual boundary entry of the guide sleeve inlet, the directional penetration entry of the rod sealing lip, and the segmented blind insertion into the cylinder. The final physical assembly result after the segmented blind insertion is output, that is, the heavy-duty hydraulic cylinder assembly entity in which the plunger rod continuously penetrates the guide sleeve, the rod sealing lip, and the cylinder in the direction of the actual mechanical inlet.

[0041] Specifically, the segmented blind insertion assembly operation abandons the vertical pressing control logic caused by forcibly locking the global coordinate system of the workstation. Since the plunger rod has already completed the attitude micro-tilt correction in the orientation correction module, directly pressing it along the global vertical axis of the workstation is very likely to cause wedging and jamming between the cylindrical metal surface and the inner wall of the guide sleeve. Spatial oblique interpolation and advancement along the feed direction axis Z of the tool rectangular coordinate system after orientation correction avoids destructive lateral shear forces and ensures that the plunger rod slides smoothly along its corrected self-center axis. The completion conditions of each entry stage abandon the timing prediction without a benchmark and are all based on the rigid mapping value of real-time physical insertion depth and three-dimensional prior geometric dimension parameters. The completion condition of the guide sleeve inlet entry stage is defined as the mapping depth showing that the leading edge of the plunger rod's leading chamfer crosses the outer edge of the guide sleeve inlet, and the actual rod contour exposed area point set. The corresponding physical entity crosses the guide sleeve inlet profile point set in the feed direction. The boundary of the two-dimensional plane. The completion condition for the directional entry stage of the rod sealing lip is defined as follows: the mapped depth shows that the chamfer of the plunger rod leader crosses the inner side of the rod sealing lip, and the working section of the plunger rod forms a circumferential physical envelopment relationship with the rod sealing lip. The completion condition for the segmented blind insertion stage inside the cylinder is defined as follows: the mapped depth shows that the working section of the plunger rod crosses the rod sealing lip area, and a physically continuous entry state is established in the guide area inside the cylinder. The above deductive logic directly applies the interference-free safety posture extracted by the machine vision recognition in the early stage to the core lip penetration stage, which has the highest risk of physical interference.

[0042] It should be noted that the third stage is defined as segmented blind insertion because the main working area of ​​the plunger rod is largely obscured by external mechanical components at this point, and the assembly field of view is outside the effective observation range of external narrow-band lighting and industrial cameras. If the dynamic visual centering adjustment based on the external residual mixed contour is reactivated during the deep blind insertion stage, the system is highly susceptible to optical misdirection due to the re-capturing of the residual eccentric oil film meniscus, which could trigger unexpected lateral servo movements, causing hard interference in the inner hole or cutting damage to the seals. Therefore, the visual closed-loop feedback is forcibly frozen throughout the blind insertion process, and a depth mapping model of three-dimensional prior geometric parameters and the displacement data of the underlying absolute encoder is used instead to eliminate the uncontrollable state of lack of position feedback in the blind zone after the loss of visual guidance. At this point, the system completes the entire closed-loop control link from the identification and extraction of the eccentric oil film coverage boundary, the quantization and inversion of multi-dimensional geometric misalignment, to the end-stage multi-axis compliant servo propulsion, truly transforming the detection and correction algorithm based on optical distortion compensation into a physical anti-collision barrier that ensures the high-precision assembly yield of heavy-duty hydraulic cylinders.

[0043] Furthermore, refer to Figure 2 As shown, an online error detection and correction method for the assembly process of mechanical parts is proposed, which is applied to the above-mentioned online error detection and correction system for the assembly process of mechanical parts, including: By combining depth constraints and mapping matrices, multi-directional illumination optical data is mapped to three-dimensional space, coordinate drift distance is calculated, and the point set of the exposed area of ​​the real rod profile and the point set of the meniscus edge covering area are split according to the rigid tolerance threshold. The point set of the guide sleeve entrance profile is also extracted. Project the point set of the exposed area of ​​the real rod profile and the point set of the crescent-locked edge-enclosed area onto the two-dimensional observation plane, and construct the first contact inversion direction vector based on the spatial opposing lateral directions of the two-dimensional geometric centers of the two point sets. In the two-dimensional observation plane, the minimum two-dimensional distance between the point set of the exposed area of ​​the real rod profile, the point set of the crescent-shaped edge covering area, and the point set of the guide sleeve inlet profile is calculated respectively. The absolute difference between the two minimum two-dimensional distances is taken as the net exposure amount. When the net exposure amount is greater than the threshold of the natural uniform oil film thickness, or when the two-dimensional projection of the point set of the exposed area of ​​the real rod profile is not completely contained within the closed polygonal area formed by the point set of the guide sleeve inlet profile, the rod is translated in the opposite direction of the first contact inversion direction vector and slightly tilted around the attitude axis perpendicular to the first contact inversion direction vector. The reconstructed real mechanical entry boundary and the oriented correction of the plunger rod assembly attitude are output. Based on the reconstructed real mechanical entry boundary and the oriented correction of the plunger rod assembly posture, the tool is interpolated and advanced along the feed direction axis of the tool rectangular coordinate system bound to the plunger rod. The real-time physical insertion depth is mapped by combining the absolute encoder position data and the three-dimensional prior geometric dimension parameters to complete the segmented blind insertion.

[0044] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An online error detection and correction system for the assembly process of mechanical parts, characterized in that, include: The boundary splitting module is used to combine depth constraints and mapping matrices to map multi-directional illumination optical data to three-dimensional space, calculate coordinate drift distance, split the real rod contour exposed area point set and the crescent lock edge covered area point set according to rigid tolerance threshold, and extract the guide sleeve entrance contour point set. The orientation inversion module is used to project the point set of the exposed area of ​​the real rod profile and the point set of the crescent-locked edge-covered area onto the two-dimensional observation plane, and construct the first contact inversion orientation vector based on the spatial opposing lateral directions of the two-dimensional geometric centers of the two point sets. The orientation correction module is used to calculate the minimum two-dimensional distance between the actual rod profile exposed area point set and the crescent edge covered area point set and the guide sleeve inlet profile point set in the two-dimensional observation plane, respectively. The absolute difference between the two minimum two-dimensional distances is taken as the net exposure amount. When the net exposure amount is greater than the natural uniform oil film thickness threshold, or when the two-dimensional projection of the actual rod profile exposed area point set is not completely contained within the closed polygonal area formed by the guide sleeve inlet profile point set, the module translates in the opposite direction of the first contact inversion direction vector and tilts slightly around the attitude axis perpendicular to the first contact inversion direction vector, and outputs the reconstructed actual mechanical entry boundary and the orientation-corrected plunger rod assembly attitude. The blind insertion execution module is used to interpolate and advance along the feed direction axis of the tool rectangular coordinate system bound to the plunger rod based on the reconstructed real mechanical entry boundary and the oriented correction of the plunger rod assembly posture. It combines the absolute encoder position data with the three-dimensional prior geometric dimension parameters to map the real-time physical insertion depth and complete the segmented blind insertion.

2. The online error detection and correction system for the assembly process of mechanical parts according to claim 1, characterized in that, include: The boundary splitting module is equipped with at least one industrial camera and multiple sets of narrowband illumination units with different incident directions that are discretely distributed around the plunger rod in the circumferential direction; Multiple narrowband illumination units with different incident directions maintain an adjacent azimuth angle difference distribution of more than 60 degrees; The mapping matrix is ​​a pre-calibrated mapping matrix that is defined before implementation; Depth constraints include the known nominal radius constraint of the piston rod and the feed axis depth feedback data of the actuator; The boundary splitting module controls multiple narrowband illumination units with different incident directions to be triggered sequentially. It uses an industrial camera to acquire original images, extracts discrete edge abrupt change pixels from the original images, and combines depth constraints and calibration mapping matrices to transform the discrete edge abrupt change pixels into three-dimensional space, generating candidate boundary point sets for the plunger rod and candidate inlet point sets for the guide sleeve, respectively.

3. The online error detection and correction system for the assembly process of mechanical parts according to claim 1, characterized in that, include: The boundary splitting module uses the same feed direction axis coordinate section or spatial nearest neighbor search as the matching benchmark, calculates the three-dimensional spatial physical Euclidean distance range of the same local edge under image mapping in different lighting directions, and uses the three-dimensional spatial physical Euclidean distance range as the coordinate drift distance. The edge points whose coordinate drift distance is less than the rigid tolerance threshold and which extend continuously along the feed direction axis are converged to generate a set of points in the actual rod profile exposed area; The edge points that are continuously attached circumferentially to the outer side of the point set whose coordinate drift distance is greater than or equal to the rigid tolerance threshold and are located in the exposed area of ​​the real rod profile are gathered to generate the crescent-shaped edge-wrapping area point set.

4. The online error detection and correction system for the assembly process of mechanical parts according to claim 1, characterized in that, include: The direction inversion module extracts the two-dimensional orthogonal coordinates of each point in the exposed area of ​​the real rod profile and the crescent edge covering area, corresponding to the lateral correction direction axis and the side correction direction axis, and performs numerical accumulation. The accumulated result of the two-dimensional orthogonal coordinates is divided by the total number of points to calculate the two-dimensional geometric center of the exposed area of ​​the real rod profile and the two-dimensional geometric center of the crescent edge covering area. When the spatial distance between the two-dimensional geometric center of the exposed area of ​​the real rod profile and the two-dimensional geometric center of the crescent-locked edge covering area is less than the preset system zero drift constant, it is determined that there is no significant asymmetric eccentricity in the current transverse section, the first contact inversion direction vector is directly assigned the value of zero vector and the current direction inversion cycle is skipped.

5. The online error detection and correction system for the assembly process of mechanical parts according to claim 4, characterized in that, include: When constructing the first contact inversion direction vector, the direction inversion module uses the spatial direction pointing from the two-dimensional geometric center of the crescent-shaped edge-covered area to the two-dimensional geometric center of the real rod contour exposed area as the biased opposing direction. The spatial vector of the biased opposing direction is normalized by dividing it by the spatial distance between the two-dimensional geometric center of the real rod contour exposed area and the two-dimensional geometric center of the crescent-shaped edge-covered area to obtain the first contact inversion direction vector.

6. The online error detection and correction system for the assembly process of mechanical parts according to claim 1, characterized in that, include: When the orientation correction module translates in the opposite direction to the first contact inversion direction vector and tilts slightly around the attitude axis perpendicular to the first contact inversion direction vector, it outputs translation servo commands and tilt servo commands based on the preset servo step size. The preset servo step size is a fixed displacement increment, or a dynamically adjustable step size based on the ratio set by the difference between the net exposure amount and the uniform assembly tolerance range.

7. The online error detection and correction system for the assembly process of mechanical parts according to claim 6, characterized in that, include: After each translation and micro-tilt combination adjustment, the orientation correction module triggers a closed-loop retest process to reacquire images and reconstruct boundary and distance data until the recalculated net exposure is less than or equal to the natural uniform oil film thickness threshold, and the two-dimensional orthogonal projection of the real rod profile exposed area point set is completely contained within the closed polygonal area formed by the guide sleeve inlet profile point set. Then, it outputs the reconstructed real mechanical entry boundary and the orientation-corrected plunger rod assembly posture.

8. The online error detection and correction system for the assembly process of mechanical parts according to claim 1, characterized in that, include: The blind insertion execution module sequentially controls the plunger rod to complete the actual boundary entry of the guide sleeve inlet, the directional penetration entry of the rod sealing lip, and the segmented blind insertion entry into the cylinder. Furthermore, the blind insertion execution module forcibly freezes the visual closed-loop feedback throughout the entire blind insertion process, and determines the completion conditions of each entry stage based on the real-time physical insertion depth.

9. The online error detection and correction system for the assembly process of mechanical parts according to claim 8, characterized in that, include: During the advancement process, the blind insertion execution module uses the absolute spatial position when the orientation correction module completes closed-loop correction as the depth zero point reference, and drives the plunger rod to perform spatial oblique interpolation advancement along the feed direction axis of the tool rectangular coordinate system tilted after orientation correction.

10. An online error detection and correction method for the assembly process of mechanical parts, applied to the online error detection and correction system for the assembly process of mechanical parts as described in any one of claims 1-9, characterized in that, Includes the following steps: By combining depth constraints and mapping matrices, multi-directional illumination optical data is mapped to three-dimensional space, coordinate drift distance is calculated, and the point set of the exposed area of ​​the real rod profile and the point set of the meniscus edge covering area are split according to the rigid tolerance threshold. The point set of the guide sleeve entrance profile is also extracted. Project the point set of the exposed area of ​​the real rod profile and the point set of the crescent-locked edge-enclosed area onto the two-dimensional observation plane, and construct the first contact inversion direction vector based on the spatial opposing lateral directions of the two-dimensional geometric centers of the two point sets. In the two-dimensional observation plane, the minimum two-dimensional distance between the point set of the exposed area of ​​the real rod profile, the point set of the crescent-shaped edge covering area, and the point set of the guide sleeve inlet profile is calculated respectively. The absolute difference between the two minimum two-dimensional distances is taken as the net exposure amount. When the net exposure amount is greater than the threshold of the natural uniform oil film thickness, or when the two-dimensional projection of the point set of the exposed area of ​​the real rod profile is not completely contained within the closed polygonal area formed by the point set of the guide sleeve inlet profile, the rod is translated in the opposite direction of the first contact inversion direction vector and slightly tilted around the attitude axis perpendicular to the first contact inversion direction vector. The reconstructed real mechanical entry boundary and the oriented correction of the plunger rod assembly attitude are output. Based on the reconstructed real mechanical entry boundary and the oriented correction of the plunger rod assembly posture, the tool is interpolated and advanced along the feed direction axis of the tool rectangular coordinate system bound to the plunger rod. The real-time physical insertion depth is mapped by combining the absolute encoder position data and the three-dimensional prior geometric dimension parameters to complete the segmented blind insertion.

Images