High-precision automatic docking method and device for connectors under vibration test conditions

Abstract

The application provides a high-precision automatic docking method and device for connectors under vibration test conditions, and belongs to the technical field of data processing under the Internet industry. The method comprises the following steps: acquiring a multi-source sensing data set with a global synchronization timestamp; performing fusion processing on the multi-source sensing data set to obtain a real-time six-degree-of-freedom dynamic pose sequence of the connector of the measured product under a docking execution end coordinate system; predicting a target plug-in pose and an optimal plug-in timing of the connector of the measured product according to the real-time six-degree-of-freedom dynamic pose sequence and vibration acceleration data; planning a docking trajectory according to the target plug-in pose and the optimal plug-in timing; controlling a docking execution mechanism to execute a plug-in action according to the docking trajectory, and correcting the docking trajectory according to the six-dimensional contact force data obtained in real time during the plug-in process, so as to complete the plug-in action. The accuracy and success rate of the automatic docking connector of the test system under complex vibration conditions can be improved.

Description

Technical Field

[0001] This application belongs to the field of data processing in the Internet industry, and specifically relates to a method and device for high-precision automatic docking of connectors under vibration testing conditions. Background Technology

[0002] In the manufacturing and factory inspection of precision optoelectronic instruments and high-precision equipment, products are typically subjected to rigorous environmental adaptability tests on a vibration table. During the automated flow of the testing process and the electrical connection stage, the vibration table itself inevitably exhibits high-frequency, low-amplitude residual vibrations. In addition, the product accumulates posture deviations after multiple tests, and elastic deformation occurs after the clamping fixtures are subjected to force. As a result, the actual spatial position of the connectors of the tested product is always in an irregular, high-frequency dynamic offset state.

[0003] Currently, most automated docking solutions on production lines rely on conventional static vision positioning combined with fixed mechanical motion trajectories for mating actions. However, when dealing with tiny connectors in dynamic vibration environments, these traditional solutions suffer from inherent time delays in visual image acquisition and algorithm processing, as well as lags in the physical response of the underlying mechanical actuators. This means that the positioning coordinates acquired by the system are often invalid by the time the robotic arm arrives, making it difficult to compensate for dynamic positional errors. This asynchrony between perception and execution makes it difficult for the system to meet the requirements of high-precision, repetitive docking. In actual operation, this can easily lead to problems such as pin collisions, pin bending and wear, and even docking failures. This not only damages high-value products but also frequently interrupts the testing process, reducing the automation continuity and overall production efficiency of the production line. Summary of the Invention

[0004] This application provides a high-precision automatic docking method and device for connectors under vibration testing conditions, which enhances the compliant adaptability of the testing system to high-frequency dynamic interference and improves the accuracy and success rate of automatic docking of connectors under complex vibration conditions.

[0005] This application provides a high-precision automatic docking method for connectors under vibration testing conditions, including: Acquire a multi-source sensor dataset with a global synchronization timestamp, wherein the global synchronization timestamp is used to unify the time series reference of the multi-source data in the multi-source sensor dataset, and the multi-source data includes real-time image data of the connector of the product under test, three-axis vibration acceleration data of the vibration table, and six-dimensional contact force data of the docking execution end. The multi-source sensor dataset is fused to obtain the real-time six-degree-of-freedom dynamic pose sequence of the connector of the tested product in the coordinate system of the docking execution end. Based on the real-time six-degree-of-freedom dynamic pose sequence and the vibration acceleration data, the target mating pose and optimal mating timing of the connector of the tested product are predicted. Plan the docking trajectory based on the target engagement pose and the optimal engagement timing; The docking actuator is controlled to perform the insertion action according to the docking trajectory, and the docking trajectory is corrected according to the six-dimensional contact force data acquired in real time during the insertion process, so as to complete the insertion action.

[0006] According to the high-precision automatic docking method for connectors under vibration testing conditions provided in this application, the step of fusing the multi-source sensor dataset to obtain the real-time six-degree-of-freedom dynamic pose sequence of the connector under test in the docking execution end coordinate system includes: extracting features from the real-time image data to obtain the initial pose of the connector under test in the camera coordinate system; fusing multi-source data with the vibration acceleration data as control input and the initial pose as the observation value to obtain the optimal pose estimate; and transforming the optimal pose estimate to the docking execution end coordinate system through a preset hand-eye transformation matrix to obtain the real-time six-degree-of-freedom dynamic pose sequence, wherein the hand-eye transformation matrix indicates the homogeneous spatial transformation relationship between the camera coordinate system and the docking execution end coordinate system.

[0007] According to the high-precision automatic docking method for connectors under vibration testing conditions provided in this application, the step of fusing multi-source data with the vibration acceleration data as control input and the initial pose as observation value to obtain the optimal pose estimate includes: constructing a system state vector containing the six-degree-of-freedom pose and pose change rate of the connector of the product under test; obtaining the prior prediction value of the system state vector based on the vibration acceleration data and a preset state transition matrix, wherein the state transition matrix indicates the dynamic evolution law of the system state vector; and completing the posterior update of the system state vector based on the prior prediction value of the system state vector, using the initial pose as observation value, to obtain the optimal pose estimate.

[0008] According to the high-precision automatic docking method for connectors under vibration test conditions provided in this application, the step of using the initial pose as the observation value and completing the posterior update of the system state vector based on the prior predicted value of the system state vector to obtain the optimal pose estimate includes: calculating the observation residual based on the preset observation matrix and the prior predicted value of the system state vector; calculating the optimal Kalman gain through the process noise covariance and the observation noise covariance; and correcting the prior predicted value of the system state vector based on the optimal Kalman gain and the observation residual to obtain the optimal pose estimate.

[0009] According to the high-precision automatic docking method for connectors under vibration testing conditions provided in this application, the step of predicting the target mating pose and optimal mating timing of the connector under test based on the real-time six-degree-of-freedom dynamic pose sequence and the vibration acceleration data includes: normalizing the real-time six-degree-of-freedom dynamic pose sequence and the corresponding frame vibration acceleration data from multiple historical consecutive frames to obtain normalized time-series data; inputting the time-series data into a pre-trained long short-term memory network model to perform time-series feature extraction and trajectory prediction, and outputting a pose prediction sequence for the future time period; determining the moment of vibration zero crossing in the pose prediction sequence as the optimal mating timing, and determining the pose corresponding to the vibration zero crossing as the target mating pose.

[0010] According to the high-precision automatic docking method for connectors under vibration test conditions provided in this application, before acquiring the multi-source sensor dataset with global synchronization timestamp, the method further includes: obtaining the hand-eye transformation matrix between the camera coordinate system and the docking execution end coordinate system based on the data after hand-eye calibration; configuring the global hardware synchronization trigger cycle; and setting the sampling frequency and triggering sequence of the multi-source data according to the global hardware synchronization trigger cycle.

[0011] According to the high-precision automatic docking method for connectors under vibration testing conditions provided in this application, the step of obtaining the hand-eye transformation matrix between the camera coordinate system and the docking execution end coordinate system based on the data after hand-eye calibration includes: controlling the docking execution end to move the calibration plate to multiple different poses, acquiring multiple sets of calibration plate images under different poses; extracting the calibration plate corner features from the calibration plate images; obtaining the camera in-camera participation distortion coefficient based on the calibration plate corner features; determining the pose of the calibration plate in the camera coordinate system based on the camera in-camera participation distortion coefficient; obtaining the end coordinate system pose of the calibration plate in each pose; obtaining the pose correspondence between the pose in the camera coordinate system and the pose in the end coordinate system; and obtaining the hand-eye transformation matrix based on the pose correspondence and the pose in the camera coordinate system using a hand-eye calibration algorithm.

[0012] According to the high-precision automatic docking method for connectors under vibration testing conditions provided in this application, the step of setting the sampling frequency and triggering sequence of the multi-source data according to the global hardware synchronization triggering cycle includes: outputting equally spaced synchronization pulse signals through a hardware synchronization triggering board according to the global hardware synchronization triggering cycle; configuring the sampling frequency of multiple sensing units that collect the multi-source data; and configuring the trigger edges of the synchronization pulse signals of the multiple sensing units to complete the triggering sequence setting of the multi-source data.

[0013] According to the high-precision automatic docking method for connectors under vibration test conditions provided in this application, after the mating action is completed, the method further includes: collecting connector pose data, steady-state contact force data, and electrical connection signal after the mating is completed; constructing docking success determination rules based on preset pose positioning deviation, preset steady-state contact force range, and preset electrical connection state; determining the docking status according to the connector pose data, the steady-state contact force data, and the electrical connection signal through the docking success determination rules; if docking is determined to be successful, outputting a docking completion signal to the upper control system.

[0014] This application also provides a high-precision automatic docking device for connectors under vibration testing conditions, including: The acquisition unit is used to acquire a multi-source sensor dataset with a global synchronization timestamp, wherein the global synchronization timestamp is used to unify the time series reference of the multi-source data in the multi-source sensor dataset, and the multi-source data includes real-time image data of the connector of the product under test, three-axis vibration acceleration data of the vibration table, and six-dimensional contact force data of the docking execution end. The processing unit is used to perform fusion processing on the multi-source sensor dataset to obtain the real-time six-degree-of-freedom dynamic pose sequence of the connector of the product under test in the coordinate system of the docking execution end. The prediction unit is used to predict the target mating pose and optimal mating timing of the connector of the product under test based on the real-time six-degree-of-freedom dynamic pose sequence and the vibration acceleration data. The planning unit is used to plan the docking trajectory based on the target docking pose and the optimal docking timing; The execution unit is used to control the docking actuator to perform the insertion action according to the docking trajectory, and to correct the docking trajectory according to the six-dimensional contact force data acquired in real time during the insertion process, so as to complete the insertion action.

[0015] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a high-precision automatic docking method for connectors under any of the vibration test conditions described above.

[0016] This application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a high-precision automatic docking method for connectors under any of the vibration test conditions described above.

[0017] This application also provides a computer program product, including a computer program that, when executed by a processor, implements a high-precision automatic docking method for connectors under any of the vibration test conditions described above.

[0018] As can be seen, compared with existing automatic docking schemes that rely on static vision and fixed trajectories, the main control device of the test system in this application first acquires a multi-source sensor dataset with a globally synchronized timestamp. Secondly, it fuses the multi-source sensor dataset to obtain a real-time six-degree-of-freedom dynamic pose sequence of the connector under test in the coordinate system of the docking execution end. Thirdly, based on the real-time six-degree-of-freedom dynamic pose sequence and vibration acceleration data, it predicts the target mating pose and optimal mating timing of the connector under test. Fourthly, it plans the docking trajectory based on the target mating pose and optimal mating timing. Finally, it controls the docking execution mechanism to perform the mating action according to the docking trajectory and corrects the docking trajectory based on the real-time acquired six-dimensional contact force data during the mating process to complete the mating action. Because the testing system can overcome system response delay and vibration offset by performing spatiotemporal fusion and dynamic pose prediction on multi-source sensor data with global synchronization timestamps, and uses six-dimensional contact force feedback for trajectory adaptive correction during the insertion execution phase, it can effectively enhance the testing system's compliant adaptability to high-frequency dynamic interference, thereby avoiding rigid collisions and improving the accuracy and success rate of automatic docking of connectors under complex vibration conditions. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the overall architecture of the test system provided in this application.

[0021] Figure 2 This is one of the flowcharts illustrating a high-precision automatic docking method for connectors under vibration testing conditions provided in this application.

[0022] Figure 3 This is the attitude positioning diagram provided in this application.

[0023] Figure 4 This is the second flowchart of a high-precision automatic docking method for connectors under vibration testing conditions provided in this application.

[0024] Figure 5 This is the third flowchart of a high-precision automatic docking method for connectors under vibration testing conditions provided in this application.

[0025] Figure 6 This is a block diagram of the functional units of a high-precision automatic docking device for connectors under vibration testing conditions provided in this application.

[0026] Figure 7 This is a schematic diagram of the main control device provided in this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0029] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0030] Existing automated docking solutions that rely on static vision and fixed trajectories cannot compensate for dynamic vibration errors due to the inherent hysteresis of both the perception and execution systems. This easily leads to collision damage to connectors, severely restricting the automation continuity and overall production efficiency of the production line.

[0031] To address the aforementioned problems, this application provides a high-precision automatic docking method and apparatus for connectors under vibration testing conditions. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0032] Please see Figure 1 , Figure 1 This is a schematic diagram of the overall architecture of the test system provided in this application. The overall test system of this application includes as follows: Figure 1 The hardware architecture shown on the left and as follows Figure 1 The software architecture shown on the right.

[0033] The hardware architecture of the testing system provided in this application includes four physical layers to support front-end operations. The first layer of the hardware architecture is the application scheduling and control layer, which includes a main control device. This main control device may have a built-in six-axis motion controller and a hardware synchronization trigger board. Its purpose is to act as the underlying scheduling hub, providing a unified global synchronization timestamp for subsequent multi-source sensor data acquisition and driving the actuators. The second layer of the hardware architecture is the multi-source sensing layer, which is specifically used to acquire the real-time environment and status input of the product under test. This includes a camera, a six-dimensional force sensor, and a three-axis micro-electro-mechanical system (MEMS). The camera, for example, is a high-speed camera with a frame rate of more than 1000 frames per second, used to rapidly capture image data under high-frequency vibration and avoid motion blur. The six-dimensional force sensor is used to acquire contact force data of the mating ends in three orthogonal force directions and three torsional moment directions in space during the mating process in real time. The three-axis MEMS, i.e., the MEMS accelerometer, is used to sense and output the three-axis vibration acceleration data of the vibration table in three-dimensional space. The third layer of the hardware architecture is the high-precision execution layer, used for the final docking and insertion actions. This includes a 6DOF servo slide and a floating guide fixture, where 6DOF indicates that the slide has six degrees of freedom (three translation axes and three rotation axes). The floating guide fixture can be a mechanical structure with passive floating margin to compensate for residual pose deviations. The fourth layer of the hardware architecture is the hardware safety protection layer, including hard limits, safety light curtains, and emergency stop circuits, used to provide purely physical-level emergency power-off and safety protection in the event of accidental collisions or boundary violations.

[0034] The software architecture of the test system provided in this application comprises five logical layers to complete data processing and business flow. The first layer is the driver layer, including device bus I / O, hardware adaptation, and acquisition communication. Its purpose is to bridge the underlying hardware and upper-layer software, completing the digital acquisition of raw electrical signals from sensors and issuing servo drive commands. The second layer is the data layer, including a real-time database and parameter configuration library, used to cache multi-source sensor datasets with globally synchronized timestamps and static parameters such as hand-eye calibration. The third layer is the core algorithm layer, including three modules: Extended Kalman Filter (EKF) pose fusion, Long Short-Term Memory (LSTM) dynamic prediction, and impedance control. These are specifically used for spatiotemporal fusion calculation of multi-source data, predicting optimal interlocking pose and timing, and generating trajectory correction based on contact force feedback. The fourth layer is the application layer, containing task scheduling, timing control, and a state machine, used to coordinate the execution order of the automatic docking process, state judgment rules, and abnormal retry loop. The fifth layer of the software architecture is the interface layer, which includes communication with the host computer based on standard protocols such as the Open Platform Communications Unified Architecture (OPC UA). After the mating action is completed and the result is determined to be successful, the interface layer outputs a handshake signal indicating that the docking is complete to the host control system.

[0035] Please see Figure 2 The high-precision automatic docking method for connectors under vibration testing conditions provided in this application includes the following steps.

[0036] S201, the main control device of the test system acquires a multi-source sensor dataset with a globally synchronized timestamp.

[0037] The global synchronization timestamp is used to unify the timing reference of the multi-source data in the multi-source sensor dataset. The multi-source data includes real-time image data of the connector of the product under test, three-axis vibration acceleration data of the vibration table, and six-dimensional contact force data of the docking execution end.

[0038] In practice, the connector of the product under test can be identified as the target docking part that experiences high-frequency, micro-amplitude dynamic positional displacement after being clamped onto the vibration table during the entire infrared product vibration test process. This displacement is caused by residual micro-vibrations from the vibration table, attitude deviations after multi-directional testing of the product, and elastic deformation of the clamping fixture. After packaging the data within a single cycle through the aforementioned microsecond-level hardware synchronous acquisition, a multi-source sensor dataset can be formed.

[0039] The triaxial vibration acceleration data of the vibration table can be sensed and reflected in real time as the vibration table is in motion. Figure 3The high-frequency vibration states in the X, Y, and Z directions of space are shown. In addition, the multi-source data also includes six-dimensional contact force data of the docking actuator end. This contact force data can indicate the real-time force data covering the connector mating condition collected by the six-dimensional torque sensor installed on the docking actuator end, and can be used to characterize the magnitude of the contact force and torsional torque of the docking actuator end in multiple dimensions of space.

[0040] S202, the main control device of the test system performs fusion processing on the multi-source sensor dataset to obtain the real-time six-degree-of-freedom dynamic pose sequence of the connector of the product under test in the coordinate system of the docking execution end.

[0041] The docking execution end coordinate system can indicate the physical space reference system of the high-precision six-degree-of-freedom (DOF) servo electric slide and its matching modular connector docking fixture. Unifying the spatial pose after multi-source data fusion to this end-point physical execution reference provides direct spatial coordinate basis for the precise drive of subsequent mechanical servo mechanisms. The six-degree-of-freedom dynamic pose sequence can indicate the spatial dynamic motion trajectory characteristics and temporal change patterns of the connector under test under continuous forced vibration. For example, the sequence can include the physical position of the connector in three-dimensional space at the current moment (three translational degrees of freedom) and the spatial attitude of the connector in three-dimensional space, and can also include the six-degree-of-freedom pose change rate describing the high-frequency micro-amplitude dynamic offset motion trend of the connector.

[0042] S203, the main control device of the test system predicts the target mating posture and optimal mating time of the connector of the product under test based on the real-time six-degree-of-freedom dynamic pose sequence and the vibration acceleration data.

[0043] The target mating pose indicates the precise physical spatial coordinates and attitude of the connector under test at the expected moment of mating. For example, the target mating pose may include the three-dimensional translational position and three-dimensional rotational deflection angle of the connector under test at the expected moment, serving as a high-precision reference for the servo docking device to plan its spatial trajectory in the next step. The optimal mating timing indicates the best future time node for the docking actuator to reach the aforementioned target pose and complete the physical mating action. By accurately predicting this timing, the main control device can issue motion control commands to the high-precision actuator in advance, enabling the mechanical clamp at the end of the actuator to synchronously mate with the connector under test at the exact moment it is in the target mating pose. This effectively avoids wear on the connector's rigid pins or leads caused by high-frequency dynamic micro-amplitude offsets remaining on the vibration table, thereby significantly improving the single-shot success rate of high-precision repeated docking under vibration testing conditions and the automation continuity of production line testing.

[0044] S204, the main control equipment of the test system plans the docking trajectory based on the target mating pose and the optimal mating timing.

[0045] The planned docking trajectory may include using the optimal docking timing predicted by the LSTM model as a strict time node constraint for the entire docking action, and using the target docking pose as the precise motion endpoint in the three-dimensional physical space. Based on this, a smooth transition S-shaped acceleration and deceleration docking motion trajectory is planned in Cartesian space.

[0046] S205, the main control device of the test system controls the docking actuator to perform the insertion action according to the docking trajectory, and corrects the docking trajectory according to the six-dimensional contact force data acquired in real time during the insertion process, so as to complete the insertion action.

[0047] The adaptive correction of the docking trajectory based on the real-time acquired six-dimensional contact force data during the mating process may include, during the physical execution phase of the mating action, the main control device designs and introduces a variable-parameter adaptive impedance control algorithm, using the real-time acquired six-dimensional environmental contact force data as a feedback input source, thereby dynamically adjusting the stiffness and damping parameters in the impedance control model. The impedance control algorithm upon which this adaptive correction step relies can be expressed as: Fenv-Fref=M×Δx¨+Bt×Δx˙+Kt×Δx, In this model, M represents the system inertia matrix, which is a constant value; Bt represents the damping matrix, which is adjusted in real time and can be adaptively updated according to changes in the actual contact force; Kt represents the stiffness matrix, which is also adjusted in real time and can be adaptively updated according to changes in the actual contact force; Δx, Δx˙, and Δx¨ represent the position deviation, velocity deviation, and acceleration deviation of the docking trajectory, respectively; Fenv represents the environmental contact force collected in real time by the six-dimensional torque sensor; and Fref represents the preset reference interlocking force setting. Through this dynamic impedance model, the main control device can calculate the corresponding trajectory correction in real time and update the control commands of the underlying high-precision servo main control device based on this correction. This allows for the dynamic correction of minute pose deviations in the interlocking trajectory as the actuator continuously approaches the optimal interlocking time.

[0048] Furthermore, during the correction process, when the real-time environmental contact force exceeds the preset safety threshold, the algorithm can automatically trigger the compliant retraction adjustment mechanism of the servo docking device. Through this compliant docking method that deeply integrates force feedback control and spatial position control, the main control device can further compensate for unforeseen physical assembly tolerances based on pre-visual prediction. This avoids rigid physical collisions of tiny connectors when facing high-frequency dynamic vibrations, thus solving the problems of connector pin collisions, pin bending, and wear. Consequently, it can reliably ensure the safety, stability, and docking success rate of each automated mating action in the entire testing process.

[0049] As can be seen, in this embodiment of the application, the main control device of the test system overcomes response delay and vibration offset by performing spatiotemporal fusion and dynamic pose prediction on multi-source sensor data with global synchronization timestamps, and uses six-dimensional contact force feedback for trajectory adaptive correction during the insertion execution stage. This can effectively enhance the system's compliant adaptability to high-frequency dynamic interference, thereby avoiding rigid collisions and significantly improving the accuracy and success rate of automatic docking of connectors under complex vibration conditions.

[0050] In one possible embodiment, the step of fusing the multi-source sensor dataset to obtain the real-time six-DOF dynamic pose sequence of the product under test connector in the docking execution end coordinate system includes: extracting features from the real-time image data to obtain the initial pose of the product under test connector in the camera coordinate system; fusing multi-source data with the vibration acceleration data as control input and the initial pose as observation value to obtain the optimal pose estimate; and transforming the optimal pose estimate to the docking execution end coordinate system using a preset hand-eye transformation matrix to obtain the real-time six-DOF dynamic pose sequence, wherein the hand-eye transformation matrix indicates the homogeneous spatial transformation relationship between the camera coordinate system and the docking execution end coordinate system.

[0051] The feature extraction of the real-time image data may include the main control device performing sub-pixel-level corner point extraction on the connector image of the current frame acquired by the high-speed visual sensing unit, thereby accurately obtaining the sub-pixel coordinates of the connector's reference feature points on the image plane. Next, calculating the initial pose of the connector under test in the camera coordinate system may include the main control device using the Perspective-n-Point (PnP) algorithm to perform spatial geometric mapping calculations based on the aforementioned obtained reference feature point coordinates, thereby directly calculating the initial pose of the connector under test in the current camera coordinate system. After obtaining the unilateral visual pose, due to the unavoidable clock delay and susceptibility to high-frequency vibration noise from the vibration table during the visual acquisition process, the main control device can use the vibration acceleration data as dynamic control input and the initial pose as static observation value, performing multi-source data spatiotemporal fusion through a Kalman filter framework to obtain the optimal pose estimate. In specific implementation, the hand-eye transformation matrix indicates the homogeneous spatial transformation relationship between the camera coordinate system and the docking execution end coordinate system. This matrix can be a four-by-four dimension matrix and can contain rotation matrices and translation vectors. Through this spatiotemporal fusion and spatial coordinate system unification, the true physical trajectory of the micro-connector under vibration can be reconstructed.

[0052] As can be seen, in this embodiment, the main control device of the test system can effectively overcome the inherent acquisition response delay of a single visual sensing device and the environmental noise interference caused by high-frequency forced vibration by combining the initial pose observation value extracted based on visual image features with the high-frequency vibration acceleration control input to achieve complementary advantages of multi-source data fusion, and combine the hand-eye conversion matrix to achieve precise unification of the spatial physical benchmark. This can improve the accuracy of the test system in real-time tracking of the six-degree-of-freedom dynamic pose of the docking plug under complex vibration conditions.

[0053] In one possible embodiment, the step of using the vibration acceleration data as control input and the initial pose as observation to perform multi-source data fusion to obtain the optimal pose estimate includes: constructing a system state vector containing the six-degree-of-freedom pose and pose change rate of the connector of the product under test; obtaining a priori predicted value of the system state vector based on the vibration acceleration data and a preset state transition matrix, wherein the state transition matrix indicates the dynamic evolution law of the system state vector; and using the initial pose as observation, completing a posterior update of the system state vector based on the prior predicted value of the system state vector to obtain the optimal pose estimate.

[0054] The construction of a system state vector containing the connector's six degrees of freedom pose and pose change rate can include building a twelve-dimensional vector at the algorithm's underlying layer, serving as the mathematical basis for EKF's data processing and iterative tracking. This system state vector indicates the core kinematic characteristic state of the tested product connector in physical space at the current moment. It can include not only the spatial physical pose of the six degrees of freedom (three-dimensional translation and three-dimensional rotation) at this moment, but also the real-time pose change rate of the aforementioned six degrees of freedom, thus comprehensively characterizing the connector's dynamic properties under high-frequency, low-amplitude forced vibration conditions.

[0055] The state transition matrix indicates the inherent dynamic evolution of the system state vector from the previous moment to the current moment. This matrix can be a 12x12 matrix, capable of directly deriving the natural evolution of the current state from the optimal estimated state of the previous moment without considering external control inputs. For example, the state transition matrix can indicate the purely physical kinematic recursive relationship between the six-DOF pose and the rate of change of the six-DOF pose of the connector under test within a global synchronization trigger cycle, without considering external control inputs. Obtaining the state transition matrix involves the master control device deriving the physical integral evolution mapping relationship in the time dimension between the six-DOF pose of the connector under test and its corresponding rate of change of the six-DOF pose, based on a pre-configured global synchronization trigger cycle, thereby constructing a 12x12 state transition matrix. Specifically, within an extremely short single microsecond-level sampling period, the spatial motion trajectory of the tiny connector can be approximated as a uniform and smooth evolution based on the physical inertia of the previous moment. Therefore, the transformation elements inside the matrix can be composed of constants and the global synchronization triggering period to characterize the natural temporal transition law of the system state when no external excitation disturbance is applied.

[0056] Obtaining the prior prediction value of the system state vector may include the main control device using the collected vibration acceleration data that reflects the real-time dynamics of the shaking table as a control input, applying the acceleration data through a preset control input matrix, and combining it with a process noise vector conforming to a Gaussian distribution, using the state prediction equation to calculate the theoretical prediction result of the current frame state. This state prediction equation can be fully expressed as: in, It can be represented as the prior predicted value of the pose state in frame t, which is also the prior predicted value of the system state vector. A can be represented as the aforementioned twelve-by-twelve-dimensional state transition matrix. B can be represented as the posterior optimal estimate of the pose state output from the previous time step, and can be represented as a 12x3 control input matrix. For example, this control input matrix can indicate how the high-frequency acceleration excitation force applied to the test system by the external vibration table accurately maps and affects the dynamic evolution of the system state vector of the connector under test. Obtaining the control input matrix can include constructing a 12x3 control input matrix based on the aforementioned global synchronization trigger period and according to Newton's laws of kinematics. By introducing first-order and second-order terms of the global synchronization trigger period into different dimensions within this matrix, the control input matrix can accurately map the three-axis vibration acceleration data into linear velocity increment compensation for the six-degree-of-freedom pose change rate of the connector, and quadratic nonlinear displacement increment compensation for the six-degree-of-freedom pose of the connector. This can be represented as the triaxial vibration acceleration data of the shaking table acquired at the current moment. It can be represented as a process noise vector at the bottom layer of the test system that conforms to a Gaussian distribution.

[0057] As can be seen, in this embodiment, the main control device of the test system can use the initial pose of the connector in the camera coordinate system obtained by image extraction and calculation as the observation benchmark, and make feedback corrections on the prior prediction values ​​obtained by the aforementioned pure kinematic recursion. This allows the system to introduce real physical space geometric constraints when fusing the underlying data, thereby effectively filtering out the cumulative drift error caused by simply relying on the integral of the acceleration signal. Finally, it can stably calculate the optimal pose estimate that accurately reflects the current real space state of the connector of the product under test.

[0058] In one possible embodiment, the step of using the initial pose as the observation value and completing the posterior update of the system state vector based on the prior predicted value of the system state vector to obtain the optimal pose estimate includes: calculating the observation residual based on a preset observation matrix and the prior predicted value of the system state vector; calculating the optimal Kalman gain through the process noise covariance and the observation noise covariance; and correcting the prior predicted value of the system state vector based on the optimal Kalman gain and the observation residual to obtain the optimal pose estimate.

[0059] The calculation of the observation residual can include using the initial pose of the connector in the camera coordinate system obtained by the PnP algorithm as the true physical observation reference, and using a 6x12 dimension observation matrix to perform linear spatial projection on the prior predicted value of the 12-dimensional system state vector, and then comparing the difference between the actual visual observation pose and the theoretically predicted projected pose. That is, subtracting the product of the observation matrix and the prior predicted value from the initial pose to calculate the geometric difference between the predicted system state and the actual visual observation state at the current moment.

[0060] The process noise covariance indicates the inherent uncertainty of the underlying control model of the master control device when performing physical inertia recursion using the state transition matrix and triaxial vibration acceleration data, as well as the degree of dispersion of the corresponding Gaussian-distributed process noise vector. Obtaining the process noise covariance can involve the master control device performing mathematical statistics and static assignment during the parameter initialization phase, based on the Gaussian distribution physical characteristics of the residual vibration of the vibration table and the inherent hardware thermal noise parameters of the MEMS accelerometer.

[0061] The observation noise covariance indicates the degree of uncertainty in visual acquisition delay and spatial pose calculation errors faced by high-speed visual sensing units during real-time image acquisition and sub-pixel-level corner feature extraction. Obtaining the observation noise covariance can involve the main control device performing offline evaluation and fixed settings of statistical variance based on the ambient lighting interference characteristics during high-frequency image acquisition and pre-calibrated intrinsic error levels of the camera's intrinsic parameters.

[0062] By combining the process noise covariance and observation noise covariance obtained above, and integrating them with the aforementioned observation matrix, the optimal Kalman gain matrix for the current moment can be dynamically calculated and output using the underlying minimum mean square error evaluation criterion. This gain matrix can act as a dynamic weight, adaptively weighing whether to place more trust in the prior prediction data based on acceleration recursion or in the geometric observation data based on the vision camera.

[0063] Correcting the prior prediction value of the system state vector to obtain the optimal pose estimate may include the main control device multiplying the calculated optimal Kalman gain with the aforementioned observation residual to obtain the system state compensation amount, and directly superimposing the state compensation amount onto the prior prediction value of the system state vector. This core feedback posterior correction step can be executed through the following observation update equation.

[0064] in, It can be represented as the posterior optimal estimate of the pose state in frame t. This can be represented as the prior prediction of the pose state in frame t. This can be represented as the Kalman gain matrix calculated above. H can be represented as the initial pose of the connector obtained from visual observation, and H can be represented as the observation matrix.

[0065] As can be seen, in this embodiment, the main control device of the test system can eliminate the spatial cumulative drift that is easily generated by relying solely on physical inertial data, as well as the high-frequency vibration noise interference that is easily affected by relying solely on visual images, through this closed-loop and adaptive posterior update method. Thus, the system can stably output the optimal six-degree-of-freedom pose state data that takes into account both high-frequency dynamic response and high-precision positioning.

[0066] In one possible embodiment, predicting the target mating pose and optimal mating timing of the connector of the product under test based on the real-time six-degrees-of-freedom dynamic pose sequence and the vibration acceleration data includes: normalizing the real-time six-degrees-of-freedom dynamic pose sequence and the corresponding frame vibration acceleration data from multiple consecutive historical frames to obtain normalized time-series data; inputting the time-series data into a pre-trained long short-term memory network model to perform time-series feature extraction and trajectory prediction, and outputting a pose prediction sequence for the future time period; determining the moment when the vibration crosses zero in the pose prediction sequence as the optimal mating timing, and determining the pose corresponding to the vibration crossing zero as the target mating pose.

[0067] The normalization process may include the main control device performing a minimum-maximum normalization (min-max normalization) on the acquired historical 20 consecutive frames of the real-time six-degree-of-freedom dynamic pose sequence and the corresponding frame's triaxial vibration acceleration data. This process effectively eliminates the dimensional differences between sensor data of different physical types caused by different units, thereby providing a standardized temporal feature input benchmark with uniform numerical distribution and stable reliability for subsequent neural network models.

[0068] Pre-training a Long Short-Term Memory (LSTM) network model may involve the testing system constructing and initializing a lightweight LSTM model at the core software algorithm layer. This model can perform efficient inference computation by loading pre-calibrated and initialized model weight parameters. The temporal feature extraction and trajectory prediction may involve the main control device inputting the aforementioned normalized temporal data containing twenty consecutive frames of multi-source states into the model. Through the model's internally coordinated forget gate, input gate, and output gate, the hidden kinematic changes of the multi-dimensional data in the time series are fully extracted, thereby accurately outputting the pose prediction sequence for the next fifteen milliseconds. This core temporal feature extraction and prediction mechanism can be fully expressed through the formulas of the various gating units within the model, specifically including: Among them, f t i t o t This can be represented by the output values ​​of the forget gate, input gate, and output gate at the current time, respectively, h. t-1 It can be represented as the hidden layer state of the previous time step, x t It can be represented as the time-series vector input at the current time, W f W i W C W o These can be represented as weight matrices within the corresponding calculation modules, b f b i b C b o These can be represented as the corresponding bias terms, C t This can be represented as the cell state at the current moment. This can be represented as the candidate cell state, σ can be represented as the sigmoid activation function, and tanh can be represented as the hyperbolic tangent activation function. Through the nonlinear recursive calculation of the above gating formula, the model can accurately capture and map the subtle nonlinear motion laws under complex forced vibration conditions.

[0069] In specific implementation, the moment of the vibration zero-crossing point can indicate the transient time node at which the spatial dynamic offset amplitude of the connector under test is exactly at its minimum or exactly crosses the physical center equilibrium position during a continuous high-frequency forced vibration trajectory. The main control device can find and determine this microsecond-level optimal moment in the pose prediction sequence within the aforementioned preset time period, and use it as the optimal mating time for the servo actuator to initiate the convergence. Simultaneously, obtaining the pose corresponding to the vibration zero-crossing point can include the main control device extracting the precise physical three-dimensional spatial translational position and rotational attitude data corresponding to the optimal mating time from the prediction sequence, and determining it as the target mating pose.

[0070] As can be seen, in this embodiment, the main control device of the test system can perform interpolation path planning by locking the zero-crossing moment with the smallest vibration offset. This can not only greatly reduce the difficulty of servo following the high-frequency dynamic pose of the docking actuator, but also overcome the docking coordinate failure problem caused by visual acquisition delay and mechanical physical response lag.

[0071] In a possible embodiment, before obtaining the multi-source sensing data set with global synchronization timestamps, the method further includes: solving for the hand-eye transformation matrix between the camera coordinate system and the docking execution end coordinate system according to the data after hand-eye calibration; configuring the global hardware synchronization trigger period; and setting the sampling frequency and trigger timing of the multi-source data according to the global hardware synchronization trigger period.

[0072] First, completing hand-eye calibration may include that the master device obtains and inputs the pre-prepared camera internal parameters, checkerboard hand-eye calibration board parameters, and motion parameters of the test system, and controls the mechanical execution component to perform multi-pose motions in space, thereby completing the basic spatial geometric feature alignment and physical calibration process.

[0073] Obtaining the hand-eye transformation matrix may include that the master device solves the homogeneous space transformation relationship between the camera coordinate system and the docking execution end coordinate system based on the above hand-eye calibration process. The solution process of this mapping relationship can be calculated by the following core calibration formula: T cam2end =T base2end ×T cam2base where T cam2end can be expressed as the homogeneous transformation matrix from the camera coordinate system to the end effector coordinate system, that is, the required hand-eye transformation matrix. This matrix can be a matrix with a size of four by four dimensions, and its internal contains the rotation matrix and translation vector for spatial transformation; T base2end can be expressed as the homogeneous transformation matrix from the robot base coordinate system to the end effector coordinate system. This matrix can be directly read and fed back in real time through the absolute value encoder internally equipped in the high-precision execution layer; T cam2base can be expressed as the homogeneous transformation matrix from the camera coordinate system to the robot base coordinate system, and this matrix can be directly solved by the foregoing basic hand-eye calibration process. By obtaining this key hand-eye transformation matrix, a static geometric bridge between the physical vision perception space and the mechanical execution end space can be successfully constructed, so that the pose calculated by the subsequent vision can be accurately mapped to the physical coordinate system of the underlying mechanical execution.

[0074] In specific implementation, the global hardware synchronization trigger cycle can instruct the underlying hardware synchronization trigger board to send equally spaced synchronization pulse signals to each distributed physical sensing unit as a basic time span reference. When configuring this cycle, it can be fixed at one millisecond, serving as the lowest-level strict clock beat for the entire high-frequency intelligent closed-loop control system. Setting the sampling frequency and trigger sequence of the multi-source data can include, based on the aforementioned global hardware synchronization trigger cycle, uniformly configuring the underlying data acquisition trigger logic and working beat of various physical hardware such as high-speed visual sensing units, force sensing units, and microelectromechanical system vibration state sensing units at the application scheduling and control level of the pure physical hardware architecture, thereby achieving microsecond-level time synchronization between various distributed sensing devices at the physical hardware level.

[0075] As can be seen, in this embodiment, the main control device of the test system, through such rigorous pre-spatiotemporal initialization and unified step configuration, can not only provide the necessary physical preconditions for accurately stamping the subsequent multi-source sensor dataset with global timestamps, but also ensure that all raw sensor data from different physical modalities are in the same defined physical three-dimensional space and time snapshot node when entering the core algorithm layer for fusion.

[0076] In one possible embodiment, the step of obtaining the hand-eye transformation matrix between the camera coordinate system and the docking execution end-effector coordinate system based on the hand-eye calibration data is described in detail below. Figure 4 As shown: S401 controls the docking execution end to move the calibration board to multiple different poses and acquire multiple sets of calibration board images under different poses.

[0077] The calibration board can be a pre-prepared checkerboard hand-eye calibration board with known and precise physical geometric dimensions. Controlling the docking execution end to move the calibration board to multiple different poses can include the system using a high-precision six-degree-of-freedom servo electric slide as a physical mechanical carrier to drive the aforementioned checkerboard hand-eye calibration board mounted on its end fixture to perform multi-dimensional, multi-pose movements such as translation and spatial rotation in three-dimensional physical space.

[0078] S402, Extract the calibration board corner features from the calibration board image.

[0079] Extracting the corner features of the calibration board image may involve the main control device using a low-level image processing module to perform sub-pixel-level precise localization and feature extraction on the black-and-white boundary corners of the checkerboard hand-eye calibration board in the multiple sets of visual images obtained above. This allows for the accurate acquisition of a high-precision set of pixel coordinates of the calibration board's reference points on a two-dimensional image plane. For example, this feature can indicate the corners at the black-and-white grid boundaries on the checkerboard hand-eye calibration board, where these corners have strictly known and fixed physical spacing parameters in real physical three-dimensional space.

[0080] S403, obtain the camera's internal distortion coefficients based on the corner features of the calibration plate.

[0081] The process of obtaining the internal distortion coefficients of the camera can include the main control device calculating and obtaining internal parameters such as focal length and optical principal point coordinates that reflect the inherent optical characteristics of the high-speed industrial camera through a basic camera calibration algorithm, based on the two-dimensional pixel coordinate set of the corner features of the calibration board extracted above, and the known physical geometric size parameters of the chessboard hand-eye calibration board itself, as well as the radial distortion coefficient and tangential distortion coefficient used to describe the degree of deformation of the camera's optical lens.

[0082] S404, determine the pose of the calibration plate in the camera coordinate system based on the in-camera distortion coefficient.

[0083] Determining the pose of the calibration board in the camera coordinate system involves first performing low-level digital image distortion correction calculations on the erroneous 2D pixel feature points extracted from the digital image based on the obtained distortion coefficients. This effectively eliminates the nonlinear spatial deformation errors caused by the physical optical lens. Next, the corrected, high-precision, distortion-free 2D pixel coordinates are substituted into the perspective projection geometric model. Combined with the actual geometric constraints of the physical calibration board, the true 3D spatial translation vector and 3D spatial rotation matrix of the calibration board relative to the camera's physical coordinate system at the moment of image acquisition are solved. This allows for the precise determination of its complete spatial physical pose in the current camera coordinate system.

[0084] S405, obtain the end coordinate system pose of the calibration plate in each pose.

[0085] After the camera's parameters are fixed, the main control device can control the docking execution end effector to move the calibration board to multiple different poses, thereby obtaining the end effector's coordinate system pose and the corresponding calibration board's pose in the camera coordinate system for each pose. The docking execution end effector can instruct a high-precision six-DOF servo electric slide and its end-mounted modular connector docking fixture with a floating guide mechanism. This fixture serves as a physical mechanical carrier that moves the physical calibration board in multiple directions within three-dimensional physical space throughout the entire hand-eye calibration process.

[0086] S406, Obtain the pose correspondence between the pose in the camera coordinate system and the pose in the end-effector coordinate system.

[0087] The pose correspondence indicates the one-to-one mapping data alignment between the digital visual perception space coordinates and the physical mechanical actuator space coordinates at the same physical moment. This correspondence reflects the relative spatial position and attitude difference of the same real physical calibration board in these two originally independent coordinate systems. The pose correspondence includes, for example, a set of core static data pairs locked by the system after each change in the spatial physical attitude of the underlying servo device.

[0088] S407, Based on the pose correspondence, the hand-eye conversion matrix is ​​obtained by using a hand-eye calibration algorithm, based on the end-effector pose in each pose and the pose of the calibration board in the camera coordinate system.

[0089] Based on the pose correspondence accumulated from a large number of samples, the hand-eye calibration algorithm can be used to obtain the hand-eye transformation matrix based on the end-effector pose and the pose of the calibration board in the camera coordinate system under each pose. This can include the main control device aligning and geometrically mapping the spatial pose of the end-effector at the same moment with the visual solution spatial pose in the camera coordinate system, thereby directly solving the homogeneous spatial transformation relationship between the camera physical coordinate system and the mechanical end-effector physical coordinate system to obtain the hand-eye transformation matrix.

[0090] As can be seen, in this embodiment, the main control device of the test system, through the above-mentioned complete and rigorous physical calibration process, can provide accurate and reliable static spatial reference parameters for subsequent multi-source data fusion and high-precision real-time spatial pose conversion under high-frequency vibration conditions.

[0091] In one possible embodiment, setting the sampling frequency and trigger timing of the multi-source data according to the global hardware synchronization trigger cycle includes: outputting equally spaced synchronization pulse signals through a hardware synchronization trigger board according to the global hardware synchronization trigger cycle; configuring the sampling frequency of multiple sensing units that collect the multi-source data; and configuring the trigger edges of the synchronization pulse signals of the multiple sensing units to complete the trigger timing setting of the multi-source data.

[0092] The hardware synchronization trigger board can be a core auxiliary component in the hardware architecture control layer. It can be defined as a hardware board capable of outputting multiple equally spaced synchronization pulse signals. This board can be used to achieve microsecond-level global clock synchronization between multiple distributed physical sensing units and mechanical execution units at the underlying level, thereby providing a unified hardware physical time reference for the entire high-frequency closed-loop test system.

[0093] In specific implementation, the multiple sensing units may include the three physical devices in the aforementioned hardware sensing layer: a high-speed visual sensing unit equipped with a global shutter, a six-dimensional force sensing unit installed at the docking execution end, and a three-axis MEMS vibration state sensing unit installed on the end face of the vibration table tooling. The sampling frequency can be a frequency value that strictly matches the aforementioned fixed value of one millisecond for the global hardware synchronization triggering period; for example, the hardware acquisition frequency of each sensing unit can be configured to 1 kilohertz.

[0094] The trigger edge of the synchronization pulse signal can be configured at the underlying driver adaptation level, precisely and uniformly binding the photosensitive exposure trigger time of the high-speed industrial camera, the force data latching time of the six-dimensional torque sensor, and the oscillation sampling time of the three-axis accelerometer to the same hardware level transition edge of the equally spaced synchronization pulse signal output by the hardware synchronization trigger board. For example, it can be uniformly configured to be synchronized triggering on the rising edge of the pulse signal. Through this fine-grained physical timing configuration of the changes in the underlying hardware electrical signal level, it is possible to ensure that the sensor state data of all different modes are frozen and synchronously acquired at the same physical microsecond.

[0095] By combining the global timestamp calculation formula tn=n×Ts, the master control device can accurately assign a strict hardware timestamp to the synchronization dataset packaged within a single trigger cycle. Here, tn can represent the global timestamp of the nth trigger, in seconds, n can represent the trigger sequence number, which is a positive integer, and Ts can represent the global synchronization trigger cycle with a fixed value of one millisecond.

[0096] As can be seen, in this embodiment, the main control device of the test system realizes the trigger timing setting through pure physical signal lines, which can eliminate the random time difference and system communication delay caused by the traditional upper-layer software polling and reading different industrial buses. In this way, it can provide an unbiased basic sensor dataset aligned on the physical time axis for subsequent high-precision, low-latency spatiotemporal fusion calculation of multi-source sensor data.

[0097] In one possible embodiment, after the mating action is completed, the method further includes: acquiring connector pose data, steady-state contact force data, and electrical connection signal after mating; constructing a successful docking determination rule based on a preset pose positioning deviation, a preset steady-state contact force range, and a preset electrical connection state; determining the docking status according to the connector pose data, the steady-state contact force data, and the electrical connection signal using the successful docking determination rule; and outputting a docking completion signal to the upper control system if the docking is determined to be successful.

[0098] The connector pose data may include the final six-degree-of-freedom physical pose of the connector under test in the coordinate system of the docking execution end effector. The steady-state contact force data may include the stable physical environment contact force data that is collected and output in real time by the six-dimensional force sensing unit on the end effector after the mating action has completely stopped. The electrical connection conduction signal may include the actual electrical conduction feedback signal generated when the internal electrical circuit is successfully connected after the physical pins of the connector under test are successfully mated.

[0099] In specific implementation, the rules for determining successful docking can include obtaining the positioning deviation threshold, the safe steady-state range of the insertion force, and the effective Boolean feedback state of the physical electrical connection, based on the spatial pose set in the task scheduling module of the software application layer, to construct a multi-dimensional comprehensive logical judgment condition.

[0100] The docking status determination process can involve the main control device using the aforementioned real-time multi-source steady-state data as input to comprehensively compare whether the final pose deviation meets the high-precision repeatable docking requirements stipulated in the original protocol. For example, whether the physical space deviation is stably controlled within ±0.02 mm, while simultaneously determining whether the steady-state contact force is within a reasonable safety threshold range and whether the electrical connection is truly conductive. This comprehensive process completes a rigorous logical determination and status confirmation of the final docking status. If docking is determined to be successful, a docking completion signal is output to the upper control system. In other words, the main control device transmits the successful docking status determination result to the main control computer of the test system through the interface layer to complete the closed loop of the current test process and save the output docking process data log.

[0101] If the docking fails, the abnormal hierarchical handling mechanism can be triggered, directly returning to the pre-step of multi-source sensor data synchronous acquisition. That is, the original multi-source sensor dataset with global hardware timestamp is re-acquired based on hardware synchronous triggering of the board, and the subsequent dynamic pose real-time solution, trajectory prediction and docking process are re-executed.

[0102] As can be seen, in this embodiment, the main control equipment of the testing system can effectively realize fully automatic unmanned docking and status self-inspection after multi-directional vibration testing of products through a strict closed-loop judgment and self-healing feedback mechanism based on multi-source state data. This eliminates the need for manual intervention and adjustment, greatly improves the automation and continuity of the infrared product testing process on the production line, and effectively shortens the single-product testing cycle.

[0103] Please see Figure 5The overall process of this application can be divided into a closed-loop control loop with six core steps. First, the system performs step 1, system calibration and parameter initialization, completing hand-eye calibration and configuring basic parameters such as the global hardware synchronization trigger cycle, providing an extremely accurate static spatial reference and clock foundation for subsequent multi-source fusion. Next, the system enters step 2, multi-source sensor data synchronous acquisition and alignment, acquiring low-level physical datasets such as visual images with global timestamps, three-axis acceleration, and six-dimensional contact force in real time through a hardware-level trigger mechanism. Subsequently, the system performs step 3, EKF multi-source fusion real-time pose calculation, using an extended Kalman filter framework to perform spatiotemporal fusion processing on the above-mentioned synchronously aligned multi-source data, filtering out high-frequency vibration noise and visual acquisition delay, and calculating a high-precision real-time spatial dynamic pose of the connector. Based on this physical tracking, the system enters step 4, temporal LSTM model dynamic trajectory prediction, inputting historical temporal poses into a long short-term memory network model, and feedforward predicting the zero-crossing point of future high-frequency vibration offset, thereby accurately locking the target mating pose and optimal mating timing of the tested connector. Next, the system executes step 5, impedance control trajectory correction and compliant mating, guiding the underlying servo motorized slide to approach the target pose along the planned trajectory. During the mating contact phase, the system dynamically corrects stiffness and damping parameters based on force feedback to avoid rigid collisions and smoothly complete the physical mating action. Finally, the system enters step 6, state determination and process closure, comprehensively examining the pose deviation, steady-state contact force, and electrical continuity status after mating for comprehensive confirmation. If the determination is successful, a completion signal is output and the process ends. Simultaneously, this mechanism has automatic self-healing capabilities. If the determination fails, a retry loop is triggered, and the system automatically reverts and seamlessly jumps to step 2 to re-acquire multi-source sensor data and iteratively track the entire subsequent process, thus forming an intelligent test closed loop that combines high precision and robustness without any manual intervention.

[0104] The following describes a high-precision automatic connector docking device under vibration testing conditions provided in this application. The high-precision automatic connector docking device under vibration testing conditions described below corresponds to the high-precision automatic connector docking method under vibration testing conditions described above.

[0105] Please see Figure 6The high-precision automatic docking device 600 for connectors under vibration testing conditions includes: an acquisition unit 601, used to acquire a multi-source sensor dataset with a global synchronization timestamp, wherein the global synchronization timestamp is used to unify the time series reference of the multi-source data in the multi-source sensor dataset, and the multi-source data includes real-time image data of the connector of the product under test, three-axis vibration acceleration data of the vibration table, and six-dimensional contact force data at the docking execution end; and a processing unit 602, used to perform fusion processing on the multi-source sensor dataset to obtain the coordinate system of the connector of the product under test at the docking execution end. The system includes a real-time six-degree-of-freedom dynamic pose sequence; a prediction unit 603, used to predict the target mating pose and optimal mating timing of the connector of the tested product based on the real-time six-degree-of-freedom dynamic pose sequence and the vibration acceleration data; a planning unit 604, used to plan the docking trajectory based on the target mating pose and the optimal mating timing; and an execution unit 605, used to control the docking execution mechanism to perform the mating action according to the docking trajectory, and to correct the docking trajectory based on the six-dimensional contact force data acquired in real time during the mating process, so as to complete the mating action.

[0106] In one possible embodiment, regarding the fusion processing of the multi-source sensor dataset to obtain the real-time six-DOF dynamic pose sequence of the tested product connector in the docking execution end coordinate system, the processing unit 602 is specifically used for: extracting features from the real-time image data to obtain the initial pose of the tested product connector in the camera coordinate system; performing spatiotemporal fusion of multi-source data using the vibration acceleration data as control input and the initial pose as the observation value to obtain the optimal pose estimate; and transforming the optimal pose estimate to the docking execution end coordinate system using a preset hand-eye transformation matrix to obtain the real-time six-DOF dynamic pose sequence, wherein the hand-eye transformation matrix indicates the homogeneous spatial transformation relationship between the camera coordinate system and the docking execution end coordinate system.

[0107] In one possible embodiment, regarding the process of fusing multi-source data using the vibration acceleration data as control input and the initial pose as observation value to obtain the optimal pose estimate, the processing unit 602 is specifically configured to: construct a system state vector containing the six-degree-of-freedom pose and pose change rate of the connector of the product under test; obtain a priori predicted value of the system state vector based on the vibration acceleration data and a preset state transition matrix, wherein the state transition matrix indicates the dynamic evolution law of the system state vector; and complete the posterior update of the system state vector based on the prior predicted value of the system state vector, using the initial pose as observation value, to obtain the optimal pose estimate.

[0108] In one possible embodiment, in the step of using the initial pose as the observation value and completing the posterior update of the system state vector based on the prior predicted value of the system state vector to obtain the optimal pose estimate, the processing unit 602 is specifically used to: calculate the observation residual based on the preset observation matrix and the prior predicted value of the system state vector; calculate the optimal Kalman gain through the process noise covariance and the observation noise covariance; and correct the prior predicted value of the system state vector based on the optimal Kalman gain and the observation residual to obtain the optimal pose estimate.

[0109] In one possible embodiment, in predicting the target mating pose and optimal mating timing of the connector of the product under test based on the real-time six-degree-of-freedom dynamic pose sequence and the vibration acceleration data, the prediction unit 603 is specifically used to: normalize the real-time six-degree-of-freedom dynamic pose sequence and the vibration acceleration data of the corresponding frames from a series of historical frames to obtain normalized time-series data; input the time-series data into a pre-trained long short-term memory network model to perform time-series feature extraction and trajectory prediction, and output a pose prediction sequence for the future time period; determine the moment when the vibration crosses zero in the pose prediction sequence as the optimal mating timing, and determine the pose corresponding to the vibration crossing zero as the target mating pose.

[0110] In one possible embodiment, before acquiring the multi-source sensor dataset with global synchronization timestamps, the acquisition unit 601 is further configured to: solve for the hand-eye transformation matrix between the camera coordinate system and the docking execution end coordinate system based on the hand-eye calibrated data; configure the global hardware synchronization trigger cycle; and set the sampling frequency and triggering sequence of the multi-source data according to the global hardware synchronization trigger cycle.

[0111] In one possible embodiment, obtaining the hand-eye transformation matrix between the camera coordinate system and the docking execution end effector coordinate system based on the hand-eye calibration data includes: controlling the docking execution end effector to move the calibration board to multiple different poses, acquiring multiple sets of calibration board images in different poses; extracting the calibration board corner features from the calibration board images; obtaining the camera in-camera participation distortion coefficients based on the calibration board corner features; determining the pose of the calibration board in the camera coordinate system based on the camera in-camera participation distortion coefficients; acquiring the end effector coordinate system pose of the calibration board in each pose; acquiring the pose correspondence between the pose in the camera coordinate system and the pose in the end effector coordinate system; and, based on the pose correspondence, obtaining the hand-eye transformation matrix using a hand-eye calibration algorithm, based on the end effector coordinate system pose in each pose and the pose of the calibration board in the camera coordinate system.

[0112] In one possible embodiment, regarding the setting of the sampling frequency and trigger timing of the multi-source data according to the global hardware synchronization trigger cycle, the acquisition unit 601 is specifically used to: output equally spaced synchronization pulse signals through the hardware synchronization trigger board according to the global hardware synchronization trigger cycle; configure the sampling frequency of the multiple sensing units that acquire the multi-source data; and configure the trigger edge of the synchronization pulse signals of the multiple sensing units to complete the trigger timing setting of the multi-source data.

[0113] In one possible embodiment, after the mating action is completed, the execution unit 605 is further configured to: collect the connector pose data, steady-state contact force data, and electrical connection signal after the mating is completed; construct a docking success determination rule based on a preset pose positioning deviation, a preset steady-state contact force range, and a preset electrical connection state; determine the docking status according to the connector pose data, the steady-state contact force data, and the electrical connection signal through the docking success determination rule; if the docking is determined to be successful, output a docking completion signal to the upper control system.

[0114] Please see Figure 7 , Figure 7 This is a structural schematic diagram of the main control device provided in this application. For example... Figure 7 As shown, the main control device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call logic instructions in the memory 730 to execute a high-precision automatic docking method for connectors under vibration testing conditions. This method includes: acquiring a multi-source sensor dataset with a global synchronization timestamp, wherein the global synchronization timestamp is used to unify the timing reference of the multi-source data in the multi-source sensor dataset; the multi-source data includes real-time image data of the connector under test, three-axis vibration acceleration data of the vibration table, and six-dimensional contact force data of the docking execution end; fusing the multi-source sensor dataset to obtain a real-time six-degree-of-freedom dynamic pose sequence of the connector under test in the coordinate system of the docking execution end; predicting the target mating pose and optimal mating timing of the connector under test based on the real-time six-degree-of-freedom dynamic pose sequence and the vibration acceleration data; planning a docking trajectory based on the target mating pose and the optimal mating timing; controlling the docking execution mechanism to perform the mating action based on the docking trajectory, and correcting the docking trajectory based on the real-time acquired six-dimensional contact force data during the mating process to complete the mating action.

[0115] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a high-precision automatic docking method for connectors under vibration testing conditions provided by the methods described above. This method includes: acquiring a multi-source sensor dataset with a globally synchronized timestamp, wherein the globally synchronized timestamp is used to unify the timing reference of the multi-source data in the multi-source sensor dataset, and the multi-source data includes real-time image data of the connector of the product under test, triaxial vibration acceleration data of the vibration table, and six-dimensional contact force data at the docking execution end; and processing the data... The multi-source sensor dataset is fused to obtain a real-time six-DOF dynamic pose sequence of the connector of the product under test in the coordinate system of the docking execution end. Based on the real-time six-DOF dynamic pose sequence and the vibration acceleration data, the target mating pose and optimal mating timing of the connector of the product under test are predicted. The docking trajectory is planned based on the target mating pose and the optimal mating timing. The docking execution mechanism is controlled to perform the mating action according to the docking trajectory, and the docking trajectory is corrected based on the six-dimensional contact force data acquired in real time during the mating process to complete the mating action.

[0117] In another aspect, this application also provides a computer program product, including a computer program that, when executed by a processor, implements a high-precision automatic docking method for connectors under any of the vibration testing conditions described above. This method includes: acquiring a multi-source sensor dataset with a globally synchronized timestamp, wherein the globally synchronized timestamp is used to unify the temporal reference of the multi-source data in the multi-source sensor dataset; the multi-source data includes real-time image data of the connector under test, three-axis vibration acceleration data of the vibration table, and six-dimensional contact force data of the docking execution end; fusing the multi-source sensor dataset to obtain a real-time six-degree-of-freedom dynamic pose sequence of the connector under test in the coordinate system of the docking execution end; predicting the target mating pose and optimal mating timing of the connector under test based on the real-time six-degree-of-freedom dynamic pose sequence and the vibration acceleration data; planning a docking trajectory based on the target mating pose and the optimal mating timing; controlling the docking execution mechanism to perform the mating action based on the docking trajectory, and correcting the docking trajectory based on the real-time acquired six-dimensional contact force data during the mating process to complete the mating action.

[0118] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A high-precision automatic docking method for connectors under vibration testing conditions, characterized in that, include: Acquire a multi-source sensor dataset with a global synchronization timestamp, wherein the global synchronization timestamp is used to unify the time series reference of the multi-source data in the multi-source sensor dataset, and the multi-source data includes real-time image data of the connector of the product under test, three-axis vibration acceleration data of the vibration table, and six-dimensional contact force data of the docking execution end. The multi-source sensor dataset is fused to obtain the real-time six-degree-of-freedom dynamic pose sequence of the connector of the tested product in the coordinate system of the docking execution end. The real-time six-degree-of-freedom dynamic pose sequence of multiple consecutive historical frames and the vibration acceleration data of the corresponding frames are normalized to obtain normalized time-series data. The time-series data is input into a pre-trained long short-term memory network model to extract time-series features and predict trajectories, and outputs a pose prediction sequence for future time periods. In the pose prediction sequence, the moment when the vibration crosses zero is determined as the optimal interpolation timing, and the pose corresponding to the vibration crossing zero is determined as the target interpolation pose. The optimal engagement timing is used as the time node constraint for the engagement action, and the target engagement pose is used as the spatial motion endpoint to plan the docking trajectory. The docking actuator is controlled to perform the insertion action according to the docking trajectory, and the docking trajectory is corrected according to the six-dimensional contact force data acquired in real time during the insertion process, so as to complete the insertion action.

2. The method according to claim 1, characterized in that, The process of fusing the multi-source sensor dataset to obtain the real-time six-degree-of-freedom dynamic pose sequence of the connector of the tested product in the coordinate system of the docking execution end point includes: Feature extraction is performed on the real-time image data to obtain the initial pose of the connector of the tested product in the camera coordinate system; Using the vibration acceleration data as the control input and the initial pose as the observation value, multi-source data fusion is performed to obtain the optimal pose estimate. By using a preset hand-eye transformation matrix, the optimal pose estimate is transformed to the docking execution end coordinate system to obtain a real-time six-degree-of-freedom dynamic pose sequence. The hand-eye transformation matrix indicates the homogeneous spatial transformation relationship between the camera coordinate system and the docking execution end coordinate system.

3. The method according to claim 2, characterized in that, The process of fusing multi-source data, using the vibration acceleration data as control input and the initial pose as observation value, to obtain the optimal pose estimate includes: Construct a system state vector containing the six-DOF pose and pose change rate of the connector of the product under test; Based on the vibration acceleration data and the preset state transition matrix, the prior prediction value of the system state vector is obtained, and the state transition matrix indicates the dynamic evolution law of the system state vector. Using the initial pose as the observation value, the posterior update of the system state vector is completed based on the prior prediction value of the system state vector to obtain the optimal pose estimate.

4. The method according to claim 3, characterized in that, The step of using the initial pose as the observation value and completing the posterior update of the system state vector based on the prior predicted value of the system state vector to obtain the optimal pose estimate includes: The observation residuals are calculated based on the preset observation matrix and the prior prediction values ​​of the system state vector. The optimal Kalman gain is calculated by using the process noise covariance and the observation noise covariance. Based on the optimal Kalman gain and the observation residual, the prior prediction value of the system state vector is corrected to obtain the optimal pose estimate.

5. The method according to any one of claims 1-4, characterized in that, Before acquiring the multi-source sensor dataset with a globally synchronized timestamp, the method further includes: Based on the data after hand-eye calibration, the hand-eye transformation matrix between the camera coordinate system and the docking execution end coordinate system is obtained; Configure the global hardware synchronization trigger cycle; set the sampling frequency and triggering sequence of the multi-source data according to the global hardware synchronization trigger cycle.

6. The method according to claim 5, characterized in that, The step of obtaining the hand-eye transformation matrix between the camera coordinate system and the docking execution end coordinate system based on the hand-eye calibration data includes: The control docking execution end drives the calibration board to move to multiple different poses, and acquires multiple sets of calibration board images under different poses; Extract the calibration board corner features from the calibration board image; The in-camera distortion coefficients are obtained based on the corner features of the calibration board. The pose of the calibration plate in the camera coordinate system is determined based on the in-camera distortion coefficients. Obtain the end coordinate system pose of the calibration plate in each pose; Obtain the pose correspondence between the pose in the camera coordinate system and the pose in the end-effector coordinate system; Based on the pose correspondence, the hand-eye transformation matrix is ​​obtained by using a hand-eye calibration algorithm, based on the end-effector pose in each pose and the pose of the calibration board in the camera coordinate system.

7. The method according to claim 5, characterized in that, The step of setting the sampling frequency and triggering sequence of the multi-source data according to the global hardware synchronization triggering cycle includes: According to the global hardware synchronization trigger cycle, the hardware synchronization trigger board outputs equally spaced synchronization pulse signals. Configure the sampling frequency of multiple sensing units that collect the multi-source data; Configure the trigger edge of the synchronization pulse signal of the multiple sensing units to complete the trigger timing setting of the multi-source data.

8. The method according to any one of claims 1-4, characterized in that, After the insertion action is completed, the method further includes: Collect the connector pose data, steady-state contact force data, and electrical connection signal after the mating is completed; Based on preset position deviation, preset steady-state contact force range and preset electrical conduction state, a docking success determination rule is constructed. The docking status is determined based on the connector pose data, the steady-state contact force data, and the electrical connection signal according to the docking success determination rules. If the docking is determined to be successful, a docking completion signal is output to the upper control system.

9. A high-precision automatic docking device for connectors under vibration testing conditions, characterized in that, include: The acquisition unit is used to acquire a multi-source sensor dataset with a global synchronization timestamp, wherein the global synchronization timestamp is used to unify the time series reference of the multi-source data in the multi-source sensor dataset, and the multi-source data includes real-time image data of the connector of the product under test, three-axis vibration acceleration data of the vibration table, and six-dimensional contact force data of the docking execution end. The processing unit is used to perform fusion processing on the multi-source sensor dataset to obtain the real-time six-degree-of-freedom dynamic pose sequence of the connector of the product under test in the coordinate system of the docking execution end. The prediction unit is used to normalize the real-time six-degree-of-freedom dynamic pose sequence of multiple consecutive historical frames and the vibration acceleration data of the corresponding frames to obtain normalized time-series data; input the time-series data into a pre-trained long short-term memory network model to perform time-series feature extraction and trajectory prediction, and output a pose prediction sequence for the future time period; determine the moment when the vibration crosses zero in the pose prediction sequence as the optimal interpolation time, and determine the pose corresponding to the vibration crossing zero as the target interpolation pose; The planning unit is used to plan the docking trajectory by taking the optimal engagement timing as the time node constraint for the engagement action and the target engagement pose as the spatial motion endpoint. The execution unit is used to control the docking actuator to perform the insertion action according to the docking trajectory, and to correct the docking trajectory according to the six-dimensional contact force data acquired in real time during the insertion process, so as to complete the insertion action.

Images