Stable lifting device for AGV under different loads and control method
By using scissor lifting mechanism and cyclic neural network control method on the AGV car, the acceleration is predicted and reduced, and the problem of unstable AGV car under different loads is solved, and the smooth lifting and lowering is achieved, which enhances the stability and work of the cargo. efficiency.
Patent Information
- Application Number
- CN202311387694.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-25
- Publication Date
- 2024-12-17
AI Technical Summary
The existing AGV trolleys are unstable during the lifting process, especially under different load conditions, which can easily lead to overturning and affect work efficiency.
A control method including a scissor lifting mechanism and a cyclic neural network is adopted to measure the acceleration curve, use the LSTM model to predict the maximum acceleration time, control the motor to slow down in advance, reduce the acceleration during contact, and achieve stable lifting and falling.
It effectively reduces the sudden acceleration during contact, enhances the stability of the cargo, and improves the stable lifting ability of AGV trolleys under different loads.
Smart Images
Figure CN119143047A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a lifting control method, more specifically, to an AGV lifting control method, and more specifically, to a control optimization method for the smooth lifting of an AGV under different loads. Background Art
[0002] AGV is a robot that can independently drive and carry objects. It is usually composed of wheels, motors, sensors and computer control systems. At present, AGV technology is developing rapidly and has a wide range of applications. It can play a huge role in automated logistics, intelligent warehousing, electronic manufacturing and other fields.
[0003] However, most of the existing AGVs are used for express delivery. They are small in size and suitable for carrying small goods. These AGVs only need to transfer the received goods to the next designated point. The AGV does not need to lift or lower the goods during the whole process. Most of the trolleys that can lift goods are similar to forklifts used in industry, but forklifts are too large and are not suitable for working in small spaces.
[0004] When lifting goods, the goods may not be on the AGV at the beginning, and the center of gravity of the goods may not be at the center of the AGV. Therefore, during the lifting and lowering process, if only the ordinary given speed or position mode is followed, the stability of the AGV will be greatly affected. The stability problem will cause the AGV and the goods above to overturn, which will reduce the working efficiency of the AGV. Maintaining the stability of the AGV mainly depends on the mechanical structure of the AGV itself. At the same time, the lifting and lowering speed of the AGV also needs to be controlled when there is a load.
[0005] In view of the above problems, the present invention proposes a control optimization method for the smooth lifting of an AGV under different loads. Summary of the invention
[0006] The purpose of the present invention is to solve the problem of instability of the existing AGV during the lifting process, and to propose a lifting algorithm to solve the problem of instability of the AGV during the lifting process, so that the AGV can obtain the optimal and most stable lifting speed under different load conditions.
[0007] The present invention achieves the above-mentioned purpose through the following technical solutions: A device and control method for smoothly lifting an AGV trolley under different loads is used, which adopts a device for smoothly lifting an AGV trolley under different loads, including an upper cover plate, a push block, two connecting rods, a base, and a scissor-type lifting mechanism; the scissor-type lifting mechanism is composed of a first movable bracket and a second movable bracket, the first movable bracket and the second movable bracket are hinged in the middle, the bottom of the first movable bracket is fixedly hinged on the chassis, and the top is in sliding contact with the upper cover plate; the bottom of the second movable bracket is in sliding contact with the base, the top is fixedly hinged on the upper cover plate, and the hinged sliding The block can move freely when opening and closing; the scissor-type lifting mechanism is connected with two connecting rods, and the two connecting rods are driven to slide by the push block. The two connecting rods are composed of a two-link front rod and a two-link rear rod, and the two-link front rod and the two-link rear rod are in curved contact; during the movement of the two connecting rods, the two-link front rod and the two-link rear rod will first rotate relative to each other. After the two-link front rod and the two-link rear rod are in curved contact, the two connecting rods are equivalent to a whole, driving the scissor-type lifting mechanism to move and thereby lifting the upper cover; the push block 2 is installed on the ball screw, and the ball screw is driven by a motor.
[0008] The specific steps include: 1) Measure the acceleration curve of the upper cover perpendicular to the ground, and pass the acceleration curve into the recurrent neural network (RNN) for training; obtain the trained LSMT model, the output gate of which is a time interval tmax1-tmax2 of the maximum acceleration value, and take the average value tmax; 2) Calculate the displacement S of the lead screw driven by the motor and the number of revolutions Q of the motor:
[0009] Ph is the lead of the screw; 3) The acceleration a is used as the raw data and normalized by z-score. Z-score normalization is to linearly transform the raw data so that the mean of the transformed data is 0 and the standard deviation is 1. 4) Divide the standardized data into multiple time windows in chronological order, and convert the data in the time windows into sequence data; the time window length is 0.1 seconds; 5) Construct the input and output of the recurrent neural network based on the sequence data; the lag observation method can be used, that is, part of the data in the time window is used as input, the time window is moved forward by one unit, and the part of the data after the shift is used as output; 6) Put the sequence data obtained in (4) onto the trained LSTM model to obtain the predicted value of each time window, expressed as the predicted acceleration a_{t}; 7) Find the maximum value of the predicted acceleration a_{t} and its location t_{max}; move the time corresponding to the maximum acceleration forward by 10 time windows (1 second) to obtain the time t_10 that is moved forward by 10 time windows; 8) According to the sample frequency, convert t_10 into the corresponding time, that is, obtain the time t10 corresponding to the first 10 time windows of the maximum acceleration value; 9) Calculate the number of revolutions Qt10 that the motor has made from 0 to t_10 based on the time t_10 obtained in step 8), and then control the motor to rotate Qt10 revolutions before decelerating, reducing the forward speed of the push block to half of the original speed. This will automatically control the motor to reduce the speed before the double-link contact, effectively reducing the acceleration caused by contact.
[0010] The working principle of the present invention is: according to structural simulation, it can be known that the maximum value of the acceleration of the cover plate occurs when the two-link front rod and the two-link rear rod are in contact. In order to make the AGV car rise and fall smoothly under different loads, reducing the running speed of the push block at the moment of contact can effectively reduce the acceleration caused by the contact, thereby achieving the purpose of smooth lifting.
[0011] The beneficial effects of the present invention are: Compared with the original contact surface of the double link being a plane, changing the contact surface of the double link to a curved surface can alleviate the sudden change in acceleration caused by plane contact, which has a great benefit in reducing the acceleration during contact. The stability of acceleration will also enhance the stability of the upper cargo during the lifting process.
[0012] The acceleration data in the whole process has positive and negative swings. The acceleration curve is standardized by z-score and the original data is linearly transformed to avoid the model being affected by the data scale.
[0013] The recurrent neural network used in this scheme can record the time corresponding to the acceleration. Compared with directly recording the maximum acceleration, this scheme can calculate the time corresponding to the maximum acceleration when any weight is lifted within the maximum load.
[0014] This solution can reversely calculate the number of revolutions the motor has made by the calculated time, and control the motor to slow down in advance based on the number of revolutions the motor has made, so that the recurrent neural network in deep learning can be suitable for each motor and screw on the AGV vehicle.
[0015] This solution adopts a delayed observation method in the recurrent neural network, and infers the data that will appear later through the existing data. When the number of circles Q corresponding to the maximum acceleration is predicted, it responds 10 time windows in advance, that is, the speed of the motor driving the screw is reduced to 50% of the original speed.
[0016] High data accuracy: In the actual acceleration curve, the absolute value of acceleration is very large. Directly inputting it into the recurrent neural network as input data will make the data scale of the model very large, causing gradient diffusion or gradient explosion, affecting the training and learning effect of the model. Perform z-score normalization before data input to convert the data into a distribution with a mean of 0 and a standard deviation of 1. This normalization method is very suitable for recurrent neural networks, which can keep the gradient within a reasonable range and avoid gradient explosion or gradient diffusion.
[0017] Strong applicability: In this solution, the recurrent neural network records the time range corresponding to the maximum acceleration and the range of the number of motor revolutions. Because the mass of the weight is inconsistent each time it is lifted, the corresponding maximum acceleration will also be inconsistent, while the time corresponding to the maximum acceleration during lifting and the number of motor revolutions are constant. Therefore, using time and number of revolutions as deep learning objects can achieve the prediction of the time when the maximum acceleration will arrive when lifting different goods.
[0018] High interchangeability: The recurrent neural network records the time range of maximum acceleration and the range of motor revolutions. When the motor or lead screw has problems and needs to be replaced, the AGV can continue testing to obtain a new time range and range of revolutions. When lifting goods in the future, the motor will be decelerated according to the new data. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 The utility model discloses a scissor-type lifting mechanism with two connecting rods.
[0020] Figure 2 This is the two-connecting rod structure of the present invention.
[0021] Figure 3 It is a flow chart of the present invention.
[0022] In the figure, 1-upper cover, 2-push block, 3-two-link front rod, 4-two-link rear rod, 5-base Implementation
[0023] The present invention will be further described below in conjunction with the accompanying drawings: like Figures 1 to 3As shown, a scissor-type lifting mechanism with two connecting rods includes an upper cover plate 1, a push block 2, a two-connecting rod front rod 3, a two-connecting rod rear rod, a base, and a scissor-type lifting mechanism; the scissor-type lifting mechanism is composed of a first movable bracket and a second movable bracket, the first movable bracket and the second movable bracket are hinged in the middle, the bottom of the first movable bracket is fixedly hinged on the chassis, and the top is in sliding contact with the upper cover plate 1; the bottom of the second movable bracket is in sliding contact with the base, and the top is fixedly hinged on the upper cover plate 1, and the hinged slider can be opened and closed. Free movement; the scissor-type lifting mechanism is connected with two connecting rods, which are driven to slide by the push block 2. The two connecting rods are composed of a two-link front rod 3 and a two-link rear rod, and the two-link front rod 3 and the two-link rear rod are in curved contact; during the movement of the two connecting rods, the two-link front rod 3 and the two-link rear rod will first rotate relative to each other. After the two-link front rod 3 and the two-link rear rod are in curved contact, the two connecting rods are equivalent to a whole, driving the scissor-type lifting mechanism to move and thereby lifting the upper cover plate 1; the push block 2 is installed on the ball screw, and the ball screw is driven by a motor.
[0024] The stable control optimization method under different loads specifically includes the following steps: 1) Measure the acceleration curve of the upper cover perpendicular to the ground, and pass the acceleration curve into the recurrent neural network (RNN) for training; obtain the trained LSMT model, the output gate of which is a time interval tmax1-tmax2 of the maximum acceleration value, and take the average value tmax; 2) Calculate the displacement S of the lead screw driven by the motor and the number of revolutions Q of the motor:
[0025] Ph is the lead of the screw; 3) The acceleration a is used as the raw data and normalized by z-score. Z-score normalization is to linearly transform the raw data so that the mean of the transformed data is 0 and the standard deviation is 1. 4) Divide the standardized data into multiple time windows in chronological order, and convert the data in the time windows into sequence data; the time window length is 0.1 seconds; 5) Construct the input and output of the recurrent neural network based on the sequence data; the lag observation method can be used, that is, part of the data in the time window is used as input, the time window is moved forward by one unit, and the part of the data after the shift is used as output, as shown below Time t x(tw) x(t-w+1) ... x(t-1) x(t) -> y(t) At time t+1 x(t-w+1) x(t-w+2) ... x(t) x(t+1) -> y(t+1) ... ... ... ... ... At time t+n x(t-n+1) x(t-n+2) ... x(t-n+w-1) x(t-n+w) -> y(t+n) Among them, x(tw) represents the data in the first w time windows at time t, x(t-w+1) to x(t) represent the data in the time window at time t, and y(t) represents the data after the time window is moved forward by one position; 6) Put the sequence data obtained in (4) onto the trained LSTM model to obtain the predicted value of each time window, expressed as the predicted acceleration a_{t}; 7) Find the maximum value of the predicted acceleration a_{t} and its location t_{max}; move the time corresponding to the maximum acceleration forward by 10 time windows (1 second) to obtain the time t_10 that is moved forward by 10 time windows; 8) According to the sample frequency, convert t_10 into the corresponding time, that is, obtain the time t10 corresponding to the first 10 time windows of the maximum acceleration value; 9) Based on the time t10 obtained in step 8), calculate the number of revolutions Qt10 that the motor has made from 0 to t10, and control the motor to rotate Qt10 revolutions before decelerating, reducing the forward speed of the push block to half of the original speed. This automatically controls the motor to reduce the speed before the double-link contact, which can effectively reduce the acceleration caused by contact.
[0026] The above embodiments are only preferred embodiments of the present invention and are not limitations of the technical solutions of the present invention. Any technical solution that can be implemented on the basis of the above embodiments without creative work should be deemed to fall within the scope of protection of the patent of the present invention.
Claims
1. Used for smooth lifting of AGV trolleys under different loads. Features: The invention comprises an upper cover plate (1), a push block (2), two connecting rods, a base (5), and a scissor-type lifting mechanism; the upper part of the scissor-type lifting mechanism is connected to the upper cover plate, and the lower end is connected to the base; the two connecting rods are composed of a two-connecting rod front rod (3) and a two-connecting rod rear rod (4), and the two-connecting rod front rod (3) and the two-connecting rod rear rod (4) are in curved surface contact; the two-connecting rod front rod (3) is connected to the slider, and the two-connecting rod rear rod (4) is connected to the scissor-type lifting mechanism; during the movement of the two connecting rods, the two-connecting rod front rod (3) and the two-connecting rod rear rod (4) will first rotate relative to each other, and after the two-connecting rod front rod (3) and the two-connecting rod rear rod (4) are in curved surface contact, the two connecting rods are equivalent to a whole, driving the scissor-type lifting mechanism to move, thereby lifting the upper cover plate (1); the push block (2) is mounted on a ball screw, and the ball screw is driven by a motor; the ball screw and the motor are mounted on the base (5).
2. According to claim 1, the device for smoothly lifting and lowering the AGV under different loads, Features: The scissor-type lifting mechanism is composed of a first movable bracket and a second movable bracket, wherein the first movable bracket and the second movable bracket are hinged in the middle, the bottom of the first movable bracket is fixedly hinged on the chassis, and the top is in sliding contact with the upper cover plate (1); the bottom of the second movable bracket is in sliding contact with the base (5), and the top is fixedly hinged on the upper cover plate (1), and the hinged slider can move freely when opening and closing.
3. A control optimization method for the smooth lifting of an AGV under different loads using the smooth lifting device as claimed in claim 1, Features: 1) measuring the acceleration curve of the upper cover plate (1) in a direction perpendicular to the ground, and passing the acceleration curve into a recurrent neural network (RNN) for training; obtaining a trained LSMT model, the output gate of which is a time interval tmax1-tmax2 of the maximum acceleration value, and taking an average value tmax; 2) Calculate the displacement S of the lead screw driven by the motor and the number of revolutions Q of the motor: ； Ph is the lead of the screw; 3) The acceleration a is used as the raw data and normalized by z-score. Z-score normalization is to linearly transform the raw data so that the mean of the transformed data is 0 and the standard deviation is 1. 4) Divide the standardized data into multiple time windows in chronological order, and convert the data in the time windows into sequence data; 5) Construct the input and output of the recurrent neural network based on the sequence data; adopt the lag observation method, that is, take part of the data in the time window as input, move the time window forward by one unit, and take part of the data after the shift as output; 6) Put the sequence data obtained in (4) onto the trained LSTM model to obtain the predicted value of each time window, expressed as the predicted acceleration a_{t}; 7) Find the maximum value of the predicted acceleration a_{t} and its location t_{max}; move the time corresponding to the maximum acceleration value forward by n time windows to obtain the time t_10 that is moved forward by n time windows; 8) According to the sample frequency, convert t_10 into the corresponding time, that is, obtain the time t10 corresponding to the first n time windows of the maximum acceleration; 9) Based on the time t10 obtained in step 8), calculate the number of revolutions Qt10 that the motor has made from 0 to t10, and control the motor to rotate Qt10 revolutions before decelerating. This will automatically control the motor to reduce its speed before the double-link contact, which can effectively reduce the acceleration caused by contact.
4. According to claim 3, a control optimization method for the smooth lifting of an AGV under different loads, Features: The time window length in step 4) is 0.1 second.
5. According to claim 3, a control optimization method for the smooth lifting of an AGV under different loads, Features: When n time windows are moved forward in step 7), n=10.
6. According to claim 3, a control optimization method for the smooth lifting of an AGV under different loads, Features: The step 9) deceleration refers to reducing the forward speed of the push block (2) to half of the original speed.