Intelligent equipment butt-assembling cooperative control system with multi-axis linkage compensation

Abstract

The invention relates to the field of intelligent equipment control and industrial automation, in particular to an intelligent equipment butt-assembling cooperative control system with multi-axis linkage compensation. The data acquisition module is used for acquiring current, force / torque, temperature, position and speed data under a unified time reference; the benchmark reconstruction module is used for generating an ideal position benchmark and an ideal force / torque benchmark on the basis of an ideal rigid body model under a preset reference coordinate system; the parameter injection module is used for injecting thermoelastic deformation, rigidity attenuation and hysteresis friction parameters to generate a theoretical simulation state; constructing a double-track difference module of a real residual error and a theoretical residual error; the coupling judgment module is used for carrying out time sequence similarity verification and generating a state judgment result; the feedback control module is used for outputting a multi-axis linkage compensation instruction or a compliant return instruction; according to the invention, inherent drift and real assembly interference of equipment can be distinguished, and stable compensation and safe concession control are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment control and industrial automation, specifically to an intelligent equipment assembly and coordination control system with multi-axis linkage compensation. Background Technology

[0002] In high-precision assembly scenarios such as semiconductor packaging, precision insertion, and multi-axis collaborative assembly, actuators are typically required to complete position alignment, contact pressing, and attitude correction under the linkage of multiple motion axes. They are also controlled by combining information such as position, force, and drive status. Therefore, multi-axis collaborative control technology has become a common means to ensure assembly accuracy and operational stability.

[0003] As the precision requirements of intelligent manufacturing increase, existing control methods are still easily affected by factors such as thermal drift, structural flexible deformation, transmission hysteresis friction, and cycle fluctuation in practical applications. This makes it difficult to effectively distinguish between the background deviation of the equipment itself and the actual assembly interference, resulting in problems such as false compensation, false alarms, or untimely retreat, which limits the stability and safety of the precision assembly process. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent equipment assembly and collaborative control system with multi-axis linkage compensation, solving the following technical problems: It avoids misjudging inherent background deviations such as thermal drift, flexible deformation, and transmission hysteresis of equipment as real assembly anomalies, and can effectively separate the equipment's own errors from the actual assembly interference from the same observation data, thereby achieving stable compensation and safe retreat control of multi-axis equipment under complex working conditions.

[0005] The objective of this invention can be achieved through the following technical solutions: An intelligent equipment assembly and coordination control system with multi-axis linkage compensation includes: An execution node includes multiple controlled motion axes and corresponding drive units; The data acquisition module is used to acquire the timing current data, force / torque data, temperature data, position data and velocity data of the execution node under a unified time reference. The benchmark reconstruction module is used to generate ideal position benchmarks and ideal force / moment benchmarks based on an ideal rigid body model in a preset reference coordinate system, according to the preset task trajectory and constraints. The parameter injection module is used to inject preset thermoelastic deformation parameters, stiffness attenuation parameters and hysteresis friction parameters into the ideal rigid body model to generate a theoretical simulation state containing theoretical position sequence and theoretical force / torque response sequence. The dual-track differential module is used to map the time-series current data, force / torque data, temperature data, position data, and velocity data into a real state vector, and to perform differential analysis with the ideal state vector corresponding to the ideal position reference and the ideal force / torque reference to generate a real residual vector sequence. It is also used to perform differential analysis with the theoretical simulation state and the ideal position reference and the ideal force / torque reference to generate a theoretical residual vector sequence. The coupling decision module is used to perform temporal similarity verification between the actual residual vector sequence and the theoretical residual vector sequence, and generate a state decision result. The feedback control module is used to output multi-axis linkage compensation commands or compliant return commands to the drive unit based on the state judgment result.

[0006] Furthermore, the data acquisition module includes: The current loop acquisition unit is used to acquire high-frequency timing current data of the servo drive link; Force sensing acquisition unit is used to acquire multi-dimensional spatial force and torque data; Thermal state acquisition unit, used to acquire distributed temperature data; The position acquisition unit is used to acquire absolute coded position data and velocity data; The time synchronization unit is used to perform unified timestamp calibration and resampling alignment on all collected data.

[0007] Furthermore, the reference reconstruction module is used to construct an ideal rigid body model without thermal drift, flexible deformation, and frictional disturbance in the preset reference coordinate system and the unified time reference based on the preset task trajectory, preset contact constraints, and preset dynamic constraints, and to calculate the ideal position reference and ideal force / torque reference corresponding to each moment based on the ideal rigid body model.

[0008] Furthermore, the parameter injection module is used to map the temperature data into a thermoelastic deformation disturbance, map the position data, velocity data, and force / torque data together with the execution node mechanism model into an equivalent dynamic stiffness disturbance, map the time-series current data, position data, and velocity data into a hysteretic friction disturbance, and couple the thermoelastic deformation disturbance, the dynamic stiffness disturbance, and the hysteretic friction disturbance into the ideal rigid body model to generate the theoretical simulation state characterizing the inherent drift of the device.

[0009] Furthermore, the dual-track differential module is used to map the time-series current data, force / torque data, temperature data, position data, and velocity data into a real-state vector, map the ideal position reference and ideal force / torque reference into an ideal-state vector, and perform differential processing on the real-state vector and the ideal-state vector to generate the real-residual vector sequence. The theoretical simulation state is mapped to a theoretical state vector and then differentially processed with the ideal state vector to generate the theoretical residual vector sequence.

[0010] Furthermore, the coupling decision module is used to perform temporal alignment between the actual residual vector sequence and the theoretical residual vector sequence using dynamic time warping; and to process the temporal alignment result using multidimensional cosine similarity to generate structural similarity. It extracts at least one peak feature from the instantaneous peak amplitude and peak duration in the real residual vector sequence, and at least one impedance jump feature from the equivalent stiffness change rate and equivalent damping change rate calculated based on the force-displacement relationship or the torque-angular displacement relationship, to generate abnormal feature results.

[0011] Furthermore, the coupling decision module is also used for: The preset decision criteria are determined by performing multiple rounds of calibration tests on the equipment under no-interference, unloaded operation, statistically analyzing historical distribution data of structural similarity and abnormal features, and using boundary values ​​covering a 95% confidence interval. Based on the preset decision criteria, preset high threshold, preset low threshold, and preset abnormal threshold are obtained. When the preset high threshold is greater than the preset low threshold, and the structural similarity is greater than or equal to the preset high threshold and the abnormal feature result is less than or equal to the preset abnormal threshold, a safe state decision result is output. When the structural similarity is less than or equal to the preset low threshold or the abnormal feature result is greater than the preset abnormal threshold, a true anomaly judgment result is output. When the structural similarity is greater than the preset low threshold and less than the preset high threshold, and the abnormal feature result is less than or equal to the preset abnormal threshold, the judgment result to be verified is output.

[0012] Furthermore, the feedback control module is used to generate a compensation vector based on the theoretical residual vector sequence when receiving the safe state decision result, and output a multi-axis linkage feedforward compensation command to the drive unit based on the compensation vector; Upon receiving the true anomaly judgment result, the multi-axis linkage feedforward compensation command is frozen, and a compliant retraction command is output to reduce the stiffness of the contact shaft and drive the execution node to retract along the preset retraction direction. Upon receiving the judgment result to be verified, a fine-tuning retry instruction and a retest trigger instruction are output to drive the execution node to re-execute the preset operation task according to the preset small pose increment and trigger the data acquisition module to retest.

[0013] Furthermore, it also includes a parameter adaptation module, which is used to iteratively update the thermoelastic deformation parameter, the stiffness attenuation parameter and the hysteresis friction parameter based on the state decision result and the principle of minimizing the difference between the actual residual vector sequence and the theoretical residual vector sequence; The feedback control module regenerates the multi-axis linkage compensation command based on the updated parameters to form a closed-loop control.

[0014] Furthermore, the execution node is an execution mechanism with multi-axis coordinated motion capability, and the execution mechanism includes multiple controlled motion axes and corresponding drive units.

[0015] The beneficial effects of this invention are: 1. This invention generates a theoretical simulation state by injecting thermoelastic deformation, stiffness attenuation and hysteretic friction disturbance into an ideal rigid body model, and uses a dual-track mechanism to perform differential comparison between the actual residual and the theoretical residual. This mechanism effectively solves the problem that it is difficult to distinguish between background deviations such as thermal drift and flexible deformation and real interference, and can accurately separate the inherent error of the equipment, avoiding false compensation and false alarm caused by spontaneous drift of the equipment. 2. This invention addresses the time misalignment problem caused by process cycle fluctuations in actual production. It utilizes dynamic time warping to align the residual sequence in time and combines multidimensional cosine similarity with local anomaly features for coupled judgment. This method eliminates the misjudgment interference caused by time offset and can accurately capture transient interference changes when the overall trend is similar, significantly improving the accuracy of assembly anomaly identification. 3. This invention constructs a hierarchical feedback control mechanism, which outputs multi-axis linkage compensation, compliant retreat, or fine-tuning retry commands according to the judgment result; it performs collaborative correction when it is safe, quickly freezes compensation and reduces stiffness retreat when an anomaly is detected, and triggers fine-tuning retest when verification is pending; this strategy effectively avoids overload damage to the device and takes into account both the stability and safety of operation under complex working conditions. 4. In response to transmission wear and environmental changes caused by long-term operation of equipment, this invention introduces a parameter adaptive module, which iteratively updates various disturbance parameters online based on the principle of minimizing residual differences. This mechanism enables the internal model of the system to be dynamically corrected as the actual physical state of the equipment evolves, maintaining a high-precision background error modeling capability over a long period of time, and ensuring the stability of closed-loop control for continuous operation of the actuator. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings; Figure 1 This is a schematic diagram of a module of an intelligent device assembly and collaborative control system with multi-axis linkage compensation provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 A smart device assembly and coordination control system with multi-axis linkage compensation includes: an execution node, which includes multiple controlled motion axes and corresponding drive units; and a data acquisition module, used to acquire the timing current data, force / torque data, temperature data, position data and speed data of the execution node under a unified time reference. The benchmark reconstruction module is used to generate ideal position benchmarks and ideal force / moment benchmarks based on an ideal rigid body model in a preset reference coordinate system, according to the preset task trajectory and constraints. The parameter injection module is used to inject preset thermoelastic deformation parameters, stiffness attenuation parameters and hysteresis friction parameters into the ideal rigid body model to generate a theoretical simulation state containing theoretical position sequence and theoretical force / torque response sequence. The dual-track differential module is used to map the time-series current data, force / torque data, temperature data, position data, and velocity data into a real state vector, and to perform differential analysis with the ideal state vector corresponding to the ideal position reference and the ideal force / torque reference to generate a real residual vector sequence. It is also used to perform differential analysis with the theoretical simulation state and the ideal position reference and the ideal force / torque reference to generate a theoretical residual vector sequence. The coupling decision module is used to perform time-series similarity verification between the actual residual vector sequence and the theoretical residual vector sequence, and generate a state decision result; the feedback control module is used to output a multi-axis linkage compensation command or a compliant return command to the drive unit according to the state decision result.

[0019] This embodiment provides an intelligent device assembly and collaborative control mechanism with multi-axis linkage compensation; the following description uses a precision insertion and alignment station in a semiconductor packaging line as a scenario; this station includes an execution node, which can be equipped with X, Y, and Z translation axes and , , The six-axis coordinated motion platform with rotating shaft carries the device to be inserted at its end, requiring the device pins to be aligned with the micro-hole array on the base. The key challenge in this scenario is not whether resistance is detected, but whether the detected resistance originates from thermal drift, flexible off-center loading, and transmission hysteresis of the device itself, or from actual assembly interference between the device and the base.

[0020] Specifically, when executing node operations, the data acquisition module synchronously collects multiple types of data under a unified time base and forms an observation sequence under the same time-series index; at time... After preprocessing, various types of data are organized into real-world observations: ; Among them, the current term Corresponding to the timing current data of each axis, force term For stress / torque data, the temperature term Corresponding temperature data, location item Corresponding location data, velocity item Corresponding velocity data; the reference reconstruction module, based on the preset task trajectory, contact constraints, and dynamic constraints, establishes an ideal rigid body model in a preset reference coordinate system to obtain the ideal position reference. Compared with ideal force / torque reference Here, "ideal" means that thermal drift, flexible deformation, and frictional disturbances are not considered.

[0021] The parameter injection module injects the known physical errors of the device itself into the ideal rigid body model; for ease of explanation, the thermoelastic deformation parameters are denoted as... Let the stiffness attenuation parameter be denoted as Let the hysteresis friction parameter be denoted as ;

[0022] In a simplified calculation example, the ideal pose at a certain moment is: um, ideal contact force is N; when the equivalent position drift caused by temperature rise is um, the equivalent compressive displacement caused by flexible eccentric loading is um, the equivalent reverse hysteresis caused by hysteresis friction is When um, the position corresponding to the theoretical simulation state is represented as: ; Among them, theoretical position perturbation It is obtained by superimposing the thermal drift term, stiffness term, and friction term, and its calculation formula is: ; Similarly, the theoretical force response can be written as: ; The theoretical simulation state obtained is not an arbitrarily assumed abnormal state, but rather the background response that the sensor should exhibit when there is no interference in the assembly itself and only the inherent drift of the equipment exists.

[0023] The dual-track difference module simultaneously constructs the actual residual and the theoretical residual; the actual state vector is represented as: ; The ideal state vector is represented as: ; The theoretical state vector is represented as: ; in, and A pre-defined diagonal scaling matrix is ​​used to perform dimensionless processing on each physical quantity, with the scaling denominator being the calibrated range of each sensor. The processed data is then concatenated into vectors according to a pre-defined fixed field order to map it to a unified state space; thus, the following is obtained: ; ; If the actual residual and the theoretical residual highly overlap throughout the entire insertion time window, it indicates that the deviation mainly comes from the inherent error of the equipment; if the two show a significant divergence, it indicates that there are still actual assembly anomalies after eliminating the equipment's own errors.

[0024] Coupled decision module and Perform temporal similarity verification; in the simplified case, take a short window of length 4, if ,and If the first three sampling points are basically consistent, but a spike appears at the fourth sampling point; such spikes usually cannot be explained by thermal drift and flexible deformation alone, so they will be judged as a real interference sign; the feedback control module outputs different control commands accordingly: if it is determined that the inherent error of the equipment is dominant, a multi-axis linkage compensation command is issued to each drive unit to make the multi-axis correct the pose in a coordinated manner; if it is determined to be a real interference, a compliant retraction command is output to reduce the stiffness of the contact axis and withdraw from the contact area in a preset direction; As an abnormal operating condition handling, if a certain type of sensor experiences a short-term sampling gap under a unified time reference, the data acquisition module will prioritize resampling from adjacent sampling points to make up the gap; if the continuous missing length exceeds the preset tolerance window, the dual-track differential module will mark the time window as a low confidence window and prohibit the feedback control module from outputting enhanced compensation within the time window, only allowing it to be maintained or rolled back. When the ideal rigid body model fails to be solved, or when there is a conflict between the contact constraints and the task trajectory, the datum reconstruction module stops generating a new round of ideal datums and outputs a locking signal to the feedback control module to avoid continuing the assembly in the absence of a datum.

[0025] For example, during the precision insertion process, the execution node is pressed into the target hole along the Z-axis; as the pressing depth increases, the motor temperature rise causes the Z-axis lead screw to thermally expand, resulting in the position observation being 1.5µm larger than the ideal value, while the cantilever section exhibits flexible sinking during lateral correction; The system obtains the theoretical residual curve through parameter injection and finds that the actual residual is highly consistent with its overall trend within 30ms. Therefore, it is determined that the current situation is not mechanical jamming, but platform drift. Based on this, the feedback control module compensates -0.8um on the X-axis and -1.4um on the Z-axis, and continues to execute the insertion action. To further clarify, in order to avoid ambiguity between the real state vector and the ideal state vector in terms of dimension, the unified state space adopts a fixed field order during implementation. For observations that are not directly solved explicitly in the ideal rigid body model, reference placeholder values ​​with the same dimension as the real state vector are used for mapping. For example, the current dimension can use the expected driving current reference obtained by converting the ideal force / torque reference and the transmission ratio of the mechanism, the temperature dimension can use the initial calibration temperature or zero bias reference, and the velocity dimension can use the difference result of the ideal position reference on the unified time base as the reference velocity. Therefore, before entering the difference Already with They exist in the same dimensional space, while the theoretical state vector It is also constructed in the same field order, so as to ensure that each dimension of the actual residual and the theoretical residual has a clear physical correspondence, rather than just performing formal difference on vectors of different lengths.

[0026] To further clarify, the timing similarity check in this embodiment is not limited to a strict one-to-one comparison of the same sampling point, but allows for a window-based comparison of the actual residual and the theoretical residual within a preset process cycle tolerance before outputting the state decision result.

[0027] When the contact start time is slightly shifted forward or backward due to upstream beat fluctuations, as long as both maintain the same structural change trend within the allowable time shift range, it can be determined that it is mainly dominated by the inherent drift of the equipment. Conversely, if a spike, abrupt change, or continuous deviation occurs within the allowable time shift range that cannot be explained by the theoretical background response, it is determined to be a real assembly anomaly.

[0028] The purpose of this step is to separate the inherent drift of the equipment from the actual assembly interference from the same observation data by comparing the actual residuals and theoretical residuals with an ideal rigid body reference, thereby achieving stable compensation and safe retreat control of multi-axis equipment under complex working conditions.

[0029] In this embodiment of the invention, the data acquisition module includes: a current loop acquisition unit for acquiring high-frequency timing current data of the servo drive link; a force sensing acquisition unit for acquiring multi-dimensional spatial force and torque data; a thermal state acquisition unit for acquiring distributed temperature data; a position acquisition unit for acquiring absolute coded position data and velocity data; and a time synchronization unit for performing unified timestamp calibration and resampling alignment on each acquired data.

[0030] The reference reconstruction module is used to construct an ideal rigid body model without thermal drift, flexible deformation, and frictional disturbance in the preset reference coordinate system and the unified time reference, based on the preset task trajectory, preset contact constraints, and preset dynamic constraints. Based on the ideal rigid body model, the module calculates the ideal position reference and ideal force / torque reference corresponding to each moment.

[0031] The parameter injection module is used to map the temperature data into a thermoelastic deformation disturbance, map the position data, velocity data, and force / torque data together with the execution node mechanism model into an equivalent dynamic stiffness disturbance, map the time-series current data, position data, and velocity data into a hysteretic friction disturbance, and couple the thermoelastic deformation disturbance, the dynamic stiffness disturbance, and the hysteretic friction disturbance into the ideal rigid body model to generate the theoretical simulation state characterizing the inherent drift of the device.

[0032] This embodiment provides a joint mechanism for data acquisition, ideal reference reconstruction, and parameter injection for high-precision assembly scenarios. Specifically, in the precision insertion station of the semiconductor packaging line, relying solely on a single force sensor for threshold determination can easily lead to misjudging background deviations after equipment heating as assembly interference. On the other hand, if only the encoder position is observed, it is difficult to identify the true mechanical state during the contact process.

[0033] To address this deficiency, an interpretable theoretical simulation state is formed through multi-source acquisition under a unified time reference, reconstruction of an unperturbed ideal reference, and injection of physical perturbations. Specifically, the data acquisition module consists of multiple sub-units working together; the current loop acquisition unit is oriented towards the servo drive link and acquires the phase current or equivalent torque current of each axis at a preset sampling frequency; Since current is most sensitive to frictional reversal, overload transients, and local jamming, its time resolution can be higher than the position sampling frequency; the force sensing acquisition unit is set on the contact path to output six-dimensional force / torque data; the thermal state acquisition unit arranges multiple temperature measurement points along the lead screw seat, drive motor housing, guide rail base, and end platform; the position acquisition unit reads the absolute encoder position and obtains speed data through adjacent sampling differentials or driver feedback; the time synchronization unit maps data of different frequencies to the same time grid, for example, to a 1ms time base; For a given original temperature sequence , , While the current sequence is sampled every 0.5ms, the time synchronization unit can perform interpolation or hold-and-resample the temperature sequence, so that the temperature information can participate in the modeling at the same time as the current, position, and force information. Based on this, the benchmark reconstruction module establishes an ideal rigid body model according to the preset task trajectory, contact constraints, and dynamic constraints. Here, the task trajectory represents the standard motion sequence of the insertion device in the three stages of alignment, contact, and pressing. The contact constraints represent the boundaries that the normal force and tangential force should satisfy after the device pins are geometrically aligned with the target hole. The dynamic constraints represent the acceleration, driving force, and allowable inertial response range.

[0034] During baseline reconstruction, the model is configured to be free of thermal drift, flexible deformation, and frictional disturbance; that is, the thermal expansion coefficient, flexible compliance, and hysteresis terms are set to zero in the model. Therefore, at each time step... The ideal position reference can be solved. Compared with ideal force / torque reference For example, the ideal force reference should be close to zero before the initial contact of the insert; after entering the introduction stage, the ideal normal force will rise smoothly at a preset slope without any abnormal spikes.

[0035] Furthermore, the parameter injection module applies physical perturbations to the ideal rigid body model to obtain the theoretical simulation state; the thermoelastic deformation perturbation can be mapped to the geometric offset of each axis based on the temperature field, and can be simplified as follows: ; in, The thermal displacement mapping relationship is obtained and characterized by calibration. This characterizes the increment of each temperature measuring point relative to the reference temperature; in this embodiment, the aforementioned thermoelastic deformation parameter That is, the mapping matrix The core linear scaling factor; the aforementioned stiffness attenuation parameter Used to characterize the current equivalent stiffness Relative to the nominal stiffness of the execution node The attenuation ratio, i.e. The aforementioned hysteresis friction parameters For function Hysteresis width gain coefficient in the model.

[0036] If, during a certain operation, the motor housing rises by 8°C and the guide rail base rises by 3°C, and the calibration results show that the combined effect on the Z-axis end position is 0.12µm per degree Celsius, then thermal factors alone may cause a positional shift of approximately 1.32µm. Dynamic stiffness disturbances are mapped using position, velocity, force / torque, and actuator mechanism models; simplified, the equivalent contact stiffness can be obtained first: ; Among them, preset constants It has the dimension of displacement to avoid the denominator being zero; when the equipment undergoes flexible sinking due to cantilever eccentric loading during the high-speed fine-tuning phase, The stiffness will be lower than the nominal stiffness, and the parameter injection module will generate a corresponding stiffness attenuation disturbance accordingly; the hysteretic friction disturbance is jointly mapped by current, position, and velocity; if the sign of the velocity on a certain axis changes, and the current response has a fixed lag relative to the ideal torque, then the hysteretic friction term can be extracted: ; Then, convert these into equivalent position or force response perturbations; couple these three types of perturbations into the ideal rigid body model to obtain the theoretical position sequence and the theoretical force / torque response sequence: ; ; in, For theoretical force / torque disturbances, This refers to the positional disturbance caused by thermoelastic deformation. The positional disturbance is caused by stiffness decay. This refers to the positional disturbance caused by hysteresis friction. The above and These represent the force increment and displacement increment relative to the reference value at the current sampling point or within the current calculation window, respectively; when calculated based on adjacent sampling points, they can also be understood as the change between adjacent sampling points; correspondingly, where This represents the piecewise mapping rule for hysteresis friction based on current, position, and velocity.

[0037] Specifically, the segmented mapping rule is set as follows: when the absolute value of the velocity is greater than a preset steady-state threshold, the friction disturbance output is a constant dynamic friction bias; when the velocity approaches zero and the sign reverses, the friction disturbance output is a hysteresis compensation amount obtained by interpolating the current and position through a preset lookup table. Its output is the equivalent friction disturbance characterization value at that moment, rather than a new independent sensing quantity; therefore, the aforementioned... , In this embodiment, each disturbance component has a clear physical correspondence.

[0038] As an extreme abnormal working condition, if a temperature node of the thermal state acquisition unit fails, the parameter injection module will prioritize the estimation and compensation using spatially adjacent nodes and historical temperature rise slopes; if the failed node is located in a highly sensitive area and the continuous failure time exceeds the threshold, the thermoelastic deformation disturbance will only be estimated according to the conservative upper bound, and the theoretical simulation state will be marked as a low confidence state. In the initial contact if If the value is approximately zero, the equivalent stiffness calculation in the above formula may be unstable. In this case, the module should be changed to use the average displacement increment of the time window, or directly use the previous effective stiffness value to avoid a step change in the stiffness disturbance term.

[0039] For example, after the insertion station ran continuously for 40 minutes, the temperature of the Z-axis motor rose from 28°C to 37°C, and the temperature of the guide rail base rose from 27°C to 31°C. The system calculated through thermal mapping that there was a downward drift of 1.7 μm at the end. At the same time, due to the frequent reverse speed of the X-axis during lateral correction, the current loop data showed obvious hysteresis friction characteristics. The reference reconstruction module first provides the ideal insertion trajectory, and then the parameter injection module superimposes thermal, flexible and frictional disturbances to generate a theoretical response caused only by the device's own state if the device is aligned; the subsequent dual-track differential can then separate the real jamming from the background noise based on this.

[0040] Furthermore, the equivalent dynamic stiffness disturbance does not require direct calculation using the contact force to displacement ratio throughout the entire time period, but rather it is mapped in segments according to the process stage: in the non-contact stage, the structural compliance background is estimated using the mechanism model and motion state; In the contact stage, the relationship between force / torque and displacement and angular displacement is introduced to estimate the contact-related stiffness changes. The above processing method can avoid amplifying noise due to the force value being close to zero in the non-contact stage, and also make the stiffness attenuation parameter have a clear physical source when entering the parameter injection, that is, characterizing the degree of deviation of the execution node from the nominal mechanism stiffness, rather than all position errors being lumped into the stiffness term.

[0041] To further clarify, when implementing the hysteresis friction disturbance, it is prioritized to extract the velocity reversal zone, low-speed crawling zone, and steady-speed zone respectively; for the steady-speed zone where the absolute value of the velocity is higher than the preset threshold, the friction term mainly manifests as a relatively stable bias; for the zone where the velocity is close to zero and reversal occurs, the friction term mainly manifests as hysteresis width and response delay. The parameter injection module uses a segmented mapping rule under the same parameter set in different intervals to ensure that the theoretical simulation state can explain the slowly changing background drift and will not mistakenly equate the real interference at the moment of reversal to simple friction. The purpose of this mechanism is to provide a reliable theoretical background response for subsequent similarity verification, thereby achieving pre-modeling of inherent equipment errors and avoiding direct misjudgment of thermal drift and flexibility deviation as assembly anomalies.

[0042] In this embodiment of the invention, the dual-track differential module is used to map the time-series current data, force / torque data, temperature data, position data, and velocity data into real-state vectors, map the ideal position reference and ideal force / torque reference into ideal-state vectors, and perform differential processing on the real-state vectors and the ideal-state vectors to generate the real-state residual vector sequence; and after mapping the theoretical simulation state into a theoretical state vector, perform differential processing on it with the ideal-state vector to generate the theoretical residual vector sequence.

[0043] The coupling decision module is used to perform time-series alignment of the actual residual vector sequence and the theoretical residual vector sequence using dynamic time warping; to process the time-series alignment result using multidimensional cosine similarity to generate structural similarity; and to extract at least one peak feature from the instantaneous peak amplitude and peak duration in the actual residual vector sequence, as well as at least one impedance jump feature from the equivalent stiffness change rate and equivalent damping change rate calculated based on the force-displacement relationship or the torque-angular displacement relationship, to generate abnormal feature results.

[0044] The coupling decision module is also used for: the preset decision criteria are determined by performing multiple rounds of calibration tests on the equipment under interference-free no-load operation, statistically analyzing historical distribution data of structural similarity and abnormal features, and using boundary values ​​covering a 95% confidence interval; obtaining preset high threshold, preset low threshold, and preset abnormal threshold according to the preset decision criteria; under the condition that the preset high threshold is greater than the preset low threshold, when the structural similarity is greater than or equal to the preset high threshold and the abnormal feature result is less than or equal to the preset abnormal threshold, outputting a safe state decision result; when the structural similarity is less than or equal to the preset low threshold or the abnormal feature result is greater than the preset abnormal threshold, outputting a true abnormality decision result.

[0045] When the structural similarity is greater than the preset low threshold and less than the preset high threshold, and the abnormal feature result is less than or equal to the preset abnormal threshold, the judgment result to be verified is output.

[0046] The feedback control module is used to generate a compensation vector based on the theoretical residual vector sequence when the safe state judgment result is received, and output a multi-axis linkage feedforward compensation command to the drive unit based on the compensation vector; when the true anomaly judgment result is received, the multi-axis linkage feedforward compensation command is frozen and a compliant retraction command is output to reduce the stiffness of the contact shaft and drive the execution node to retract along the preset retraction direction. Upon receiving the judgment result to be verified, a fine-tuning retry instruction and a retest trigger instruction are output to drive the execution node to re-execute the preset operation task according to the preset small pose increment and trigger the data acquisition module to retest.

[0047] This embodiment provides a collaborative mechanism based on dual-track differential, similarity judgment, and hierarchical feedback control. Specifically, in the aforementioned workstation, even if the theoretical simulation state has been obtained, without a stable residual construction method and hierarchical handling logic, it will still fail in two extreme scenarios: First, assembly cycle fluctuations will cause the same physical processes to be misaligned in time, and direct point-by-point comparison will lead to misjudgment; Second, some real interferences are similar to background drift in overall trend, but will be accompanied by local spikes or impedance abrupt changes, and it is easy to miss the judgment if only a single similarity index is observed. To solve the above problems, a complete closed loop is formed by dual-track differential, dynamic time warping, multi-dimensional cosine similarity, abnormal feature extraction, and three-part control output. Specifically, the dual-track difference module constructs a unified-dimensional state vector. The actual state vector can be represented as follows: ; Here, the capped variables represent data that have undergone dimension normalization and weight scaling; the ideal state vector and the theoretical state vector are represented as follows: ; ; Thus, we can obtain the actual residual vector sequence and the theoretical residual vector sequence: ; ; in, This represents the state reference after extending the ideal pose and ideal force field to the same dimension as the actual state. If a certain dimension has no corresponding quantity in the ideal model, such as temperature itself, which does not belong to the explicit state of an ideal rigid body, then its ideal reference value can be set as the initial calibration value or the zero bias reference.

[0048] Considering that the device contact time may deviate from the preset tolerance (ms) due to upstream cycle fluctuations during actual insertion, the coupling decision module first performs dynamic time warping on the two residual sequences; let the warped sequence be... and Structural similarity is calculated using multidimensional cosine similarity. ; when The closer the value is to 1, the more consistent the actual residuals and theoretical residuals are in overall shape; however, structural similarity alone is insufficient, so it is also necessary to extract anomalous feature results from the actual residuals; spike features should at least include the instantaneous peak amplitude. and peak duration .

[0049] Specifically, the instantaneous peak amplitude Defined as the difference between the maximum absolute value of the residuals within the analysis window and the baseline mean; the peak duration Defined as the time span during which the residual amplitude continuously exceeds a preset multiple of the baseline mean; The impedance jump characteristic can include at least the equivalent stiffness change rate and the equivalent damping change rate; simplified, if within a certain short window: ; ; Then it can be calculated: ; ; The peak characteristics and impedance jump characteristics are fused according to a preset rule to obtain the abnormal characteristic results. Specifically, the preset rules are integrated into a weighted summation formula: ; in, These are weighting coefficients pre-calibrated based on each feature to assess the sensitivity to true interference.

[0050] In a simplified calculation example, if we set the high threshold to 0.92, the low threshold to 0.75, and the outlier threshold to 0.40, then the structural similarity of a certain insertion... and When, output the safe state; when or When, output a true exception; when and At that time, the output will be pending verification. and These are preset positive numbers with dimensions of displacement and velocity, respectively, to avoid the denominator being zero; , and These represent the force increment, displacement increment, and velocity increment relative to the reference value at the current sampling point or within the current analysis window, respectively. thus, , , , , and The meaning remains fixed in this embodiment.

[0051] The feedback control module executes different strategies based on the decision result; for the safe state, the system generates a compensation vector based on the theoretical residual vector. The compensation vector can be written as: ; Among them, the multi-axis coupling gain matrix This formula describes the compensating coupling relationship between axes; the negative sign in the formula indicates that the theoretical background deviation is counteracted; if there is coupling between the X and Z axes, then... The off-diagonal elements are not zero so that the lateral deviation can be corrected synchronously during Z-axis compensation; For genuine anomalies, the feedback control module freezes the current feedforward compensation command and outputs a compliant backoff command, such as reducing the Z-axis contact stiffness and maintaining X / Y attitude stability with a 30µm backward retraction. For anomalies to be verified, it outputs a fine-tuning retry command and a retest trigger command, such as executing X-axis +1µm, The axis was fine-tuned by -0.003°, and the short-range insertion was repeated and data was collected again to confirm whether it was a transient unstable contact.

[0052] As an abnormal operating condition, when the optimal path after dynamic time warping is too long, it indicates that the time misalignment between the two sequences has exceeded the normal process cycle range. At this time, even if the cosine similarity is high, the structural similarity confidence weight can be directly reduced to prevent incorrect classification to a safe state. When a local peak in the actual residual is caused by a single distorted sampling point, its peak duration is usually less than the preset time threshold and is inconsistent with the force-displacement relationship. The system can identify it as sensor pulse noise and require verification within two consecutive windows before upgrading it to a true anomaly. For example, if the state to be verified occurs continuously for a preset number of times, it indicates that the system has entered the unstable boundary. The feedback control module will no longer retry indefinitely, but will switch to a conservative backoff process to avoid repeated impact on the device.

[0053] For example, in a certain device insertion process, the theoretical residual shows that due to platform heating and flexible off-center loading, the normal force should smoothly increase from 1.2N to 1.8N. After dynamic time warping, the actual residual has an overall shape similar to the theoretical residual, with a structural similarity of 0.94. However, in another batch, although the actual residual is also relatively close in the early stages, a narrow pulse peak of 3.6N appears at the indentation depth of 18µm, lasting for 4ms, while the equivalent stiffness change rate suddenly increases. Based on this, the system determines that it is a true anomaly, immediately freezes the compensation and instructs the Z-axis to smoothly retract. If there is another working condition with a similarity of 0.83 and no significant spike, the system first performs a µm-level realignment on the X-axis and then triggers a retest to distinguish whether there is only a slight offset in the initial contact posture.

[0054] Furthermore, the dimension normalization and weight scaling are implemented by first making each dimension dimensionless according to its calibrated range or safe operating range, and then allocating weights according to process sensitivity. For different dimensions such as current, force / torque, temperature, position, and speed, it is not required that their values ​​be of the same absolute order of magnitude, but rather that they have comparable relative deviations in a unified state space. This can prevent a certain type of original quantity from unreasonably dominating the results in the cosine similarity calculation due to its excessively large numerical magnitude, and can also prevent slow variables such as temperature from being completely submerged due to their small magnitude.

[0055] To further explain, the results of abnormal features In implementation, it can be obtained by fusing peak features and impedance jump features through rules. Its core is not to pursue complex algorithmic forms, but to ensure that local mutations are not masked by the overall similarity trend. For example, when the structural similarity is high but... If the peak duration exceeds the preset limit, or the peak duration is significantly mismatched with the contact duration, or the equivalent stiffness change rate suddenly increases to more than a preset multiple of the background change rate within a very short window, the system can mark the window as an abnormal enhancement window and give the window a higher weight in the final decision. After this processing, even if the actual residual is close to the theoretical residual in most sections, it will not be misjudged as a safe state as long as there are strong anomalies in some areas that cannot be explained by the inherent drift of the equipment.

[0056] Furthermore, the multi-axis linkage feedforward compensation commands, compliant back-off commands, and fine-tuning retry commands output by the feedback control module are all constrained by axis travel boundaries, velocity boundaries, and attitude change boundaries. The compensation vector in the safe state undergoes amplitude clipping and smoothing before being issued to prevent multi-axis overcompensation due to single-window residual fluctuations; in true anomalies, the back-off path prioritizes de-contact from the contact area rather than continuing to seek optimal alignment; fine-tuning retry under verification conditions uses a preset small pose increment smaller than the normal alignment step size, and the number of retryes is limited; through these constraints, the hierarchical control logic can be kept consistent with the actual controllable range of the actuator.

[0057] The purpose of this mechanism is to improve the accuracy of identifying real assembly interference by using the dual constraints of overall structural similarity and local anomaly features, and to make interpretable control switching between continued compensation, compliant retreat and fine-tuning retest through a hierarchical feedback strategy, thereby achieving a balance between stability and safety in the assembly process.

[0058] In this embodiment of the invention, a parameter adaptive module is further included, which is used to iteratively update the thermoelastic deformation parameters, the stiffness attenuation parameters, and the hysteresis friction parameters based on the state decision result and the principle of minimizing the difference between the actual residual vector sequence and the theoretical residual vector sequence; wherein, the feedback control module regenerates the multi-axis linkage compensation command based on the updated parameters to form closed-loop control.

[0059] The execution node is an execution mechanism with multi-axis coordinated motion capability, and the execution mechanism includes multiple controlled motion axes and corresponding drive units.

[0060] This embodiment provides a parameter adaptive closed-loop control mechanism suitable for multi-axis actuators. Specifically, although the aforementioned scheme can use theoretical residuals for compensation and decision-making during a single insertion process, if the equipment operates for a long time, the ambient temperature changes, or the transmission wear intensifies, the theoretical simulation state of fixed parameters may gradually deviate from the actual equipment characteristics, reducing the compensation accuracy. To solve this problem, this embodiment introduces a parameter adaptive module, which enables the thermoelastic deformation parameters, stiffness attenuation parameters, and hysteresis friction parameters to be updated online in combination with the decision results and residual differences, and then applied to the next round of control of the multi-axis actuator.

[0061] Specifically, the execution node remains an actuator with multi-axis collaborative capabilities, which can be a six-axis precision platform or other automated execution platforms with multiple controlled motion axes and corresponding drive units; the parameter adaptation module maintains a parameter vector to be updated. ,in Characterizing parameters related to thermoelastic deformation, Parameters characterizing stiffness attenuation Characterize hysteresis-related parameters; calculate the cost of the difference between the actual residual and the theoretical residual after each insertion cycle or a fixed-length time window. ; The basic idea behind parameter updates is to gradually reduce the cost; simplified, it can be updated iteratively as follows: ; Among them, step size Used to control the update amplitude in each round; since the ideal rigid body model and the physical perturbation injection process contain complex nonlinear coupling characteristics, analytical differentiation is difficult, therefore, in this embodiment... The numerical gradient is approximated using the finite difference method or by mapping based on a preset parameter sensitivity matrix. The system iteratively corrects the parameters according to the residual difference and the calculated numerical gradient, so that the theoretical simulation state gradually approximates the actual background response of the device.

[0062] To avoid abnormal assembly events contaminating parameter updates, this embodiment incorporates the state judgment result into the update strategy; if the current cycle is determined to be in a safe state, it means that most of the actual residual is composed of inherent equipment errors, and at this time, preset high-weight coefficients are allowed to participate in parameter updates; If the anomaly is determined to be genuine, it indicates the presence of external interference. In this case, the periodic sample should not be directly used to update the thermoelastic deformation, stiffness attenuation, and hysteretic friction parameters, or it should only be used for edge correction with a preset minimal weighting coefficient. If the anomaly is determined to be pending verification, the periodic data should be cached first, and a decision on whether to include it in the update should be made after the retest results stabilize. Using simplified weighting, it can be written as: ; Among them, weight The weighting is determined by the state judgment result. A safe state corresponds to a higher weight, while a true anomaly corresponds to zero or near-zero weight.

[0063] To further clarify, in the two summation equations above... Both represent the total number of sampling points included in the parameter update calculation within the current insertion cycle or the current fixed-length time window, thus maintaining consistency with the aforementioned time window range of the residual sequence; correspondingly, Indicates the first The parameter vector used in each round of updates. This represents the parameter vector after the next round of updates; and These represent the unweighted difference cost and the difference cost weighted according to the decision result, respectively. The meanings of the relevant symbols remain fixed in this embodiment.

[0064] In a simplified calculation example, let the initial thermal parameters be... stiffness parameters Friction parameters When the theoretical residual is found to consistently underestimate the actual residual by approximately 0.4 μm in the Z-axis direction for 20 consecutive safe-state cycles, it indicates that the thermoelastic deformation model is weak. The parameter adaptive module can then address this. Updated to 1.06; if the theoretical current lag amplitude is found to be too small during speed reverse switching, then... The parameters are gradually updated to 0.15; the updated parameters are re-entered into the parameter injection module to generate a new theoretical simulation state; the feedback control module then generates a new multi-axis linkage compensation command based on the updated theoretical residual, thus forming a closed loop of acquisition-modeling-decision-compensation-remodeling.

[0065] As an abnormal operating condition handling feature, to prevent control oscillations caused by excessively rapid parameter drift, the parameter adaptive module can be configured with update upper limits and rate of change limits; for example, within a single cycle. The change range does not exceed the preset ratio, and cumulative updates are only allowed after multiple consecutive safe state cycles; for example, if updating a parameter causes the structural similarity to decrease, the system can roll back to the previous parameter set; for batches that are in a true abnormal state for a long time, since their data mainly reflects the part fitting problem rather than the characteristics of the equipment itself, the module directly skips parameter learning, only retains the abnormal log, and does not include the batch in the model convergence process.

[0066] For example, when the packaging line is running at night, the ambient temperature in the workshop drops from 24°C during the day shift to 20°C, while the heating mode of the motor inside the equipment remains unchanged, causing the original thermal mapping parameters to be unable to accurately describe the end drift. The parameter adaptive module detects that the theoretical residual and the actual residual have a stable deviation in the X-axis direction in multiple safe-state insertion cycles, and then gradually corrects the thermoelastic deformation parameters.

[0067] At the same time, due to the change in the lubrication state of the guide rail, the hysteresis friction parameters are also fine-tuned; the feedback control module generates a new coupling compensation vector based on the updated parameters, so that the six-axis actuator can be restored to a more stable infeed force curve and a smaller lateral deviation in subsequent insertion.

[0068] The purpose of this mechanism is to enable the system not only to identify and compensate for the inherent errors of the equipment at the current moment, but also to continuously correct the internal model parameters as the equipment state evolves, thereby achieving long-term stable closed-loop control for multi-axis actuators.

[0069] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent equipment assembly and collaborative control system with multi-axis linkage compensation, characterized in that, include: An execution node includes multiple controlled motion axes and corresponding drive units; The data acquisition module is used to acquire the timing current data, force / torque data, temperature data, position data and velocity data of the execution node under a unified time reference. The benchmark reconstruction module is used to generate ideal position benchmarks and ideal force / moment benchmarks based on an ideal rigid body model in a preset reference coordinate system, according to the preset task trajectory and constraints. The parameter injection module is used to inject preset thermoelastic deformation parameters, stiffness attenuation parameters and hysteresis friction parameters into the ideal rigid body model to generate a theoretical simulation state containing theoretical position sequence and theoretical force / torque response sequence. The dual-track differential module is used to map the time-series current data, force / torque data, temperature data, position data, and velocity data into a real state vector, and to perform differential analysis with the ideal state vector corresponding to the ideal position reference and the ideal force / torque reference to generate a real residual vector sequence. It is also used to perform differential analysis with the theoretical simulation state and the ideal position reference and the ideal force / torque reference to generate a theoretical residual vector sequence. The coupling decision module is used to perform temporal similarity verification between the actual residual vector sequence and the theoretical residual vector sequence, and generate a state decision result. The feedback control module is used to output multi-axis linkage compensation commands or compliant return commands to the drive unit based on the state judgment result.

2. The intelligent equipment assembly and collaborative control system with multi-axis linkage compensation according to claim 1, characterized in that, The data acquisition module includes: The current loop acquisition unit is used to acquire high-frequency timing current data of the servo drive link; Force sensing acquisition unit is used to acquire multi-dimensional spatial force and torque data; Thermal state acquisition unit, used to acquire distributed temperature data; The position acquisition unit is used to acquire absolute coded position data and velocity data; The time synchronization unit is used to perform unified timestamp calibration and resampling alignment on all collected data.

3. The intelligent equipment assembly and collaborative control system with multi-axis linkage compensation according to claim 1, characterized in that, The reference reconstruction module is used to construct an ideal rigid body model without thermal drift, flexible deformation, and frictional disturbance in the preset reference coordinate system and the unified time reference, based on the preset task trajectory, preset contact constraints, and preset dynamic constraints, and to calculate the ideal position reference and ideal force / torque reference corresponding to each moment based on the ideal rigid body model.

4. The intelligent equipment assembly and collaborative control system with multi-axis linkage compensation according to claim 1, characterized in that, The parameter injection module is used to map the temperature data into a thermoelastic deformation disturbance, map the position data, velocity data, and force / torque data together with the execution node mechanism model into an equivalent dynamic stiffness disturbance, map the time-series current data, position data, and velocity data into a hysteretic friction disturbance, and couple the thermoelastic deformation disturbance, the dynamic stiffness disturbance, and the hysteretic friction disturbance into the ideal rigid body model to generate the theoretical simulation state characterizing the inherent drift of the device.

5. The intelligent equipment assembly and collaborative control system with multi-axis linkage compensation according to claim 1, characterized in that, The dual-track differential module is used to map the time-series current data, force / torque data, temperature data, position data, and velocity data into a real-state vector, map the ideal position reference and ideal force / torque reference into an ideal-state vector, and perform differential processing on the real-state vector and the ideal-state vector to generate the real-residual vector sequence. The theoretical simulation state is then mapped to a theoretical state vector and then differentially processed with the ideal state vector to generate the theoretical residual vector sequence.

6. The intelligent equipment assembly and collaborative control system with multi-axis linkage compensation according to claim 1, characterized in that, The coupling decision module is used to perform time alignment between the actual residual vector sequence and the theoretical residual vector sequence using dynamic time warping; Multidimensional cosine similarity is used to process the temporal alignment results to generate structural similarity; It extracts at least one peak feature from the instantaneous peak amplitude and peak duration in the real residual vector sequence, and at least one impedance jump feature from the equivalent stiffness change rate and equivalent damping change rate calculated based on the force-displacement relationship or the torque-angular displacement relationship, to generate abnormal feature results.

7. The intelligent equipment assembly and collaborative control system with multi-axis linkage compensation according to claim 6, characterized in that, The coupling decision module is also used for: The preset decision criteria are determined by performing multiple rounds of calibration tests on the equipment under no-interference, unloaded operation, statistically analyzing historical distribution data of structural similarity and abnormal features, and using boundary values ​​covering a 95% confidence interval. Based on the preset decision criteria, preset high threshold, preset low threshold, and preset abnormal threshold are obtained. When the preset high threshold is greater than the preset low threshold, and the structural similarity is greater than or equal to the preset high threshold and the abnormal feature result is less than or equal to the preset abnormal threshold, a safe state decision result is output. When the structural similarity is less than or equal to the preset low threshold or the abnormal feature result is greater than the preset abnormal threshold, a true anomaly judgment result is output. When the structural similarity is greater than the preset low threshold and less than the preset high threshold, and the abnormal feature result is less than or equal to the preset abnormal threshold, the judgment result to be verified is output.

8. The intelligent equipment assembly and collaborative control system with multi-axis linkage compensation according to claim 7, characterized in that, The feedback control module is used to generate a compensation vector based on the theoretical residual vector sequence when it receives the safe state decision result, and output a multi-axis linkage feedforward compensation command to the drive unit based on the compensation vector. Upon receiving the true anomaly judgment result, the multi-axis linkage feedforward compensation command is frozen, and a compliant retraction command is output to reduce the stiffness of the contact shaft and drive the execution node to retract along the preset retraction direction. Upon receiving the judgment result to be verified, a fine-tuning retry instruction and a retest trigger instruction are output to drive the execution node to re-execute the preset operation task according to the preset small pose increment and trigger the data acquisition module to retest.

9. The intelligent equipment assembly and collaborative control system with multi-axis linkage compensation according to claim 1, characterized in that, It also includes a parameter adaptation module, which is used to iteratively update the thermoelastic deformation parameter, the stiffness attenuation parameter and the hysteresis friction parameter based on the state decision result and the principle of minimizing the difference between the actual residual vector sequence and the theoretical residual vector sequence; The feedback control module regenerates the multi-axis linkage compensation command based on the updated parameters to form a closed-loop control.

10. The intelligent equipment assembly and collaborative control system with multi-axis linkage compensation according to claim 1, characterized in that, The execution node is an execution mechanism with multi-axis coordinated motion capability, and the execution mechanism includes multiple controlled motion axes and corresponding drive units.

Images