TP lead FPC automatic insertion force and posture coordination control method and device

Abstract

This invention relates to the field of posture control technology and discloses a method and device for automatic force-posture collaborative control of TP lead-wire FPC insertion. It identifies installation misalignment by constructing a touch memory map and establishes a visual compensation plan library. The initial visual positioning matrix is ​​fused with the compensation plan to generate a fused positioning matrix. The insertion process is divided into three stages: exploration, alignment, and insertion. In the exploration stage, the fused matrix guides the robotic arm to approach and records the actual posture in real time, comparing it with the theoretical contact posture to obtain the posture deviation. In the alignment stage, the optimal insertion angle is calculated based on this. In the insertion stage, insertion is completed at this angle, and force feedback closed-loop correction is initiated, instantly identifying and handling jamming. This invention uses force-touch information to compensate for visual uncertainty online, achieving vision-force control closed-loop self-calibration, continuously converging positioning errors, and ensuring accurate insertion even in visually degraded scenarios such as reflective surfaces and confined spaces, thus improving FPC assembly yield and cycle stability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pose control, and more particularly, to a TP lead FPC automatic insertion force-pose collaborative control method and device. BACKGROUND

[0002] In electronic devices such as smart phones, car machines, tablet computers, etc., the TP (touch panel) lead FPC (flexible printed circuit board) is the core component for realizing touch signal transmission, and the precise insertion of the TP lead FPC and the connector directly determines the stability and reliability of the touch function of the device. With the rapid development of electronic devices towards miniaturization and integration, the FPC assembly space is increasingly compact, and higher requirements are put forward for the precision, stability and anti-interference ability of the automatic insertion process.

[0003] A Chinese patent with the publication number CN106877113B discloses a method and system for controlling FPC connector pins. By setting a main shaft motor, a material stirring motor and a stepping motor, within the rotation period of the main shaft motor, the material stirring motor is controlled to take the pins from the material belt according to the real-time rotation angle, and the stepping motor drives the FPC connector to step to the next pin insertion position. The timing of the motor is coordinated to realize accurate and rapid control of the pin insertion process. This scheme focuses on the optimization of the motor action logic of the FPC connector pin insertion process, and the core goal is to improve the pin insertion efficiency and pin assembly precision, but it does not involve the visual positioning precision fluctuation problem in the FPC and connector insertion stage, and does not establish a correlation compensation mechanism between the force control signal and the visual positioning deviation.

[0004] A Chinese patent with the publication number CN109041451B provides an assembly method of an FPC assembly and an FPC assembly. The spacing of multiple connectors is fixed by a positioning cap, and the connectors are synchronously attached to the FPC pre-coated with tin paste. After the tin paste is hot-melted and solidified, the positioning cap is removed, and the FPC with the fixed connectors is correspondingly pressed into the plastic part mounting hole. This scheme focuses on solving the spacing consistency problem in the FPC and connector attachment stage, and ensures the connector assembly position precision through mechanical positioning structure, but does not consider the visual positioning error caused by the narrow assembly space and the reflection of surrounding metal parts in the subsequent FPC and external connector insertion, and does not design a force control correction strategy for such error.

[0005] The aforementioned existing technologies and current mainstream FPC automatic mating solutions have failed to address the cascading problem of "positioning error - force control misalignment - mating failure" in scenarios of visual degradation. In conditions such as inside a vehicle's infotainment system (where space is limited and multiple components obstruct the view) and around a smartphone motherboard (where the connector is surrounded by a metal shield, easily causing glare), the host computer's visual measurement is prone to occasional offset errors due to limited field of view and light interference, resulting in a deviation between the output initial pose and the actual pose of the FPC connector. Due to the lack of a physical reference based on force-touch characteristics, the force control mechanism cannot recognize this visual offset. If the visual offset causes the FPC end to align with the connector edge instead of the mating interface, the force control mechanism will mistakenly interpret the minute impact force generated by edge contact as normal mating. If the contact signal is not received and the FPC continues to advance along the preset trajectory, it will trigger a hard collision, causing the FPC pins to bend or the connector interface to deform. If the visual deviation causes the FPC end to be horizontally / vertically skewed to the interface, the force control link will judge it as "virtual contact" because the contact force does not reach the expected threshold, and repeated advancement will still fail to complete effective insertion. Even if the FPC is forced into the interface, the jamming caused by the initial posture deviation will cause the insertion resistance to increase sharply. The existing technology lacks a closed-loop adjustment strategy that combines the "peak value of the micro-impact force at the moment of initial contact (force control synapse signal)" with the posture deviation, and cannot accurately correct the insertion angle to eliminate jamming. This ultimately leads to a decrease in FPC assembly yield, excessive fluctuation in production cycle, and even increased material loss costs due to component damage. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of existing technologies, this invention provides a method and apparatus for automatic insertion force-position coordination control of TP leaded FPCs. By constructing a touch memory map to identify FPC connector installation misalignment and establishing a visual compensation pre-setup library, a fused positioning matrix is ​​generated by fusing the initial visual positioning matrix. Combined with segmented force-position coordination control, a vision-force control closed-loop self-calibration is achieved, effectively eliminating visual positioning errors in confined / reflective environments and ensuring accurate FPC insertion on the first attempt. A force feedback closed-loop correction mechanism is activated during the insertion segment to instantly identify and address jamming points, significantly improving FPC assembly yield and production cycle stability, and solving the insertion failure problem caused by visual degradation in traditional solutions.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The automatic insertion force and pose coordination control method for TP lead FPC includes:

[0009] Construct the touch memory pattern of the target FPC connector;

[0010] Based on touch memory maps, identify the installation misalignment status of the target FPC connector;

[0011] Establish a visual compensation plan library based on the installation misalignment status of the target FPC connector;

[0012] Obtain the initial positioning matrix of the target FPC connector, and generate a fused positioning matrix by combining it with the visual compensation plan library;

[0013] The insertion process is divided into an exploration segment, an alignment segment, and an insertion segment. In the exploration segment, a fusion positioning matrix is ​​used to guide the robotic arm forward, the actual pose of the robotic arm end effector is recorded, the theoretical contact pose of the robotic arm end effector is obtained, and the pose deviation between the actual pose and the theoretical contact pose is calculated. In the alignment segment, the optimal insertion angle is determined based on the pose deviation. In the insertion segment, the insertion is performed using the optimal insertion angle, and a closed-loop correction mechanism based on force feedback is constructed to identify and handle jamming points.

[0014] The method for constructing the touch memory map of the target FPC connector includes:

[0015] The installation status parameters of the target FPC connector are obtained. Based on the installation status parameters, the force sensor at the end of the robotic arm is controlled to approach the target FPC connector from three initial test angles. The force-touch feature triplets of each initial test angle are recorded to construct the touch memory map of the target FPC connector. The force-touch feature triplets include the contact position coordinates, the peak contact force, and the vibration characteristic frequency.

[0016] The time when the force contact feature triplet for each initial test angle is recorded is when the force sensor detects the initial contact force.

[0017] The criterion for detecting initial contact force is: the real-time force value collected by the force sensor exceeds the force value baseline in the non-contact state for 5 consecutive sampling points.

[0018] The method for identifying the installation misalignment state of the target FPC connector includes:

[0019] Based on the touch memory map, the three initial test angles are expanded into nine fine test angles. An unloaded pre-touch test is performed on each fine test angle, and the expanded force touch feature dataset of each fine test angle is recorded. The installation misalignment state of the target FPC connector is identified based on the expanded force touch feature dataset.

[0020] The extended force-contact feature dataset includes the coordinates of the no-load contact position, the peak value of the no-load contact force, and the characteristic frequency of the no-load vibration at each fine test angle.

[0021] The method for identifying the installation misalignment state of the target FPC connector based on the extended force contact feature dataset includes:

[0022] The system identifies whether there is installation misalignment based on the extended force contact feature dataset, determines the direction of installation misalignment, and calculates the misalignment angle of the installation misalignment direction. The installation misalignment direction and the corresponding misalignment angle form the installation misalignment state of the target FPC connector.

[0023] The installation misalignment direction includes horizontal misalignment and vertical misalignment;

[0024] The nine precision test angles include one straight-ahead angle, two left-leaning angles in the horizontal direction, two right-leaning angles in the horizontal direction, two upward-tilting angles in the vertical direction, and two downward-tilting angles in the vertical direction.

[0025] The method for identifying horizontal skew includes: calculating the average of the peak unloaded contact forces corresponding to two left skew angles in the horizontal direction, as the average contact force on the left side; calculating the average of the peak unloaded contact forces corresponding to two right skew angles in the horizontal direction, as the average contact force on the right side; calculating the difference ΔF between the average contact force on the left side and the average contact force on the right side; and determining the horizontal skew threshold ΔF when the absolute value of ΔF is greater than the horizontal skew threshold ΔF. threshold At that time, it was determined that there was a horizontal deviation.

[0026] The method for obtaining the initial positioning matrix of the target FPC connector includes:

[0027] The original field-of-view image of the target FPC connector in the image coordinate system is obtained by the vision measurement component, and the initial positioning matrix of the target FPC connector is obtained by analyzing the original field-of-view image.

[0028] The method for analyzing the original field-of-view image includes: performing target segmentation on the original field-of-view image in the image coordinate system to obtain a segmented connector image; performing feature detection on the segmented connector image to obtain the coordinates of the target FPC connector feature points in the image coordinate system; mapping the coordinates of the target FPC connector feature points in the image coordinate system to the actual physical coordinates of the feature points in the robotic arm base coordinate system; and performing pose calculation of the target FPC connector based on the actual physical coordinates of the feature points in the robotic arm base coordinate system to obtain an initial positioning matrix.

[0029] The method for generating the fused positioning matrix includes:

[0030] The image sharpness index of the original field of view image is calculated, the visual confidence level is evaluated based on the image sharpness index, the perception weights of the visual measurement component and the force sensor are dynamically adjusted based on the visual confidence level, and a fusion positioning matrix is ​​generated based on the adjusted perception weights and the initial positioning matrix, combined with the visual compensation plan library.

[0031] The method for determining the optimal insertion angle includes:

[0032] In the alignment segment, the pose deviation is decomposed into horizontal, vertical and rotational components, and it is determined whether each component exceeds the corresponding allowable deviation threshold.

[0033] For components exceeding the corresponding allowable deviation threshold, perform a swing test in the corresponding plane and record the contact force sequence at different swing angles;

[0034] Analyze the changing trend of the contact force sequence and determine the optimal adjustment angle for the corresponding component direction using the gradient descent method;

[0035] By combining the optimal adjustment angles of each component direction, the optimal insertion angle in three-dimensional space is generated.

[0036] A TP lead FPC automatic insertion force-position coordination control device, used to implement the above-mentioned TP lead FPC automatic insertion force-position coordination control method, the device comprising:

[0037] Touch memory map construction module: used to construct the touch memory map of the target FPC connector;

[0038] Installation Misalignment Recognition Module: Based on touch memory maps, it identifies the installation misalignment status of the target FPC connector;

[0039] Compensation plan library generation module: used to create a visual compensation plan library based on the installation misalignment status of the target FPC connector;

[0040] Fusion positioning matrix generation module: used to obtain the initial positioning matrix of the target FPC connector, and generate the fusion positioning matrix by combining it with the visual compensation plan library;

[0041] Force-position coupling control module: The insertion process is divided into exploration, alignment and insertion segments. In the exploration segment, the fused positioning matrix guides the robot arm forward, records the actual pose of the robot arm end effector, obtains the theoretical contact pose of the robot arm end effector, and calculates the pose deviation between the actual pose and the theoretical contact pose. In the alignment segment, the optimal insertion angle is determined based on the pose deviation. In the insertion segment, the optimal insertion angle is used to perform insertion, and a closed-loop correction mechanism based on force feedback is constructed to identify and handle jamming points.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] This invention constructs a touch memory map of the target FPC connector to accurately identify its installation misalignment state, and establishes a visual compensation plan library based on this map. This provides targeted compensation for visual positioning and can effectively address occasional offset errors in upper computer visual measurement under confined spaces or reflective conditions. By fusing the initial visual positioning matrix with the visual compensation plan library to generate a fused positioning matrix, combined with the three-stage refined control of the insertion process—exploration, alignment, and insertion—the exploration stage guides the robotic arm forward through the fused positioning matrix and calculates the deviation between the actual pose and the theoretical contact pose. The alignment stage determines the optimal insertion angle based on this deviation. The insertion stage executes the insertion at the optimal angle and initiates a force feedback closed-loop correction mechanism. This ensures that the force-controlled synaptic signal and the pose deviation given by vision form an effective closed-loop coordination, achieving automatic online self-calibration. This gradually eliminates visual errors, ensures the consistency of contact point detection in the force-controlled process, and ultimately ensures the accuracy and stability of the FPC insertion process. This effectively improves the FPC assembly yield and cycle stability, avoids insertion failures caused by visual offset, and improves overall assembly efficiency and reliability. Attached Figure Description

[0044] To more clearly illustrate the technical solutions 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.

[0045] Figure 1 This is a flowchart of a method for automatic insertion force and pose coordination control of TP lead FPC provided in an embodiment of the present invention;

[0046] Figure 2 A flowchart illustrating the principle of analyzing the original field-of-view image provided in this embodiment of the invention;

[0047] Figure 3 This is a schematic diagram illustrating the principle of dividing the insertion process into an exploration segment, an alignment segment, and an insertion segment, as provided in an embodiment of the present invention.

[0048] Figure 4 A flowchart illustrating a method for determining the optimal insertion angle provided in an embodiment of the present invention;

[0049] Figure 5 This is a functional block diagram of an automatic insertion force and posture coordination control device for TP lead FPC provided in an embodiment of the present invention. Detailed Implementation

[0050] 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.

[0051] Example 1

[0052] Please see Figure 1 As shown, this embodiment provides a method for automatic insertion force-pose coordination control of TP lead FPC, including:

[0053] Step S10: Construct a touch memory map of the target FPC connector; based on the touch memory map, identify the installation misalignment state of the target FPC connector; establish a visual compensation plan library based on the installation misalignment state of the target FPC connector; obtain the initial positioning matrix of the target FPC connector; and generate a fusion positioning matrix by combining the visual compensation plan library.

[0054] Further, step S10 includes:

[0055] Step S11: Obtain the installation status parameters of the target FPC connector. Based on the installation status parameters, control the force sensor at the end of the robotic arm to approach the target FPC connector from three initial test angles. Record the force-touch feature triplets at each initial test angle to construct the touch memory map of the target FPC connector. The force-touch feature triplets include contact position coordinates, peak contact force, and vibration characteristic frequency.

[0056] Installation status parameters include the interface width W, insertion depth D, and tilt tolerance angle θ. These parameters are obtained through a combination of preliminary mechanical measurements and design parameter verification: the interface width W is measured three times along the horizontal centerline of the interface using high-precision calipers, and the average value is taken to eliminate random errors. The interface width reflects the physical boundary of the connector in the horizontal direction. The insertion depth D is measured using a depth gauge to measure the vertical distance from the bottom surface of the connector interface to the mounting surface, and the average value is also taken from three measurements, characterizing the effective depth range during insertion. The tilt tolerance angle θ is measured using an angle meter to determine the maximum allowable tilt angle between the connector body and the mounting reference surface, based on the mechanical tolerance parameters in the connector design manual, reflecting the connector's tolerance to installation posture. These parameters need to be stored in the system database as the underlying physical constraints for subsequent detection and insertion control.

[0057] The selection of the three initial test angles is determined based on the interface width W in the installation status parameters, including the front angle, the left deflection angle α, and the front angle. left and right deflection angle α rightThe forward angle is the initial positioning direction of the robotic arm, collinear with the centerline of the connector interface; the leftward angle α... left and right deflection angle α right The angle value must be chosen to ensure that the detection point covers more than 80% of the interface width W. For example, when W is large, the angle value should be slightly larger, and when W is small, the angle value should be smaller to avoid the detection point exceeding the interface edge and causing accidental contact with non-target structures. The force sensor at the end of the robotic arm is controlled to detect at three initial test angles with a detection speed V. probe When the force sensor detects the initial contact force F as it approaches the target FPC connector... initial At that time, record the contact position coordinates P under the initial test angle. contact Peak contact force F peak The characteristic frequency f of vibration at the instant of contact vibration The contact position coordinates P contact Peak contact force F peak The characteristic frequency f of vibration at the instant of contact vibration This constitutes a force-contact characteristic triplet. Detection velocity V probe The setting is based on the tilt tolerance angle θ and the connector material characteristics. The smaller θ is (the more sensitive the connector is to tilt) or the more brittle the material, the lower the V value. probe The lower the value, the better. The specific correspondence is determined through multiple preliminary experiments. For example, when θ ≤ 3°, V... probe When the speed range is 3 mm / s to 8 mm / s and θ > 3°, V probe The range is from 8 mm / s to 15 mm / s, thus balancing detection efficiency and contact safety.

[0058] The force sensor uses a strain gauge high-precision sensor with a sampling frequency set to 1kHz to ensure the capture of minute force signals at the moment of contact, detecting the initial contact force F. initial The criterion is: the real-time force value collected by the force sensor exceeds the baseline force value F in the non-contact state for 5 consecutive sampling points. baseline F baseline This determination method, obtained by averaging 100 force values ​​under non-contact conditions, effectively filters out false signals caused by environmental vibrations. When F is detected... initial Immediately upon activation, data recording is triggered, recording the contact position coordinates P at the initial test angle. contact Peak contact force F peak The characteristic frequency f of vibration at the instant of contact vibration Contact position coordinates P contact The peak contact force F is obtained through the encoder at the end of the robotic arm. peak The maximum value of the internal force data within 100ms after contact was obtained through real-time data caching and extreme value extraction algorithms; vibration characteristic frequency f vibrationBy performing a Fast Fourier Transform on the force signal within 50 ms of the contact instant, the frequency component with the largest amplitude after the transform is extracted. This parameter reflects the structural response characteristics at the moment of contact. The touch memory map is constructed using a structured dataset, storing the corresponding P values ​​according to the three initial test angles. contact F peak f vibration Simultaneously, it associates and stores the detection parameters for each initial test angle, such as V. probe wait.

[0059] Step S11 addresses the core issue of blind detection and signal misinterpretation caused by the lack of a stable physical reference in the force control stage when visual measurement fails due to reflections or confined spaces. The introduction of installation state parameters provides physical constraints on the detection range and speed, preventing the robotic arm from deviating from the target without a reference or causing structural damage due to improper speed, thus solving the problem of unfounded setting of traditional force control detection parameters. As a fundamental constraint, the installation state parameters ensure that subsequent detection parameter settings conform to the physical characteristics of the connector, significantly reducing the risk of accidental contact during detection. The selection of three initial test angles enables the construction of a force perception profile at the lowest detection cost, solving the problem of low efficiency in all-angle detection. The multi-dimensional recording of the force-touch feature triplet avoids the susceptibility of single force value parameters to interference, solving the problem of unreliable reference caused by incomplete force perception feature description.

[0060] Step S11 provides the force characteristic baseline and spatial anchor point for step S12. The peak contact force and vibration characteristic frequency at the moment of contact in S11 provide a reference for the abnormal data judgment in S12, and the contact position coordinates provide a basis for limiting the detection range in S12. Without this step, the nine fine test angles in step S12 will have no distribution basis, the detection speed of the nine fine test angles cannot be reasonably set, the collected data lacks a valid verification standard, resulting in the loss of reference for skew state identification, and the subsequent visual compensation plan library will also be unable to be established due to the lack of quantitative force perception basis. The entire force-position coordination control process will fall into a chaotic state without a benchmark.

[0061] Step S12: Based on the touch memory map, the three initial test angles are expanded into nine fine test angles. An unloaded pre-touch test is performed on each fine test angle, and the expanded force touch feature dataset of each fine test angle is recorded. The installation misalignment state of the target FPC connector is identified based on the expanded force touch feature dataset.

[0062] Further, step S12 includes:

[0063] Step S121: Extract the force-touch feature triplets of three initial test angles from the touch memory map. Based on the force-touch feature triplets and installation state parameters, determine the spatial distribution scheme of nine fine test angles. The nine fine test angles include one front angle, two left-leaning angles in the horizontal direction, two right-leaning angles in the horizontal direction, two upward-tilting angles in the vertical direction, and two downward-tilting angles in the vertical direction.

[0064] Step S122: Control the robotic arm to perform no-load pre-contact test according to the spatial distribution scheme of nine fine test angles, record the no-load contact position coordinates, no-load contact force peak value and no-load vibration characteristic frequency under each fine test angle, and form an extended force contact feature dataset;

[0065] Step S123: Identify whether there is installation misalignment based on the extended force contact feature dataset, determine the direction of installation misalignment, and calculate the misalignment angle of the installation misalignment direction. The installation misalignment direction and the corresponding misalignment angle form the installation misalignment state of the target FPC connector.

[0066] The extraction of the force-touch feature triplets is achieved through database queries, with a focus on extracting the contact position coordinates at the front angle as the spatial origin reference. The spatial distribution scheme of nine refined test angles is designed based on the feature differences reflected by the force-touch feature triplets: In the horizontal direction, the left-leaning angle is refined into α1 and α2, and the right-leaning angle into β1 and β2. The angle interval is determined according to the interface width W; the larger W is, the smaller the interval. For example, the interval is 2° when W ≥ 15mm and 3° when W < 15mm, ensuring coverage of horizontal force gradient changes. In the vertical direction, upward tilt angles γ1 and γ2 and downward tilt angles δ1 and δ2 are added. The angle range is determined according to the tilt tolerance angle θ, not exceeding 50% of θ to avoid exceeding the connector's mechanical tolerance range. This distribution scheme ensures increased test density in force-sensitive areas and reduced redundant testing in non-sensitive areas.

[0067] The execution process of the no-load pre-contact test is as follows: the robotic arm moves forward at a preset speed V_test at each fine test angle, and the force sensor continuously samples at a frequency of 1kHz. When the force value at 5 consecutive sampling points exceeds F, the test is initiated. baseline When the force baseline of step S11 is reached, it is determined to be the initial contact. Immediately stop advancing and record the contact position coordinates and the peak value of the no-load contact force F at the time of initial contact. empty The vibration characteristic frequency and data are recorded and associated with the corresponding fine test angles to form an extended force-contact feature dataset containing 9 sets of triplets. V_test is set to the detection velocity V in step S11. probe Half of the force is based on the fact that S11 has already acquired the general force characteristics. Reducing the speed allows the force sensor to more accurately capture the subtle changes in force at the moment of contact, reducing the force overshoot error caused by high-speed impact.

[0068] The extended force contact feature dataset is classified. The peak values ​​of the unloaded contact force corresponding to α1, α2, β1, and β2 in the horizontal direction are classified into the horizontal force value group, and the peak values ​​of the unloaded contact force corresponding to γ1 and γ2 in the upward direction and δ1 and δ2 in the downward direction are classified into the vertical force value group.

[0069] Installation misalignment includes horizontal and vertical misalignment. Horizontal misalignment includes leftward and rightward misalignment, and vertical misalignment includes upward and downward misalignment. The method for identifying horizontal misalignment includes calculating the average of the peak values ​​of the unloaded contact force corresponding to the two leftward misalignment angles α1 and α2, which is used as the average left-side contact force F. left,avg Calculate the average of the peak values ​​of the unloaded contact force corresponding to the two right-hand deflection angles β1 and β2 in the horizontal direction, and use this as the average contact force F on the right side. right,avg ; Calculate the difference ΔF between the average contact force on the left and the average contact force on the right. When the absolute value of ΔF is greater than the horizontal skew threshold ΔF threshold When ΔF is positive, a horizontal deviation is determined; a positive ΔF indicates a leftward deviation, while a negative ΔF indicates a rightward deviation. threshold The setting is based on the peak contact force F in step S11. peak The mean is determined to be 15% to 25% of the mean. According to F... left,avg With F right,avg The ratio is used to calculate the horizontal deflection angle θ. skew θ skew The specific calculation formula is as follows Calculate, where k skew The angle conversion coefficient was obtained through preliminary calibration tests. Connectors with different known skew angles were selected, and the corresponding F values ​​were measured. left,avg With F right,avg The ratio was determined by fitting the relationship between the logarithm of the ratio and the skew angle using linear regression, thus establishing k. skew The value ranges from 1.2 to 2.5. The logic of this formula lies in the logarithmic relationship between the deflection angle and the force value. When the left and right force values ​​are equal, the ratio is 1, and the logarithm is 0. skew A value of 0 indicates an unskewed state; as the degree of skew increases, the deviation of the ratio from 1 increases, and θ... skew The force increases linearly, consistent with physical laws. The same logic is used for vertical skew identification, calculating the average of the peak unloaded contact force corresponding to the two upward tilt angles γ1 and γ2 in the vertical direction, which is taken as the average upward contact force F. up,avg Calculate the average of the peak values ​​of the unloaded contact force corresponding to the two downward angles δ1 and δ2 in the vertical direction, and use this as the average downward contact force F. down,avg ; Calculate F up,avg With F down,avgThe difference ΔFz, when the absolute value of ΔFz is greater than the vertical skew threshold ΔFz threshold (Settings based on the same criteria as ΔF) threshold When ΔFz is positive, it is determined that there is a vertical deviation. If ΔFz is positive, the deviation is upward; if it is negative, the deviation is downward. The deviation angle of the vertical deviation is calculated, and finally, the installation deviation state is formed by combining the deviation angles of the horizontal deviation, the horizontal deviation, and the vertical deviation.

[0070] Step S12 addresses the core issue of the initial three-angle detection failing to cover three-dimensional spatial force characteristics and the difficulty in quantifying connector installation misalignment. Step S121's spatial distribution scheme resolves the problem of test redundancy or feature omission caused by the lack of a basis for distributing fine test angles; the nine-angle distribution scheme achieves omnidirectional force detection in three-dimensional space, compensating for the limitation of the initial three angles only covering the horizontal direction, enabling the identification of vertical misalignment, and expanding the dimensional completeness of force characteristics. Step S122's low-speed no-load test solves the problem of high-speed detection failing to capture subtle force changes; Step S123's difference and ratio calculation solves the problem of the inability to accurately quantify the misalignment state. Without step S12, vertical misalignment cannot be identified using only three initial test angles, and the misalignment angle cannot be quantified. This renders the visual compensation plan library in S13 a vague framework due to the lack of precise force basis, and the fusion positioning matrix in step S14 unable to correct visual errors due to the lack of quantified compensation parameters. The closed-loop coordination between the force control and vision stages will lose its core bridge, ultimately failing to achieve self-calibration.

[0071] Step S13: Establish a visual compensation plan library based on the installation misalignment status of the target FPC connector;

[0072] When a leftward skew is detected, the compensation angle δ is adjusted to the right. right =θ skew ×k comp The compensation logic for left and right skew is symmetrical, and the compensation calculation for upward and downward skew follows the same logic. Compensation coefficient k comp The method for determining k is as follows: Select connectors with different known skew angles, record the minimum compensation angle that eliminates the visual positioning error through multiple insertion tests, and use linear regression to fit the relationship between the skew angle and the compensation angle to determine k. comp The range of values ​​should be such that the compensation angle is slightly smaller than the skew angle to avoid overcompensation, while ensuring the effectiveness of the compensation.

[0073] The visual compensation plan library is constructed using a key-value pair structured storage, with "installation tilt state combination" as the key and "compensation angle set (δ)" as the value. right / δ left / δ up / δ downThe value is δ. left To compensate for the angle to the left, δ up To compensate for the upward angle, δ down For downward compensation angle. "Installation Misalignment Combination" refers to the comprehensive attitude description of the FPC connector in three-dimensional space, encompassing both horizontal and vertical misalignment. It must include two core elements: "misalignment direction" and "corresponding misalignment angle." Because assembly errors in confined spaces often result in connector misalignment not only in a single plane (horizontal or vertical), but also in complex deviations across multiple three-dimensional directions, such as "horizontal misalignment to the left while vertical misalignment to the upward" or "horizontal misalignment to the right only," its attitude must be fully defined through a "combination" method. This combination is the core index of the visual compensation plan library, directly associated with the corresponding multi-directional compensation parameters.

[0074] For example, assume the compensation coefficient k comp =1.5, Table 1 is an example table of key-value pairs:

[0075] Table 1. Example table of key-value pairs

[0076]

[0077] Step S13 addresses the core issue that existing technologies cannot quantify and compensate for visual positioning errors using force information, leading to blind setting of compensation parameters and uncontrollable correction effects. The correlation calculation between the compensation angle and k_comp allows for precise matching of skew correction to the actual deviation, resolving the lack of quantitative basis for skew compensation. The quantification calculation of the compensation angle significantly improves the angle error correction effect of visual positioning, greatly reducing the overcompensation risk compared to traditional empirical compensation. The structured storage of the pre-set library allows for rapid retrieval of compensation parameters, solving the problem of inefficient switching of compensation strategies under multiple skew conditions.

[0078] Step S14: Obtain the original field-of-view image of the target FPC connector in the image coordinate system through the vision measurement component; analyze the original field-of-view image to obtain the initial positioning matrix of the target FPC connector; and calculate the image sharpness index of the original field-of-view image.

[0079] The acquisition of the raw field-of-view image is achieved through a combination of an industrial camera and lens. During the acquisition process, the industrial camera automatically generates an image coordinate system—a two-dimensional pixel coordinate system with the upper left corner of the camera's image sensor as the origin, the horizontal axis as the U-axis, and the vertical axis as the V-axis, with the unit being pixels. This system describes the position of each pixel in the image and serves as the fundamental reference frame for the industrial camera and subsequent image processing algorithms (such as filtering and feature extraction). Please refer to [link to relevant documentation]. Figure 2 As shown, the image analysis process proceeds according to the progressive logic of "preprocessing - feature point extraction - coordinate transformation - pose calculation". Each step is based on a clear coordinate system relationship, as detailed below:

[0080] In the preprocessing stage, noise reduction and target segmentation are performed on the original field-of-view image in the image coordinate system: Gaussian filtering algorithm is used to filter out image noise caused by environmental interference. The size of the filter kernel is determined by calculating the standard deviation of image gray level. The standard deviation of gray level reflects the noise intensity. The larger the standard deviation, the more obvious the noise. A larger filter kernel is selected accordingly to enhance the noise reduction effect. After filtering, an adaptive threshold segmentation algorithm is used to separate the FPC connector from the background region to obtain the segmented connector image. The threshold calculation is based on the mean and variance of gray level in the local region of the image. By dynamically adjusting the segmentation threshold, it is ensured that the connector target region in the image coordinate system can be accurately extracted under different lighting conditions, thus limiting the effective range for subsequent feature point extraction.

[0081] The feature point extraction stage involves key feature localization based on the image coordinate system: The SIFT algorithm is used to detect features in the segmented connector image, obtaining the coordinates of the target FPC connector feature points in the image coordinate system. Corner points and inflection points of the interface edge are selected as key feature points—the pixel coordinates of these feature points in the image coordinate system can be directly extracted by the algorithm, and their corresponding physical locations are clearly marked in the FPC connector design manual, i.e., the theoretical three-dimensional coordinates of the feature points in the robotic arm's base coordinate system. The robotic arm's base coordinate system is a three-dimensional spatial coordinate system with a preset point on the robotic arm's fixed base as the origin, containing X (horizontal), Y (horizontal and vertical), and Z (vertical) axes, with units of length (e.g., millimeters). It is the core reference system for the robotic arm's motion control. The extracted feature points need to be filtered for mismatches using the RANSAC algorithm: the algorithm fits the spatial distribution pattern of the feature points, filtering out false feature points caused by image noise and edge blurring, ensuring that the pixel coordinates of the remaining feature point set in the image coordinate system correspond one-to-one with the theoretical physical coordinates in the robotic arm's base coordinate system.

[0082] In the coordinate conversion stage, a mapping relationship is established between the image coordinate system and the robot arm base coordinate system. The coordinates of the target FPC connector feature points in the image coordinate system are mapped to the actual physical coordinates of the feature points in the robot arm base coordinate system. The specific implementation process is as follows: the calibration plate is fixed on the mounting reference surface of the FPC connector, and the industrial camera is controlled to take images of the calibration plate from multiple different poses. The pixel coordinates of the corner points of the calibration plate in the image coordinate system are extracted in each image, and the three-dimensional coordinates of these corner points in the robot arm base coordinate system are known quantities (obtained through previous precise measurements). Based on multiple sets of "pixel coordinate-physical coordinate" corresponding data, the perspective transformation algorithm (such as the pinhole model of a monocular camera) is used to calculate the mapping matrix Mmap. The mapping matrix Mmap contains spatial transformation parameters such as translation and rotation, which can convert the pixel coordinates of any feature point in the image coordinate system into three-dimensional physical coordinates in the robot arm base coordinate system, realizing the cross-system conversion from "two-dimensional image information" to "three-dimensional spatial information".

[0083] In the attitude calculation stage, the pose of the target FPC connector is calculated based on the actual physical coordinates of the feature points in the robot arm's base coordinate system. The actual physical coordinates of the feature points in the robot arm's base coordinate system obtained after coordinate transformation are compared with the standard coordinates of the feature points in the design manual, i.e., the theoretical three-dimensional coordinates in the robot arm's base coordinate system. The pose transformation matrix is ​​then fitted using the least squares method; this matrix is ​​the initial positioning matrix T. vision This includes the spatial position parameters (X, Y, Z coordinates) and attitude parameters of the FPC connector in the robot arm's base coordinate system, such as tilt angle and rotation angle. These parameters can be directly output to the robot arm control system, providing a pose reference for subsequent motion control. The robot arm's base coordinate system is preset at the factory and maintains a fixed relative position with the installation environment. The standard coordinates are theoretical coordinates defined during the FPC connector design phase, and their relative relationship with the robot arm's base coordinate system can be determined through pre-installation calibration to ensure that the attitude calculation results meet the robot arm's control requirements.

[0084] Image sharpness index I clarity The calculation employs the Laplacian operator method, performing Laplacian convolution on the preprocessed image and extracting the gray-level variance of the convolution result as I. clarity The larger the variance, the clearer the image edges and the more reliable the visual positioning.

[0085] Step S14 addresses the core issue that the generation process of the visual positioning matrix in existing technologies is unclear, and the reliability of visual measurement results cannot be quantitatively assessed, leading to a lack of reliable visual input and quality judgment basis for subsequent collaborative control. Image preprocessing ensures that feature point extraction is not affected by noise, solving the problem of mis-extracted feature points; the combination of SIFT and RANSAC algorithms ensures the robustness of feature point matching, solving the problem of feature point loss caused by occlusion or reflection; perspective transformation realizes cross-coordinate conversion, solving the problem of the inability to correlate image coordinates with robotic arm coordinates; the Laplacian operator method quantifies sharpness, solving the problem of the inability to judge visual reliability. The combination of preprocessing and feature point extraction significantly improves the accuracy of feature point matching and effectively controls positioning errors; perspective transformation coordinate conversion achieves accurate conversion from image coordinates to physical coordinates; I clarity The quantization significantly improves the accuracy of visual failure identification and avoids incorporating low-quality visual data into positioning calculations.

[0086] Step S15: Assess visual confidence based on image clarity index, dynamically adjust the perception weights of visual measurement component and force sensor based on visual confidence, and generate fusion positioning matrix based on the adjusted perception weights and initial positioning matrix, combined with visual compensation plan library.

[0087] Visual confidence level C vision The assessment was conducted by I clarity Based on this as the core criterion, a sharpness threshold I is set.min I min The determination method is as follows: Multiple sets of tests are conducted under different lighting, reflection, and occlusion conditions. The critical value of the image sharpness index when the visual positioning error exceeds the preset accuracy requirement is recorded. The average of multiple critical values ​​is taken as I. min Exemplary I min This can be set to a value corresponding to a grayscale variance between 100 and 200. Perceptual weights include visual weight W. vision With force perception weight W force The sum of these two is 1, and their function is to quantify the contribution ratio of visual and force information in localization. When I clarity Less than I min When visual measurement is affected by environmental interference, C vision In a low confidence state, C vision Follow me clarity Decrease and decrease synchronously, for example, let C vision =I clarity / I min W force The formula for calculating W is force =(C min -C vision ) / C min W vision =C vision / C min The design of this formula follows linear dynamic adaptation logic, when C vision Infinitely close to C min At that time, vision had just reached the failure boundary and was slightly unreliable. force =(C min -C vision ) / C min Approaching 0, W vision Approaching 1, at this point, vision remains primary with slight assistance from force perception, avoiding over-reliance on force perception which could lead to decreased efficiency; when C vision Significantly lower than C min This means severe visual impairment, such as extremely low clarity under strong glare. force Will follow C vision W increases linearly as it decreases. vision Then, the synchronous linear decrease achieves precise adaptation where "the more severe the visual failure, the more dominant the force compensation," ensuring that the force-controlled synaptic signal can offset visual pose deviations to the greatest extent. The "force-controlled synaptic signal" refers to the force-touch characteristic signal that reflects the contact state, collected by the force sensor at the end of the robotic arm at the moment of initial contact with the target FPC connector. Its core includes the peak value of the minute impact force at the initial contact, along with force-touch information such as contact position coordinates and the vibration characteristic frequency at the moment of contact. When I... clarity Greater than or equal to I min At that time, C visionIn a high confidence state, I clarity The standard is met; the visual positioning error is within the preset accuracy, and the visual information is completely reliable. Therefore, C... vision Take the maximum value of 1, at which point W force The value is between 0.1 and 0.3, W vision The value is between 0.7 and 0.9, and the specific value is determined through preliminary testing based on the connector material and mating accuracy requirements.

[0088] C min It is the critical boundary value of visual confidence, representing the confidence threshold at which a visual measurement switches from a "high confidence state" to a "low confidence state." Its physical meaning is that when the image sharpness index I... clarity This is exactly equal to the critical sharpness at which the visual positioning error exceeds the preset accuracy, i.e., the sharpness threshold I. min At that time, the corresponding visual confidence level C vision Value, i.e., C min This is the minimum acceptable level for visual reliability, when the actual C vision ≥C min When C is within acceptable limits, the visual positioning error is within acceptable limits; when C is within acceptable limits, the visual positioning error is within acceptable limits. vision <C min At this point, visual positioning is no longer sufficient to meet the accuracy requirements, and force compensation must be relied upon. The process of integrating the visual compensation plan library is as follows: The installation skew state combination index in the visual compensation plan library is used to query the key-value pairs in the library, the corresponding compensation angle set is called, and the compensation angles are converted into attitude correction parameters in the robot arm's base coordinate system, forming the compensation matrix T. comp T comp Structure and T vision Consistent, both containing translation and rotation parameters. Fusion positioning matrix T fusion The generating formula is T fusion =T vision ×W vision +T comp ×W force This formula achieves parameter fusion for visual positioning and force compensation through weight allocation. min with I min This forms a chain of correlation between "sharpness boundary → confidence boundary," where the image sharpness index I... clarity Exactly equal to I min At that time, the corresponding visual confidence level C vision Value as C min The core of this design logic is to use the actual accuracy requirements of visual positioning as the anchor point, allowing C... min with I min Together, they constitute the "dual judgment boundary" of visual reliability, avoiding the disconnect between the subjectively set confidence threshold and the actual positioning effect.

[0089] High confidence levels use a "fixed numerical range" (W) vision 0.7-0.9, W force In the low-confidence state (0.1-0.3), "dynamic formula calculation" is used, which is essentially a "differentiated adaptation of efficiency and accuracy": In the high-confidence state, the visual positioning error already meets the insertion accuracy requirements. At this time, there is no need for complex dynamic calculation. Fixed high visual weights can directly take vision as the lead to ensure insertion efficiency and avoid process delays caused by frequent force intervention. Formula calculation is used in the low-confidence state because there is a continuously changing "gradient difference" in visual reliability. For example, slight reflection may cause C to be affected. vision =0.8C min Severe occlusion leads to C vision =0.2C min A fixed value cannot accommodate different degrees of visual impairment; if a fixed W is used... force For example, a value of 0.5 would lead to over-reliance on force perception when visual impairment is slight, while the force perception weight would be insufficient when visual impairment is severe, thus failing to guarantee accuracy. The formula, however, allows W to... force Follow C vision The continuous linear adjustment precisely matches each degree of failure, ensuring that force compensation is "put into operation as needed".

[0090] The visual compensation plan library for step S13 is T. comp Provide quantization parameters, I in step S14 clarity To provide a basis for adjusting perception weights, step S15 addresses the core issues including: the positioning contradictions caused by the lack of quantitative coordination between vision and force perception in traditional methods, and the problem of low-quality visual data misleading control. Dynamic perception weight adjustment ensures that the system prioritizes vision to guarantee efficiency when vision is reliable, and prioritizes force perception compensation to guarantee accuracy when vision fails, avoiding the limitations of a single perception source. The index call of the visual compensation plan library enables rapid conversion of force perception information into visual correction parameters, solving the problem of discontinuity in the correlation between the two parameters. Without step S15, the visual compensation plan library in step S13 and the visual positioning data in step S14 cannot form a closed loop, the coordination relationship between force control synaptic signals and visual pose deviation is broken, the system cannot adapt the perception strategy according to environmental changes, resulting in decreased positioning accuracy in reflective or confined spaces, and even splice damage.

[0091] Step S10 addresses the core issues of distortion and misjudgment by the visual measurement components in confined spaces or under strong reflective conditions, leading to a lack of positional references and the self-calibration obstacle preventing the force-controlled synaptic signals and visual pose deviations from forming a closed-loop coordination. Step S10 is implemented through a progressive logic of "force reference construction - skew quantification recognition - compensation parameter generation - visual reliability assessment - multi-source information fusion": First, step S11 constructs a touch memory map, using a force-touch feature triplet independent of vision as a physical reference to solve the problem of reference inequality in reflective or confined spaces; then, step S12 expands the test angle to achieve three-dimensional quantification recognition of skew states, transforming physical states into calculable parameters; next, step S13 establishes a compensation plan library to convert force information into visual compensation parameters; subsequently, step S14 introduces an image sharpness index to assess visual reliability, providing a basis for perception weight allocation; finally, step S15 dynamically adjusts perception weights to generate a fusion positioning matrix, completing the closed-loop coordination of vision and force control. The construction of the touch memory map utilizes a structured dataset to store force-touch feature triples and associated detection parameters. This serves as a force-sensory benchmark throughout the entire process, providing a feature baseline for subsequent skew identification, a quantitative basis for compensation plans, and a data carrier for self-learning. Skew identification is achieved through multi-angle force value differences and gradient changes, overcoming the limitations of traditional qualitative judgment and enabling precise quantification of skew angles. The visual compensation plan library stores compensation angles categorized by skew state, transforming abstract force-sensory feedback into specific visual correction parameters. The initial positioning matrix is ​​obtained through image acquisition and positioning algorithm calculations by the visual measurement component, while image sharpness indicators are calculated using gradient operators. The combination of these two methods completes a quantitative assessment of visual reliability. The fused positioning matrix integrates the initial visual positioning and force-sensory compensation parameters based on dynamic perception weights, achieving complementary advantages from multiple information sources.

[0092] The construction of the touch memory map enables the system to maintain physical reference even when vision fails, significantly improving the reference loss situation compared to traditional pure vision positioning systems. Quantitative identification of skew states provides accurate input for compensation parameters, greatly improving angle compensation accuracy. The establishment of a visual compensation plan library enables parameterized correlation between force and vision, solving the problem of blindly setting compensation parameters in traditional collaborative control. The introduction of image sharpness indicators allows for quantifiable assessment of visual reliability, avoiding the inclusion of failed visual data in positioning calculations. The generation of the fusion positioning matrix enables dynamic adaptation of perception weights, prioritizing vision to ensure efficiency when visual confidence is high, and prioritizing force to ensure accuracy when visual confidence is low. The "force-vision" collaborative framework constructed in step S10 not only serves the current insertion task, but its generated touch memory map and compensation plan library can also serve as a self-learning basis. Through subsequent updates and optimizations in steps S24 and S25, the system's insertion accuracy for similar FPC connectors gradually improves with the number of uses, breaking through the limitation of traditional control methods where "one calibration only serves a single task". Step S10 provides the fusion positioning matrix as the initial reference for insertion in step S20, and the touch memory map as the basis for dynamic correction. The successful insertion samples and jamming data in step S20 feed back into step S10, enabling iterative optimization of the map and the pre-set library. Without step S10, the system will lack a physical reference for visual failure, the force control signal and visual deviation cannot be correlated, the compensation parameters have no basis for generation, and the fusion positioning has no computational basis. This would cause the segmented insertion in step S20 to proceed blindly due to the lack of a reliable initial pose, and the problem of inconsistent contact point detection in the force control link could not be solved, ultimately failing to achieve self-calibration of visual errors.

[0093] Step S20: The insertion process is divided into an exploration segment, an alignment segment, and an insertion segment. In the exploration segment, a fusion positioning matrix is ​​used to guide the robotic arm forward, the actual pose of the robotic arm end effector is recorded, the theoretical contact pose of the robotic arm end effector is obtained, and the pose deviation between the actual pose and the theoretical contact pose is calculated. In the alignment segment, the optimal insertion angle is determined based on the pose deviation. In the insertion segment, the insertion is performed using the optimal insertion angle, a closed-loop correction mechanism based on force feedback is constructed, and jamming points are identified and handled.

[0094] Further, step S20 includes:

[0095] Step S21: Divide the insertion process into an exploration segment, an alignment segment, and an insertion segment. In the exploration segment, use a fusion positioning matrix to guide the robot arm forward. When the force sensor detects the initial contact force, record the actual pose of the robot arm's end effector. Calculate the theoretical contact pose of the robot arm's end effector based on the fusion positioning matrix, and calculate the pose deviation between the actual pose and the theoretical contact pose.

[0096] Step S21 uses the fused positioning matrix T generated in step S15 fusionUsing the initial pose reference and the insertion depth D obtained in step S11 as the segmentation basis, precise initialization and deviation capture of the insertion process are achieved through extremely low-speed detection and pose quantization calculation. The fused positioning matrix T... fusion It includes the position parameters (X, Y, Z) and attitude parameters (tilt angle, rotation angle) in the robot arm's base coordinate system, T fusion It directly correlates with the motion trajectory of the robotic arm's end effector, providing precise guidance for the direction of movement during the exploration phase.

[0097] The segmentation logic of the mating process strictly relies on the mating depth D in the installation status parameters of the target FPC connector. The mating depth D is a physical reference reflecting the effective depth range of the connector interface. Please refer to [link / reference]. Figure 3 As shown, the exploration segment length is set as the product of D and the first coefficient k1. The value of k1 is determined based on the connector material and the flexibility characteristics of the FPC, and is usually 0.1. This setting ensures that the exploration segment only covers the insertion interface area, avoiding premature collision of the robotic arm with the connector body due to excessive segment length, or failure to capture the initial contact due to insufficient segment length. The alignment segment length is set as the product of D and the second coefficient k2, with k2 also set to 0.1. This length allows sufficient space to perform swing alignment actions. If the length is too short, the adjustment range will be limited; if it is too long, the insertion time will be increased. The insertion segment length is set as the product of D and the third coefficient k3, with k3 set to 0.8, serving as the length benchmark for the main insertion process. k1+k2+k3=1, ensuring that the end of the robotic arm can push the FPC to the predetermined insertion depth D_target, where D_target=D, to avoid substandard insertion or over-insertion that could damage the FPC.

[0098] The exploration segment's robotic arm forward speed is set to an extremely low V. explore V explore V_test is the product of V_test and the fourth coefficient k4. For example, k4 has a value of 0.1. explore The setting is based on the fact that V_test is already adapted to low speeds with a tilt tolerance angle θ, and further reduced to V explore This design minimizes the impact force at the moment of contact between the robotic arm's end effector and the connector, preventing FPC deformation or distortion of the contact force signal collected by the force sensor due to impact. It also ensures that the force sensor has sufficient time to capture minute force changes at the moment of contact. The force sensor at the robotic arm's end effector continuously monitors the contact force. When the initial contact force is detected, the robotic arm immediately stops moving forward, and the actual position P at this moment is recorded by the encoder at the robotic arm's end effector. real Theoretical contact pose P predict The calculation is based on the fused localization matrix T fusion Based on the kinematic model of the robotic arm, the forward kinematics algorithm in the robotic arm control system is first called to convert T... fusionThe attitude parameters (tilt angle, rotation angle) are converted into joint angles of the robotic arm's end effector. Then, through the mapping relationship between the joint angles and the end effector's spatial coordinates, the three-dimensional coordinates of the end effector reaching the theoretical contact point of the connector interface are calculated. These coordinates are P. predict During the calculation process, it is necessary to ensure that the parameters of the kinematic model match the actual structure of the robotic arm to avoid P being affected by model errors. predict The deviation from the theoretical position. The pose deviation ΔP is calculated as the vector difference between the three-dimensional coordinates in the robot arm's base coordinate system, i.e., ΔP = P real -P predict The X component of ΔP corresponds to the horizontal deviation ΔP_x, the Y component corresponds to the vertical deviation ΔP_y, the Z component corresponds to the vertical deviation ΔP_z, and the Y component corresponds to the forward and backward deviation. Since the insertion direction is fixed at the Y-axis, this component is usually small and can be ignored or included in the rotational deviation consideration. Each component of ΔP directly quantifies the combined influence of visual positioning error and force control contact deviation, providing a clear quantitative target for deviation correction in the subsequent step S22.

[0099] Step S21 employs a segmentation strategy based on the insertion depth D to ensure that the length of each segment has clear physical constraints, avoiding inaccurate contact capture caused by traditional empirical segmentation; extremely low speed V explore The settings reduce contact impact and solve the problems of FPC damage and force signal distortion caused by high-speed detection. The quantization calculation of ΔP transforms the abstract pose deviation into a directly processable three-dimensional vector, solving the problem of visual errors and the inability to correct inconsistencies in force control contact points. The ΔP output in step S21 directly provides the core input for the deviation component decomposition and swing alignment in step S22. Without step S21, step S22 will be unable to determine the alignment direction and amplitude due to the lack of quantized deviation data, resulting in blind execution of the alignment action, which may ultimately cause FPC insertion jamming or damage, and break the entire force-position coordination control process.

[0100] Step S22: In the alignment segment, the pose deviation is decomposed into horizontal, vertical and rotational components, and swing alignment is performed on each component to determine the optimal insertion angle.

[0101] Please see Figure 4 As shown, step S22 further includes:

[0102] Step S221: In the alignment segment, the pose deviation is decomposed into horizontal, vertical and rotational components, and it is determined whether each component exceeds the corresponding allowable deviation threshold.

[0103] Step S222: Perform a rocking test in the corresponding plane on the component that exceeds the corresponding allowable deviation threshold, and record the contact force sequence at different rocking angles;

[0104] Step S223: Analyze the changing trend of the contact force sequence, and determine the swing angle with the minimum contact force as the optimal adjustment angle for the corresponding component direction using the gradient descent method.

[0105] Step S224: Combine the optimal adjustment angles of each component direction to generate the optimal insertion angle in three-dimensional space.

[0106] Specifically, step S22 takes the pose deviation ΔP calculated in step S21 as input. Through deviation component decomposition, targeted sway testing, contact force sequence analysis, and multi-component angle synthesis, it achieves precise optimization of the insertion angle, providing the optimal attitude reference for the insertion segment. First, step S221 performs deviation component decomposition and threshold determination. This process, based on the spatial dimensional characteristics of the robotic arm's base coordinate system, decomposes the three-dimensional vector of ΔP into a horizontal component ΔP_x, a vertical component ΔP_z, and a rotational component ΔP_rot. ΔP_x, along the X-axis, corresponds to the left-right deviation of the connector interface; ΔP_z, along the Z-axis, corresponds to the up-down deviation of the connector interface; and ΔP_rot, rotating around the Y-axis, corresponds to the angular skew of the connector interface. During decomposition, a coordinate transformation algorithm is called to ensure that each component accurately reflects the deviation characteristics in different directions. The allowable deviation threshold is set based on the installation state parameters obtained in step S11, with the horizontal allowable deviation threshold ΔP... x,allow The value is set as the product of the interface width W and the sixth coefficient k6. Typically, k6 ranges from 0.05 to 0.1 to ensure that complex alignment is unnecessary when the horizontal deviation does not exceed 5% to 10% of the interface width, thus avoiding over-adjustment. The vertical tolerance threshold ΔP... z,allow The value is set to the product of the insertion depth D and the seventh coefficient k7, where k7 ranges from 0.05 to 0.1, consistent with the horizontal threshold logic, ensuring that the process is simplified when the vertical deviation is within an acceptable range; the rotational tolerance ΔP rot,allow The tilt tolerance angle θ is set as the product of the eighth coefficient k8, with k8 ranging from 0.3 to 0.5. This setting is based on the connector's tolerance to rotational deviations, avoiding unnecessary adjustments triggered by slight rotational deviations. During the judgment process, ΔP_x, ΔP_z, and ΔP_rot must be compared with their corresponding thresholds one by one to identify the deviation components exceeding the thresholds, thus determining the target direction for subsequent sway tests.

[0107] The sway test in step S222 is performed on components that exceed the corresponding allowable deviation threshold. Taking the horizontal sway test when the horizontal component ΔP_x exceeds the limit as an example, the sway center O is first determined, and point O is set as the actual pose P recorded in step S21. real The corresponding center point of the robotic arm's end effector is set to ensure that the swing motion unfolds around the actual contact point, avoiding data distortion due to swing center offset. The initial value γ of the swing angle is calculated using trigonometric functions. init , i.e. γinit =arctan(ΔP_x / L), where L is the vertical distance from the center point of the robotic arm's end effector to the connector interface, and γ init This ensures that the initial swing covers the angular range corresponding to the deviation, avoiding inefficient testing due to an initial angle that is too small or exceeding the physical range of the connector due to an initial angle that is too large. The robotic arm swings γ to the left along the swing center O. init The left swing angle γ_left is obtained and held for 0.5 to 1 second. Simultaneously, the contact force sequence collected by the force sensor is recorded. This contact force sequence is a set of force values ​​at a sampling frequency of 1 kHz within this time period, including the initial impact peak and subsequent stable force values. Then, the force is swung to the right by γ_left. init Obtain the right swing angle γ_right, and repeat the contact force sequence recording process. The vertical swing test logic when the vertical component ΔP_z exceeds the limit is the same as that in the horizontal direction, only the swing plane is changed to a vertical plane. The rotational swing test when the rotational component ΔP_rot exceeds the limit is performed around the Y-axis, and the contact force sequence at different rotation angles is recorded.

[0108] Step S223, contact force sequence analysis and optimal adjustment angle determination, employs the gradient descent method. First, the contact force sequence at each swing angle is preprocessed to remove the initial impact peak, and the average contact force within that time period is calculated, forming a dataset corresponding to the swing angle and average contact force. The gradient descent method uses minimizing the average contact force as its objective function, with the initial search step size set to 0.1 times γ. init This step size ensures that the initial iterations quickly approach the optimal angle, avoiding either missing the optimal value due to an excessively large step size or inefficient iteration due to an excessively small step size. During the iteration process, if the average contact force F_left corresponding to the left swing angle γ_left is less than F_right corresponding to the right swing angle γ_right, the optimal angle is determined to be on the left. In the next iteration, the step size is adjusted to the left to obtain a new angle γ_left1, and its average contact force F_left1 is calculated. If F_left1 is less than F_left, the step size is adjusted to the left until the change in average contact force between two adjacent iterations is less than the set force deviation threshold ΔF_stop. The angle at this point is the optimal adjustment angle γ in the horizontal direction. x,opt ΔF_stop is set to 0.01 times F peak The optimal adjustment angle γ in the vertical and rotational directions. z,opt γ rot,opt Obtained through the same logical calculations, this process ensures that the optimal adjustment angle can accurately match the posture with the minimum contact force, thereby minimizing insertion resistance.

[0109] The optimal insertion angle in step S224 is synthesized based on the optimal adjustment angle of each component direction and the kinematic model of the robotic arm. First, γ x,opt γ z,optγ_rot,opt is converted into attitude adjustment parameters in the robot arm's base coordinate system, where γ x,opt The angle correction corresponding to the X-axis direction, γ z,opt The angle correction corresponding to the Z-axis direction, γ rot,opt The rotation correction around the Y-axis is then applied; subsequently, the inverse kinematics algorithm of the robotic arm is called to fuse the attitude adjustment parameters in the three directions into the three-dimensional attitude parameters of the end effector, which is the optimal insertion angle γ. optimal During the integration process, it is necessary to ensure that there is no conflict between the adjustment angles. The multi-directional adjustments are quantified into unified posture parameters through the coordinate transformation matrix to ensure that the end effector of the robotic arm can be accurately aligned with the connector interface at that angle.

[0110] Step S22 addresses several key issues, including the blind adjustments caused by the lack of directional targeting in existing deviation correction techniques, the inability to determine the optimal angle due to the lack of a quantification method for the relationship between contact force and angle, and the mismatch in insertion posture caused by the inability to coordinate the correction of multiple deviation components. By decomposing deviation components, adjustments can be precisely targeted at the direction of the deviation, avoiding redundant operations caused by traditional overall adjustments. Swing testing and contact force sequence analysis establish a direct correlation between angle and contact force, solving the problem of subtle posture deviations that cannot be captured by visual methods alone. The application of gradient descent ensures efficient finding of the optimal angle with the minimum contact force, avoiding insufficient accuracy caused by empirical adjustments. Multi-component angle synthesis achieves posture optimization in three-dimensional space, solving the problem that single-direction adjustments cannot cover complex deviations. The output γ of step S22... optimal This directly provides an attitude reference for the insertion segment in step S23. If step S22 is missing, step S23 will use the T from step S15. fusion Insertion in the initial posture fails to correct the pose deviation detected in step S21, leading to excessive contact force that causes FPC jamming or damage. This renders the force control closed-loop correction mechanism ineffective, ultimately preventing self-calibration of visual errors. Furthermore, the rocking test in step S22 indirectly reflects the flexible deformation state of the FPC. When the contact force sequence fluctuates significantly in a certain direction, it indicates localized deformation of the FPC, providing a predictive basis for the subsequent flattening action in step S23. This expands the system's ability to perceive the FPC's state and further improves insertion reliability.

[0111] Step S23: In the insertion segment, the insertion is performed using a determined optimal insertion angle, and a closed-loop correction mechanism based on force feedback is constructed.

[0112] Step S23 uses the optimal insertion angle γ generated in step S22. optimal Using the attitude reference and the contact force fed back in real time by the force sensor as the core control basis, a closed-loop correction mechanism for the insertion section is constructed to solve the dynamic deviation problem caused by the flexible deformation of FPC and the tolerance of connector during the insertion process.

[0113] During the insertion action, the force sensor continuously collects the contact force F. contact When F contact Greater than the expected contact force F expect Multiply by k high Upon arrival, the collision is determined to be a hard collision, and a retreat correction is immediately initiated; the expected contact force F expect The reference value for the reasonable range of contact force is determined based on the previous force characteristics and the current insertion posture; hard impact refers to an abnormal contact state where the contact force exceeds the mechanical tolerance limit of the connector and FPC; expected contact force F expect The acquisition requires the fusion of multi-source data. First, the peak contact force F at three initial test angles is extracted from the touch memory map. peak Calculate its mean F peak,avg Then extract the optimal insertion angle γ. optimal The corresponding average contact force F opt F opt It is for the optimal insertion angle γ optimal The corresponding swing test process involves arithmetically averaging the contact force data collected by the force sensor during the stable contact phase to obtain the force value; this is combined with the connector material hardness coefficient k. hard With FPC flexibility coefficient k flex F is calculated using a weighted average algorithm. expect The formula is F expect =F peak,avg ×k hard +F opt ×k flex , where k hard +k flex =1. k hard According to the material handbook, metal materials have higher values, while plastic materials have lower values; k flex The value is determined based on the FPC thickness and flexural modulus; the smaller the thickness, the higher the value. This calculation method ensures that F... expect It conforms to the basic force characteristics of connectors and adapts to the actual contact requirements under the current optimal posture, avoiding the shortcomings of traditional fixed thresholds that cannot adapt to dynamic changes. For example, if F peak,avg For 5N, k hard It is 0.6, F opt For 3N, k flex If it is 0.4, then F expect The result is 5 × 0.6 + 3 × 0.4 = 4.2 N.

[0114] The retraction correction includes: first controlling the retraction distance d of the robotic arm. back d back=Current insertion depth × 0.1. The current insertion depth is obtained by comparing the displacement data recorded by the encoder at the end of the robotic arm with the initial position. This retreat distance ensures that the hard collision contact state is eliminated, while avoiding excessive retreat that would lead to re-entry into the exploration phase, thus balancing correction efficiency and process continuity. After retreating, the basic compensation angle that matches the current installation misalignment state in the visual compensation plan library must be called first. This basic compensation angle is a static correction parameter pre-generated in step S13 for the known installation misalignment of the connector. The core reason for calling the basic compensation angle is that the occurrence of a hard collision may be superimposed with the influence of the initial installation misalignment of the connector. If this known misalignment is not corrected first, it is difficult to completely eliminate the collision problem through subsequent dynamic compensation alone. For example, if the connector itself is horizontally misaligned to the left, the hard collision may be caused by the FPC being squeezed by the edge of the misaligned interface. This misalignment must be corrected to the right first by the basic compensation angle to clear known interference for subsequent dynamic adaptation. After calling the basic compensation angle, a flexible compensation angle φ must be superimposed on this angle. soft This forms the final correction angle, which is: final correction angle = basic compensation angle + φ soft , where φ soft =(F contact / F expect -k high )×k13, where k13 is the angle correction coefficient, determined through preliminary experiments, to ensure that the larger the contact force ratio, the greater the φ soft The larger the adjustment range, the more gradually it adapts to the connector's minute deformations, avoiding further hard collisions. Example values ​​for k13 are 0.5°-1.5°. The essence of this superposition logic is to first correct known system errors (installation misalignment) and then compensate for unknown random errors (dynamic interference), avoiding residual deviations caused by a single correction. In the formula, F... contact / F expect By normalizing the process, the collision severity of different connector models is standardized, avoiding issues caused by F... expect The difference leads to the same difference corresponding to different collision severity, (F contact / F expect -k high Only keep those exceeding k high In the hard collision section, "×k13" converts the normalized collision severity into an executable angle adjustment, ensuring that the more severe the collision (the larger this ratio), the greater the φ. soft The larger the adjustment range, the more perfectly it matches the quantization characteristics of dynamic interference.

[0115] After the final correction angle is determined, the system controls the robotic arm to adjust the end effector's attitude according to that angle. After the attitude adjustment is completed, the insertion action is re-executed. During the insertion process, the force sensor continuously monitors the contact force. If the contact force falls back to the range [F], the system will initiate an insertion. expect ×k low ,F expect ×khigh If the contact force is still greater than F, the hard collision problem is considered resolved, and the insertion segment procedure continues; if the contact force is still greater than F, the hard collision problem is considered resolved. expect ×k high If the above-mentioned backtracking correction process is repeated, a complete closed-loop correction mechanism is formed, but the continuous correction shall not exceed 3 times to avoid falling into an infinite loop. If the number of corrections is exceeded, a "connector abnormality" alarm will be triggered. In the entire backtracking correction logic, calling the base compensation angle that matches the current installation misalignment state is a key prerequisite. Its function is to first eliminate known installation misalignment interference, create conditions for the flexible compensation angle to focus on handling dynamic interference, and avoid the superposition of base misalignment and dynamic interference, which would cause the correction to fail. The superposition of the flexible compensation angle and the base compensation angle achieves a comprehensive correction of "static error + dynamic error", ensuring that the contact posture of the FPC and the connector is completely adapted to the actual state when reinserted, fundamentally avoiding hard collisions from happening again.

[0116] When the contact force F contact Less than the expected contact force F expect Multiply by k low When this occurs, it is determined to be a virtual contact state, k low The lower limit coefficient is k. low The method for determining this is as follows: record the minimum contact force and the corresponding F under the effective insertion state. expect The ratio of multiple ratios is taken as k. low The exemplary range is 0.3 to 0.5. A false contact refers to a contact force that does not reach the level required for effective mating, typically caused by FPC bending, resulting in incomplete contact. When a false contact is determined, a flattening action is performed: the end effector of the robotic arm applies a preload F in a direction perpendicular to the FPC surface. pre F pre =(F expect -F contact )×k flatten , where k flatten k is the flattening coefficient. flatten The value of k depends on the thickness and material of the FPC; the smaller the thickness and the softer the material, the higher the k value. flatten The lower the value, for example, from 0.8 to 1.2, the better. This calculation method ensures that the pre-pressure can effectively flatten the bent FPC while avoiding damage to the FPC due to excessive pressure. The flattening action duration t_flatten is set to 0.5 to 1 second to ensure that the FPC is fully flattened before insertion, avoiding repeated false contacts caused by incomplete flattening in a short time.

[0117] Step S23 addresses several core issues, including damage from hard collisions due to the lack of dynamic force feedback in the insertion segment in existing technologies, insertion failure due to the inability to identify virtual contacts, and blind adjustments due to the lack of quantitative basis for correction strategies. This is achieved through F... expectThe dynamic calculation enables the contact force judgment benchmark to adapt to different insertion postures and material characteristics, solving the problem of poor adaptability of traditional fixed thresholds; the combination of retreat after hard collision and flexible compensation not only avoids structural damage, but also adapts to dynamic deviations through posture fine adjustment, solving the problem that single retreat cannot completely eliminate collisions; the flattening action of virtual contact is ensured by quantitative pre-pressure calculation to ensure that the FPC is effectively flattened, solving the problems of low efficiency and uncontrollable effect of traditional manual adjustment.

[0118] Step S24: Monitor the insertion resistance curve in real time during the insertion segment, identify the jamming point based on the insertion resistance curve, and apply the retraction-rotation-advance strategy to handle the jamming point.

[0119] Step S24 is based on the insertion process in step S23. By monitoring the insertion resistance curve in real time, it captures the jamming characteristics, adopts the "retract-turn-advance" strategy to handle the jamming, and integrates the jamming data into the touch memory map to achieve the prediction and avoidance of insertion obstacles.

[0120] The insertion resistance curve F(t) refers to the continuous curve of contact force changing with time t during insertion; the monitoring and storage of the insertion resistance curve F(t) are synchronized with the contact force acquisition in step S23, and the force sensor acquires F... contact Then, the data is stored in a real-time database according to the time series, forming a continuous F(t) curve. The rate of change dF / dt of the insertion resistance curve F(t) is calculated using a numerical differential algorithm, taking the ratio of the force difference to the time difference between two adjacent sampling points. If dF / dt > k... spike If a bottleneck is detected, a retreat-turn-advance strategy is immediately executed. Where k spike k is the mutation threshold. spike The determination of the critical damage value needs to be combined with the critical damage state of the connector and FPC: by gradually increasing the insertion resistance and recording dF / dt, when scratches are observed on the FPC or deformation of the connector interface is observed, the corresponding dF / dt is the critical damage value. The product of this critical value and the safety factor k14 is taken as k. spike The value of k14 ranges from 0.7 to 0.8 to ensure that the jamming process is triggered before damage is caused.

[0121] The method for implementing the "retreat-turn-advance" strategy includes: first, controlling the robotic arm to retreat a small distance d. micro d micro The retraction distance is equal to 0.5 times the FPC thickness, which is obtained through mechanical measurement. This retraction distance ensures that the FPC is completely disengaged from the jamming position while avoiding excessive retraction that would increase retry time. Then, the installation misalignment status identified in step S12 is read. If an installation misalignment exists, such as a horizontal leftward misalignment θ',... skew Then rotate a small angle θ in the opposite direction of the deflection (to the right). micro For example, θ micro =θ'skew ×k15, where k15 is the rotation angle adjustment coefficient, ranging from 0.2 to 0.3. This rotation angle can fine-tune the relative posture of the FPC and the interface to avoid jamming positions; finally, insertion is performed.

[0122] When a jam is detected n times consecutively at the same spatial location, the spatial coordinates P of that location are... jam The corresponding force characteristics are associated and stored in the touch memory map constructed in step S11, and marked as interference regions. Spatial coordinates P jam The force characteristics, obtained through the encoder at the end of the robotic arm, refer to the force contact value and dF / dt during jamming. The method for determining the number of attempts (n) is as follows: statistically analyze historical jamming data to determine the critical number between a single successful adjustment after jamming and multiple unsuccessful attempts. This critical number is taken as n, with an exemplary range of 3 to 5 attempts, to avoid accidental marking of interference areas. The marking of the interference area must include its position coordinates, force characteristics, and corresponding installation misalignment state, providing obstacle avoidance information for subsequent insertions. When the next insertion approaches this area, the system can adjust its posture in advance to avoid repeated jamming.

[0123] Step S24 addresses key issues including structural damage caused by the inability to predict jamming in real time, low efficiency due to the lack of standardized jamming handling strategies, and recurring failures caused by the lack of recording jamming locations. Real-time monitoring and abrupt change determination of dF / dt solves the problem of lag in traditional jamming detection relying on contact force thresholds; the standardized "retreat-turn-advance" strategy addresses the issues of reliance on manual experience and blind adjustment of direction in jamming handling; and the marking and updating of interference regions solves the problem of repeated jamming in similar connections. The force characteristics of the interference region can be used to analyze connector wear conditions. When the force characteristics of a certain region gradually increase, it indicates that there may be connector interface deformation in that region, requiring maintenance, thus achieving an upgrade from passive handling to proactive prediction. Without step S24, forced insertion when jamming occurs will lead to FPC tearing or connector damage, and no obstacle avoidance experience can be accumulated, resulting in repeated jamming in similar connections, reducing overall connection efficiency and reliability.

[0124] Step S20 uses segmented force-position coupling control of “exploration segment-alignment segment-insertion segment”, combined with closed-loop correction and self-learning mechanism, to solve the dynamic deviation problem in actual insertion after pre-calibration in step S10, forming a complete control chain of “deviation identification-attitude optimization-dynamic correction-experience accumulation”.

[0125] Step S20 addresses the following shortcomings in existing technologies: traditional insertion methods, lacking segmentation, lead to accumulated deviations, with initial positioning errors amplifying with insertion depth; the absence of targeted attitude alignment, relying solely on overall adjustment resulting in insufficient accuracy; the lack of dynamic correction during insertion, making it unable to handle real-time interference such as FPC flexibility deformation; and the lack of an experience accumulation mechanism, leading to repeated problems in similar insertion methods. To address these shortcomings, step S20 employs a multi-pronged approach: segmented control breaks down the insertion process into independent stages, each focusing on a specific task (exploration stage for deviation capture, alignment stage for attitude optimization, and insertion stage for execution and correction), preventing deviation accumulation across stages. For example, the exploration stage only covers a depth of D×0.1, allowing for timely correction of even minor deviations during alignment; component-based alignment, based on the dimensional characteristics of the robotic arm's base coordinate system, decomposes deviations into horizontal, vertical, and rotational components. Each component is adjusted individually before being combined, preventing deviations in one dimension from being masked by other dimensions in traditional overall adjustment. For example, horizontal deviation is aligned only by horizontal swaying to ensure adjustment accuracy; the closed-loop correction mechanism, with real-time feedback from the force sensor as its core, forms a closed loop of "set posture - actual contact - deviation correction". For example, the retreat and flexible compensation during hard collisions and the flattening action during false contact are all based on real-time dynamic adjustment of force values, avoiding the inability of fixed control parameters to adapt to dynamic changes; the self-learning mechanism improves the system's insertion accuracy for similar connectors with the number of uses by updating the graph of the parameter averaging of successful samples and the jamming data. For example, when inserting the same type of FPC again, the ΔP and γ values ​​of historical successful samples can be directly called. optimal By eliminating parameters such as multi-angle testing in step S12, the process time is shortened. Segmented and component control keeps the mating deviation within millimeters, solving the core problems of visual error and inconsistent force control contact points. Closed-loop correction and jamming treatment prevent mechanical damage to the FPC and connector, extending the equipment's lifespan. Through force feedback and dynamic adjustment, it can adapt to FPCs of different thicknesses and materials and connectors of different models without rewriting the control program.

[0126] Example 2

[0127] This embodiment, based on embodiment 1, provides a TP lead FPC automatic insertion force and posture coordination control device, such as... Figure 5 As shown, it includes:

[0128] Touch memory map construction module: used to construct the touch memory map of the target FPC connector;

[0129] Installation Misalignment Recognition Module: Based on touch memory maps, it identifies the installation misalignment status of the target FPC connector;

[0130] Compensation plan library generation module: used to create a visual compensation plan library based on the installation misalignment status of the target FPC connector;

[0131] Fusion positioning matrix generation module: used to obtain the initial positioning matrix of the target FPC connector, and generate the fusion positioning matrix by combining it with the visual compensation plan library;

[0132] Force-position coupling control module: The insertion process is divided into exploration, alignment and insertion segments. In the exploration segment, the fused positioning matrix guides the robot arm forward, records the actual pose of the robot arm end effector, obtains the theoretical contact pose of the robot arm end effector, and calculates the pose deviation between the actual pose and the theoretical contact pose. In the alignment segment, the optimal insertion angle is determined based on the pose deviation. In the insertion segment, the optimal insertion angle is used to perform insertion, and a closed-loop correction mechanism based on force feedback is constructed to identify and handle jamming points.

[0133] Furthermore, in the touch memory map construction module, the method for constructing the touch memory map of the target FPC connector includes: obtaining the installation state parameters of the target FPC connector, controlling the force sensor at the end of the robotic arm to approach the target FPC connector from three initial test angles based on the installation state parameters, recording the force-touch feature triplet of each initial test angle, and constructing the touch memory map of the target FPC connector.

[0134] Furthermore, the method for identifying the installation misalignment state of the target FPC connector in the installation misalignment state identification module includes:

[0135] Step S121: Extract the force-touch feature triplets of three initial test angles from the touch memory map. Based on the force-touch feature triplets and installation state parameters, determine the spatial distribution scheme of nine fine test angles. The nine fine test angles include one front angle, two left-leaning angles in the horizontal direction, two right-leaning angles in the horizontal direction, two upward-tilting angles in the vertical direction, and two downward-tilting angles in the vertical direction.

[0136] Step S122: Control the robotic arm to perform no-load pre-contact test according to the spatial distribution scheme of nine fine test angles, record the no-load contact position coordinates, no-load contact force peak value and no-load vibration characteristic frequency under each fine test angle, and form an extended force contact feature dataset;

[0137] Step S123: Identify whether there is installation misalignment based on the extended force contact feature dataset, determine the direction of installation misalignment, and calculate the misalignment angle of the installation misalignment direction. The installation misalignment direction and the corresponding misalignment angle form the installation misalignment state of the target FPC connector.

[0138] Furthermore, in the force-position coupling control module, the method for determining the optimal insertion angle includes:

[0139] Step S221: In the alignment segment, the pose deviation is decomposed into horizontal, vertical and rotational components, and it is determined whether each component exceeds the corresponding allowable deviation threshold.

[0140] Step S222: Perform a rocking test in the corresponding plane on the component that exceeds the corresponding allowable deviation threshold, and record the contact force sequence at different rocking angles;

[0141] Step S223: Analyze the changing trend of the contact force sequence, and determine the swing angle with the minimum contact force as the optimal adjustment angle for the corresponding component direction using the gradient descent method.

[0142] Step S224: Combine the optimal adjustment angles of each component direction to generate the optimal insertion angle in three-dimensional space.

[0143] The methods and apparatus of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated.

[0144] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0145] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for coordinated control of force and position of automatic insertion of TP lead FPC, characterized in that, The method comprises: constructing a touch memory map of the target FPC connector; identifying the installation skew state of the target FPC connector based on the touch memory map; establishing a visual compensation plan library according to the installation skew state of the target FPC connector; obtaining an initial positioning matrix of the target FPC connector, combining the visual compensation plan library, and generating a fusion positioning matrix; dividing the insertion process into an exploration segment, an alignment segment and an insertion segment, using the fusion positioning matrix to guide the mechanical arm to advance in the exploration segment, recording the actual pose of the mechanical arm end, obtaining the theoretical contact pose of the mechanical arm end, and calculating the pose deviation between the actual pose and the theoretical contact pose; in the alignment segment, based on the pose deviation, the optimal insertion angle is determined; in the insertion segment, the optimal insertion angle is used to perform insertion, a closed-loop correction mechanism based on force feedback is constructed, and the jamming point is identified and processed.

2. The TP lead FPC automatic plug-in force position cooperative control method according to claim 1, characterized in that, The method for constructing the touch memory map of the target FPC connector comprises: obtaining the installation state parameters of the target FPC connector, controlling the mechanical arm end force sensor to approach the target FPC connector from three initial test angles based on the installation state parameters, recording the force touch feature triplets of each initial test angle, and constructing the touch memory map of the target FPC connector; the force touch feature triplet includes contact position coordinates, contact force peak value and vibration characteristic frequency.

3. The TP lead FPC automatic plug-in force position cooperative control method according to claim 2, characterized in that, The time for recording the force touch feature triplet of each initial test angle is when the force sensor detects the initial contact force; The judgment standard for detecting the initial contact force is that the real-time force value collected by the force sensor exceeds the force value baseline in the non-contact state for 5 consecutive sampling points.

4. The TP lead FPC automatic plug-in force position cooperative control method according to claim 3, characterized in that, The method for identifying the installation skew state of the target FPC connector comprises: based on the touch memory map, expanding the three initial test angles into nine fine test angles, performing no-load pre-touch test on each fine test angle, recording the expanded force touch feature data set of each fine test angle, and identifying the installation skew state of the target FPC connector according to the expanded force touch feature data set.

5. The TP lead FPC automatic plug-in force position cooperative control method according to claim 4, characterized in that, The expanded force touch feature data set includes the no-load contact position coordinates, the no-load contact force peak value and the no-load vibration characteristic frequency at each fine test angle; The method for identifying the installation skew state of the target FPC connector according to the expanded force touch feature data set comprises: identifying whether there is installation skew according to the expanded force touch feature data set, determining the installation skew direction, calculating the skew angle of the installation skew direction, and the installation skew direction and the corresponding skew angle form the installation skew state of the target FPC connector.

6. The TP lead FPC automatic plug-in force position cooperative control method according to claim 5, characterized in that, The installation skew direction includes horizontal direction skew and vertical direction skew; The nine fine test angles include one straight-ahead angle, two left skew angles in the horizontal direction, two right skew angles in the horizontal direction, two upward angles in the vertical direction, and two downward angles in the vertical direction; The identification method of the horizontal direction deflection comprises: calculating the average value of the peak values of the no-load contact force corresponding to two left deflection angles in the horizontal direction as the average touch force on the left side; calculating the average value of the peak values of the no-load contact force corresponding to two right deflection angles in the horizontal direction as the average touch force on the right side; calculating the difference ΔF between the average touch force on the left side and the average touch force on the right side; and when the absolute value of ΔF is greater than the horizontal deflection threshold ΔF threshold , it is determined that there is horizontal direction deflection.

7. The TP lead FPC automatic plug-in force position cooperative control method according to claim 6, characterized in that, The method for obtaining the initial positioning matrix of the target FPC connector comprises: obtaining the original field image of the target FPC connector in the image coordinate system through the visual measurement assembly, analyzing the original field image, and obtaining the initial positioning matrix of the target FPC connector; The method for analyzing the original field-of-view image comprises: performing target segmentation on the original field-of-view image in an image coordinate system to obtain a segmented connector image; performing feature detection on the segmented connector image to obtain target FPC connector feature point coordinates in the image coordinate system; mapping the target FPC connector feature point coordinates in the image coordinate system into actual physical coordinates of the feature points in a mechanical arm base coordinate system; and performing pose solving of the target FPC connector based on the actual physical coordinates of the feature points in the mechanical arm base coordinate system to obtain an initial positioning matrix.

8. The TP lead FPC automatic plug-in force position cooperative control method according to claim 7, characterized in that, The method for generating the fused positioning matrix comprises: calculating an image definition index of the original field-of-view image, evaluating a visual confidence according to the image definition index, dynamically adjusting a perception weight of the visual measurement assembly and the force sensor according to the visual confidence, generating the fused positioning matrix based on the adjusted perception weight and the initial positioning matrix, and combining a visual compensation plan library.

9. The TP lead FPC automatic plug-in force position cooperative control method according to claim 8, characterized in that, The method for determining the optimal insertion angle comprises: in the alignment segment, decomposing the pose deviation into a horizontal component, a vertical component and a rotation component, and determining whether each component exceeds a corresponding allowable deviation threshold; performing a swing test in a corresponding plane on the component that exceeds the corresponding allowable deviation threshold, and recording a contact force sequence under different swing angles; analyzing the change trend of the contact force sequence, and determining the optimal adjustment angle of the corresponding component direction by the gradient descent method; generating the optimal insertion angle in the three-dimensional space by comprehensively considering the optimal adjustment angles of the component directions.

10. A TP lead FPC automatic insertion force and pose collaborative control device, which is used to implement the TP lead FPC automatic insertion force and pose collaborative control method in any one of claims 1-9, characterized in that, The device comprises: a touch memory map construction module for constructing a touch memory map of the target FPC connector; an installation skew state recognition module for recognizing the installation skew state of the target FPC connector based on the touch memory map; a compensation plan library generation module for establishing a visual compensation plan library according to the installation skew state of the target FPC connector; a fused positioning matrix generation module for obtaining an initial positioning matrix of the target FPC connector, and generating a fused positioning matrix by combining the visual compensation plan library; a force-position coupling control module for dividing the plug-in process into an exploration segment, an alignment segment and an insertion segment, using the fused positioning matrix to guide the mechanical arm to move forward in the exploration segment, recording the actual pose of the mechanical arm end, obtaining the theoretical contact pose of the mechanical arm end, calculating the pose deviation between the actual pose and the theoretical contact pose, determining the optimal insertion angle based on the pose deviation in the alignment segment, and executing the insertion by using the optimal insertion angle in the insertion segment, constructing a closed-loop correction mechanism based on force feedback, and identifying and processing the jamming point.

Images