A 3D positioning control method and system for a full-viewpoint ultra-micro chip spacer

Abstract

The present invention discloses a three-dimensional positioning control method and system for a full-viewpoint ultra-micro chip spacer. By establishing a first trajectory iteration model for the manipulator to grasp the spacer, constructing a pseudo-pseudo partial derivative optimization function, optimizing the pseudo-pseudo partial derivative, constructing a system input optimization function, optimizing the system input, obtaining the system time-domain compensation control component of the manipulator by using a prediction compensation algorithm, and establishing an iterative prediction trajectory model according to the optimized pseudo-pseudo partial derivative, system input, and the system time-domain compensation control component of the manipulator, the optimal spacer position trajectory for the manipulator to grasp the spacer is obtained, solving the technical problem of low three-dimensional positioning accuracy of the existing ultra-micro chip spacer. The iterative prediction trajectory model established by the optimized pseudo-pseudo partial derivative, system input, and the predicted system time-domain compensation control component can obtain the optimal spacer position trajectory for the manipulator to grasp the spacer, greatly improving the three-dimensional positioning accuracy of the ultra-micro chip spacer.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of eutectic chip bonding, and particularly refers to a three-dimensional positioning control method and system for a full-viewpoint ultra-micro chip spacer block. Background Art

[0002] In the process of semiconductor chip production, since the eutectic chip bonding technology automatically realizes the eutectic chip bonding process by placing the package or substrate on a thermally controllable surface or moving the package or substrate to this surface, and has high accuracy and energy utilization rate, it has been widely used in the electronic device production industry. However, when operating chips of different specifications, due to differences such as different heights and different surface inclination angles of ultra-micro chips, the chip bonding quality of ultra-micro chips will be seriously affected. Therefore, in order to ensure that chips of different specifications can operate stably and efficiently under the same chip bonding operation, it is necessary to use a robotic arm to move the chip spacer block to raise the ultra-micro chip on the substrate, so that its top surface is stably at a given operation height. However, since the chip bonding operation has high requirements for the position accuracy of the spacer block and requires certain self-learning and anti-interference capabilities, traditional robotic arm control cannot meet the accuracy requirements. For this reason, a new robotic arm control method needs to be designed, which can accurately move the spacer block with given position information without manual monitoring, providing guarantee for the stable operation of the subsequent ultra-micro chip bonding process and improving the yield.

[0003] Currently, the methods for moving the spacer block can be mainly divided into manual operation methods and traditional open-loop motor control methods.

[0004] The manual operation method can operate the robotic arm motor manually and calibrate the position of the spacer block through manual visual inspection. Although this calibration method is simple to operate and has low cost, due to the defects that all mechanical operation processes are manually controlled and visual observation is inaccurate, the actual position of the final ultra-micro chip spacer block is quite different from the given position, and manual operation cannot summarize the law of position error, which is not conducive to the continuous improvement of the control process. Therefore, the accuracy cannot be improved.

[0005] The traditional open-loop motor control inputs the target spacer block position information into the robotic arm control system, and the system function automatically generates the target control strategy and executes the corresponding operation. This method solves the system error caused by human factors and can realize the statistics of the final target position error. However, it is an open-loop control process in the control process, with poor anti-interference ability, and cannot self-optimize the same control process during the control process, resulting in large environmental errors. Therefore, the above methods are not applicable to the displacement control process of high-precision ultra-micro chips.

[0006] The patent publication number CN115309044A patent application is a manipulator angular velocity control algorithm based on a predictive control model. It is characterized in that within the specified range is the response surface obtained according to the nonlinear model in the operation space, different stages of predictive control commands are divided, and by adjusting the predictive control quantity and incrementally searching upward for switching until the corresponding control mode that meets the control requirements is reached. Using this control method, the influence caused by interference factors such as motor errors and pad specifications can be overcome during the same control process. And it can still make timely adjustments to the control strategy when other unknown interference quantities are added, ensuring a high degree of accuracy. Moreover, this method also overcomes the pressure fluctuations and hydraulic vibrations existing in the manipulator device itself, and preferably avoids the risk of nonlinear overshoot, and can make timely adjustments to various interference factors. However, since the predictive model itself does not have a learning law, the model optimization of multiple control processes cannot be effectively utilized, resulting in a long model establishment time and a large amount of calculation, making the control process slow and the effectiveness will decrease.

[0007] The patent publication number CN115107034A patent application is a quantization iterative learning control method for a single manipulator. This method uses a lifting technique to convert the operation of a single manipulator performing repetitive control tasks into a matrix model on the iterative axis, and designs an encoder-decoder based on a finite uniform quantizer to achieve information interaction in the communication scenario. Under the norm optimization framework, a performance index function is designed by the scaling method, and further a quantization iterative learning algorithm is obtained. The tracking optimization control problem of the manipulator in multiple identical control processes can be solved through the iterative learning control learning law, and a selection scheme for the quantization level in the finite uniform quantizer is given. This control method can continuously reduce the system error as the number of iterations increases, and has an obvious optimization effect on the process with the same control logic. But for the sudden interference quantity in the same manipulator control process, it cannot make timely feedback and adjustment, resulting in insufficient system temporary response ability, easy to cause control logic chaos and lack of stability. Summary of the Invention

[0008] The three-dimensional positioning control method and system for full-view ultra-micro chip pads provided by the present invention solve the technical problem of low three-dimensional positioning accuracy of existing ultra-micro chip pads.

[0009] To solve the above technical problems, the three-dimensional positioning control method for full-view ultra-micro chip pads proposed by the present invention includes:

[0010] Establish a first trajectory iteration model for the manipulator to grasp the pad, and the first trajectory iteration model is specifically:

[0011]

[0012] Δq k (t + 1) = q k(t + 1)-q k-1 (t + 1),

[0013] Δu k (t)=u k (t)-u k-1 (t),

[0014] where Δq k (t + 1) represents the difference between the output pad position trajectory at the (k - 1)-th iteration and the output pad position trajectory at the (k - 2)-th iteration at time (t + 1), and Δu k (t) represents the difference between the system input at the k-th iteration and the system input at the (k - 1)-th iteration at time t, q k (t + 1) and q k-1 (t + 1) represent the output pad position trajectories at the k-th and (k - 1)-th iterations at time (t + 1) respectively, represents the estimated value of the pseudo-pseudo partial derivative at the k-th iteration, u k (t) and u k-1 (t) represent the system inputs at the k-th and (k - 1)-th iterations at time t respectively;

[0015] Construct a pseudo-pseudo partial derivative optimization function and optimize the pseudo-pseudo partial derivative according to the pseudo-pseudo partial derivative optimization function;

[0016] Construct a system input optimization function and optimize the system input according to the system input optimization function;

[0017] Based on the manipulator state increment model, adopt a predictive compensation algorithm to obtain the system time-domain compensation control component of the manipulator;

[0018] According to the optimized pseudo-pseudo partial derivative and system input, and the system time-domain compensation control component of the manipulator, establish an iterative prediction trajectory model;

[0019] Based on the iterative prediction trajectory model, obtain the optimal pad position trajectory for the manipulator to grasp the pad.

[0020] Furthermore, the calculation formula of the pseudo-pseudo partial derivative optimization function is:

[0021]

[0022] where Δq k-1 (t + 1) represents the difference between the output pad position trajectory at the (k - 1)-th iteration and the output pad position trajectory at the (k - 2)-th iteration at time (t + 1), and Δu k-1 (t) represents the difference between the system input at the (k - 1)-th iteration and the system input at the (k - 2)-th iteration at time t, and α represents a weight factor, represents the estimated value of the pseudo-pseudo partial derivative at the (k - 1)-th iteration.

[0023] Furthermore, the calculation formula of the system input optimization function is as follows:

[0024]

[0025]

[0026] where e k (t + 1) represents the iteration error of the k-th iteration at the (t + 1)-th moment, γ represents the system influence factor, u k (t) and u k-1 (t) respectively represent the system inputs of the k-th and (k - 1)-th iterations at the t-th moment, represents the system forgetting factor, ω represents the initial balance factor, and m represents the system adjustment control rate.

[0027] Furthermore, based on the manipulator state increment model, the system time-domain compensation control component of the manipulator obtained by using the predictive compensation algorithm includes:

[0028] Predict the manipulator position output for the future preset steps based on the predictive compensation algorithm and the current state quantity and position function of the manipulator to obtain a prediction sequence;

[0029] Solve for the minimum tracking error under the constraints of the manipulator system to obtain the manipulator motor control quantity at the next moment, and obtain the optimal predictive compensation control component.

[0030] Furthermore, the calculation formula of the system time-domain compensation control component is as follows:

[0031] δU * = (φ c T Vφ c + R) -1 φ c T V(V r - F c x c (t)),

[0032] where U * is the predicted optimal control quantity, δU * represents the optimal predictive compensation control quantity, φ c and F c are the first coefficient increment matrix and the second coefficient increment matrix respectively, R and V represent the first weight matrix and the second weight matrix respectively, and φ c T represents the transpose of φ c .

[0033] Further, based on the optimized pseudo - quasi - partial derivative, the system input, and the system time - domain compensation control component of the robotic arm, the specific formula for establishing the iterative prediction trajectory model is as follows:

[0034]

[0035] where e k-1 (t + 1) represents the iterative error of the (k - 1)-th iteration at the (t + 1)-th moment, and p represents the prediction compensation adjustment weight.

[0036] The three - dimensional positioning and control system for full - view ultra - micro chip pads provided by the present invention includes:

[0037] A memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the three - dimensional positioning control method for full - view ultra - micro chip pads provided by the present invention.

[0038] The present invention proposes a three - dimensional positioning control method and system for full - view ultra - micro chip pads. By establishing a first - trajectory iterative model for the robotic arm to grasp the pads, constructing a pseudo - quasi - partial derivative optimization function, optimizing the pseudo - quasi - partial derivative according to the pseudo - quasi - partial derivative optimization function, constructing a system - input optimization function, optimizing the system input according to the system - input optimization function, based on the robotic - arm state increment model, using the prediction compensation algorithm to obtain the system time - domain compensation control component of the robotic arm, establishing an iterative prediction trajectory model based on the optimized pseudo - quasi - partial derivative, the system input, and the system time - domain compensation control component of the robotic arm, and obtaining the optimal pad - position trajectory for the robotic arm to grasp the pads, the technical problem of low three - dimensional positioning accuracy of existing ultra - micro chip pads is solved. The iterative prediction trajectory model established by the optimized pseudo - quasi - partial derivative, the system input, and the predicted system time - domain compensation control component can obtain the optimal pad - position trajectory for the robotic arm to grasp the pads, greatly improving the three - dimensional positioning accuracy of ultra - micro chip pads.

[0039] The beneficial effects of the present invention specifically include:

[0040] (1). Based on the determination of the target - position information of the substrate pads of the eutectic mounter, input the target - position coordinate information into the variable - forgetting - factor non - parametric adaptive iterative learning control model, and continuously adjust the forgetting factor according to the control process, so that the control - model effect has a good fitting effect and a fast convergence speed in the dynamic stage, and has strong anti - system - random - noise interference performance in the steady state, and continuously update the system pseudo - quasi - partial derivative in the iterative learning, overcoming the condition that the traditional control algorithm overly relies on the system - structure parameters and the constant target trajectory, and the control effect is prominent. The closed - loop control process of the control motor is realized, and it has strong anti - interference ability.

[0041] (2) Based on the determination of the target position information of the substrate pads on the eutectic mounter, a predictive compensation control algorithm is introduced into the variable forgetting factor non - parametric adaptive iterative learning model. According to the predicted control quantity at the future moment, the compensation value of the control quantity in the next time series is continuously optimized, and real - time high - precision prediction compensation is performed on the single - time random noise of the system, significantly improving the anti - real - time random interference ability of the algorithm model in the single - time control process, achieving the error unification of the model in the iterative domain and the time domain, and thus significantly improving the position tracking performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is the algorithm flowchart of the three - dimensional positioning control method for the full - view ultra - micro chip pads in the second embodiment of the present invention;

[0043] Figure 2 It is the overall flowchart of the three - dimensional positioning control method for the full - view ultra - micro chip pads in the second embodiment of the present invention;

[0044] Figure 3 It is the flowchart of the three - dimensional positioning control method for the full - view ultra - micro chip pads in the third embodiment of the present invention;

[0045] Figure 4 It is the hardware example diagram of the three - dimensional positioning control method for the full - view ultra - micro chip pads in the third embodiment of the present invention;

[0046] Figure 5 It is the system block diagram of the three - dimensional positioning control method for the full - view ultra - micro chip pads in the embodiment of the present invention.

[0047] REFERENCE SIGNS:

[0048] 1. Robotic arm drive motor; 2. Pad; 3. Infrared calibration camera; 10. Memory; 20. Processor. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] For the convenience of understanding the present invention, the following will describe the present invention more comprehensively and meticulously in combination with the accompanying drawings of the specification and preferred embodiments, but the protection scope of the present invention is not limited to the following specific embodiments.

[0050] The following will detail the embodiments of the present invention with reference to the accompanying drawings, but the present invention can be implemented in many different ways defined and covered by the claims.

[0051] Embodiment 1

[0052] The three - dimensional positioning control method for the full - view ultra - micro chip pads provided in Embodiment 1 of the present invention includes:

[0053] Step S101, establishing a first trajectory iteration model for the robotic arm to grasp the pads, and the first trajectory iteration model is specifically:

[0054]

[0055] Δq k (t + 1) = q k (t + 1) - q k-1 (t + 1),

[0056] Δu k (t) = u k (t) - u k-1 (t),

[0057] Among them, Δq k (t + 1) represents the difference between the output pad position trajectory at the (k)-th iteration at time t + 1 and the output pad position trajectory at the (k - 1)-th iteration, and Δu k (t) represents the difference between the system input at the (k)-th iteration at time t and the system input at the (k - 1)-th iteration, q k (t + 1) and q k-1 (t + 1) respectively represent the output pad position trajectories at the k-th and (k - 1)-th iterations at time t + 1, represents the estimated value of the pseudo quasi-partial derivative at the k-th iteration, u k (t) and u k-1 (t) respectively represent the system inputs at the k-th and (k - 1)-th iterations at time t;

[0058] Step S102, construct a pseudo quasi-partial derivative optimization function, and optimize the pseudo quasi-partial derivative according to the pseudo quasi-partial derivative optimization function;

[0059] Step S103, construct a system input optimization function, and optimize the system input according to the system input optimization function;

[0060] Step S104, based on the manipulator state increment model, use the predictive compensation algorithm to obtain the system time-domain compensation control component of the manipulator;

[0061] Step S105, according to the optimized pseudo quasi-partial derivative and system input, and the system time-domain compensation control component of the manipulator, establish an iterative prediction trajectory model;

[0062] Step S106, based on the iterative prediction trajectory model, obtain the optimal pad position trajectory for the manipulator to grasp the pad.

[0063] The three-dimensional positioning control method for the all-viewpoint ultra-micro chip cushion block provided by the embodiment of the present invention solves the technical problem of low three-dimensional positioning accuracy of the existing ultra-micro chip cushion block. By establishing a first trajectory iteration model for the robotic arm to grasp the cushion block, constructing a pseudo-partial derivative optimization function, optimizing the pseudo-partial derivative according to the pseudo-partial derivative optimization function, constructing a system input optimization function, and optimizing the system input according to the system input optimization function. Based on the robotic arm state increment model, a predictive compensation algorithm is used to obtain the system time-domain compensation control component of the robotic arm. According to the optimized pseudo-partial derivative and system input, as well as the system time-domain compensation control component of the robotic arm, an iterative predictive trajectory model is established, and based on the iterative predictive trajectory model, the optimal cushion block position trajectory for the robotic arm to grasp the cushion block is obtained. The iterative predictive trajectory model established by the optimized pseudo-partial derivative, system input, and predicted system time-domain compensation control component can obtain the optimal cushion block position trajectory for the robotic arm to grasp the cushion block, greatly improving the three-dimensional positioning accuracy of the ultra-micro chip cushion block.

[0064] Embodiment 2

[0065] The embodiment of the present invention provides a three-dimensional positioning control method for eutectic patch substrate cushion blocks with variable forgetting factor non-parametric adaptive iterative learning and predictive compensation, which solves the problems of high difficulty in the cushion block displacement control process and uncertainty of the desired trajectory function and system structure parameters due to the small size of the cushion block and high control precision requirements, and realizes real-time controllability of the cushion block position.

[0066] Referring to Figure 1 , the algorithm flowchart of the three-dimensional positioning control method for the all-viewpoint ultra-micro chip cushion block proposed by the embodiment of the present invention is as follows:

[0067] First, set the input control quantity of the k-th iteration in the control system as u k (t), the system output cushion block position trajectory is q k (t), the system desired output cushion block position trajectory is q d (t), so the k-th iteration error can be defined as e k (t) = q d (t) - q k (t). Since the system satisfies generalized Lipschitz continuity, there exists an estimated value of the pseudo-partial derivative such that:

[0068]

[0069] In the formula, Δq k (t + 1) = q k (t + 1) - q k-1 (t + 1), Δu k (t) = u k (t) - u k-1(t) is the difference between two consecutive iteration processes at the same moment. Since this iteration model does not require any structural model parameter information of the system, it only estimates the pseudo partial derivative based on the system input and output This makes the algorithm iteration change approximate the system gradient change, and finally makes the actual output of the control system approximate the expected output trajectory, realizing the complete tracking in the system iteration domain. To make the system estimate the pseudo partial derivative gradually approach the actual value ψ k (t), so the setting of the optimization function always hopes that the time-domain input-output error and the iteration-domain partial derivative error reach the minimum value, so that the estimated pseudo partial derivative can describe the input-output relationship of the control system to the greatest extent. Therefore, based on this design idea and according to the optimization index, ψ k (t) The optimization function is:

[0070]

[0071] Taking the partial derivative stationary point of the optimization function, the optimal pseudo partial derivative is obtained as:

[0072]

[0073] This model-free control algorithm successfully overcomes the problem in the traditional iterative learning control idea that requires the system initial conditions and the expected output to remain unchanged during the iteration process. That is, by estimating the optimal pseudo partial derivative, the system structure is simulated with minimum error, and the fitting effect and application value are prominent.

[0074] To further improve the control performance of the system and make it have strong convergence tracking ability and anti-system random noise interference ability in dynamic regulation and steady-state maintenance respectively during the system control process. The present invention hopes that the optimal output quantity can always minimize the output error and the iteration-domain error. Therefore, a non-parametric adaptive iterative learning controller system input optimization function with a variable forgetting factor is designed as:

[0075]

[0076] where γ is the system influence factor, is the system forgetting factor. The design of the optimization function aims to adjust the system forgetting factor, change the dependence of the system output on the control quantity at the corresponding moment of the previous iteration process, so that the system not only has fast convergence and good fitting ability in dynamic regulation, but also has good anti-interference ability in the steady-state process, meeting the design requirements. The change rule of the forgetting factor is as follows:

[0077]

[0078] Where m is the system adjustment control rate and ω is the system initial balance factor. With such a design, when the system is in the dynamic adjustment stage, the forgetting factor is small, and the system has better tracking ability and convergence effect for the desired trajectory, but sacrifices part of the anti-noise ability. After that, as the number of iterations increases, the system gradually converges to the vicinity of the target curve. At this time, the forgetting factor is large, which increases the dependence of the system output on the control quantity at the previous iteration moment, improves the anti-random noise interference ability of the system under steady state, and meets the steady state requirements of the system in the iteration domain and time domain. The fitting effect is significantly better than that of the traditional iterative learning algorithm. After that, the partial derivative stationary point of the system optimization function is obtained, and the optimal control quantity of the preliminary algorithm is:

[0079]

[0080] Thus, the non-parametric adaptive iterative learning model of the forgetting factor is established, and the perfect tracking of the pad target trajectory in the iteration domain by the system is realized. To further optimize the fitting performance of the system in the time domain, the present invention proposes to add a prediction compensation algorithm to the above model. Based on the manipulator state increment model, the future manipulator output position recurrence function can be obtained:

[0081]

[0082] Writing the manipulator position output quantity and the control quantity in matrix form, the position output function can be determined as:

[0083] Q = F c x c + φ c δU (8)

[0084] F c and φ c are the first coefficient increment matrix and the second coefficient increment matrix respectively. Based on the predicted position output equation, the optimization function can be defined:

[0085]

[0086] Where V and R are the first weight matrix and the second weight matrix respectively, and Q is the system output matrix. Under the unconstrained condition, according to the optimal condition judgment of the prediction compensation algorithm, the system time domain compensation control component can be calculated as:

[0087] δU * =(φ c T Vφ c + R) -1 φ c T V(Q - F c u k (t)) (10)

[0088] In summary, by combining the predicted compensation component to establish the system control equation, the non-parametric adaptive iterative learning and predictive compensation optimal algorithm model with variable forgetting factor for the system is obtained as follows:

[0089]

[0090] where p is the predicted compensation adjustment weight, and by adjusting the fixed weight factors α and γ, the model is made to conform to the actual displacement process of the robotic arm. The control algorithm estimates the pseudo-partial derivative of the system through the input and output variables of the system so as to repeatedly iterate and update the system control quantity. In this way, the system control rate simultaneously includes the iterative learning experience and the predicted compensation error, successfully realizing the error unification in the iterative domain and the time domain of the algorithm, with good stability and convergence, and overcoming the dependence of the traditional control algorithm on the model structure parameters and the constancy of the control trajectory, significantly improving the position tracking performance of the model.

[0091] The overall flowchart of the non-parametric adaptive iterative learning and predictive compensation control method with variable forgetting factor proposed in the embodiment of the present invention is as shown in Figure 2 and the specific steps are detailed as follows:

[0092] 1. Determine the estimated value of the pseudo-partial derivative of the algorithm based on the target trajectory of the spacer block, the initial input, and the model parameters.

[0093] 2. Obtain the forgetting factor based on the system error and substitute it into the non-parametric iterative learning rate equation of the system with variable forgetting factor to obtain the optimal control quantity sequence for the next iteration process of the model.

[0094] 3. Predict the robotic arm position output for the next l steps based on the prediction model, the current state quantity and position function of the robotic arm, to obtain the prediction sequence {q(t + 1|t), q(t + 2|t), …, q(t + l|t)}.

[0095] 4. Solve the minimization of the tracking error under the constraints of the robotic arm system to obtain the control quantity u of the robotic arm motor at the next moment * , and obtain the optimal predicted compensation sequence.

[0096] 5. Superimpose the control quantities of the two algorithms, input the obtained motor control quantity into the control port of the robotic arm drive motor, measure the state quantity of the robotic arm motor and the position output function at the next moment, and shift the current moment backward by one unit, and repeat steps 1 to 5.

[0097] The purpose of the embodiments of the present invention is to design a measurement method that is accurate, feasible, and reliable for three-dimensional positioning of the substrate pads of a multi-viewpoint eutectic mounter based on infrared three-dimensional imaging technology. The purpose of the embodiments of the present invention is to design a variable forgetting factor non-parametric adaptive iterative learning and predictive compensation control algorithm based on a state space model to provide an accurate and efficient control method for the substrate pad displacement control process.

[0098] The three-dimensional positioning control method for the full-view ultra-micro chip pads provided by the embodiments of the present invention solves the technical problem of low three-dimensional positioning accuracy of existing ultra-micro chip pads. By establishing a first trajectory iteration model for the manipulator to grasp the pads, constructing a pseudo-partial derivative optimization function, optimizing the pseudo-partial derivative according to the pseudo-partial derivative optimization function, constructing a system input optimization function, and optimizing the system input according to the system input optimization function. Based on the manipulator state increment model, a predictive compensation algorithm is used to obtain the system time-domain compensation control component of the manipulator. According to the optimized pseudo-partial derivative and system input, as well as the system time-domain compensation control component of the manipulator, an iterative prediction trajectory model is established, and based on the iterative prediction trajectory model, the optimal pad position trajectory for the manipulator to grasp the pads is obtained. The iterative prediction trajectory model established by the optimized pseudo-partial derivative, system input, and predicted system time-domain compensation control component can obtain the optimal pad position trajectory for the manipulator to grasp the pads, greatly improving the three-dimensional positioning accuracy of the ultra-micro chip pads.

[0099] Specifically, based on the determination of the target position information of the substrate pads of the eutectic mounter, the target position coordinate information is input into the variable forgetting factor non-parametric adaptive iterative learning control model, and the forgetting factor is continuously adjusted according to the control process, so that the control model has a good fitting effect and a fast convergence speed in the dynamic stage, and has strong anti-system random noise interference performance in the steady state stage. And continuously update the system pseudo-partial derivative in the iterative learning, overcoming the over-reliance of traditional control algorithms on system structure parameters and the condition of constant target trajectory, and the control effect is prominent. The closed-loop control process of the control motor is realized, and it has strong anti-interference ability.

[0100] In addition, based on the determination of the target position information of the substrate pads of the eutectic mounter, the predictive compensation control algorithm is introduced into the variable forgetting factor non-parametric adaptive iterative learning model. According to the predicted control quantity at the future moment, the control quantity compensation value of the next time series is continuously optimized, and real-time high-precision prediction compensation is performed on the single-system random noise, significantly improving the anti-real-time random interference ability of the algorithm model in a single control process, realizing the error unification of the model in the iterative domain and the time domain, and thus significantly improving the model position tracking performance.

[0101] Embodiment III

[0102] Combined with specific control methods such asFigure 3 As shown in the hardware example diagram Figure 4 The following further describes Embodiment 3 of the present invention. The three-dimensional positioning control method for the all-viewpoint ultra-micro chip pad of the embodiments of the present invention includes:

[0103] Step S301: Establish a spatial three-dimensional positioning system inside the eutectic stage, and use the eutectic bonding three-dimensional all-viewpoint infrared camera to continuously obtain the three-dimensional height position of the pad 2 to be moved and store it in the feedback loop and the algorithm iterative learning law.

[0104] Step S302: Calculate the height compensation of the chip on the eutectic stage to obtain the optimal matching strategy for pads of each specification and the target position coordinates of each pad, and set a position infrared calibration camera 3 at the bottom to continuously obtain error calibration information.

[0105] Step S303: Construct the optimal path planning for the spatial movement of the pad based on the historical control process, and input it as the objective function to the input end of the variable forgetting factor non-parametric adaptive iterative learning and predictive compensation control system.

[0106] Step S304: Set a robotic arm control system, which randomly changes with the requirements of the target position change and mechanical disturbances during the displacement control process of each batch.

[0107] Step S305: Use non-parametric adaptive iterative learning to mathematically describe the changes in the operation process and disturbance amounts of each batch during the robotic arm control process, and perform high-precision compensation for the control process and random errors of each batch based on the time-domain prediction model mechanism.

[0108] Step S306: Based on the non-parametric adaptive iterative control rate and the predictive compensation optimization function, use the learning weights of each historical batch of the robotic arm iterative learning predictive compensation control model for each historical data as optimization variables, and use the alternating optimization method to realize the optimization and upgrade of each learning control structure.

[0109] Step S307: Based on the compensated historical data, construct a two-dimensional prediction model for the robotic arm displacement control system. Set the optimization objective function of the variable forgetting factor non-parametric adaptive iterative learning model predictive compensation control as the quadratic form of the difference between the system control trajectory and the pad target movement position function trajectory within the remaining operation interval. Continuously optimize the objective function in the iterative learning rate, and input the optimized control result to the control circuit of the robotic arm drive motor 1 for picking up the pad until the pad position is within the given error interval.

[0110] The three-dimensional positioning control method for the full-viewpoint ultra-micro chip cushion block provided by the embodiment of the present invention solves the technical problem of low three-dimensional positioning accuracy of the existing ultra-micro chip cushion block. By establishing a first trajectory iteration model for the robotic arm to grasp the cushion block, constructing a pseudo partial derivative optimization function, optimizing the pseudo partial derivative according to the pseudo partial derivative optimization function, constructing a system input optimization function, and optimizing the system input according to the system input optimization function. Based on the robotic arm state increment model, a predictive compensation algorithm is used to obtain the system time-domain compensation control component of the robotic arm. According to the optimized pseudo partial derivative and system input, as well as the system time-domain compensation control component of the robotic arm, an iterative predictive trajectory model is established, and based on the iterative predictive trajectory model, the optimal cushion block position trajectory for the robotic arm to grasp the cushion block is obtained. The iterative predictive trajectory model established by the optimized pseudo partial derivative, system input, and predicted system time-domain compensation control component can obtain the optimal cushion block position trajectory for the robotic arm to grasp the cushion block, greatly improving the three-dimensional positioning accuracy of the ultra-micro chip cushion block.

[0111] Referring to Figure 5 , the three-dimensional positioning and control system for the full-viewpoint ultra-micro chip cushion block proposed by the embodiment of the present invention includes a memory 10, a processor 20, and a computer program stored on the memory 10 and executable on the processor 20. When the processor 20 executes the computer program, the steps of the three-dimensional positioning control method for the full-viewpoint ultra-micro chip cushion block proposed in this embodiment are implemented.

[0112] For the specific working process and working principle of the three-dimensional positioning and control system for the full-viewpoint ultra-micro chip cushion block in this embodiment, reference can be made to the working process and working principle of the three-dimensional positioning control method for the full-viewpoint ultra-micro chip cushion block in this embodiment.

[0113] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A three-dimensional positioning control method for a full-viewpoint ultra-micro chip spacer, characterized in that The method includes: Establishing a first trajectory iteration model for the robotic arm to grasp the cushion block, and specifically, the first trajectory iteration model is: Δq k q(t + 1) = q k q(t + 1) - q k-1 q(t + 1), Δu k (t) = u k (t) - u k-1 (t), where, Δq k (t + 1) represents the difference between the output pad position trajectories of the k-th iteration at time t + 1 and the (k - 1)-th iteration, Δu k (t) represents the difference between the system inputs of the k-th iteration and the (k - 1)-th iteration at time t, q k (t + 1) and q k-1 (t + 1) represent the output pad position trajectories of the k-th and (k - 1)-th iterations at time t + 1 respectively, represents the pseudo-pseudo partial derivative estimation value of the k-th iteration, u k (t) and u k-1 (t) represent the system inputs of the k-th and (k - 1)-th iterations at time t respectively; Constructing a pseudo-quasi partial derivative optimization function and optimizing the pseudo-quasi partial derivative according to the pseudo-quasi partial derivative optimization function; Constructing a system input optimization function and optimizing the system input according to the system input optimization function; Based on the robotic arm state increment model, obtaining the system time-domain compensation control component of the robotic arm by using the predictive compensation algorithm, wherein obtaining the system time-domain compensation control component of the robotic arm by using the predictive compensation algorithm based on the robotic arm state increment model includes: Predicting the robotic arm position output for a future preset number of steps based on the predictive compensation algorithm, the current state quantity of the robotic arm, and the position function to obtain a prediction sequence; Solving for the minimization of the tracking error under the robotic arm system constraint conditions to obtain the robotic arm motor control quantity at the next moment, and obtaining the optimal predictive compensation control component; Based on the optimized pseudo-quasi partial derivative and system input, and the system time-domain compensation control component of the robotic arm, establishing an iterative predictive trajectory model, wherein the calculation formula for the system time-domain compensation control component is: δU * =(φ c T Vφ c +R) -1 φ c T V(Q - F c u k (t)) Among them, U * predicts the optimal control quantity, δU * represents the optimal predictive compensation control quantity, φ c and F c are the first coefficient increment matrix and the second coefficient increment matrix respectively, R and V represent the first weight matrix and the second weight matrix respectively, φ c T represents the transpose of φ c and the specific formula for establishing the iterative predictive trajectory model according to the optimized pseudo quasi-partial derivative, system input, and the system time-domain compensation control component of the robotic arm is: where e k-1 (t + 1) represents the iteration error of the (k - 1)-th iteration at the (t + 1)-th moment, and p represents the prediction compensation adjustment weight; Based on the iterative predictive trajectory model, obtaining the optimal cushion block position trajectory for the robotic arm to grasp the cushion block.

2. The three-dimensional positioning control method of the all-viewpoint ultra-micro chip pad according to claim 1, characterized in that The calculation formula for the pseudo-quasi partial derivative optimization function is: where, Δq k-1 (t + 1) represents the difference between the output pad position trajectory at the (k - 1)-th iteration at time t + 1 and the output pad position trajectory at the (k - 2)-th iteration, and Δu k-1 (t) represents the difference between the system input at the (k - 1)-th iteration at time t and the system input at the (k - 2)-th iteration, and α represents a weighting factor, represents the estimated value of the pseudo-pseudo partial derivative at the (k - 1)-th iteration.

3. The three-dimensional positioning control method of the full-viewpoint ultra-micro chip spacer according to claim 1 or 2, characterized in that, The calculation formula for the system input optimization function is: Among them, e k (t + 1) represents the iteration error at the (k)-th iteration at time (t + 1), γ represents the system influence factor, u k (t) and u k-1 (t) represent the system inputs at time (t) for the k-th and (k - 1)-th iterations respectively, represents the system forgetting factor, ω represents the initial balance factor, and m represents the system adjustment control rate.

4. A three-dimensional positioning and control system for a full-viewpoint ultra-micro chip cushion block, the system comprising: A memory (10), a processor (20), and a computer program stored on the memory (10) and executable on the processor (20), wherein when the processor (20) executes the computer program, the steps of the method according to any one of claims 1 to 3 above are implemented.

Images