阅读说明：本技术 一种机器人精密装配系统及方法 (Robot precision assembling system and method ) 是由 宋锐 崔涛 李凤鸣 付天宇 王艳红 田新诚 于 2021-09-07 设计创作，主要内容包括：本公开提出了一种机器人精密装配系统及方法,所述装配过程分为外部接触阶段和内部接触阶段,包括：在外部接触阶段,基于环境反馈的力觉信息,通过直觉搜索策略,发出配对指令,逐步引导装配插头与插座达到配对位姿,实现对准操作；在内部接触阶段,获取卡扣力序列,将卡扣力序列输入训练好的卡扣状态感知网络,输出装配状态值,根据装配状态值判断检测装配动作是否完成,实现机器人对工件的装配。(The present disclosure provides a robot precision assembly system and method, wherein the assembly process is divided into an external contact stage and an internal contact stage, and the assembly process comprises the following steps: in the external contact stage, based on force sense information fed back by the environment, a pairing instruction is sent out through an intuition search strategy, the assembly plug and the socket are gradually guided to achieve a pairing pose, and alignment operation is achieved; and in the internal contact stage, acquiring a buckle force sequence, inputting the buckle force sequence into a trained buckle state perception network, outputting an assembly state value, judging whether the detection assembly action is completed according to the assembly state value, and realizing the assembly of the robot on the workpiece.)

1. A method for precision assembly of a robot, the assembly process being divided into an external contact phase and an internal contact phase, comprising:

in the external contact stage, based on force sense information fed back by the environment, a pairing instruction is sent out through an intuition search strategy, the assembly plug and the socket are gradually guided to achieve a pairing pose, and alignment operation is achieved;

and in the internal contact stage, acquiring a buckle force sequence, inputting the buckle force sequence into a trained buckle state perception network, outputting an assembly state value, judging whether the detection assembly action is completed according to the assembly state value, and realizing the assembly of the robot on the workpiece.

2. The robotic precision assembly method of claim 1, wherein the intuitive search strategy is established by the following process:

acquiring an initial target pose of a workpiece to be assembled;

and performing an approaching action, judging whether the assembly parts are in contact or not according to the initial pose, if so, performing a searching action, and otherwise, controlling the assembly parts to reach the contact positions through the Cartesian space positions by the robot.

3. The robotic precision assembly method of claim 2, wherein the step of performing a search action comprises:

and judging whether the assembly workpieces reaching the contact positions are successfully matched, if so, controlling the revolution and the rotation of the robot by using the force-position mixed model until the assembly workpieces are successfully matched.

4. The robot precision assembling method according to claim 3, wherein said judging whether the assembling work pieces reaching the contact position are successfully paired comprises:

and force control is carried out on the Z axis in the assembly direction by adopting an admittance model, constant force contact is carried out, position control is carried out on the other axial directions, when the target torque of the Z axis of the tool coordinate is greater than the target torque constant force, the X axis force of the tool coordinate is greater than the X axis constant force, and the Y axis force of the tool coordinate is greater than the Y axis constant force, the pairing is judged to be successful.

5. The robot precision assembling method according to claim 3, wherein the motion path of the search motion is an Archimedes spiral, the motion plane is an XY plane in a world coordinate system, and the span of the robot during the motion process is smaller than the assembling precision, and the span includes a long axis direction continuous turn span and a short axis direction continuous turn span.

6. The robotic precision assembly method of claim 3, wherein the snap state aware network training process comprises the steps of:

acquiring a Z-axis force sequence of a tool coordinate system, and establishing a sample database;

randomly initializing a network weight value;

determining a learning rate, carrying out periodic training, and updating a network weight value according to a loss value obtained by each training;

and (4) carrying out real-time force perception by using the trained network model, and outputting an assembly state value to control whether the robot stops applying pressing force.

7. The robot precision assembling method of claim 3, wherein the snap state perception network adopts a circular network structure, and comprises an input layer, a hidden layer and an output layer.

8. A robotic precision assembly system, comprising:

an alignment module configured to: in the external contact stage, a pairing instruction is sent out based on force sense information fed back by the environment, and the assembly plug and the socket are gradually guided to achieve a pairing pose through an intuition search strategy, so that alignment operation is realized;

and the state perception module is configured to acquire a buckling force sequence at an internal contact stage, input the buckling force sequence into a trained buckling state perception network, output an assembly state value, judge whether the detection assembly action is finished according to the assembly state value, and realize the assembly of the robot on the workpiece.

9. An assembly robot, comprising: robot body and controller, characterized in that the controller comprises a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of a robot precision assembly method according to any of claims 1-7 when executing the program.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of a robot precision assembly method according to any one of claims 1-7.

Technical Field

The disclosure relates to the technical field of control of assembly robots, in particular to a robot precision assembly system and method.

Background

The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.

In the production of 3C field, flexible circuit board (FPC) assembly utilizes the board to connect board connector (BTB) usually, and its inside is "returning" shape structure, and it is small to have the structure complicacy, the assembly is accurate, the shallow characteristics of assembly depth, and current BTB assembly system is mostly through the completion assembly operation of manual fit cartesian robot and peculiar frock, because production update iteration accelerates gradually, the frock need in time be changed to the production line, and the artifical work demand is big. However, with the high standards and high requirements of industrial production, the problems of low labor efficiency, low precision and weak production flexibility are urgently needed to be solved. The current robot assembly system mainly uses a vision-based method as a main part, but the assembly of BTB requires high-precision positioning of vision, and is easily influenced by problems such as vision occlusion, visual angle deviation and the like in the assembly process, so that assembly data is lost, and the assembly system is easily interfered by environmental noise. Different from the traditional shaft hole assembly, in the micro-precision assembly of the BTB connector, the assembly depth is only 0.5mm, the problems of blocking, wedging and the like caused by large initial attitude difference are almost avoided, but usually, the visual sensing ignores the internal information at the moment of clamping, and only a force threshold value is set to judge whether the assembly is completed or not, so that the potential risk exists, and the force in the insertion process needs to be sensed.

Disclosure of Invention

The purpose of the implementation mode of the specification is based on a robot precision assembling method and system, in order to enable the robot to realize the assembling of BTB, the invention establishes a robot two-stage force perception model, and in the first stage, based on force perception information fed back by the environment, a plug can be guided to reach a state of aligning with a socket through an intuition search strategy; in the second stage, due to the structural influence, force needs to be applied in the assembling direction to obtain a force sequence collected in real time, and whether assembling action is completed or not is detected through the buckle state perception network, so that the BTB assembling of the robot is realized.

The invention provides a robot precision assembling method, which is realized by the following technical scheme:

the assembly process is divided into an external contact phase and an internal contact phase, and comprises:

in the external contact stage, based on force sense information fed back by the environment, a pairing instruction is sent out through an intuition search strategy, the assembly plug and the socket are gradually guided to achieve a pairing pose, and alignment operation is achieved;

and in the internal contact stage, acquiring a buckle force sequence, inputting the buckle force sequence into a trained buckle state perception network, outputting an assembly state value, judging whether the detection assembly action is completed according to the assembly state value, and realizing the assembly of the robot on the workpiece.

In a further technical solution, the process of establishing the intuitive search strategy is as follows:

acquiring an initial target pose of a workpiece to be assembled;

and performing an approaching action, judging whether the assembly parts are in contact or not according to the initial pose, if so, performing a searching action, and otherwise, controlling the assembly parts to reach the contact positions through the Cartesian space positions by the robot.

According to a further technical scheme, the step of executing the search action comprises:

and judging whether the assembly workpieces reaching the contact positions are successfully matched, if so, controlling the revolution and the rotation of the robot by using the force-position mixed model until the assembly workpieces are successfully matched.

According to a further technical scheme, the judging whether the assembly workpieces reaching the contact position are successfully paired comprises the following steps:

and force control is carried out on the Z axis in the assembly direction by adopting an admittance model, constant force contact is carried out, position control is carried out on the other axial directions, when the target torque of the Z axis of the tool coordinate is greater than the target torque constant force, the X axis force of the tool coordinate is greater than the X axis constant force, and the Y axis force of the tool coordinate is greater than the Y axis constant force, the pairing is judged to be successful.

According to the further technical scheme, the motion path of the search action is an Archimedes spiral line, the motion plane is an XY plane under a world coordinate system, the span of the robot in the motion process is smaller than the assembly precision, and the span comprises a long-axis-direction continuous turn span and a short-axis-direction continuous turn span.

According to a further technical scheme, the buckling state perception network training process comprises the following steps:

acquiring a plurality of force sequences and establishing a sample database;

randomly initializing a network weight value;

determining a learning rate, carrying out periodic training, and updating a network weight value according to a loss value obtained by each training;

and (4) carrying out real-time force perception by using the trained network model, and outputting an assembly state value to control whether the robot stops applying pressing force.

According to the further technical scheme, the buckle state perception network adopts a circulating network structure and comprises an input layer, a hidden layer and an output layer.

The second aspect of the invention provides a robot precision assembly system, which is realized by the following technical scheme:

an alignment module configured to: in the external contact stage, a pairing instruction is sent out based on force sense information fed back by the environment, and the assembly plug and the socket are gradually guided to achieve a pairing pose through an intuition search strategy, so that alignment operation is realized;

and the state perception module is configured to acquire a buckling force sequence at an internal contact stage, input the buckling force sequence into a trained buckling state perception network, output an assembly state value, judge whether the detection assembly action is finished according to the assembly state value, and realize the assembly of the robot on the workpiece.

A third aspect of the present invention provides a precision assembling robot comprising: the robot comprises a robot body and a controller, wherein the controller comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, and the robot body is characterized in that the processor executes the program to realize the steps of the robot precision assembling method.

A fourth aspect of the invention provides a computer-readable storage medium, on which a computer program is stored, which program, when being executed by a processor, carries out the steps of a robot precision assembly method as described above.

Compared with the prior art, the beneficial effect of this disclosure is:

(1) the method comprises the steps that mechanical characteristics of each stage of assembly are analyzed, a two-stage force sensing model of the robot is established, and in the first stage, a plug can be guided to reach a state of being aligned with a socket through an intuition search strategy based on force sense information fed back by the environment; in the second stage, due to the structural influence, force needs to be applied in the assembling direction to obtain a force sequence collected in real time, and whether assembling action is completed or not is detected through the buckle state perception network, so that the BTB assembling of the robot is realized.

(2) According to the robot assembly method, the force position hybrid control of the robot is achieved by acquiring the six-dimensional force information of the tail end, an intuition search strategy and a buckle state sensing method are provided, position control and force and moment control are combined to form assembly actions in different stages, and the mechanical arm completes the assembly actions by using the force position hybrid control, so that the robot can rapidly achieve assembly in a small range.

(3) The intuition tactics of this application robot can adapt to multiple assembly parts to utilize the deep learning to master assembly state perception skill, can initiatively adapt to the change of environment, thereby make the assembly operation more possess flexible generalization ability.

Drawings

The accompanying drawings, which are included to provide a further understanding of the disclosure, illustrate embodiments of the disclosure and together with the description serve to explain the disclosure and are not to limit the disclosure.

FIG. 1 is a flow chart of a robot precision assembly method disclosed by the present invention;

FIG. 2A is a board-to-board connector receptacle model;

FIG. 2B is a board-to-board connector plug model;

FIG. 3 is a schematic diagram of a revolution search path according to an embodiment of the present invention;

FIG. 4 is a search action model based on an intuitive search strategy according to an embodiment of the present invention;

FIG. 5 is a flowchart illustrating a buckling state awareness network training process according to an embodiment of the present invention;

fig. 6 is a block diagram illustrating a buckle status awareness network structure according to an embodiment of the present invention.

Detailed Description

It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present disclosure. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.

Example one

As described in the background art, in the micro precision assembly of the BTB connector, the assembly depth is only 0.5mm, and the problems of jamming, wedging and the like caused by a large initial attitude difference hardly exist, but usually visual sensing ignores internal information at the moment of clamping, and only a force threshold is set to judge whether the assembly is completed or not, so that a potential risk exists, and the force in the insertion process needs to be sensed.

According to the BTB assembly process flow characteristics, the assembly process is divided into 2 stages of an external contact stage and an internal contact stage, and a two-stage force sensing model of the robot is established by analyzing and assembling mechanical characteristics of each stage. The robot force position hybrid control is realized by acquiring the six-dimensional force information of the tail end, and an intuition search strategy and a buckling state sensing method are provided. In the first stage, based on the force sense information fed back by the environment, the plug can be guided to reach the state of aligning with the socket through an intuition search strategy; in the second stage, due to the structural influence, force needs to be applied in the assembling direction to obtain a force sequence collected in real time, and whether assembling action is completed or not is detected through the buckle state perception network, so that the BTB assembling of the robot is realized.

As shown in fig. 1, the present application provides a robot precision assembly method, which includes the steps of:

according to the process flow characteristics of board-to-board connector (BTB) assembly, the assembly process is divided into an external contact stage and an internal contact stage;

step A1: in the external contact stage, based on force sense information fed back by the environment, a pairing instruction is sent out through an intuition search strategy, the assembly plug and the socket are gradually guided to achieve a pairing pose, and alignment operation is achieved;

step A2: in the internal contact stage, force is applied in the assembly direction to obtain a buckle force sequence, a trained buckle state perception network is input, an assembly state value is output, whether the detection assembly action is completed or not is judged according to the assembly state value, and the assembly of the robot on the board-to-board connector workpiece is realized, wherein fig. 2A is a board-to-board connector socket model, and fig. 2B is a board-to-board connector plug model.

In step a1, in the external contact phase, the intuitive search strategy is established as follows:

acquiring an initial target pose of a workpiece to be assembled;

specifically, the step A1 is to obtain the assembling force through the analysis of an assembling model before assemblingAnd pairing torqueObtaining an initial target pose P of a workpiece to be loaded through a vision sensor or a known CAD model₀＝(x₀,y₀,z₀,a₀,b₀,c₀) An initial search range length a and an initial search range width b.

Performing an approaching action, judging whether the assembly parts are in contact or not according to the initial pose, if so, performing a searching action, otherwise, controlling the assembly parts to reach the contact positions through the Cartesian space positions by the robot;

wherein the step of executing the search action comprises:

and judging whether the assembly workpieces reaching the contact position are successfully matched, if so, entering an internal contact stage, and otherwise, performing force-position hybrid model control on the revolution and rotation of the robot until the assembly workpieces are successfully matched.

Specifically, the robot reaches the contact position through cartesian space position control assembly spare, specifically includes: the robot obtains a set theta of each joint of the robot joint through inverse kinematics calculation, wherein theta is (theta)₁,θ₂,θ₃,θ₄,θ₅,θ₆)，θ_iEach joint value of six joints is obtained when the contact force is larger than the constant forceWhen the contact state is reached.

The method for controlling the revolution and the rotation of the robot by the force position hybrid model comprises the following steps:

fig. 3 is a schematic diagram of a revolution search path, as shown in fig. 3, the search path is an archimedes spiral, and the span of the robot during the movement process is smaller than the assembly precision, and the span includes a long axis direction continuous turn span and a short axis direction continuous turn span.

The motion plane is an XY plane under a world coordinate system and is defined as revolution motion:

the cartesian space coordinate equation is:

where max (α) is a, max (β) is b, angular velocityThe span of the continuous turns in the long axis direction is e_aThe span of the continuous turns in the minor axis direction is e_b；

Obtaining the continuous turn span in the major axis direction and the continuous turn span in the minor axis direction according to the following steps:

fig. 4 is a search motion model based on an intuitive search strategy, and as shown in fig. 4, in order to ensure that the matching point is on the motion path, the span of the robot during the motion process is smaller than the assembly precision, and the span comprises a long axis direction continuous turn span and a short axis direction continuous turn span.

The span e should be less than the assembly accuracy e_rI.e. byWhile the robot is based on the tool coordinate system O_eZ of (A)_eThe shaft direction carries out autorotation motion, the autorotation motion is periodic rotation motion, and the maximum motion angle isAngular velocityAnd carrying out reverse motion in the maximum angle state, and reciprocating in such a way.

Wherein the judging whether the assembly workpieces reaching the contact position are successfully paired comprises:

in the searching process, the robot is controlled by using a force position mixed model, an admittance model is adopted for force control on an assembly direction Z axis, and constant force is adoptedMaking contact, and adopting position control for other axial directions, and when the tool coordinate is Z-axis torque tau_zGreater than target Z-axis torqueTool coordinate X-axis force f_xGreater than target X-axis forceTool coordinate Y-axis force f_yGreater than target Y-axis forceAnd judging that the pairing is successful.

Fig. 5 is a flowchart of the buckle state awareness network training, and as shown in fig. 6, in step a1, in the external contact phase, the buckle state awareness network training process includes the following steps:

(1) acquiring a Z-axis force sequence of 100 tool coordinate systems, and establishing sample data S ═ P, y ];

in the formula, P is a value sequence after the window truncation and filtering, y is a sample label, y is {0,1}, 0 indicates incomplete assembly, and 1 indicates complete assembly.

Fitting contact force information s_f＝(f_x,f_y,f_z,τ_x,τ_y,τ_z) With snap-fit force sequence F ═ s_f(t₁),s_f(t₂),s_f(t₃),…s_f(t_n) After window interceptionThe window width is T, and the window width is obtained through filtering:

P_j＝A_j+D_j，

wherein A is_jLow frequency subband signal for the j-th sample, D_jThe high frequency subband signal of the j sample.

(2) Training a buckle state perception network; the method specifically comprises the following steps:

the specific structure of the buckle state perception network is shown in fig. 6, and the buckle state perception network comprises an input layer, a hidden layer and an output layer;

the buckle state perception network adopts a circulating network (RNN) structure;

inputting an assembly contact force information sequence by an input layer;

normalizing the assembly contact force information sequence to obtain:

where μ is the mean of all sample values, σ is the standard deviation of all sample values,

randomly initializing network weight value W_i；

Will U_i＝W_i-1·h^(i-1)+x⁽ⁱ⁾Input hidden layer, U only₁＝x⁽¹⁾Activating a functionEach node of the hidden layer is of a residual error structure, and is only U₁＝x⁽¹⁾Output is h⁽ⁱ⁾＝tanh(U_i)+x⁽ⁱ⁾Output h of only the T-th layer hidden node^(T)As influencing the output layer, there is o^(T)＝h^(T)V, output layer output prediction value y^∧＝o^(T)。

Wherein, the number of nodes of the input layer and the number of nodes of the hidden layer are both the window width T.

Fourthly, the output assembly state predicted value y^∧And comparing the loss value with a sample label y, and calculating through cross entropy to obtain a loss value. Wherein the cross entropy is:wherein N is the number of sample label types, y_iIs a class i tag value, p_iIs a probability value predicted as class i.

(3) And determining the learning rate r, carrying out m times of periodic training, and updating the network weight value according to the loss value obtained by each training.

(4) And (4) carrying out real-time force perception by using the trained network model, controlling whether the robot stops applying pressure or not by outputting a state value, and leaving the assembly part to finish the assembly task.

In the embodiment, the robot precision assembly method is adopted, and the robot is independently and minutely assembled by combining the model structure according to the force characteristics of different assembly stages; the position control and the force and moment control are combined to form assembling actions in different stages, and the mechanical arm completes the assembling actions by using force and position mixed control, so that the robot can quickly realize assembling in a small range. The intuition strategy of the robot can be adapted to various assembly parts, and the assembly state perception skill can be mastered by deep learning, so that the robot can actively adapt to the change of the environment, and the assembly operation has more flexible generalization capability.

Example two

The embodiment of the specification provides a robot precision assembling system, which is realized by the following technical scheme:

the method comprises the following steps:

a guide alignment module configured to: in the external contact stage, a pairing instruction is sent out based on force sense information fed back by the environment, and the assembly plug and the socket are gradually guided to achieve a pairing pose through an intuition search strategy, so that alignment operation is realized;

and the buckle sensing module is configured to acquire a buckle force sequence at an internal contact stage, input the buckle force sequence into a trained buckle state sensing network, output an assembly state value, judge whether the detection assembly action is completed according to the assembly state value, and realize the assembly of the robot on the workpiece.

EXAMPLE III

An embodiment of the present specification provides a precision assembly robot, including: the robot comprises a robot body and a controller, wherein the controller comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, and when the processor executes the program, the steps of the robot precision assembling method in the first embodiment are realized.

Example four

The implementation manner of the present specification provides a computer-readable storage medium, on which a computer program is stored, wherein the program, when executed by a processor, implements the steps of a robot precision assembly method in the first embodiment.

It is to be understood that throughout the description of the present specification, reference to the term "one embodiment", "another embodiment", "other embodiments", or "first through nth embodiments", etc., is intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, or materials described may be combined in any suitable manner in any one or more embodiments or examples.

The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure, and various modifications and changes may be made to the present disclosure by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present disclosure should be included in the protection scope of the present disclosure.