Robot flexible assembly method and system based on assembly state cognition

Abstract

The invention belongs to the technical field of robot program control, and particularly relates to a robot flexible assembly method and system based on assembly state cognition, and the method comprises the steps: collecting force and torque signals, and carrying out the filtering processing, and obtaining a smooth force vector and a smooth torque vector; calculating coordinates of equivalent contact points based on a wrench decomposition principle; according to the force vector sequence, determining a normal vector of the contact surface based on a projection variance minimum principle, and decomposing the force vector into normal supporting force and friction force; obtaining the instability index of the contact state according to the equivalent contact point position and the change degree of the normal vector of the contact surface; and when the instability index is greater than a preset threshold value, executing a self-adaptive evasion strategy to remove the jam risk. The technical problems that in robot flexible assembly, the contact state is not clearly recognized, and the jamming precursor is difficult to recognize are solved, and the stability and the success rate of the assembly process are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot program control. More particularly, the present application relates to a robot flexible assembly method and system based on assembly state cognition. BACKGROUND

[0002] In automated production, robot flexible assembly is a key technology to realize intelligent manufacturing. Its core is to equip robots with force and torque sensors, giving them human-like tactile sensation, so that they can perceive and adapt to the changes in contact force caused by factors such as part tolerance and positioning error. When performing precise insertion and assembly tasks, robots monitor the interaction force and torque between the end and the environment in real time, dynamically adjust their motion posture to avoid part jamming or damage, and thus complete the assembly.

[0003] However, the existing technology still has systematic problems in understanding and applying the high-dimensional, coupled data output by force and torque sensors. The original force and torque signals are essentially abstract numerical streams, and traditional methods cannot accurately invert the complete physical image of the contact event, such as the specific location of contact occurrence and the normal direction of the contact surface. Moreover, due to the lack of deep understanding of the contact state, the existing technology cannot effectively evaluate the stability of the assembly process, especially for the critical stage of evolution from stable single-point contact to unstable multi-point contact, lacking reliable identification mechanisms.

[0004] To solve the above problems, some researches in the industry try to analyze the force and torque data through algorithm models. For example, based on the principle of rigid body dynamics, the position of the equivalent action point can be calculated from the force and torque data. However, this model is usually based on the single-point contact assumption, and in the complex scenario of contact state transition from single-point to multi-point, the calculation result will produce dramatic jumps or even fail, making it difficult to stably reflect the real contact geometry. In addition, some methods try to decouple the normal force and friction force through force projection variance analysis, which assumes that the normal force remains stable while the friction force changes with the movement direction during sliding. However, when the contact surface itself changes rapidly or enters the jamming state, this assumption no longer holds, causing the algorithm to fail to accurately separate the normal force and friction force, and thus affecting the subsequent control strategy. Therefore, robots still rely on passive force threshold judgment for exploratory error correction, making it difficult to achieve high success rate of fine assembly under complex working conditions. SUMMARY

[0005] To solve the above technical problems of robots being difficult to achieve high success rate of fine assembly under complex working conditions, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides a robot flexible assembly method based on assembly state awareness, comprising: acquiring and smoothing force vectors and torque vectors generated when a robot end effector contacts a workpiece during the robot performing an assembly task; based on the wrench decomposition principle, converting the smoothed force vectors and torque vectors into a spatial equivalent force system to obtain the coordinate vector of the equivalent point of action of the contact force and torque sensed by the sensor; based on the projection variance minimum principle, determining the contact surface normal vector of the robot end effector and the workpiece according to the force vector sequence collected in the continuous control period, and decomposing the force vector into a normal support force perpendicular to the contact surface and a friction force along the contact surface; obtaining an instability index of the contact state according to the change of the equivalent contact point position and the change of the contact surface normal vector in the adjacent control period; presetting a threshold value of the instability index, and in response to the instability index being less than or equal to the threshold value, the controller performs compliant assembly control, and in response to the instability index being greater than the threshold value, the controller performs an adaptive avoidance strategy to ensure the stability of the assembly process.

[0007] The present application realizes accurate recognition of the physical geometric characteristics of the contact state by analyzing the collected force and torque signals into the equivalent contact point position and the contact surface normal vector, evaluates the stability of the contact state by calculating the change rate of the equivalent contact point and the normal vector, obtains an instability index that can represent the transition degree from stable sliding to unstable multi-point contact as an effective precursor indicator of the jamming risk, enables the system to identify the unstable trend of the assembly process in advance before the high force jamming occurs, and enables the system to actively perform normal retreat, lateral shaking and other adaptive avoidance strategies when the index exceeds the preset threshold value to break the forming jamming condition and restore stability. This closed-loop control based on accurate state recognition and precursor identification solves the technical problems that the traditional method is difficult to effectively evaluate the stability and avoid jamming under complex working conditions, thereby improving the stability and success rate of the robot flexible assembly.

[0008] Preferably, the acquisition of the force vectors and torque vectors generated when the robot end effector contacts the workpiece comprises: in each control period, the system collects the force vectors and torque vectors generated when the robot end effector contacts the workpiece, and performs smoothing processing on the data by using the moving average filtering method to obtain the smoothed force vectors and torque vectors generated when the robot end effector contacts the workpiece.

[0009] The present application uses moving average filtering to process the original signal, providing a stable and reliable data basis for subsequent accurate analysis of the contact state, and by averaging the data of the recent control period, the interference of the inherent high-frequency noise of the sensor on the force and torque signals can be suppressed, ensuring the stability and accuracy of the subsequent calculation, and avoiding model calculation failure or result oscillation caused by data jump.

[0010] Preferably, the coordinate vector of the equivalent point of contact of the contact force and the moment felt by the sensor satisfies the expression: ; wherein, represents the coordinate vector of the equivalent contact point; represents the force vector; represents the moment vector; represents the dot product operation of vectors, represents the vector length.

[0011] The present application provides a calculation method for real-time analysis of the position of the equivalent contact point based on the wrench decomposition principle in rigid body dynamics, can convert the abstract, high-dimensional force and moment signals output by the sensor into an intuitive, specific spatial coordinate in real time, and can visualize the complex force sensation information into a geometric position, so that the system can clearly recognize the specific part where the contact occurs, lay a foundation for subsequent evaluation of the stability of the contact state, and is a key step for realizing the transition from abstract force perception to specific state cognition.

[0012] Preferably, the contact surface normal vector of the robot end and the workpiece satisfies the expression: ; wherein, represents the contact surface normal vector; represents the candidate normal vector; represents the candidate normal vector that minimizes the expression in the parentheses ; represents the variance; is any time within the time window ; represents the force vector collected at any time within the time window ; represents the current time; represents the time window length for calculating the variance, represents the dot product operation of vectors.

[0013] Preferably, the force vector is decomposed into a normal support force perpendicular to the contact surface and a friction force along the contact surface, comprising: ; ; wherein, represents the normal force vector of the robot end effector in contact with the workpiece; represents the contact surface normal vector; represents the friction force vector of the robot end effector in contact with the workpiece; represents the force vector; represents the dot product operation of vectors.

[0014] Preferably, the instability index of the contact state satisfies the expression: ; wherein,​ is the time; represents the instability index at the time; respectively represent the current time and the equivalent contact point coordinate vector parsed out in the last control cycle; respectively represent the current time and the contact surface normal vector parsed out in the last control cycle; represents the length of a control cycle; represents a first hyperparameter for balancing position changes; represents a second hyperparameter for balancing attitude changes; represents the dot product operation of vectors, represents the vector length.

[0015] The instability index of the present application comprehensively considers the position change rate of the equivalent contact point and the change rate of the contact surface normal vector, so as to evaluate the stability of the assembly process in real time and objectively. Since the physical precursor of the jam is the transition from continuous smoothness to dramatic and discontinuous jump of the contact geometry, the index is highly sensitive to the trend of deviating from the stable state, and thus can be used as a reliable precursor index for identifying the jam risk. By calculating the index, the system can provide an early warning and identify the potential assembly failure risk before the contact force increases sharply and a complete jam is formed, thereby providing a key decision basis for executing the active avoidance strategy.

[0016] Preferably, the threshold value of the preset instability index comprises: long-time operation of the system under multiple normal working conditions to collect a large amount of historical instability indexes, arrangement of the historical instability indexes from small to large, and setting the quantile as the threshold value.

[0017] The present application determines the threshold value of the instability index by using a statistical learning method, improves the reliability and seriousness of the avoidance strategy triggering, reduces false positives, ensures that the threshold value is data-driven and objective, so that the system only starts the avoidance action when an extremely probable abnormal event occurs, thereby effectively distinguishing between normal signal fluctuations and real jam precursors, reducing system misjudgments and unnecessary interruptions caused by normal noise, and ensuring the smoothness and efficiency of the entire assembly process.

[0018] Preferably, the first hyperparameter is set to 1.0.

[0019] Preferably, the second hyperparameter is set to 0.5 m / rad.

[0020] In a second aspect, the present application provides a robot flexible assembly system based on assembly state cognition, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned robot flexible assembly method based on assembly state cognition is realized.

[0021] By adopting the above technical solution, the above-mentioned robot flexible assembly method based on assembly state cognition is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that a terminal device is manufactured according to the memory and the processor, and use is facilitated.

[0022] The present application has the following beneficial effects: The present application pre-processes the original signal by using the sliding average filter, provides a stable and reliable data basis for subsequent accurate calculation, analyzes the force and torque signals into the spatial coordinates of an equivalent contact point, visualizes the complex force sensation into a geometric position, determines the unit normal vector of the contact surface based on the principle of minimum projection variance, and further decomposes the total contact force into the normal support force and the friction force with clear physical meaning. A series of steps jointly construct a complete physical image of the contact event occurrence position, attitude information and force property, and solve the problem that the traditional method is difficult to inverse the contact information from the high-dimensional coupled data.

[0023] The present application establishes a mechanism capable of early warning of jamming risk, changes the system from passive force threshold judgment to active risk prediction, obtains the instability index of the current contact state stability, and obtains the change rate of the equivalent contact point position and the contact surface normal vector through real-time calculation, which can sensitively capture the physical precursor of jamming in the assembly process. In order to ensure the reliability of the prediction, the historical data under normal working conditions are collected to set the trigger threshold of the instability index, so as to effectively avoid false alarms caused by normal working condition fluctuations.

[0024] The present application integrates accurate state cognition and risk prediction capability into the robot control loop, forms a set of efficient adaptive assembly strategy, when the contact state is stable, the controller can use the decoupled normal force and friction force information to perform fine control, so as to ensure the smooth and efficient of the assembly process. Once the system identifies the precursor of jamming through the instability index, it will immediately suspend the current task and actively execute the adaptive avoidance strategy including normal retreat, lateral shaking or rotary relaxation, and the robot can continue to perform the assembly task after the instability index falls below the threshold, so that the robot can autonomously cope with the uncertainty in the assembly process, and significantly improve the success rate and stability of flexible assembly under complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1is a flow chart illustrating a robot flexible assembly method based on assembly state cognition in the present application; Figure 2 is a schematic diagram illustrating comparison of instability index and threshold value in the present application. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0027] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0028] The embodiments of the present application disclose a robot flexible assembly method based on assembly state cognition, referring to Figure 1 , comprising steps S1-S5: S1, during execution of an assembly task by a robot, acquiring and smoothing force vectors and moment vectors generated when an end effector of the robot contacts a workpiece.

[0029] Specifically, a robot system establishes a tool coordinate system at a tool center point before execution of an assembly task; a force and moment sensor is installed between a wrist flange of the robot and an end effector; during the assembly process, the force and moment sensor collects force and moment information generated when the end effector of the robot contacts a workpiece environment in real time at a preset control period. The control period length of the present embodiment is , which can be selected by an implementer according to actual conditions.

[0030] Further, in each control period, the system collects force vectors and moment vectors generated when the end effector of the robot contacts the workpiece, and performs smoothing processing on the data by using a sliding average filtering method. The size of a sliding window is , which is preferably . When is less than , the filtering effect is not obvious, and noise cannot be effectively suppressed; when is greater than , the signal is smoother, but a larger delay is introduced, which affects the real-time response capability of the system. The present embodiment selects , which can achieve a good balance between noise reduction effect and system real-time performance. The force vectors and moment vectors generated when the smoothed end effector of the robot contacts the workpiece are simply referred to as force vectors and moment vectors.

[0031] S2, based on the wrench decomposition principle, convert the smoothed force vector and the moment vector into a spatial equivalent force system to obtain a coordinate vector of an equivalent action point of the contact force and the moment sensed by the sensor.

[0032] It should be noted that after obtaining the force vector and the moment vector, the present application first confirms the contact position of the robot and the contact surface in order to solve the technical problem that the robot is difficult to achieve high success rate of fine assembly under complex working conditions. According to the wrench decomposition principle in rigid body dynamics, any spatial force system can be equivalent to a force acting on a spatial equivalent force action line and a moment parallel to the force. The object of the present application is to inverse a representative action point in the equivalent force system, i.e. the equivalent contact point.

[0033] Specifically, the coordinate vector of the equivalent action point of the contact force and the moment sensed by the sensor satisfies the expression:

[0034] wherein, represents the coordinate vector of the equivalent contact point; represents the force vector; represents the moment vector; represents the dot product operation of the vector, represents the vector module length.

[0035] In the formula, the vector of the nearest point on the equivalent force action line to the origin of the coordinate system is calculated . a displacement vector along the force vector direction is calculated, and the size is determined by the projection of the moment in the force direction, and the sum of the two parts ultimately confirms the unique position of the equivalent contact point.

[0036] When the contact state of the robot end and the environment changes, the force and the moment measured by the sensor will change, and the present application can convert these dynamic changes of force sensation information into changes of a direct geometric position in space in real time. For example, when the side of the pin contacts the edge of the hole, the generated moment will make the calculated equivalent contact point located on the side of the pin; when the top of the pin contacts the bottom of the hole, the moment is close to zero, and the calculated equivalent contact point will be located on the axis of the pin, and the present application abstractly converts the force perception into the concrete contact position cognition.

[0037] It should be added that a small threshold for the magnitude of the force vector is set, such as 0.01 Newtons. When the magnitude of the force vector is greater than the preset threshold, the equivalent contact point expression is used for calculation; when its magnitude is less than or equal to the threshold, it is considered that there is no effective contact. At this time, the equivalent contact point is meaningless, and the robot continues to move according to the original assembly program path.

[0038] S3. Based on the force vector sequence collected during the continuous control cycle, determine the normal vector of the contact surface between the robot end and the workpiece according to the principle of minimizing the projection variance, and decompose the force vector into a normal support force perpendicular to the contact surface and a frictional force along the contact surface.

[0039] It should be noted that after resolving the coordinate vectors of the equivalent contact points, the system also needs to understand the nature of the contact, i.e., how the robot should adjust its posture. The force generated when the robot's end effector contacts the workpiece is the vector sum of the normal force vector perpendicular to the contact surface and the frictional force vector tangential to the contact surface. This invention separates the normal force vector and the frictional force vector from the mixed force vector.

[0040] Specifically, the expression for the normal vector of the contact surface between the robot end effector and the workpiece is as follows:

[0041] in, Represents the normal vector of the contact surface; Represents the candidate normal vector; This indicates the search for the candidate normal vector that minimizes the value of the expression within the parentheses. ; Indicates variance; It is a time window At any time within; Indicates the time window Internal acquisition at any time The force vector; Indicates the current moment; This indicates the length of the time window used to calculate the variance. This represents the dot product operation of vectors.

[0042] The system iterates through all possible unit directions, ranging from 1 degree to 360 degrees, in increments of 1 degree. And for each Calculation in the past Force vector sequence within time t. Projection value in this direction variance Find the direction that minimizes variance. As the most reliable contact surface normal vector currently available , because in the process of micro sliding assembly, as long as the contact does not jump, for example, from one face to another face, the normal direction of the contact surface is relatively stable; while the size and direction of the friction force will change significantly with the change of the robot motion trend, so the projection of the force vector on the real normal direction should have the minimum fluctuation.

[0043] Further, once the contact surface normal vector is determined, the normal force and friction force of the robot end effector in contact with the workpiece can be decoupled by vector projection, specifically, ; ; wherein, represents the normal force vector of the robot end effector in contact with the workpiece; represents the contact surface normal vector; represents the friction force vector of the robot end effector in contact with the workpiece; represents the force vector; represents the dot product operation of the vector.

[0044] In the embodiment, the time window length is preferably to . If the time window length is too short, for example, less than , the data sample is too small, and the variance calculation is easily disturbed by accidental noise, resulting in unstable estimation; if the time window length is too long, for example, greater than , the system will become sluggish in response to the real change of the contact surface. The embodiment selects the time window length as , which can achieve a good balance between estimation stability and dynamic responsiveness. The implementer can select other time window lengths according to the actual situation.

[0045] S4, obtaining an instability index of the contact state according to the change of the equivalent contact point position and the change of the contact surface normal vector in the adjacent control period.

[0046] It should be noted that on the basis of obtaining the equivalent contact point and the contact surface normal vector, the present application evaluates the stability of the current contact state and gives an early warning of the impending jamming risk, and the physical basis is that the evolution of the equivalent contact point and the contact surface normal vector of a stable single-point sliding contact process should be continuous and smooth; on the contrary, when the assembly process transitions from stable single-point contact to unstable multi-point contact, the equivalent contact point and the contact surface normal vector will jump sharply and discontinuously.

[0047] Specifically, in order to evaluate the degree of such jump, the present application obtains an instability index of the contact state, which satisfies the expression:

[0048] wherein, is the time instant; denotes the instability index at time instant; denotes the equivalent contact point coordinate vector resolved at the current time instant and the last control cycle respectively; denotes the contact surface normal vector resolved at the current time instant and the last control cycle respectively; denotes the length of a control cycle; denotes the first hyper-parameter for balancing the position change; denotes the second hyper-parameter for balancing the posture change; denotes the dot product operation of vectors, denotes the vector norm.

[0049] wherein, is the weighted value of the moving speed of the equivalent contact point, which matches the nominal motion speed of the robot when stable sliding occurs; when the equivalent contact point slides from one plane to another, the value of this term instantaneously increases.

[0050] wherein, is the weighted value of the rotational angular speed of the contact surface normal vector, which changes slowly when moving along a smooth surface; when the equivalent contact point encounters a sharp corner or forms a new equivalent contact point, causing the equivalent normal to change dramatically, the value of this term instantaneously increases.

[0051] The first hyper-parameter range is [0.7, 1.3], which is used to balance the weight of the equivalent contact point position change rate. If the first hyper-parameter is set too small, the system will not be sensitive to the position mutation of the contact point, which may lead to the failure to identify the jamming precursor caused by the dramatic change in position in time. If the first hyper-parameter is set too high, the system may overreact to the normal jitter or smooth movement of the contact position, resulting in misjudgment of normal assembly as unstable, thus frequently triggering the avoidance strategy, reducing the assembly efficiency. The first hyper-parameter of the embodiment is set to The second hyper-parameter range is [0.2 m / rad, 0.8 m / rad]. If the second hyper-parameter is set too low, the system will become insensitive to the dramatic change in contact posture, resulting in the system's inability to identify such serious jamming risks, thus missing the opportunity to execute the adaptive avoidance strategy, reducing the assembly success rate. If the second hyper-parameter is set too high, the system will overreact to any small change in the contact surface normal vector, including the normal continuous change of the robot when moving along a smooth surface. As a result, the system will constantly misjudge and frequently trigger unnecessary avoidance actions, seriously affecting the smoothness and efficiency of assembly. The second hyper-parameter of the embodiment is set to = 0.5 m / rad. In other embodiments, the implementer can select the values of the first hyper-parameter and the second hyper-parameter according to the actual situation.

[0052] Instability Index It integrates the degree of drastic change in contact geometry and attitude, becoming an indicator that is highly sensitive to any deviation from a stable and continuous evolutionary trend, and is the core index for recognizing precursors to catastrophic events.

[0053] S5. A threshold for the instability index is preset. When the instability index is less than or equal to the threshold, the controller executes compliant assembly control. When the instability index is greater than the threshold, the controller executes an adaptive avoidance strategy to ensure the stability of the assembly process.

[0054] Specifically, the system operates under various normal conditions for extended periods, collecting a large number of historical instability indices, which are then arranged from smallest to largest. quantiles are thresholds This ensures that the system only triggers the avoidance strategy when an abnormal event with a very high probability occurs, thus avoiding false alarms caused by normal noise.

[0055] Furthermore, the system monitors in real time during the assembly process. The value is then used to execute adaptive control logic: In response to The system determines that the current contact state is stable. At this point, the robot controller can use the decoupled force information to perform fine control. For example, a force controller can be implemented to fine-tune the robot's posture so that the normal force... Maintain at a level such as A smaller expected value is needed to achieve a smooth fit between the robot and the contact surface; this can be achieved based on friction. The magnitude of the value is used to feedforward compensation of the driving force of the robot along the direction of motion to overcome sliding resistance.

[0056] In response to The system determines that a significant anomaly requiring immediate intervention has occurred, indicating an impending jam. At this point, the system immediately suspends the current propulsion or insertion action and actively executes a pre-set avoidance strategy aimed at breaking the emerging self-locking or wedge-like conditions.

[0057] For example, the avoidance strategy may include one or more combinations: the controller instructs the robot to move along the currently estimated contact surface normal vector. Make a small backward movement in the opposite direction, for example, move... This action reduces contact pressure and releases stress; while pausing the spindle advance, it applies a force perpendicular to the spindle direction, such as... High frequency, such as A slight translational vibration. This action helps to loosen the equivalent contact points that can have been slightly stuck; a slight rotational moment around the main axis is applied, trying to change the contact angle by slightly rotating, thus breaking the wedging condition.

[0058] Further, after performing one or more avoidance actions, the system re-enters the state-aware mode and analyzes the current contact state and instability index again If Falls below the threshold, indicating that the jamming risk has been resolved, the system will try to continue the assembly task with the corrected pose or path. If Still very high, the system may try another avoidance strategy, or after multiple attempts fail, eventually stop the task and request human intervention.

[0059] For example, a schematic diagram of the comparison between the instability index and the threshold is shown in Figure 2 As shown in Figure 2 The blue curve is the curve of the instability index of the contact state changing with time, the red dotted line is the threshold, and the red dot is the unstable point when the instability index exceeds the threshold. When the instability index exceeds the threshold, the preset avoidance strategy is executed, and the instability index is reduced until it is less than the threshold.

[0060] So far, a robot flexible assembly method based on assembly state awareness has been realized.

[0061] The embodiment of the present application also discloses a robot flexible assembly system based on assembly state awareness, comprising a processor and a memory, and the memory stores computer program instructions. When the computer program instructions are executed by the processor, the robot flexible assembly method based on assembly state awareness according to the present application is realized.

[0062] The above system also includes a communication bus and a communication interface and other components familiar to those skilled in the art. The settings and functions of these components are known in the art, and therefore will not be described here.

Claims

1. A robot flexible assembly method based on assembly state awareness, characterized by, The method comprises: acquiring and smoothing force vectors and torque vectors generated when a robot end effector contacts a workpiece during execution of an assembly task by the robot; based on a wrench decomposition principle, converting the smoothed force vectors and torque vectors into a spatial equivalent force system to obtain a coordinate vector of an equivalent action point of contact force and torque sensed by a sensor; based on a projection variance minimum principle, determining a normal vector of a contact surface between the robot end effector and the workpiece according to a sequence of force vectors collected in a continuous control period, and decomposing the force vectors into a normal support force perpendicular to the contact surface and a friction force along the contact surface; acquiring an instability index of a contact state according to a change in position of the equivalent contact point and a change in the normal vector of the contact surface in adjacent control periods; presetting a threshold value of the instability index, and in response to the instability index being less than or equal to the threshold value, a controller executing compliant assembly control, and in response to the instability index being greater than the threshold value, the controller executing an adaptive avoidance strategy to ensure stability of the assembly process.

2. The robot flexible assembly method based on assembly state awareness according to claim 1, characterized in that, The method of acquiring and smoothing force vectors and torque vectors generated when a robot end effector contacts a workpiece comprises: In each control period, the system collects force vectors and torque vectors generated when the robot end effector contacts the workpiece, and uses a moving average filtering method to smooth the data to obtain smoothed force vectors and torque vectors generated when the robot end effector contacts the workpiece. 3.The robot flexible assembly method based on assembly state awareness according to claim 1, wherein, The coordinate vector of the equivalent action point of the contact force and torque sensed by the sensor satisfies the expression: ； wherein, represents a coordinate vector of the equivalent contact point; represents a force vector; represents a moment vector; represents a dot product operation of vectors, represents a vector length.

4. The robot flexible assembly method based on assembly state awareness according to claim 1, characterized in that, The normal vector of the contact surface between the robot end effector and the workpiece satisfies the expression: ； wherein, represents a contact surface normal vector; represents a candidate normal vector; represents finding a candidate normal vector that minimizes the value of the expression within the parentheses ; represents a variance; is any time within a time window ; represents a force vector collected at any time within a time window ; represents a current time; represents a time window length for calculating variance, represents a dot product operation of vectors.

5. The robot flexible assembly method based on assembly state awareness according to claim 1, characterized in that, The method of decomposing the force vectors into a normal support force perpendicular to the contact surface and a friction force along the contact surface comprises: ； ； wherein, represents a normal force vector of the robot end effector in contact with the workpiece; represents a contact surface normal vector; represents a friction force vector of the robot end effector in contact with the workpiece; represents a force vector; represents a dot product operation of vectors.

6. The robot flexible assembly method based on assembly state awareness according to claim 1, wherein, The instability index of the contact state satisfies the expression: ； wherein, is the time instant; denotes the instability index at time instant; , denote the current time instant and the equivalent contact point coordinate vector resolved at the previous control period, respectively; , denote the current time instant and the contact surface normal vector resolved at the previous control period, respectively; denotes the length of a control period; is a first hyperparameter for balancing position variations; is a second hyperparameter for balancing attitude variations; denotes the dot product operation of vectors, denotes the vector norm.

7. The robot flexible assembly method based on assembly state awareness according to claim 1, wherein, The method of presetting a threshold value of the instability index comprises: The acquisition system collects a large number of historical instability indexes under various normal working conditions for a long time, arranges the historical instability indexes from small to large, sets The quantile is a threshold value.

8. The robot flexible assembly method based on assembly state awareness according to claim 6, wherein, The first hyperparameter is set to 1.

0.

9. The robot flexible assembly method based on assembly state awareness according to claim 6, wherein, The second hyperparameter is set to 0.5 m / rad.

10. A robot flexible assembly system based on assembly state awareness, characterized by, The method comprises: a processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a robot flexible assembly method based on assembly state cognition according to any one of claims 1-9.

Images