An assembly robot for use in an industrial production line

Abstract

The application discloses an assembling robot applied to an industrial production line, which synchronously acquires workpiece position, vibration frequency and contact force change data through an object detection unit, generates a high-precision displacement vector in combination with parallel processing technology, dynamically calculates a gain correction coefficient based on a sliding mode controller, effectively deals with nonlinear fluctuation of workpiece displacement, significantly improves system robustness, predicts future displacement trajectory of the workpiece through historical data analysis, provides a control instruction for the robot, reduces operation failure caused by lag, and realizes real-time adjustment of grabbing or assembling action of the robot when the workpiece is displaced due to vibration or conveying belt error, thereby solving the technical problem that the robot cannot real-time adjust grabbing or assembling action when the workpiece is displaced due to vibration or conveying belt error, and task failure or fault is caused.

Description

Technical Field

[0001] This invention relates to the field of assembly robot technology, and more specifically to an assembly robot applied to industrial production lines. Background Technology

[0002] Assembly robots on industrial production lines are high-precision automated devices composed of robotic arms, sensors, and control systems, possessing multi-degree-of-freedom flexible movement capabilities. They are programmed to perform complex assembly tasks and are widely used in industries such as automotive and electronics, completing processes such as welding, handling, and sorting. Their advantages include 24-hour continuous operation, significantly improved production efficiency and product consistency, while reducing labor costs and safety risks, making them core equipment in intelligent manufacturing.

[0003] However, current industrial assembly robots rely on preset programs and fixed trajectories, making it difficult to cope with the dynamically changing environment on the production line. For example, when a workpiece shifts due to vibration or conveyor belt errors, the robot cannot adjust its gripping or assembly actions in real time, leading to task failure or malfunction. Therefore, there is an urgent need to design an assembly robot for industrial production lines with reinforcement learning capabilities. Summary of the Invention

[0004] The purpose of this invention is to provide an assembly robot for use in industrial production lines, in order to solve the technical problem that existing industrial robots cannot adjust their gripping or assembly actions in real time when the workpiece is displaced due to vibration or conveyor belt errors, resulting in task failure or malfunction.

[0005] The technical solution of this invention is implemented as follows:

[0006] An assembly robot for use on an industrial production line includes an object detection unit, a parameter adjustment unit, a path prediction unit, and an anomaly handling unit.

[0007] The object detection unit is used to acquire raw data of the target workpiece position, vibration frequency, and contact force changes, and to process the raw data in parallel to generate the displacement vector of the target workpiece.

[0008] The parameter adjustment unit is used to calculate the displacement vector gain correction coefficient using a sliding mode controller based on the displacement vector of the target workpiece.

[0009] The path prediction unit is used to predict the displacement position of the target workpiece within a preset time based on the historical data of the displacement vector.

[0010] The anomaly handling unit is used to detect whether the target workpiece exceeds the displacement threshold within a preset number of times. If it does, the task is paused, the production line is notified to stop the conveyor belt, and the backup robot end effector is activated to assist in resetting the target workpiece.

[0011] A further technical solution is that the object detection unit includes:

[0012] The multi-angle scanning module uses a camera and a vibration sensor to capture raw data of the target workpiece's position, vibration frequency, and contact force.

[0013] The data alignment module is used to unify the original data onto the same time axis, perform parallel processing, and generate unified data.

[0014] The dynamic analysis module is used to identify abnormal vibration modes and sudden changes in contact force of the target workpiece based on the unified data.

[0015] The displacement calculation module is used to calculate the displacement vector of the target workpiece based on the multi-angle scanning module, the data alignment module, and the dynamic analysis module.

[0016] A further technical solution is that the parameter adjustment unit specifically performs the following steps:

[0017] Step S11: Compare the actual displacement of the target workpiece with the preset standard displacement to generate an error signal;

[0018] Step S12: Define a virtual sliding surface, wherein the virtual sliding surface serves as the displacement reference trajectory of the target workpiece;

[0019] Step S13: Determine the displacement velocity of the target workpiece based on the error signal and the deviation of the virtual sliding surface;

[0020] Step S14: Combine the error signal, the preset virtual sliding surface, and the displacement velocity data to calculate the displacement vector gain correction coefficient used to adjust the target workpiece.

[0021] A further technical solution is that step S13 specifically includes the following steps:

[0022] Step S131: Correlate the error signal with the deviation of the virtual sliding surface to generate a deviation scaling factor;

[0023] Step S132: Set the upper and lower limits of the displacement velocity according to the deviation scaling factor;

[0024] Step S133: Obtain the changing trend of the error signal and adjust the changing range of the displacement speed. If the error signal shows a rapid decay trend, the upper limit of the displacement speed is reduced; if the error signal fluctuates repeatedly, the upper limit of the displacement speed is increased.

[0025] Step S134: Calculate the real-time displacement velocity of the target workpiece by combining the scaling factor, the range of displacement velocity variation, and the dynamic calibration results.

[0026] A further technical solution is that step S14 specifically includes the following steps:

[0027] Step S141: Dynamically weight the error signal and the deviation of the virtual sliding surface to generate the correlation degree between the error signal and the virtual sliding surface;

[0028] Step S142: Combine the displacement velocity data with the correlation degree to calculate the coupling coefficient between the displacement velocity and the correlation degree;

[0029] Step S143: Based on the coupling coefficient, proportionally allocate the gain correction coefficient to different control dimensions.

[0030] A further technical solution is that the path prediction unit includes:

[0031] The feature extraction module is used to arrange the historical displacement vector data of the target workpiece in a time series and extract features such as displacement trend, periodic fluctuation, and abrupt change point.

[0032] The path simulation module, based on the displacement trend, periodic fluctuation, and abrupt change point feature extraction results, constructs a baseline to simulate the expected displacement path of the target workpiece within a preset time.

[0033] The data compensation module is used to correct the disturbance of the baseline by combining real-time monitoring data from the production line.

[0034] The interval prediction module is used to expand the corrected dynamic baseline into a probability interval, representing the possible displacement range of the target workpiece within a preset time.

[0035] The model parameter adjustment module is used to continuously collect the actual displacement data of the target workpiece and compare it with the predicted value during the prediction process, and adjust the model parameters through comparison error feedback.

[0036] A further technical solution is that the path simulation module specifically executes the following steps:

[0037] Step S21: Separate the displacement trend from the periodic fluctuation;

[0038] Step S22: Based on the mutation point detection results, insert correction points into the baseline;

[0039] Step S23: The displacement trend component, periodic component, and correction point are fused according to weights to generate the baseline;

[0040] Step S24: Based on the baseline, simulate the displacement path of the target workpiece in the next N seconds, and compare it with the actual displacement data in real time. If the deviation between the displacement path and the actual displacement exceeds a threshold, the baseline parameter is triggered to reduce the weight of the periodic component.

[0041] A further technical solution is that the specific execution steps of the model parameter adjustment module include:

[0042] Step S31: During the prediction execution, the actual displacement data and predicted value of the target workpiece are collected synchronously and aligned by timestamp;

[0043] Step S32: Divide the deviation between the predicted value and the actual displacement data into trend error and sudden error;

[0044] Step S33: Adjust the sensitivity of the model parameters based on the stratification results of the trend error and the sudden error;

[0045] Step S34: Apply the adjusted model parameters to the next round of prediction and monitor the matching degree between the actual displacement of the target workpiece and the predicted value.

[0046] A further technical solution is that step S33 specifically includes the following steps:

[0047] Step S331: Set the priority of the model parameter adjustment according to the ratio of the trend error to the sudden error;

[0048] Step S332: Set dynamic response thresholds for the trend error and the sudden error respectively;

[0049] Step S333: Based on the priority and the dynamic response threshold, allocate the adjustment range of the model parameters proportionally.

[0050] A further technical solution is that the exception handling unit includes:

[0051] The threshold trigger detection module is used to monitor in real time whether the displacement of the target workpiece exceeds a set threshold within the preset number of times.

[0052] The production line linkage pause module is used to trigger an abnormal signal based on the threshold trigger detection module, send a pause command to the production line control system, and synchronously stop the operation of the conveyor belt.

[0053] The backup actuator intervention module is used to activate the backup robot end effector and plan a reset path based on the current displacement state of the target workpiece.

[0054] The beneficial effects of this invention are as follows:

[0055] The object detection unit synchronously acquires workpiece position, vibration frequency, and contact force change data, and generates a high-precision displacement vector by combining parallel processing technology. The parameter adjustment unit dynamically calculates the gain correction coefficient based on the sliding mode controller, effectively dealing with nonlinear fluctuations in workpiece displacement and significantly improving system robustness. The path prediction unit predicts the future displacement trajectory of the workpiece through historical data analysis, providing control commands to the robot and reducing operational errors caused by lag. The anomaly handling unit monitors the displacement threshold in real time, triggering production line pause and backup robot reset operations to form a closed-loop emergency mechanism, ensuring equipment safety and production line continuity. This solves the technical problem that existing industrial robots cannot adjust their gripping or assembly actions in real time when workpieces are displaced due to vibration or conveyor belt errors, leading to task failure or malfunction. Attached Figure Description

[0056] Figure 1 This is a block diagram of an assembly robot system applied to an industrial production line according to the present invention;

[0057] Figure 2 This is a block diagram of an object detection unit system in an assembly robot applied to an industrial production line according to the present invention.

[0058] Figure 3 This is a flowchart illustrating the specific execution steps of a parameter adjustment unit in an assembly robot applied to an industrial production line according to the present invention.

[0059] Figure 4 This is a system block diagram of a path prediction unit in an assembly robot applied to an industrial production line according to the present invention.

[0060] Figure 5 This is a flowchart illustrating the specific execution steps of a path simulation module in an assembly robot applied to an industrial production line according to the present invention.

[0061] Figure 6 This is a flowchart illustrating the specific execution steps of a model parameter adjustment module in an assembly robot applied to an industrial production line according to the present invention.

[0062] Figure 7 This is a system block diagram of an abnormality handling unit in an assembly robot applied to an industrial production line according to the present invention. Detailed Implementation

[0063] To better understand the technical content of this invention, specific embodiments are provided below, and the invention will be further described in conjunction with the accompanying drawings.

[0064] See Figures 1 to 7This invention provides an assembly robot for industrial production lines, comprising an object detection unit, a parameter adjustment unit, a path prediction unit, and an anomaly handling unit. The object detection unit acquires raw data on the target workpiece's position, vibration frequency, and contact force changes, and processes this raw data in parallel to generate a displacement vector for the target workpiece. The parameter adjustment unit calculates a displacement vector gain correction coefficient using a sliding mode controller based on the target workpiece's displacement vector. The path prediction unit predicts the target workpiece's displacement position within a preset time based on historical displacement vector data. The anomaly handling unit detects whether the target workpiece exceeds a displacement threshold within a preset number of attempts; if it does, it pauses the task, notifies the production line to stop the conveyor belt, and activates a backup robot end effector to assist in resetting the target workpiece.

[0065] It should be noted that the displacement vector includes the vector representing the change in position of the target workpiece in three-dimensional space, consisting of displacement components in the X / Y / Z axis directions and rotation angles, used to quantify the dynamic motion state of the target workpiece. The displacement position within the preset time includes the future position of the target workpiece predicted by modeling based on historical displacement data, serving as a reference trajectory for the robot to perform operations such as grasping and assembly.

[0066] In this embodiment of the invention, by integrating four functional modules—object detection unit, parameter adjustment unit, path prediction unit, and anomaly handling unit—real-time perception, dynamic control, prediction, and proactive anomaly response of the target workpiece displacement state can be achieved.

[0067] Specifically, the object detection unit uses laser rangefinders, accelerometers, and force sensors to simultaneously collect raw data such as the target workpiece's position coordinates, vibration frequency, and contact force changes. This data is then processed in parallel to generate a high-precision displacement vector, enabling multi-dimensional modeling of the workpiece's dynamic state. The parameter adjustment unit calculates the displacement vector gain correction coefficient based on a sliding mode controller. By dynamically adjusting the control gain, it compensates for nonlinear fluctuations in the workpiece's displacement, ensuring the robustness and real-time performance of the control system. The path prediction unit uses historical displacement vector data to construct a time-series model. Combined with trend analysis and error stratification algorithms, it predicts the target workpiece's displacement position within a preset timeframe, providing control commands to the robot's end effector. The anomaly handling unit detects a displacement threshold within a preset number of cycles to determine if the workpiece has experienced abnormal displacement. If the threshold is triggered, the production line control system pauses the conveyor belt operation and activates a backup robot end effector (such as a robotic arm) to perform a reset operation, forming a closed-loop emergency mechanism. This addresses the technical problem of existing industrial robots failing to adjust their gripping or assembly actions in real time when the workpiece is displaced due to vibration or conveyor belt errors, leading to task failure or malfunction.

[0068] Furthermore, the object detection unit includes: a multi-angle scanning module, which uses a camera and vibration sensor to capture raw data of the target workpiece's position, vibration frequency, and contact force; a data alignment module, which unifies the raw data onto the same time axis for parallel processing to generate unified data; a dynamic analysis module, which identifies abnormal vibration patterns and sudden changes in contact force of the target workpiece based on the unified data; and a displacement calculation module, which calculates the displacement vector of the target workpiece based on the multi-angle scanning module, data alignment module, and dynamic analysis module.

[0069] In this embodiment of the invention, the multi-angle scanning module works collaboratively with an industrial camera, a triaxial accelerometer, and a force sensor to collect the position coordinates of the target workpiece, the harmonic vibration frequency within the 0-2000Hz range, and the instantaneous fluctuations in contact force of clamping or friction, generating three types of raw data. The camera extracts the target workpiece's contour and center coordinates using image recognition technology, while the vibration sensor quantifies vibration signals and contact force data using the piezoelectric effect or strain gauge principle. The camera provides high-resolution visual positioning, and the vibration sensor captures microscopic mechanical properties; the two complement each other to comprehensively perceive the target workpiece's state. The data alignment module employs a timestamp hardware-triggered synchronization algorithm, eliminating multi-sensor time-series offsets through hardware triggering or software interpolation, ensuring subsequent analysis is based on synchronized data and avoiding misjudgments caused by timestamp misalignment. Then, image frames, vibration signals, and force data are mapped to a unified time axis to eliminate time-series offsets caused by differences in sensor sampling rates, forming a unified dataset that provides a benchmark for subsequent parallel processing. The dynamic analysis module, based on a unified dataset, extracts the periodic characteristics of vibration frequencies through sliding window Fourier transform and identifies abnormal vibration modes such as high-frequency flutter and nonlinear harmonics using sliding mode control theory. Simultaneously, it utilizes a differential threshold algorithm to detect abrupt changes in contact force, determining whether abnormal contact such as jamming or slippage has occurred in the workpiece. The displacement calculation module integrates workpiece position data from multi-angle scans and aligned vibration and force data, using the three-dimensional coordinate difference formulas: ΔX = X2 - X1, ΔY = Y1 - Y2, ΔZ = Z1 - Z2 to calculate the displacement vector. It then combines the dynamic analysis results to correct displacement errors, ultimately outputting a high-precision target workpiece displacement vector containing both direction and amplitude.

[0070] Preferably, the parameter adjustment unit specifically performs the following steps:

[0071] Step S11: Compare the actual displacement of the target workpiece with the preset standard displacement to generate an error signal;

[0072] Step S12: Define a virtual sliding surface, whereby the virtual sliding surface serves as the displacement reference trajectory of the target workpiece;

[0073] Step S13: Determine the displacement velocity of the target workpiece based on the error signal and the deviation of the virtual sliding surface;

[0074] Step S14: Combine the error signal, preset virtual sliding surface and displacement velocity data to calculate the displacement vector gain correction coefficient used to adjust the target workpiece.

[0075] It should be noted that the virtual sliding surface includes the mathematical reference surface defined in sliding mode control, which characterizes the dynamic properties of the ideal displacement trajectory. A reaching law design ensures that the actual displacement converges quickly to this surface and remains stable. The preset standard displacement includes the theoretical trajectory coordinates.

[0076] In this embodiment of the invention, the actual displacement of the target workpiece is compared with a preset standard displacement in real time, and the deviation between the two in the spatial dimension is calculated to generate an error signal. This error signal is used to quantify the degree of deviation between the current displacement of the target workpiece and the target trajectory. For example, if the actual displacement is 5.2mm, -0.3mm, 1.8mm, and the preset standard displacement is 5.0mm, 0.0mm, 2.0mm, then the error signal is 0.2mm, -0.3mm, -0.2mm. Next, a virtual sliding surface is defined, which is represented by the mathematical function S(t) = e(t) + λ∫e(t)dt, where e(t) is the error signal, and λ is a positive definite constant representing the ideal displacement reference trajectory of the target workpiece. The design of the sliding surface follows sliding mode control theory, and the reaching law ensures that the actual displacement quickly converges to the sliding surface and maintains stable motion. Based on the deviation between the error signal and the virtual sliding surface (such as the sliding surface function value S(t)), and combined with the time difference calculation of Δt, the displacement velocity of the target workpiece is derived, i.e., the rate of displacement change per unit time (dx / dt, dy / dt, dz / dt). For example, if the error signal increases linearly with time, the displacement velocity v(t) can be estimated and quantified by the slope Δe / Δt, reflecting the trend strength of the target workpiece deviating from the trajectory. Finally, by integrating the error signal, the virtual sliding surface function value, and the displacement velocity data, the gain correction coefficient K is designed using the sliding mode reaching law, as shown in the following formula: Where η is the switching gain, used to suppress chattering, and κ is the approach velocity coefficient, which determines the system convergence speed. Substituting the above formula into the sliding surface dynamic equation, the gain correction coefficient is derived: Here, ε is a small positive number to prevent the denominator from being zero. This coefficient is used to dynamically adjust the output gain of the controller to compensate for displacement deviations of the target workpiece caused by disturbances such as vibration and sudden changes in contact force. For example, when the error signal increases and the sliding surface deviation exceeds the threshold, the gain correction coefficient is adjusted to K = 1.2 to enhance the controller's response strength.

[0077] Furthermore, step S13 specifically includes the following steps:

[0078] Step S131: Correlate the error signal with the deviation of the virtual sliding surface to generate a deviation scaling factor;

[0079] Step S132: Set the upper and lower limits of the displacement velocity according to the deviation proportional factor;

[0080] Step S133: Obtain the changing trend of the error signal and adjust the changing range of the displacement speed. If the error signal shows a rapid decay trend, the upper limit of the displacement speed is reduced; if the error signal fluctuates repeatedly, the upper limit of the displacement speed is increased.

[0081] Step S134: Calculate the real-time displacement velocity of the target workpiece by combining the scaling factor, the range of displacement velocity variation, and the dynamic calibration results.

[0082] In the embodiments of the invention, the error signal of the target workpiece is correlated with the deviation of the virtual sliding surface, and a deviation scaling factor K is generated by the PI algorithm in sliding mode control. P For example, if the error signal is e(t), and the virtual sliding surface deviation is... The scaling factor can then be expressed as: The upper limit V of the displacement velocity is set according to the deviation proportional factor. max With lower limit V min The specific formula is:

[0083] V max =V0+α·K P V min =V0-β·K p

[0084] Where V0 is the reference speed, and α and β are adjustment coefficients. When K P When the value increases, the system raises the upper limit V. max and lowering the lower limit V min To address larger deviations, the speed control range is expanded. The trend of the error signal e(t) is analyzed using a sliding window Fourier transform or moving average algorithm: a rapid decay trend is observed. This indicates that the actual displacement converges rapidly to the virtual sliding surface, at which point the upper velocity limit V is reduced. max To V′ max =V max •(1-γ) to avoid overshooting. Repeated fluctuation trend. Furthermore, the frequent sign changes of e(t) indicate that there is a disturbance or chattering in the actual displacement. In this case, the upper limit of velocity V should be increased. max To V′ max =V max • (1+δ) is used to enhance response capability. Where θ, φ, γ, and δ are preset threshold coefficients. The comprehensive deviation scaling factor K... P Adjusted speed range [V′] min,V′ max The real-time displacement velocity v(t) is calculated using the sliding mode reaching law formula, along with the dynamic calibration results (such as the sensor noise compensation value η).

[0085]

[0086] Here, sat(·) is the boundary layer saturation function, used to suppress high-frequency chattering. This formula ensures that the speed is dynamically adjusted within a set range, while also eliminating measurement errors by incorporating calibration data.

[0087] Furthermore, step S14 specifically includes the following steps:

[0088] Step S141: Dynamically weight the error signal and the deviation of the virtual sliding surface to generate the correlation degree between the error signal and the virtual sliding surface;

[0089] Step S142: Combine displacement velocity data with correlation degree to calculate the coupling coefficient between displacement velocity and correlation degree;

[0090] Step S143: Based on the coupling coefficient, proportionally allocate the gain correction coefficient to different control dimensions.

[0091] In this embodiment of the invention, the error signal of the target workpiece and the deviation of the virtual sliding surface are dynamically weighted in real time to generate a correlation between the two. The dynamic weighting employs a sliding window exponential decay algorithm, and the weight w(t) changes over time:

[0092]

[0093] Where α > 0, the formula for calculating the correlation degree C(t) is:

[0094] C(t)=w(t)·e(t)+(1-w(t))·s(t)

[0095] Where e(t) and s(t) represent the error signal and the virtual sliding surface deviation, respectively. The weight w(t) is dynamically adjusted based on the amplitude of the sliding surface deviation s(t): when s(t) is large, the weight w(t) increases, strengthening the influence of the error signal on the correlation; when s(t) is small, the weight w(t) decreases, weakening the effect of the error signal. Dynamic weighting ensures that the correlation C(t) can reflect the coupling strength between the error signal and the sliding surface deviation in real time, providing a quantitative basis for subsequent gain allocation. Combining the displacement velocity data v(t) and the correlation C(t), the coupling coefficient between the two is calculated through covariance analysis, as shown in the following formula:

[0096]

[0097] Where Cov(·,·) represents the covariance, and Var(·) represents the variance. If KC A value close to 1 indicates a strong positive correlation between displacement velocity and correlation degree, necessitating an increase in the gain correction coefficient to suppress error propagation; K C A value close to -1 indicates a negative correlation between the two, requiring a reduction in the gain correction factor to avoid overshoot. This is based on the coupling coefficient K. C The gain correction factor K is proportionally allocated to different control dimensions (such as the X / Y / Z axes).

[0098] Furthermore, the path prediction unit includes: a feature extraction module, used to arrange the historical displacement vector data of the target workpiece in a time series and extract displacement trend, periodic fluctuation, and abrupt change features; a path simulation module, used to construct a baseline based on the displacement trend, periodic fluctuation, and abrupt change feature extraction results to simulate the expected displacement path of the target workpiece within a preset time; a data compensation module, used to correct the baseline for disturbances by combining real-time monitoring data from the production line; an interval prediction module, used to expand the corrected dynamic baseline into a probability interval, representing the possible displacement range of the target workpiece within a preset time; and a model parameter adjustment module, used to continuously collect the actual displacement data of the target workpiece and compare it with the predicted value during the prediction process, and adjust the model parameters through error feedback.

[0099] It should be noted that the baseline includes the expected displacement path constructed by combining trend, periodicity, and abrupt change characteristics, reflecting the law of motion of the target workpiece. The probability interval includes the prediction boundary extended based on statistical methods, representing the possible range of displacement of the target workpiece. The model can be a neural network model or an AI model, and the model parameter adjustment includes the process of optimizing the model parameters through error feedback to ensure that the prediction model adapts to actual working conditions.

[0100] In this embodiment of the invention, the feature extraction module performs time-series analysis on the historical displacement vector data of the target workpiece, extracting features such as uniform, accelerated, or decelerated displacement trends, periodic fluctuations due to mechanical resonance or disturbance, and sudden stops or rapid acceleration abrupt changes. Subsequently, the path simulation module constructs a baseline based on the above features, simulating the expected displacement path of the workpiece through trend fitting, periodic compensation, and abrupt change correction. Next, the data compensation module combines real-time monitoring data from the production line (such as vibration noise and external disturbances) and dynamically corrects the baseline using a weighted algorithm to adapt to environmental changes. Based on this, the interval prediction module uses statistical methods to expand the corrected baseline into a probability interval, quantifying the distribution of prediction errors and forming a confidence range for workpiece displacement. Finally, the model parameter adjustment module continuously collects actual displacement data and dynamically optimizes the model parameters using the gradient descent method driven by error feedback, enabling the prediction model to adapt to changes in actual working conditions. This technology achieves high-precision prediction and robust control of the target workpiece displacement path through feature-driven modeling, dynamic compensation, probability boundary expansion, and online parameter optimization.

[0101] Furthermore, the specific execution steps of the path simulation module include:

[0102] Step S21: Separate the displacement trend from the periodic fluctuations;

[0103] Step S22: Based on the mutation point detection results, insert correction points in the baseline;

[0104] Step S23: Combine the displacement trend component, periodic component, and correction point according to their weights to generate a baseline.

[0105] Step S24: Based on the baseline, simulate the displacement path of the target workpiece in the next 3 to 5 seconds, and compare it with the actual displacement data in real time. If the deviation between the displacement path and the actual displacement exceeds the threshold, the baseline parameters will be triggered to reduce the weight of the periodic component.

[0106] In this embodiment of the invention, the displacement trend and periodic fluctuations of the target workpiece are decoupled through time-frequency analysis (e.g., separation of low-frequency and high-frequency signals), and their features are extracted separately. Then, based on the results of abrupt change point detection (e.g., sudden pauses, rapid acceleration, or rapid deceleration events), anomaly locations are identified using a threshold detection algorithm, and correction points are inserted into the baseline to compensate for abrupt disturbances. Next, the separated displacement trend components, periodic components, and correction points are fused with dynamic weights to generate a baseline, where the weights are dynamically adjusted based on trend stability, periodic intensity, and abrupt change frequency to balance the contribution of each component. Finally, the displacement path of the workpiece in the next N seconds is simulated based on the baseline, and the actual displacement data is compared in real time. If the deviation between the predicted path and the actual path exceeds a preset threshold, the weight of the periodic component is automatically reduced, and the influence of the trend component is prioritized, thereby improving the model's robustness to environmental disturbances. This method, through multi-component decoupling, dynamic correction, and adaptive adjustment, achieves high-precision prediction of the target workpiece's displacement path and real-time optimization of model parameters.

[0107] Furthermore, the specific execution steps of the model parameter adjustment module include:

[0108] Step S31: During the prediction execution, the actual displacement data and predicted values ​​of the target workpiece are collected synchronously and aligned by timestamp;

[0109] Step S32: Divide the deviation between the predicted value and the actual displacement data into trend error and sudden error;

[0110] Step S33: Adjust the sensitivity of the model parameters based on the stratification results of trend error and sudden error;

[0111] Step S34: Apply the adjusted model parameters to the next round of prediction and monitor the matching degree between the actual displacement of the target workpiece and the predicted value.

[0112] In this embodiment of the invention, during the prediction execution, the actual displacement data and predicted value (output by the model) of the target workpiece are collected synchronously, and the difference in data collection time is eliminated by timestamp alignment to ensure that the two are compared under the same time reference. The deviation between the predicted value and the actual displacement is divided into two categories: trend error (such as the model's prediction deviation of the long-term acceleration / deceleration trend of the workpiece) and sudden error (such as short-term drastic deviation caused by external impact or sensor noise). The former is extracted by low-pass filtering to extract long-term deviation features, and the latter is identified by threshold detection to identify sudden events. The sensitivity of the model parameters is adjusted according to the error type stratification results: if the trend error is significant, the response weight of the model to the displacement trend (such as the slope of linear regression) is enhanced, and the sensitivity to high-frequency disturbances is reduced; if sudden errors are frequent, the sensitivity of sudden point detection is increased and the threshold for correction point insertion is dynamically adjusted to compensate for short-term disturbances. The adjusted model parameters are applied to the next round of prediction, and the matching degree between the actual displacement and the predicted value is continuously monitored (such as by quantifying the accuracy through root mean square error). If the matching degree is lower than the preset threshold, a new round of error analysis and parameter adjustment is triggered to form a closed-loop feedback mechanism. This method enables the model to maintain high-precision prediction capabilities under complex working conditions through synchronous data alignment, error hierarchical analysis, and dynamic parameter optimization.

[0113] Preferably, step S33 specifically includes the following steps:

[0114] Step S331: Based on the proportion of trend error and sudden error, set the priority of model parameter adjustment;

[0115] Step S332: Set dynamic response thresholds for trend errors and sudden errors respectively;

[0116] Step S333: Based on priority and dynamic response threshold, allocate the adjustment range of model parameters proportionally.

[0117] In this embodiment of the invention, the ratio of the potential error generated by the model's prediction deviation of the long-term acceleration / deceleration trend of the target workpiece to the sudden error generated by the short-term drastic deviation caused by external impact or sensor noise (e.g., the trend error ratio P) is calculated. t The proportion of sudden errors P s The priority of model parameter adjustment is set through weighted rules: if P t >P S If p S >p t Then, priority should be given to optimizing the model's ability to correct for mutation points (e.g., increasing the sensitivity of the mutation point detection threshold). Dynamic response thresholds are set separately for trend errors and sudden errors: trend error threshold θ tThe long-term fluctuation range of historical deviations is calculated by using a sliding window averaging or low-pass filtering, i.e.:

[0118] θ e =μ e +k·σ e

[0119] Where μ is the mean, σ is the standard deviation, and k is the amplification factor. The sudden error threshold θ... s By dynamically adjusting the frequency and amplitude of mutation events in real time, that is:

[0120] θ s =θ b ·(1+α·f j )

[0121] Among them, f j Let α be the mutation frequency and α be the adjustment coefficient. Based on priority and dynamic response threshold, the model parameter adjustment magnitude is allocated proportionally: if the trend error has a high priority, the parameter adjustment magnitude Δw is based on the proportion of the trend error P. t Allocation, Δw tr =Δw to ·P t And limit the adjustment range Δw for sudden errors. s The adjustment weight should not exceed the remaining proportion; conversely, if sudden errors have a high priority, the adjustment magnitude should be allocated in reverse, and the adjustment weight of trend errors should be dynamically compressed. This method enables the model to achieve precise parameter adjustment and adaptive optimization under complex working conditions through error proportion-driven priority setting, dynamic threshold response, and proportion allocation strategy.

[0122] Furthermore, the anomaly handling unit includes: a threshold trigger detection module, used to monitor in real time whether the displacement of the target workpiece exceeds a set threshold within a preset number of times; a production line linkage pause module, used to send a pause command to the production line control system based on the anomaly signal triggered by the threshold trigger detection module, and synchronously stop the conveyor belt operation; and a backup actuator intervention module, used to activate the backup robot end effector and plan a reset path based on the current displacement state of the target workpiece.

[0123] In this embodiment of the invention, the threshold trigger detection module continuously acquires the displacement of the target workpiece through a sensor or vision system within a preset number of times (e.g., multiple consecutive sampling cycles) and compares it with the threshold set by the process. When the displacement exceeds the threshold range, the module triggers an abnormal signal and starts the subsequent process. After receiving the abnormal signal, the production line linkage pause module immediately sends a pause command to the production line control system through the industrial communication protocol, synchronously stopping the operation of the conveyor belt and other related equipment to ensure that the position of the target workpiece does not change when the abnormality occurs, avoiding secondary failures. After the production line is paused, the backup actuator intervention module activates the backup robot end effector (e.g., a robotic arm or gripper) and dynamically plans the reset path according to the current displacement state of the target workpiece (e.g., offset direction and magnitude): if the offset is small, the grasping path is calculated through a kinematic algorithm; if the offset is severe, a preset safety path is called (e.g., returning to the initial position). The entire process achieves closed-loop control from abnormality detection to reset through the coordinated cooperation of sensor data acquisition, industrial control signal transmission, and intelligent path planning, ensuring the stability and safety of the production line operation.

[0124] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An assembly robot for use on an industrial production line, characterized in that, It includes an object detection unit, a parameter adjustment unit, a path prediction unit, and an anomaly handling unit; The object detection unit is used to acquire raw data of the target workpiece position, vibration frequency, and contact force changes, and to process the raw data in parallel to generate the displacement vector of the target workpiece. The parameter adjustment unit is used to calculate the displacement vector gain correction coefficient of the target workpiece using a sliding mode controller based on the displacement vector of the target workpiece. The specific execution steps of the parameter adjustment unit include: Step S11: Compare the actual displacement of the target workpiece with the preset standard displacement to generate an error signal; Step S12: Define a virtual sliding surface, wherein the virtual sliding surface serves as the displacement reference trajectory of the target workpiece; Step S13: Determine the displacement velocity of the target workpiece based on the error signal and the deviation of the virtual sliding surface; Step S14: Combine the error signal, the preset virtual sliding surface, and the displacement velocity data of the target workpiece to calculate the displacement vector gain correction coefficient for adjusting the target workpiece; The specific steps of step S13 include: Step S131: Correlate the error signal with the deviation of the virtual sliding surface to generate a deviation scaling factor; Step S132: Set the upper and lower limits of the displacement velocity according to the deviation scaling factor; Step S133: Obtain the changing trend of the error signal and adjust the changing range of the displacement speed. If the error signal shows a rapid decay trend, the upper limit of the displacement speed of the target workpiece is reduced. If the error signal fluctuates repeatedly, the upper limit of the displacement speed of the target workpiece is increased. Step S134: Calculate the real-time displacement velocity of the target workpiece by combining the scaling factor, the range of displacement velocity of the target workpiece, and the dynamic calibration results. The specific steps of step S14 include: Step S141: Dynamically weight the error signal and the deviation of the virtual sliding surface to generate the correlation degree between the error signal and the virtual sliding surface; Step S142: Combine the displacement velocity data of the target workpiece with the correlation degree to calculate the coupling coefficient between the displacement velocity of the target workpiece and the correlation degree of the virtual sliding surface; Step S143: Based on the coupling coefficient, proportionally allocate the gain correction coefficient to different control dimensions; The path prediction unit is used to predict the displacement position of the target workpiece within a preset time based on the historical data of the displacement vector. The anomaly handling unit is used to detect whether the target workpiece exceeds the displacement threshold within a preset number of times. If it does, the task is paused, the production line is notified to stop the conveyor belt, and the backup robot end effector is activated to assist in resetting the target workpiece.

2. The assembly robot for use in an industrial production line according to claim 1, characterized in that, The object detection unit includes: The multi-angle scanning module uses a camera and a vibration sensor to capture raw data of the target workpiece's position, vibration frequency, and contact force. The data alignment module is used to unify the original data onto the same time axis, perform parallel processing, and generate unified data. The dynamic analysis module is used to identify abnormal vibration modes and sudden changes in contact force of the target workpiece based on the unified data. The displacement calculation module is used to calculate the displacement vector of the target workpiece based on the multi-angle scanning module, the data alignment module, and the dynamic analysis module.

3. The assembly robot for use in an industrial production line according to claim 1, characterized in that, The path prediction unit includes: The feature extraction module is used to arrange the historical displacement vector data of the target workpiece in a time series and extract features such as displacement trend, periodic fluctuation, and abrupt change point. The path simulation module, based on the displacement trend, periodic fluctuation, and abrupt change point feature extraction results, constructs a baseline to simulate the expected displacement path of the target workpiece within a preset time. The data compensation module is used to correct the disturbance of the baseline by combining real-time monitoring data from the production line. The interval prediction module is used to expand the corrected baseline into a probability interval, representing the possible displacement range of the target workpiece within a preset time. The model parameter adjustment module is used to continuously collect the actual displacement data of the target workpiece and compare it with the predicted value during the prediction process, and adjust the model parameters through comparison error feedback.

4. An assembly robot for use in an industrial production line according to claim 3, characterized in that, The specific execution steps of the path simulation module include: Step S21: Separate the displacement trend from the periodic fluctuation; Step S22: Based on the feature extraction results of the mutation points, insert correction points into the baseline; Step S23: The displacement trend, the periodic fluctuation, and the correction point are weighted and integrated to generate the baseline; Step S24: Based on the baseline, simulate the displacement path of the target workpiece in the next N seconds, and compare it with the actual displacement data in real time. If the deviation between the displacement path and the actual displacement exceeds a threshold, the parameters of the baseline are triggered to reduce the periodic fluctuation weight.

5. An assembly robot for use in an industrial production line according to claim 3, characterized in that, The specific execution steps of the model parameter adjustment module include: Step S31: During the prediction execution, the actual displacement data and predicted value of the target workpiece are collected synchronously and aligned by timestamp; Step S32: Divide the deviation between the predicted value and the actual displacement data into trend error and sudden error; Step S33: Adjust the sensitivity of the model parameters based on the stratification results of the trend error and the sudden error; Step S34: Apply the adjusted model parameters to the next round of prediction and monitor the matching degree between the actual displacement of the target workpiece and the predicted value.

6. An assembly robot for use in an industrial production line according to claim 5, characterized in that, The specific steps of step S33 include: Step S331: Set the priority of the model parameter adjustment according to the ratio of the trend error to the sudden error; Step S332: Set dynamic response thresholds for the trend error and the sudden error respectively; Step S333: Based on the priority and the dynamic response threshold, allocate the adjustment range of the model parameters proportionally.

7. An assembly robot for use in an industrial production line according to claim 1, characterized in that, The exception handling unit includes: The threshold trigger detection module is used to monitor in real time whether the displacement of the target workpiece exceeds a set threshold within the preset number of times. The production line linkage pause module is used to trigger an abnormal signal based on the threshold trigger detection module, send a pause command to the production line, and synchronously stop the operation of the conveyor belt; The backup actuator intervention module is used to activate the backup robot end effector and plan a reset path based on the current displacement state of the target workpiece.

Images