Method and system for automatically planning oiling path of oiling robot

Abstract

The invention relates to the technical field of route control, in particular to an automatic planning method and system for an oiling path of an oiling robot, and the method comprises the steps: constructing a theoretical passage section sequence of a slideway of the oiling robot; performing information association storage on the path distance variable of the same deformation characteristic sampling point and the two types of deviations to generate a discrete deviation data set; based on a deformation trend deduction model constructed by the discrete deviation data set, outputting a predicted deformation mean value and a predicted risk variance of each position; performing rigid transformation correction on the theoretical passage cross section sequence according to the predicted deformation mean value, and further applying dynamic contraction to the corrected cross section boundary based on the predicted risk variance to generate an adaptive safety cross section sequence; generating a time domain track of the oiling robot based on the self-adaptive safe section sequence; and moreover, a delay compensation flow instruction synchronous with the speed change is generated, so that the safety, the process consistency and the production adaptability of the oiling operation are remarkably improved.

Description

An automatic planning method and system for oiling robot oiling path Technical Field

[0001] This invention relates to the field of route control technology, and specifically to an automatic planning method and system for an oiling robot's oiling path. Background Technology

[0002] In the automotive parts manufacturing industry, large injection-molded parts such as air conditioner housings typically have complex spatial track structures. Due to uneven shrinkage of the injection-molded material during cooling and the release of internal stress during demolding, the molded workpiece often undergoes nonlinear geometric deformation. To achieve automated oiling operations, oiling robots are commonly used in industrial settings, with path planning primarily relying on teach-and-playback or offline programming. For the aforementioned deformed workpieces, to balance cost and practicality, current industrial practices tend to use contact probes or single-point laser sensors to sparsely measure several feature points on the track and perform local compensation on the preset path based on the obtained data.

[0003] Especially in the early stages of mold changeover for multi-variety, small-batch production, or in processing scenarios where the use of low-cost recycled materials leads to significant fluctuations in batch-to-batch thermal shrinkage rates, the workpiece deformation pattern is difficult to compensate for through fixed molds or reuse of historical data, further exacerbating the uncertainty of path planning. To reduce modeling complexity, existing methods typically simplify the three-dimensional spatial deformation of the slide into a one-dimensional feature distribution along the theoretical path centerline, that is, establishing a mapping relationship from path distance to three-dimensional deviations (including positional offsets and cross-sectional torsion), and relying on data-driven methods (such as linear interpolation or spline fitting) to reconstruct the entire path.

[0004] However, existing compensation algorithms are generally based on the linear interpolation assumption, which assumes that the deformation between measurement points is a smooth linear transition. When the injection molded part exhibits nonlinear warping or local abrupt changes in the unmeasured area (blind zone), the robot end effector running along this path is highly susceptible to collision with the contracting slide sidewall, posing a safety hazard. Furthermore, existing path planning methods typically only correct for the robot end effector's positional deviation, failing to consider the spatial attitude changes caused by the slide cross-section torsion. This leads to a mismatch between the coating nozzle and the workpiece surface normal, affecting coating quality. Simultaneously, the transmission delay of the fluid pipeline in the coating system is not incorporated into the control model, causing the actual oil output to be out of sync with the movement speed, further reducing coating uniformity.

[0005] In summary, existing technologies, under conditions of only sparse measurement data, are insufficient to effectively avoid blind zone collision risks, adaptively compensate for complex nonlinear deformations (including displacement and torsion), and lack synergistic optimization of fluid control hysteresis effects. There is an urgent need for a robust, low-cost, and multi-variety production-applicable intelligent path planning method for oiling robots. Summary of the Invention

[0006] To address the technical problems of existing oiling robot path planning methods, which struggle to simultaneously guarantee blind zone collision safety, adaptive compensation capability for complex nonlinear deformation, and coating uniformity under fluid control lag when relying solely on sparse measurement data, this invention aims to provide an automatic oiling path planning method and system for an oiling robot. The specific technical solution adopted is as follows: One embodiment of this invention provides an automatic oiling path planning method for an oiling robot, comprising the following steps: constructing a theoretical passage section sequence of the oiling robot's slide based on the standard design model of the workpiece, wherein the sections in the theoretical passage section sequence are safe passage spaces with directional adaptability to the nonlinear deformation of the injection molded part; determining the path distance variables, spatial position deviations, and axial torsional deviations of several deformation feature sampling points on the slide; and... The path distance variable of the same deformation feature sampling point is associated with and stored with two types of deviations to generate a discrete deviation dataset. Based on the discrete deviation dataset, a deformation trend inference model is constructed to infer the spatial position deviation and axial torsional deviation on the slide, and the predicted mean deformation and predicted risk variance of each position are output simultaneously. The theoretical passage section sequence is rigidly transformed and corrected according to the predicted mean deformation, and dynamic contraction is applied to the corrected section boundary based on the predicted risk variance to generate an adaptive safety section sequence. The time-domain trajectory of the oiling robot is generated based on the adaptive safety section sequence, and the time-domain trajectory includes the position, attitude and speed of the oiling robot. Based on the preset pipeline transmission delay parameters, the flow command is advanced and compensated on the time axis to generate a delay-compensated flow command synchronized with the speed change.

[0007] Furthermore, the construction of the theoretical passage section sequence of the oiling robot slide based on the standard design model of the workpiece includes: parameterizing the arc length of the center line of the oiling robot slide to obtain an arc length parameterized path, and establishing a local coordinate system along the arc length parameterized path with tangential, normal and subnormal as basis vectors; performing discrete sampling along the arc length parameterized path with a fixed step size, and intercepting the inner wall contour of the slide based on the local coordinate system at each sampling point to generate each original section bound to the path distance; according to the physical characteristics of the workpiece sidewall, setting an avoidance distance greater than the depth distance in the width direction of each original section, and performing non-uniform inward offset based on the avoidance distance to generate the theoretical passage section sequence.

[0008] Further, obtaining the arc length parameterized path includes: taking the starting point of the slide centerline as the zero point, calculating the cumulative arc length extending along the slide centerline as the path distance variable, and taking the slide centerline with the path distance variable as the arc length parameterized path.

[0009] Further, determining the spatial position deviation and axial torsional deviation of several deformation feature sampling points on the slide includes: sampling the arc-length parameterized path according to the geometric features of the slide to obtain each deformation feature sampling point, and obtaining the measured spatial coordinates and measured normal vectors at each deformation feature sampling point. The geometric features include at least the starting point, the ending point, and the curvature extremum point; determining the spatial position deviation of each deformation feature sampling point based on the difference between the measured spatial coordinates and the theoretical spatial coordinates at the same deformation feature sampling point; and analyzing the direction and degree of torsional deformation based on the measured normal vector and the theoretical normal vector at the same deformation feature sampling point to determine the axial torsional deviation of each deformation feature sampling point.

[0010] Further, determining the axial torsional deviation of each deformation feature sampling point includes: calculating the cross product vector of the measured normal vector and the theoretical normal vector for each deformation feature sampling point; projecting the cross product vector onto the tangential direction of the guide path at the sampling point, and determining the direction of torsional deformation based on the sign of the projection value; calculating the angle between the theoretical normal vector and the measured normal vector; and generating a signed axial torsional deviation based on the direction of the torsional deformation and the angle.

[0011] Further, obtaining the deformation trend inference model includes: selecting a Gaussian process regression model, initializing the kernel function and hyperparameters of the Gaussian process regression model to obtain an initial model; and using the discrete deviation dataset to train the initial model to obtain a trained model, which serves as the deformation trend inference model.

[0012] Further, obtaining the adaptive safety cross-section sequence includes: for the j-th cross-section in the theoretical passage cross-section sequence, its distance from the path variable... Unique binding, the execution steps are as follows: based on the path distance The predicted position deviation at a certain point is used to perform a translation correction on the j-th section, and the correction is based on the path distance. The predicted torsional deviation at that location is corrected by rotation about the tangential direction of the guide path at that location, resulting in the cross-section after rigid transformation; based on the path distance... The initial shrinkage amount is calculated based on the predicted risk variance at the location, and the initial shrinkage amount is clamped according to the preset minimum safe passage width threshold to obtain the actual shrinkage amount; taking the cross section after rigid transformation as the reference, the actual shrinkage amount is used to perform inward offset on the boundary of the j-th cross section after rigid transformation to generate the corresponding adaptive safe cross section.

[0013] Further, generating the time-domain trajectory of the oiling robot based on the adaptive safety section sequence includes: determining guide points within each section of the adaptive safety section sequence and fitting and generating a smooth motion path located inside all adaptive safety sections; dynamically planning the running speed based on the geometric curvature of the smooth motion path; mapping the path parameters to a time function according to the smooth motion path and the running speed to generate a time-domain trajectory containing position, attitude, and velocity; wherein the attitude is determined according to the corrected section normal, so that the axis of the oiling nozzle is perpendicular to the bottom surface of the slide after torsional deformation.

[0014] Furthermore, the process of generating a delayed flow command synchronized with speed changes by compensating for the flow command ahead of time based on preset pipeline transmission delay parameters includes: calculating the theoretical flow rate required to maintain a constant coating amount per unit length based on the robot's running speed; shifting the theoretical flow rate forward on the time axis by a duration equal to the pipeline transmission delay to generate an advanced flow command; adding a preparatory segment at the beginning of the trajectory that is not less than the pipeline transmission delay, during which the robot remains stationary while the flow control system executes the advanced flow command; determining the advance pump stop distance based on the current running speed and pipeline transmission delay before the trajectory ends, and issuing a pump stop command to close the liquid supply valve when the robot reaches the position corresponding to the advance pump stop distance; and in the remaining trajectory segment, the coating robot runs at the current running speed, utilizing the residual pressure in the pipeline system to complete the tail-end coating.

[0015] Another embodiment of the present invention provides an automatic oiling path planning system for an oiling robot, including a processor and a memory, wherein the processor is used to process instructions stored in the memory to implement an automatic oiling path planning method for an oiling robot.

[0016] This invention offers the following advantages: It provides an automatic path planning method and system for an oiling robot. This method constructs a theoretical passable cross-section sequence that adapts to the nonlinear deformation of injection-molded parts. Furthermore, it dynamically shrinks the corrected cross-section boundaries based on the predicted risk variance, generating an adaptive safety cross-section sequence. This allows the safety passage to automatically tighten in unmeasured areas and remain loose in high-confidence areas, effectively avoiding collision risks caused by nonlinear warping even with only sparse measurement data. By determining spatial position deviation and axial torsional deviation, and storing them in association with path distance variables to form a discrete deviation dataset, the two types of deviations along the entire path are deduced based on this dataset. Finally, the position and attitude are simultaneously planned when generating the time-domain trajectory, ensuring that the oiling nozzle attitude is always aligned with the normal of the torsional deformed slide surface, solving the problem of coating quality degradation caused by attitude mismatch. By using preset pipeline transmission delay parameters to advance the flow command on the time axis, the oiling action is synchronized with the robot's actual movement speed, overcoming the insufficient glue volume in the acceleration section or the accumulation in the deceleration section caused by pipeline delay in traditional methods, significantly improving oiling uniformity. In summary, this invention significantly improves the safety, process consistency, and production adaptability of oiling operations. Attached Figure Description

[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 is a flowchart of the automatic planning method for the oiling path of an oiling robot according to the present invention; Figure 2 is a flowchart of the steps for constructing the theoretical passage cross section sequence of the oiling robot slide based on the standard design model of the workpiece in an embodiment of the present invention. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solution proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] To address the technical problem mentioned in the background art that existing oiling robot path planning methods, when relying solely on sparse measurement data, cannot simultaneously guarantee blind zone collision safety, adaptive compensation capability for complex nonlinear deformation, and coating uniformity under fluid control hysteresis, an embodiment of the present invention provides an automatic oiling path planning method for an oiling robot, as shown in Figure 1, which includes the following steps: S1, constructing a theoretical passage cross-section sequence of the oiling robot slide based on the standard design model of the workpiece.

[0022] Here, the cross-sections in the theoretical passage cross-section sequence are safe passage spaces that are directionally adaptable to the nonlinear deformation of the injection molded part, i.e., theoretical cross-sections with directional safety constraints. The workpiece can be an injection molded part, and the standard design model is usually a three-dimensional digital model in STEP or IGES format.

[0023] By constructing a theoretical sequence of passage sections for the oiling robot slide, a safe passage prior model that integrates manufacturing prior knowledge can be provided for the entire adaptive path planning system, enabling high-precision, high-safety, and high-efficiency oiling operations to be achieved with only sparse measurements.

[0024] As an exemplary implementation, a theoretical passage section sequence of the oiling robot slide is constructed based on the standard design model of the workpiece, as shown in Figure 2. This includes: S11, the arc length parameterization of the slide centerline in the standard design model is performed to obtain the arc length parameterization path, and a local coordinate system with tangential, normal and subnormal as basis vectors is established along the arc length parameterization path.

[0025] To achieve unified alignment and retrieval of data throughout the entire process, a scalar reference system with path extension length as the unique index needs to be established, which requires arc length parameterization of the slide centerline in the standard design model.

[0026] In this embodiment, the starting point of the slide centerline is taken as the zero point. The cumulative arc length extending along the slide centerline is calculated as the path distance variable, and the slide centerline with the path distance variable is used as the arc length parameterized path. Specifically, in the standard design model, the system extracts the geometric centerline of the slide bottom surface using a geometric algorithm and defines it as the slide centerline; the path distance variable ranges from 0 to... , The total length of the slide centerline is given. The spatial coordinates of any point on the slide centerline can be determined using the path distance variable, and can be uniquely represented as... .

[0027] Of course, the path distance variable can also be a normalized relative length, with a value range between 0 and 1, or a discrete path point index. This embodiment takes the cumulative arc length as an example.

[0028] To distinguish between the depth and width directions of the track, the system calculates a local coordinate system along the track centerline that continuously varies with the path distance variable, known as the Frenet frame. This coordinate system consists of three mutually orthogonal unit vectors: a tangential vector (direction along the tangent of the arc-length parameterized path, indicating the direction of the oiling operation); a normal vector (perpendicular to the tangent and pointing towards the track opening (or the center of curvature), clearly indicating the depth direction of the track); and a binormal vector (defined by the cross product of the tangential and normal vectors, clearly indicating the width direction of the track).

[0029] It is worth noting that for line segments, the normal vector of the previous valid node remains unchanged; or the Rotation-Minimizing Frame (RMF) is explicitly used.

[0030] It should be noted that by establishing a local coordinate system, a unified mathematical benchmark can be provided for the subsequent separation and quantification of the bending deformation and axial torsional deformation of the slide.

[0031] S12 performs discrete sampling along the arc-length parameterized path with a fixed step size, and at each sampling point, it extracts the inner wall contour of the slide based on the local coordinate system to generate each original cross section that is bound to the path distance.

[0032] To achieve discretization of continuous geometric data, and since subsequent deformation corrections need to be performed at specific locations, the subsequent slide can be transformed into a sequence of finite cross sections.

[0033] In this embodiment, the system performs discrete sampling along the arc-length parameterized path with a fixed step size to obtain several sampling points. The fixed step size can be set to 1 mm, and the number of sampling points can be set to M. For each sampling point, a two-dimensional plane composed of a normal vector and a binormal vector is constructed. This two-dimensional plane is then used to extract the inner wall contour curve of the slide in the standard design model, obtaining the original inner wall contour, i.e., the original cross-section, at each sampling point location. Each original cross-section is bound to a unique path distance to serve as the basic unit for subsequent data processing.

[0034] S13. Based on the physical properties of the workpiece sidewall, set a clearance distance greater than the depth distance in the width direction of each original section, and perform non-uniform inward offset based on the clearance distance to generate a theoretical passable section sequence.

[0035] Here, the physical properties of the sidewall of the injection molded part can be based on preset empirical deformation amount or finite element simulation deformation data.

[0036] Considering that during the demolding and cooling process of injection molded parts, the sidewalls are prone to inward shrinkage, flattening, or elliptical distortion due to the release of internal stress, while the deformation of the bottom surface is relatively small; and the probability of the oiling nozzle spatially overlapping with the sidewalls during movement is significantly higher than that of the bottom surface, a uniform path distance cannot be simply used. Instead, non-uniform directional clearance spacing can be set in the width and depth directions of the original cross-section.

[0037] In this embodiment, based on the typical physical deformation characteristics of injection molded parts, non-uniform directional clearance spacing is set in the width and depth directions of the original cross-section to construct a theoretical passable cross-section sequence with structural adaptability tolerance.

[0038] The first step is to set the sidewall clearance distance in the width direction indicated by the subnormal vector, which corresponds to the sidewall of the slide. The range of its values ​​is significantly larger than the bottom clearance distance.

[0039] In this embodiment, the setting of the sidewall clearance spacing needs to comprehensively consider robot positioning errors, inward shrinkage caused by the release of internal stress during the demolding and cooling process of the thin-walled sidewall structure, and typical nonlinear deformation modes such as elliptic or trapezoidal distortion, reflecting the active tolerance reservation for high-risk areas. Therefore, the sidewall clearance spacing is usually taken as 0.5 mm to 2.0 mm to cover the most unfavorable deformation situation in high-risk areas.

[0040] The second step is to set the bottom clearance distance in the depth direction indicated by the normal vector (corresponding to the bottom of the slide).

[0041] In this embodiment, since the bottom surface of the injection molded part is usually equipped with reinforcing ribs or structural supports, it has high rigidity and small and stable deformation during demolding and cooling. The main source of error is robot positioning deviation. Therefore, the bottom surface clearance distance is relatively small, usually 0.1 mm to 0.5 mm.

[0042] Based on the aforementioned sidewall clearance and bottom clearance, a non-uniform inward offset operation is performed on each original cross-section, that is, offset along the width direction and offset along the depth direction to generate a closed polygonal region, which can be defined as the theoretical passage cross-section. Then, the theoretical passage cross-section of each sampling point is stored in the path order to form a theoretical passage cross-section sequence, which can serve as the basis for subsequent data analysis.

[0043] It should be noted that the theoretical passable section can characterize the structural robustness design that has been embedded with actual manufacturing deviations under the ideal design geometric constraints. That is, the center of the oiling nozzle can effectively avoid the collision risk caused by the indentation of the side wall while meeting the safety space requirements, and at the same time avoid sacrificing process efficiency due to excessive conservative offset.

[0044] S2, determine the path distance variable, spatial position deviation and axial torsional deviation of several deformation feature sampling points on the slide, and store the information association between the path distance variable and the two types of deviations for the same deformation feature sampling point to generate a discrete deviation dataset.

[0045] By using contact sensors mounted on the robot to perform sparse measurements on the slide, deformation information of the workpiece deviating from the ideal geometric model under actual manufacturing conditions is obtained, i.e., a discrete deviation dataset. The aforementioned path distance variable, spatial position deviation, and axial torsional deviation are bound and stored to form a structured discrete deviation dataset. The discrete deviation dataset uses the path distance variable as a unique index, realizing accurate alignment and efficient retrieval of deformation information under a unified scalar reference system, which can provide a high signal-to-noise ratio input basis for subsequent processing.

[0046] As an exemplary implementation, determining the spatial position deviation and axial torsional deviation of several deformation feature sampling points on the slide includes: First, sampling the arc-length parameterized path according to the geometric characteristics of the slide to obtain each deformation feature sampling point, and obtaining the measured spatial coordinates and measured normal vectors at each deformation feature sampling point.

[0047] To balance production cycle time and measurement accuracy, the system does not perform full-field scanning, but instead selects a small number of key locations for contact measurement, that is, analyzes the discrete deviation data of several deformation feature sampling points.

[0048] In this embodiment, based on the centerline of the slide, several key locations are selected as deformation feature sampling points (such as the starting point, the ending point, and the curvature extreme point), and the path distance variable corresponding to each deformation feature sampling point is recorded; then, the measured spatial coordinates and measured normal vector of each deformation feature sampling point are obtained by detecting with an end sensor (such as a contact probe).

[0049] It should be noted that when acquiring deformation feature sampling points, for long straight line segments, sampling points can be added at preset intervals (e.g., every 50mm).

[0050] The second step is to determine the spatial position deviation of each deformation feature sampling point based on the difference between the measured spatial coordinates and the theoretical spatial coordinates at the same deformation feature sampling point.

[0051] Here, the spatial position deviation reflects the coordinate offset of the sampling points of the deformation characteristics of the injection molded part in three-dimensional space.

[0052] As an example, the formula for calculating the spatial position deviation of the k-th deformation feature sampling point can be: In the formula, This represents the spatial position deviation of the k-th deformation feature sampling point. This represents the measured spatial coordinates of the k-th deformation feature sampling point. This represents the theoretical spatial coordinates of the k-th deformation feature sampling point.

[0053] The third step is to analyze the direction and extent of torsional deformation based on the measured normal vector and theoretical normal vector at the same deformation feature sampling point, and to determine the axial torsional deviation of each deformation feature sampling point.

[0054] Since injection molded parts may undergo torsional deformation around their axis, and this deformation has a direction (clockwise or counterclockwise), it is necessary to accurately calculate the signed angular deviation. However, in order to avoid the loss of directional information or the generation of computational singularities at specific angles by the traditional inverse cosine function, this embodiment can adopt a calculation method based on cross product projection.

[0055] As an exemplary implementation, for each deformation feature sampling point, the axial torsional deviation is determined, including: calculating the cross product vector of the measured normal vector and the theoretical normal vector; here, the cross product vector is perpendicular to the plane containing the theoretical normal and the measured normal, and its direction reflects the direction of the rotation axis.

[0056] As an example, the formula for calculating the cross product vector of the k-th deformation feature sampling point can be: In the formula, This represents the cross product vector of the k-th deformation feature sampling point. This represents the measured normal vector of the k-th deformation feature sampling point. This represents the theoretical normal vector of the k-th deformation feature sampling point.

[0057] The cross product vector is projected onto the tangential direction of the guide path at the sampling point, and the direction of torsional deformation is determined according to the sign of the projection value. In this embodiment, the cross product vector is projected onto the tangential vector at the sampling point, and the projection value is calculated. The calculation formula is as follows: In the formula, Indicates the projection value. This represents the cross product vector of the k-th deformation feature sampling point. This represents the tangential vector of the k-th deformation feature sampling point.

[0058] when When the direction of torsional deformation is determined to be positive torsion (e.g., clockwise); when When the torsional deformation is determined to be in the negative direction (e.g., counterclockwise), the direction of the torsional deformation is determined to be negative.

[0059] Calculate the angle between the theoretical normal vector and the measured normal vector. In this embodiment, the absolute value of the angle between the theoretical and measured normal vectors is calculated. However, to prevent division by zero errors, a minor protection logic is added. Specifically, if the measured normal vector and the theoretical normal vector are extremely close, such as their dot product being close to 1, or their cross product magnitude being less than 1, the calculation is performed as follows: If the result is zero, the torsion angle is directly determined to be zero; otherwise, the angle is calculated using the atan2 function or the protected acos function.

[0060] Based on the direction and angle of torsional deformation, a signed axial torsional deviation is generated.

[0061] In this embodiment, the formula for calculating the axial torsional deviation of the kth deformation feature sampling point can be: In the formula, This represents the axial torsional deviation at the k-th deformation feature sampling point, where sign represents the sign function. Indicates the projection value. This represents the angle between the theoretical normal vector and the measured normal vector.

[0062] It should be noted that in this embodiment, the complex physical deformation is orthogonally decoupled into two independent data sequences: spatial position deviation and axial torsional deviation. The spatial position deviation and axial torsional deviation are then bound to their corresponding path distance variables to generate a discrete deviation dataset containing several sets of data, which can be used as training samples for the subsequent deformation trend inference model.

[0063] S3 constructs a deformation trend inference model based on discrete deviation dataset, infers the spatial position deviation and axial torsional deviation on the slide, and simultaneously outputs the mean of predicted deformation and the variance of predicted risk at each position.

[0064] Because contact measurements are performed at only a limited number of key locations on the slide, the resulting deviation dataset is highly sparse and cannot be directly used for continuous correction across the entire path. Forcing linear interpolation or spline fitting into unmeasured areas ignores the nonlinear characteristics of the molded part's deformation, easily leading to uncontrollable collision risks in blind spots. Therefore, this embodiment introduces a deformation trend prediction model based on a discrete deviation dataset. Its core principle is to treat physical deformation as a stochastic process that smoothly changes with path distance, and to model the subsequent distribution of this stochastic process using known sampling points, thereby simultaneously outputting the most probable deformation and its confidence level at any location.

[0065] It should be noted that for slide structures with one-dimensional extension characteristics in spatial orientation, their three-dimensional deformation is simplified to the distribution of empirical data features along the path centerline, and a data-driven approach is used to construct a mapping from one-dimensional distance to three-dimensional deviation.

[0066] As an exemplary implementation, obtaining a deformation trend inference model includes: First, selecting a Gaussian process regression model, initializing the kernel function and hyperparameters of the Gaussian process regression model, and obtaining an initialized model.

[0067] In this embodiment, in order to infer the deformation trend of the unmeasured area from a limited number of sampling points, Gaussian Process Regression (GPR) is selected as the core inference algorithm. First, the kernel function and hyperparameters of the GPR model are initialized, and the model with completed parameter initialization is used as the initialization model.

[0068] Considering the physical characteristics of locally smooth and non-abrupt deformation of injection molded parts, the Matern3 / 2 kernel function is selected to describe the correlation of deformation between different locations along the path. Its calculation formula is as follows: In the formula, Represents path distance variable and The correlation of deformation between corresponding sliding positions, Indicates the signal variance. This represents the length scale parameter. exp represents the absolute value of the difference between two path distance variables, and exp represents an exponential function with the natural constant as the base.

[0069] Among them, the hyperparameter signal variance This determines the amplitude of deformation fluctuations. In the absence of historical statistical data, the variance of the deviation values ​​of all current deformation characteristic sampling points is calculated and denoted as... , ,in, To preset the minimum system noise variance, such as ; An empirical value of 1.5 can be used, which ensures that the model is capable of covering a fluctuation range that is slightly larger than the current measured deviation; the hyperparameter length scale parameter The length scale parameter determines the frequency of deformation fluctuations. A larger length scale parameter means that the pseudo-deformation changes of the model are relatively gentle. The average path distance between adjacent deformation feature sampling points can be calculated and used as the length scale parameter.

[0070] It should be noted that, in order to avoid the initial model predicting zero when it is far from the sampling point, the prior mean function of GPR is set to the linear interpolation function of the adjacent sampling points, or simply set to 0 (assuming that the deviation fluctuates around 0).

[0071] The second step is to use the discrete deviation dataset to train the initial model to obtain a trained model, which serves as the deformation trend inference model.

[0072] In this embodiment, the initialization model training process is the process of solving for the optimal posterior distribution of hyperparameters.

[0073] After model training is complete, for any path distance variable within the path range, the deformation trend inference model outputs two continuous function values: the predicted mean deformation and the predicted risk variance. The predicted mean deformation represents the most likely geometric deformation of the workpiece at different path distance variables, including predicted positional deviation and torsional deviation. The predicted risk variance is the posterior variance of the regression model, directly quantifying the risk level of the prediction blind zone. This value has a clear physical meaning: near the sampling point, due to the support of measured data, the predicted risk variance approaches zero; however, in the blind zone between two sampling points, as the distance increases, the uncertainty caused by missing data increases, and the predicted risk variance increases significantly.

[0074] It should be noted that these refer to the three components of spatial position deviation. Axial torsional deviation and axial torsion deviation can be used to construct an independent deformation trend inference model.

[0075] S4. Based on the predicted mean deformation, the theoretical passable cross-section sequence is rigidly transformed and corrected. Furthermore, based on the predicted risk variance, dynamic contraction is applied to the corrected cross-section boundary to generate an adaptive safe cross-section sequence.

[0076] Here, the adaptive safety section sequence is a safety passage that is both aligned with the actual workpiece and has risk perception capabilities.

[0077] During the demolding and cooling process of injection molded parts, the slideway may undergo translational movement (spatial position deviation) and torsion around its axis (axial torsional deviation). This is rigid body motion, which does not change the relative structure inside the cross-section, but only its position in space. Therefore, the theoretical passage cross-section can be rigidly transformed by translation and tangential rotation to align it with the local geometry of the actual workpiece. When implementing automatic oiling path planning, introducing uncertainty quantification can mitigate blind zone risks to some extent. The predicted risk variance can be used to extrapolate uncertainties. The larger the risk variance, the farther the location is from the sampling point, and the less reliable the extrapolation result. Therefore, it can be used to achieve dynamic shrinkage.

[0078] As an exemplary implementation, obtaining the adaptive safety cross-section sequence includes: for the j-th cross-section in the theoretical passage cross-section sequence, its distance from the path variable... Unique binding, the execution steps are as follows: First, based on the path distance The predicted position deviation at point j is used to perform translation correction for the j-th section, and based on the path distance... The predicted torsional deviation at that location is corrected by rotation around the tangential direction of the guide path at that location to obtain the cross section after rigid transformation.

[0079] In this embodiment, the path distance is obtained. To correct the predicted position deviation, the coordinates of all vertices of the j-th section are translated along the vector, so that the j-th section is moved to the position after the actual displacement of the workpiece, thus completing the translation correction; the path distance is obtained. The predicted torsional deviation at the point of rotation is calculated based on the center of rotation, with the translated section around the path distance. The tangential vector at a certain point is rotated to predict the torsional deviation, thereby correcting the attitude error caused by the workpiece's torsion around the axis. This completes the rotation correction, resulting in the j-th cross-section after rigid transformation. Here, the current path distance is used as the reference. The corresponding theoretical path point is the center of rotation.

[0080] Rigid correction alone is insufficient to mitigate risk because the predicted mean deformation may contain errors. To ensure high reliability of the safe cross-section, the predicted risk variance is used to dynamically shrink the cross-section boundary, thereby constructing a risk-adaptive safe space, as shown in steps two and three below: Step two, based on the path distance... The initial contraction amount is calculated based on the predicted risk variance at the location, and the initial contraction amount is clamped according to the preset minimum safe passage width threshold to obtain the actual contraction amount.

[0081] The larger the variance of the predicted risk, that is, the more uncertain the area, the greater the safety distance that needs to be contracted inward, that is, the greater the initial contraction amount, and the greater the space that needs to be reserved for avoidance.

[0082] As an example, the formula for calculating the initial shrinkage of the j-th section can be: In the formula, This represents the initial shrinkage of the j-th cross section; This represents the confidence coefficient, which can be taken as an empirical value of 2. Indicates path distance The predicted risk variance at a given location is the predicted risk variance corresponding to the spatial position deviation output by the deformation trend extrapolation model. It is worth noting that the predicted risk variance of the axial torsional deviation is only used for the reliability assessment of attitude setting and is not included in the calculation of cross-sectional boundary shrinkage.

[0083] Meanwhile, to avoid excessive cross-sectional shrinkage or even cross-sectional closure (channel breakage) due to excessively large predicted risk variance, the system introduces a minimum safe channel width threshold to obtain the actual shrinkage amount.

[0084] First, set a minimum safety passage width threshold, which is typically set to the nozzle diameter plus a necessary small margin, such as... , This represents the minimum safe passage width threshold. Indicates the nozzle diameter.

[0085] Secondly, before performing the shrinkage operation, the remaining width of the cross-section after shrinkage is pre-calculated. The remaining width after shrinkage can be twice the shortest Euclidean distance from the geometric center of the cross-section to its boundary.

[0086] If the remaining width is less than the minimum safe passage width threshold, the shrinkage amount can be limited to (current cross-section width - minimum safe passage width threshold) / 2, and the area can be marked as a high-risk area.

[0087] Finally, the area that triggered the clamping logic is recorded, and warning messages can be selectively output to alert the operator that the corresponding area is at high risk.

[0088] The third step involves using the rigidly transformed section as a reference and applying an inward offset to the boundary of the j-th section after the rigid transformation using the actual shrinkage amount to generate the corresponding adaptive safety section.

[0089] In this embodiment, based on the actual shrinkage amount, an inward offset operation is performed on the boundary of the j-th cross section after rigid transformation, and the region formed after shrinkage can be used as the j-th adaptive safety cross section.

[0090] Referring to the process of obtaining the j-th adaptive safety section, each adaptive safety section can be obtained. Connecting all M adaptive safety sections in sequence forms a closed risk adaptive channel. This channel maintains the design width near the sampling point, while automatically narrowing in the unmeasured middle blind zone (but ensuring minimum passage), which can be used as the geometric constraint boundary for path planning.

[0091] It should be noted that by using rigid transformation to achieve geometric alignment and risk variance to drive dynamic safety margin, the two are combined to construct an adaptive safety channel that is both accurate and robust, effectively solving the contradiction between blind zone collision risk and process efficiency under sparse measurement.

[0092] S5 generates the time-domain trajectory of the oiling robot based on the adaptive safety cross-section sequence.

[0093] Here, the time-domain trajectory includes the position, orientation, and velocity of the oiling robot.

[0094] Due to the nonlinear deformation of injection molded parts and the sparse measurement data, traditional path planning methods struggle to balance obstacle avoidance safety and process accuracy. This invention constructs an adaptive safety section sequence, integrating physical deformation, manufacturing priors, and deductive uncertainties into a dynamic feasible domain. Within this domain, a smooth path is solved, thereby achieving high-precision and high-efficiency oiling operations while ensuring absolute safety.

[0095] As an exemplary implementation, determining the temporal trajectory of the oiling robot includes: a first step of determining guide points within each section of the adaptive safety section sequence and fitting and generating a smooth motion path located within all adaptive safety sections.

[0096] The adaptive safety cross section sequence is a dynamic safety channel that changes with the path after rigid transformation correction and risk perception contraction. The robot end effector's motion path must be strictly located inside all cross sections, otherwise there is a risk of collision. Therefore, path generation requires the adaptive safety cross section sequence as a hard geometric constraint.

[0097] As an exemplary implementation, obtaining a smooth motion path includes: first, extracting the geometric center point of each adaptive safety section as a guide point to form an initial guide point column.

[0098] Since the adaptive safety section is a flexible contraction that completes the rigid transformation correction and risk-driven process, its geometric center is naturally located in the high-confidence safety region, and it can serve as a reliable initial guide for path planning.

[0099] Subsequently, parametric curve fitting algorithms such as cubic B-spline or NURBS are used to smoothly interpolate the initial guide point sequence, generating a continuous spatial curve. The implementation process of the curve fitting algorithm is existing technology and will not be elaborated here. However, simple fitting may cause local trajectories to cross the cross-sectional boundary due to abrupt curvature changes (such as sharp bends), posing a collision risk. Therefore, a constraint verification loop mechanism is introduced: discrete sampling is performed along the spatial curve at steps no larger than a preset step size (e.g., 0.5 mm); for each sampling point, it is determined whether it is strictly located inside the adaptive safety cross-sectional polygon at the corresponding path distance variable; if a point crosses the boundary, a correction force pointing inwards towards the normal direction of the cross-section is applied to its neighboring spline control points, iteratively adjusting the curve shape; the direction of the correction force is the vector pointing towards the nearest point on the adaptive safety cross-sectional inward boundary to the current boundary-crossing point; the above process is repeated until the maximum number of iterations is reached, such as 50 times. If the constraint is still not satisfied after reaching the maximum number of iterations, a channel adaptive relaxation strategy is triggered: the confidence coefficient corresponding to the current boundary-crossing region is gradually reduced by a fixed step size (e.g., 0.1). The actual shrinkage is recalculated to locally relax the safety section until the spline curve meets the inclusion constraint, and the reference running speed of this section is reduced proportionally at the same time.

[0100] It should be noted that the smooth motion path is forcibly restricted to the safe region after risk perception contraction. The blind zone does not depend on the assumptions. Even if the initial fit goes out of bounds, it can regress to the safe region through iterative correction.

[0101] The second step is to dynamically program the running speed based on the geometric curvature of the smooth motion path.

[0102] The robot cannot maintain a constant speed in complex, curved tracks. Excessive cornering speed can cause the robot joints to exceed their limits or generate excessive centrifugal force, affecting the accuracy of the oil application. Therefore, this embodiment plans the speed based on the geometry of the path.

[0103] In straight sections, the robot operates at full speed at the reference speed; in curved sections, the speed is automatically reduced to limit the centripetal acceleration to no more than the robot's maximum permissible centripetal acceleration, in order to ensure trajectory accuracy.

[0104] In this embodiment, the smooth motion path is traversed, the geometric curvature of each point on the path is calculated, and a formula for calculating the running speed is established, which can be: In the formula, This represents the running speed located at the path distance variable l, and min represents the function for finding the minimum value. This indicates the preset straight-line reference speed (e.g., 200 mm / s). This indicates the robot's maximum permissible centripetal acceleration (e.g., 500 mm / s²). This represents the geometric curvature located at the path distance variable l; This represents a tiny constant to prevent the denominator from being zero, such as 0.001.

[0105] The third step is to map the path parameters into a time function based on the smooth motion path and running speed, generating a time-domain trajectory that includes position, attitude and velocity.

[0106] To transform the spatial path into a time-series instruction executable by the robot controller, a mapping relationship between path distance variables and time is established, and the cumulative time required to reach different sampling points on the path is calculated through numerical integration. The expression can be: In the formula, This represents the cumulative time required to be located at the path distance variable. This represents the running speed at a path distance x.

[0107] For each point on the path, not only the position and speed of the oiling robot are needed, but also the orientation of the nozzle. Specifically, the tangent direction of the path at each point is calculated, and the corrected cross-sectional normal is read to construct a rotation matrix so that the nozzle axis is aligned with the corrected cross-sectional normal, thereby ensuring that the nozzle is always perpendicular to the bottom surface of the slide after torsional deformation.

[0108] S6, based on preset pipeline transmission delay parameters, advances the flow command on the time axis to generate a delay-compensated flow command that is synchronized with the speed change.

[0109] In actual operation, there is usually a long hose between the oil pump and the nozzle. When the robot moves at varying speeds, if a command to change the flow rate is issued at time t, the oil will actually flow at time t. The time lag between the oil flow from the nozzle and its exit point causes a mismatch between the amount of oil applied and the speed of application; for example, during acceleration, the oil flow lags behind, resulting in a thinner coating. Therefore, this embodiment introduces a pre-determined pipeline transport delay constant. The system generates a delay-compensated flow command. Specifically, it uses a high-speed camera to capture the time difference between the nozzle opening moment and the actual oil outflow moment, calculating the average of these two time differences as the pipeline transmission delay constant.

[0110] As an exemplary implementation, obtaining the delay compensation flow instruction includes: First, calculating the theoretical flow required to maintain a constant coating amount per unit length based on the robot's running speed.

[0111] As an example, the formula for calculating theoretical flow rate can be: In the formula, This represents the theoretical flow rate at time t. This represents the oil application rate coefficient per unit length. This represents the running speed at time t.

[0112] The second step is to shift the theoretical flow rate forward on the time axis by a duration equal to the pipeline transmission delay, thereby generating a lead flow command.

[0113] In this embodiment, to compensate for the delay, the control command is preemptively executed, that is, the flow command is shifted forward on the time axis to generate a leading flow command, the expression of which can be: In the formula, issuing a command at the current time t is actually aimed at a future time. The speed.

[0114] Because of the reference to future moments, it is necessary to handle the boundary overflow problem at the beginning and end of the trajectory, as shown in steps three and four below: Step three, add a preparation segment at the beginning of the trajectory that is no less than the pipeline transmission delay. During the preparation segment, the robot remains stationary while the flow control system executes the advance flow command.

[0115] In this embodiment, a period of at least [duration] is added before the robot begins to move. In the preparatory phase, the robot remains stationary, but the flow control system begins to execute advance flow commands to facilitate the pressure build-up within the pipeline.

[0116] The fourth step is to determine the early pump stop distance based on the current operating speed and pipeline transmission delay before the trajectory ends, and issue a pump stop command to close the liquid supply valve when the pump reaches the position corresponding to the early pump stop distance.

[0117] In this embodiment, the distance is equal to the distance before the trajectory ends. At the designated location, a pump stop command is issued to close the liquid supply valve.

[0118] Fifth, within the remaining trajectory segment, the oiling robot operates at its current speed, utilizing the residual pressure within the pipeline system to complete the tail-end coating.

[0119] In this embodiment, within the remaining trajectory segment, the robot maintains the planned speed and continues to move, using the residual pressure within the pipeline system to extrude excess adhesive and complete the tail-end coating. The planned speed can be the real-time operating speed within the time-domain trajectory.

[0120] Furthermore, when generating the time-domain trajectory, the running speed of the trajectory endpoint is also constrained to be 0.

[0121] Finally, the generated time-domain trajectory and the delayed-compensated flow instructions are packaged together to generate standard code that the robot controller can recognize, and then sent out for execution to drive the movement of each joint motor.

[0122] Another embodiment of the present invention provides an automatic oiling path planning system for an oiling robot, including a processor and a memory, wherein the processor is used to process instructions stored in the memory to implement an automatic oiling path planning method for an oiling robot.

[0123] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An automatic planning method for the oiling path of an oiling robot, characterized in that, Includes the following steps: A theoretical passage section sequence for the oiling robot slideway is constructed based on the standard design model of the workpiece. The sections in this theoretical passage section sequence represent safe passage spaces that adapt to the nonlinear deformation of the workpiece. The path distance variable, spatial position deviation, and axial torsional deviation of several deformation feature sampling points on the slideway are determined. The path distance variable and the two types of deviations at the same deformation feature sampling point are associated and stored to generate a discrete deviation dataset. A deformation trend prediction model is constructed based on this discrete deviation dataset to predict the spatial position deviation and axial torsional deviation on the slideway, and the predicted deformation mean and predicted risk variance at each position are output simultaneously. The theoretical passage section sequence is rigidly modified according to the predicted deformation mean, and further, dynamic contraction is applied to the modified section boundaries based on the predicted risk variance to generate an adaptive safe section sequence. The time-domain trajectory of the oiling robot is generated based on the adaptive safety section sequence. The time-domain trajectory includes the position, attitude and speed of the oiling robot. Based on the preset pipeline transmission delay parameters, the flow command is advanced and compensated on the time axis to generate a delay-compensated flow command that is synchronized with the speed change.

2. The automatic planning method for the oiling path of an oiling robot according to claim 1, characterized in that, The construction of the theoretical passage section sequence of the oiling robot slide based on the standard design model of the workpiece includes: parameterizing the arc length of the slide centerline of the oiling robot to obtain the arc length parameterized path, and establishing a local coordinate system along the arc length parameterized path with tangential, normal and subnormal as basis vectors; performing discrete sampling along the arc length parameterized path with a fixed step size, and intercepting the inner wall contour of the slide based on the local coordinate system at each sampling point to generate each original section bound to the path distance; according to the physical characteristics of the workpiece sidewall, setting an avoidance distance greater than the depth distance in the width direction of each original section, and performing non-uniform inward offset based on the avoidance distance to generate the theoretical passage section sequence.

3. The automatic planning method for the oiling path of an oiling robot according to claim 2, characterized in that, Obtaining the arc length parameterized path includes: taking the starting point of the slide centerline as the zero point, calculating the cumulative arc length extending along the slide centerline as the path distance variable, and taking the slide centerline with the path distance variable as the arc length parameterized path.

4. The automatic planning method for the oiling path of an oiling robot according to claim 3, characterized in that, Determining the spatial position deviation and axial torsional deviation of several deformation feature sampling points on the slide includes: sampling the arc-length parameterized path according to the geometric features of the slide to obtain each deformation feature sampling point, and obtaining the measured spatial coordinates and measured normal vectors at each deformation feature sampling point. The geometric features include at least the start point, end point, and curvature extremum point; determining the spatial position deviation of each deformation feature sampling point based on the difference between the measured spatial coordinates and theoretical spatial coordinates at the same deformation feature sampling point; and analyzing the direction and degree of torsional deformation based on the measured normal vector and theoretical normal vector at the same deformation feature sampling point to determine the axial torsional deviation of each deformation feature sampling point.

5. The automatic planning method for the oiling path of an oiling robot according to claim 4, characterized in that, Determining the axial torsional deviation of each deformation feature sampling point includes: calculating the cross product vector of the measured normal vector and the theoretical normal vector for each deformation feature sampling point; projecting the cross product vector onto the tangential direction of the guide path at the sampling point, and determining the direction of torsional deformation based on the sign of the projection value; calculating the angle between the theoretical normal vector and the measured normal vector; and generating a signed axial torsional deviation based on the direction of the torsional deformation and the angle.

6. The automatic planning method for the oiling path of an oiling robot according to claim 1, characterized in that, Obtaining the deformation trend inference model includes: selecting a Gaussian process regression model, initializing the kernel function and hyperparameters of the Gaussian process regression model to obtain an initial model; and using the discrete deviation dataset to train the initial model to obtain a trained model, which serves as the deformation trend inference model.

7. The automatic planning method for the oiling path of an oiling robot according to claim 1, characterized in that, Obtaining the adaptive safety cross-section sequence includes: for the j-th cross-section in the theoretical passage cross-section sequence, its distance from the path variable... Unique binding, the execution steps are as follows: based on the path distance The predicted position deviation at a certain point is used to perform a translation correction on the j-th section, and the correction is based on the path distance. The predicted torsional deviation at that location is corrected by rotation about the tangential direction of the guide path at that location, resulting in the cross-section after rigid transformation; based on the path distance... The initial shrinkage amount is calculated based on the predicted risk variance at the location, and the initial shrinkage amount is clamped according to the preset minimum safe passage width threshold to obtain the actual shrinkage amount; taking the cross section after rigid transformation as the reference, the actual shrinkage amount is used to perform inward offset on the boundary of the j-th cross section after rigid transformation to generate the corresponding adaptive safe cross section.

8. The automatic planning method for the oiling path of an oiling robot according to claim 1, characterized in that, The step of generating the time-domain trajectory of the oiling robot based on the adaptive safety section sequence includes: determining a guide point within each section of the adaptive safety section sequence and fitting and generating a smooth motion path located within all adaptive safety sections; dynamically planning the running speed based on the geometric curvature of the smooth motion path; mapping the path parameters to a time function according to the smooth motion path and the running speed to generate a time-domain trajectory containing position, attitude, and velocity; wherein the attitude is determined according to the corrected section normal, so that the axis of the oiling nozzle is perpendicular to the bottom surface of the slide after torsional deformation.

9. The automatic planning method for the oiling path of an oiling robot according to claim 1, characterized in that, The flow command, based on a preset pipeline transmission delay parameter, is advanced and compensated on the time axis to generate a delay-compensated flow command synchronized with speed changes. This includes: calculating the theoretical flow rate required to maintain a constant coating amount per unit length based on the robot's running speed; shifting the theoretical flow rate forward on the time axis by a duration equal to the pipeline transmission delay to generate an advanced flow command; adding a preparatory segment at the beginning of the trajectory that is not less than the pipeline transmission delay, during which the robot remains stationary while the flow control system executes the advanced flow command; determining the early pump stop distance based on the current running speed and the pipeline transmission delay before the trajectory ends, and issuing a pump stop command to close the liquid supply valve when the early pump stop distance is reached; and within the remaining trajectory segment, the coating robot runs at the current running speed, utilizing the residual pressure in the pipeline system to complete the tail-end coating.

10. An automatic planning system for the oiling path of an oiling robot, characterized in that, It includes a processor and a memory, the processor being used to process instructions stored in the memory to implement an automatic oiling path planning method for an oiling robot as described in any one of claims 1-9.

Images