Robot dynamic speed optimization method with sucker tool as end effector
By combining genetic algorithms, particle swarm optimization and deep neural networks to optimize robot speed, the problem of low robot efficiency under fixed speed control is solved, and efficient robot transportation in multi-variety small batch production is achieved.
Patent Information
- Application Number
- CN202510798926.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-16
- Publication Date
- 2025-09-26
AI Technical Summary
In the existing technology, industrial robots use a fixed speed control mode, which results in low working efficiency for robots with suction cup tools as end effectors in multi-variety, small-batch, and flexible production, and they are difficult to adapt to workpieces of different weights, materials, or process requirements.
A hybrid genetic algorithm and particle swarm algorithm are combined with deep neural networks and digital twin models to dynamically optimize the speed at which robots transport objects. By calculating fitness and constraints, the optimal speed set is generated to prevent objects from falling.
It improves the efficiency of robots in transporting objects, enhances the adaptability of speed and objects, reduces the number of iterations, and improves the overall efficiency and safety of the equipment.
Smart Images
Figure CN120705702A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a robot speed method, in particular to a robot dynamic speed optimization method using a suction cup tool as an end effector. Background Art
[0002] In the field of industrial automation, industrial robots serve as core execution units, and their speed control accuracy directly determines production efficiency, product quality, and equipment reliability. Currently, mainstream industrial robots generally use a fixed speed control mode, where a single speed parameter is preset during the design or commissioning phase and this value is applied throughout subsequent tasks. This mode, with its advantages of simple control logic and high program stability, has long supported the development of industrial automation.
[0003] However, as modern manufacturing transitions toward high-variety, low-batch, and flexible production, the limitations of fixed speed control are becoming increasingly apparent. Robots using suction cups as end-effectors, which grasp and transport objects using the suction force of the cups, often rely on fixed speeds for handling workpieces of varying weights, materials, or process requirements. Current industrial robots often rely on on-site programming with teach pendants, resulting in rigid speed parameters for specific process actions, reducing the robot's efficiency. Summary of the Invention
[0004] The purpose of the present invention is to provide a robot dynamic speed optimization method using a suction cup tool as an end effector to solve the problem of low working efficiency caused by the robot operating at a fixed speed.
[0005] The purpose of the present invention is achieved like this:
[0006] A method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector comprises the following steps:
[0007] S1. Obtaining the operating parameters and vacuum pressure difference of the robot with the suction cup tool as the end effector;
[0008] S2. Under the constraints, the robot's operating parameters and vacuum pressure difference are used to calculate the robot's fitness for each speed within the preset speed range for transporting the object to be transported;
[0009] S3. Select a speed based on the fitness, use the selected speed as the parent individual, perform a genetic algorithm calculation operation, and obtain a first candidate speed set;
[0010] S4. Generate a second set of candidate speeds using the DNN, input the second set of candidate speeds into the robot's digital twin model for simulation, and eliminate speeds that do not conform to actual working conditions from the second set of candidate speeds to obtain a third set of candidate speeds;
[0011] S5. Use PSO to optimize the first candidate speed set and the third candidate speed set to obtain the updated speed;
[0012] S6. Loop through steps S3-S5 based on the data from the optimization process until the loop termination condition is met, and use the updated speed as the speed at which the robot transports the object to be transported.
[0013] Furthermore, the specific method of determining the adaptability of speed is:
[0014] S2-1. Calculate the friction force corresponding to the speed based on the robot's suction cup contact area and vacuum pressure difference.
[0015] S2-2. Calculate the tangential force corresponding to the speed based on the speed of the robot and the weight of the object to be transported;
[0016] S2-3. Calculate the centrifugal force corresponding to the speed based on the robot's arm length and speed;
[0017] S2-4. Calculate the mechanical constraints corresponding to the speed based on the friction, tangential, and centrifugal forces corresponding to the speed;
[0018] S2-5. Calculate the gas leakage constraint corresponding to the speed based on the pressure loss caused by gas leakage;
[0019] S2-6. Calculate the speed adaptability based on the mechanical constraints and gas leakage constraints corresponding to the speed.
[0020] Furthermore, the constraint condition is that the resultant force of the tangential force and the centrifugal force is less than or equal to the friction force.
[0021] Furthermore, the specific method of calculating the first candidate speed set using the genetic algorithm is as follows:
[0022] S3-1. Two parents use a simulated binary crossover operation to produce two offspring individuals;
[0023] S3-2. Use the mutation operation of differential evolution on the generated offspring individuals to generate a first candidate speed set.
[0024] Furthermore, the two offspring individuals generated are:
[0025] c1=0.5·[(1+β)·p1+(1-β)·p2]
[0026] c2=0.5·[(1-β)·p1+(1+β)·p2]
[0027] Among them, p1 and p2 are parent individuals, c1 and c2 are offspring individuals, and β is a random variable that controls the deviation of the crossover process.
[0028] Furthermore, the first candidate speed m in the first candidate speed set i for:
[0029] m i =c1+F·(c2-c3)
[0030] Where F is the scaling factor, c1, c2, and c3 are three randomly selected offspring individuals.
[0031] Furthermore, the specific method of generating the second candidate speed set is:
[0032] The vacuum pressure difference and the weight of the object to be transported are input into a DNN to obtain a second set of candidate velocities; the DNN is pre-trained.
[0033] Furthermore, the specific method of calculating the first candidate speed set and the third candidate speed set is:
[0034] S5-1. For each speed in the first candidate speed set and the third candidate speed set, take it as the initial position of a particle, calculate the fitness corresponding to the initial position, and use the initial position as the optimal position of the individual;
[0035] S5-2 traverses the individual optimal positions of all particles and takes the individual optimal position with the largest fitness as the global optimal position;
[0036] S5-3. Iteratively update the positions of all particles and calculate the fitness of the updated positions. Compare the fitness of the updated positions with the fitness of the optimal position. If the fitness of the updated particle position is greater than the fitness of the individual optimal position, the updated position is used as the individual optimal position.
[0037] S5-4. Traverse the individual optimal positions of all particles after iteration, and take the individual optimal position with the largest fitness as the global optimal position;
[0038] S5-6 repeats steps S5-3-S5-4 until the termination condition is reached, and outputs the global optimal position as the speed at which the robot transports the object to be transported.
[0039] The hybrid genetic algorithm of the present invention combines a genetic algorithm with a particle swarm optimization (PSO) to optimize the speed of robot transports. This algorithm combines the powerful global search capabilities of a genetic algorithm (GA) with the efficient local search and convergence capabilities of a particle swarm optimization (PSO). This algorithm effectively balances exploration and exploitation, avoids premature convergence, and increases the probability of finding the global optimal speed. The present invention optimizes speed by calculating fitness. Constraints and fitness ensure that the calculated transport speed is high and prevents the object from falling. The robot's transport speed varies according to the object's weight. A mathematical basis for determining suction reliability is established based on the friction, tangential force, and centrifugal force of the suction cup.
[0040] The training data used in the DNN of this invention is experimentally derived. Under the same conditions, a higher speed is selected. The DNN learns the characteristics of the input data, resulting in a higher adaptability of the speed output by the DNN. Simulation using a digital twin model filters out speeds that do not conform to mechanical laws, further improving the adaptability of the speed to the object being transported. DNNs can handle complex nonlinear problems and high-dimensional data. In robot speed optimization, based on multiple input parameters, the DNN can effectively extract correlation information between these parameter characteristics, thereby generating a more accurate second set of candidate speeds. The digital twin system provides strong support for the optimization model, helping to enhance the reliability, security, and processing efficiency of model optimization. It also helps address the discrepancy between conventional model optimization results and the actual physical system. The DNN and digital twin model are combined to assist the hybrid genetic algorithm, reducing the number of iterations and improving real-time performance to meet rapid response requirements in industrial sites. Based on the DNN and hybrid genetic algorithm model, the present invention outputs the robot's operating speed in real time under different vacuum pressure differences and different object weights, optimizing the overall efficiency of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION
[0042] The present invention will be further described below in conjunction with the accompanying drawings.
[0043] like Figure 1 As shown, the robot dynamic speed optimization method using a suction cup tool as an end effector of the present invention includes the following steps:
[0044] S1. Obtain the operating parameters and vacuum pressure difference of the robot with a suction cup tool as the end effector.
[0045] Among them, the working parameters include the friction coefficient, the length of the robotic arm, the contact area of the suction cup and the weight of the object to be transported.
[0046] The robot targeted by this invention uses a suction cup tool as its end effector and needs to transport objects. Therefore, the invention requires calculating the maximum speed at which the robot can transport objects without them falling. The robot's operating parameters can be directly retrieved from a database storing these parameters, using sensors to directly measure the vacuum pressure difference. Before the robot transports an object, it is weighed. For products of varying weights, the weight is determined using a weighing module within the production line.
[0047] S2. Under the constraints, use the robot's operating parameters and the vacuum pressure difference to calculate the fitness of the robot at each speed within a preset speed range for transporting the object to be transported.
[0048] First, according to historical data, determine the preset speed range [v min ,v max ], the present invention sets the minimum speed v min is 2m / s, the maximum speed v max 6m / s.
[0049] The constraints are:
[0050] F tangent (v)+F centrifugal (v)≤F friction (v)
[0051] Among them, F centrifugal (v) is the centrifugal force, F tangent (v) is the tangential force, F friction For friction.
[0052] Friction force F friction is the sliding friction between the suction cup and the object:
[0053] F friction (v) = μ·A·ΔP·g
[0054] Where μ is the friction coefficient, A is the contact area of the suction cup, ΔP is the vacuum pressure difference, and g is the acceleration due to gravity.
[0055] The vacuum pressure difference is obtained through a negative pressure gauge, which measures the vacuum pressure difference of the suction cup.
[0056] Tangential force F tangent (v) is:
[0057]
[0058] Centrifugal force F centrifugal (v) is:
[0059]
[0060] The centrifugal force F corresponding to the speed centrifugal (v) Friction force F friction and tangential force F tangent (v), calculate the mechanical constraint P of the velocity force :
[0061] P force =[max(0,(F tangent (v)+F centrifugal (v)) / (F friction (v))-1)] 2
[0062] Calculate the gas leakage constraint P of the velocity based on the vacuum pressure differenceleak :
[0063] P force =[max(0,(pressure loss due to leakage (v) / allowable pressure loss)-1)] 2
[0064] Leakage causes pressure loss = (P 设计值 -ΔP)·A
[0065]
[0066] Among them, P 设计值 is the preset value, The measured ΔP is in the interval middle.
[0067] It is only necessary to calculate the fitness of the speed under the constraints. According to the mechanical constraints and gas leakage constraints, the fitness of the speed (V) is calculated:
[0068] Fitness(v)=v-λ1·P force (v)-λ2·P leak (v)
[0069] Wherein, λ1 and λ2 are penalty coefficients, both greater than zero, and the value of the present invention is 100.
[0070] S3. Select a speed based on the fitness, use the selected speed as the parent individual, perform a calculation operation of the genetic algorithm, and obtain a first candidate speed set.
[0071] Use the tournament to select the speed of the parent individual. First, divide the speed into n groups. Select at least one speed with the highest fitness in each group as the parent individual. Cross the parent individuals to obtain the offspring individuals:
[0072] c1=0.5·[(1+β)·p1+(1-β)·p2]
[0073] c2=0.5·[(1-β)·p1+(1+β)·p2]
[0074] Among them, p1 and p2 are parent individuals, c1 and c2 are offspring individuals, and β is a random variable that controls the deviation of the crossover process.
[0075] The mutation operation uses the Differential Evolution (DE) strategy to perform mutation operations on the offspring after crossover to obtain the mutated speed m i :
[0076] m i =c1+F·(c2-c3)
[0077] Where F is the scaling factor, c1, c2, and c3 are three randomly selected offspring individuals.
[0078] Finally, the first candidate speed set is obtained.
[0079] The performance improvement of genetic algorithms is closely related to parameter settings: crossover probability P c The probability of the crossover operation of the parent individual is used to update the population. The present invention sets P C is 0.6; mutation probability P m In order to control the probability of individual random variation and maintain population diversity, and to avoid falling into the optimal solution to a certain extent, the present invention sets P m is 0.1; the population size N, the number of individuals in the population, is set in combination with the computational cost and convergence speed, and needs to be set in combination with the situation of key variables.
[0080] S4. Use DNN to generate a second candidate speed set, input the second candidate speed set into the digital twin model of the robot for simulation, eliminate the speeds that do not conform to the actual working conditions from the second candidate speed set, and obtain a third candidate speed set.
[0081] The DNN is pre-trained. Specifically, it obtains historical vacuum pressure differences, historical weights of objects to be transported, and corresponding historical speeds as a data set. The historical vacuum pressure differences are cleaned to remove outliers and the pre-processed data is normalized:
[0082]
[0083] Among them, X is the original eigenvalue, X min and X max are the minimum and maximum values in the training set, X normalized is the normalized eigenvalue.
[0084] The range of vacuum pressure difference is -140kPa to -90kPa, and historical vacuum pressure differences outside this range are abnormal values.
[0085] Each historical speed corresponds to a historical vacuum pressure difference and a historical weight of the object to be transported. The dataset obtained is based on previous robot experiments. One historical weight of an object to be transported corresponds to at least two historical vacuum pressure differences. First, determine the historical vacuum pressure difference corresponding to each historical weight of the object to be transported, as well as the speed corresponding to each historical weight and vacuum pressure difference. Given a fixed historical weight of the object to be transported and historical vacuum pressure difference, at least two speeds are corresponding. Under the same conditions, the top k speeds with the largest values are selected, where the value of k can be defined. The historical weight of the object to be transported, the normalized historical vacuum pressure, and the corresponding speeds are used as the training dataset.
[0086] 80% of the dataset is used as a training set and 20% is used as a test set.
[0087] The normalized value of the current vacuum pressure difference obtained in step S1 and the weight of the object to be transported are input into the DNN to obtain a second candidate speed set. The second candidate speed set is input into the digital twin model for simulation, and the speeds that will cause the object to fall are filtered out. The remaining speeds are used as the third candidate speed set.
[0088] The robot's physical parameters are collected to establish a digital twin model. The digital twin model and deep neural network (DNN) are both implemented in Visual Studio. The two interact with data through local method calls, mapping the robot's operating parameters and environmental parameters (i.e., vacuum pressure difference) to the digital twin model in real time.
[0089] S5. Use PSO to iteratively update the first candidate speed set and the third candidate speed set to generate new speeds.
[0090] S5-1. Take each speed in the first candidate speed set and the third candidate speed set as the initial position of a particle, update the initial position of the particle, calculate the fitness corresponding to the initial position, and use the initial position x_i(0) as the individual optimal position pbest_i.
[0091] Each speed in the first candidate speed set and the third candidate speed set is used as the initial position of a particle in the PSO algorithm, that is, one speed represents the initial position of a particle.
[0092] Calculate the fitness f(x_i(0)) of the particle's initial position using the same method as in step S2.
[0093] S5-2 traverses the individual optimal positions of all particles and takes the individual optimal position with the largest fitness as the global optimal position.
[0094] Each particle has an individual optimal position. The fitness of the individual optimal positions of all particles is compared, and the individual optimal position with the largest fitness is taken as the global optimal position.
[0095] S5-3. Iteratively update the positions of all particles and calculate the fitness of the updated positions. Compare the fitness of the updated positions with the fitness of the optimal position. If the fitness of the updated position of the particle is greater than the fitness of the individual optimal position, the updated position is used as the individual optimal position.
[0096] In each step of the evolution process, PSO is used to perform local optimization on the individuals with high fitness in the current population. Through local optimization, particle swarm optimization can further improve the fitness of individuals. The speed and position update formula of particle swarm optimization is:
[0097] v i (t+1)=w·v i (t)+d1·r1·(p best,i -x i (t))+d2·r2·(g best -x i (t))
[0098] x i (t+1)=x i (t)+v i (t+1)
[0099] Among them, w is the inertia weight, c1 and c2 are acceleration factors, is a random number, and p best,i is the individual optimal position of particle i, g best is the global optimal position.
[0100] At each time step t, the particle moves to a new position x_i(t), calculates the fitness of the new position f(x_i(t)), and compares the fitness of the new position with the fitness of the currently recorded individual optimal position f(pbest_i). If the fitness of the new position is greater than the fitness of the particle's individual optimal position, the new position is used as the particle's individual optimal position, that is, pbest_i = x_i(t); if the fitness of the new position is less than or equal to the fitness of the individual optimal position, the particle's individual optimal position remains unchanged.
[0101] S5-4. Traverse the individual optimal positions of all particles after iteration, and take the individual optimal position with the largest fitness as the global optimal position.
[0102] After all particles have iterated, the individual optimal positions of all particles in the entire population are traversed, and the individual optimal position with the largest fitness value among all pbest_i is found and set as the new global optimal position g best .
[0103] S5-6 repeats steps S5-3-S5-4 until the termination condition is reached.
[0104] When 50 generations are passed, the global optimal position fitness (g best )(t) and the global optimal position fitness of the last optimization (Fitness(g best The difference between )(t-1) and the global optimal position fitness of the last optimization (Fitness(g bestWhen the ratio of (|Fitness(g best )(t)-Fitness(g best )(t-1)|) / Fitness(g best )<0.001, or when the calculation time exceeds 1 second, the iteration is stopped.
[0105] S6. Execute steps S3-S5 according to the iteratively updated data until the termination condition is met, and use the iteratively updated speed as the speed at which the robot transports the object to be transported.
[0106] When for 50 consecutive generations, the difference between the fitness of the global optimal position after the nth loop iteration, Fitness(n-1) after the n-1th loop iteration, and the fitness of the global optimal position after the n-1th loop iteration, Fitness(n-1) after the n-1th loop iteration, is less than 0.001, that is, (|Fitness(n)-Fitness(n-1)|) / Fitness(n-1)<0.001, the iteration is stopped, or when the calculation time exceeds 1 second, the updated global optimal position is output as the speed of the robot, and the output data is the global optimal position obtained from the last iterative update as the speed of the robot transporting the object to be transported.
[0107] When the termination condition is not met, the speed obtained by the PSO optimization process continues to execute steps S3-S5, and the speed obtained by the PSO optimization process is screened using the genetic algorithm and fitness using tournament selection to obtain a first candidate speed set. The second candidate speed set is generated using DNN, and the second candidate speed set is input into the digital twin model of the robot for simulation. The second candidate speed set is screened to obtain a third candidate speed set; the first candidate speed set calculated by the genetic algorithm and the third candidate speed set generated by the digital twin model are iteratively updated using PSO to generate a new speed until the termination condition is met. The output data is the global optimal position obtained by the last iterative update as the speed for the robot to transport the object to be transported.
Claims
1. A method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector, characterized in that: The steps include: S1. Obtaining the operating parameters and vacuum pressure difference of the robot with the suction cup tool as the end effector; S2. Under the constraints, the robot's operating parameters and vacuum pressure difference are used to calculate the robot's fitness for each speed within the preset speed range for transporting the object to be transported; S3. Select a speed based on the fitness, use the selected speed as the parent individual, perform a genetic algorithm calculation operation, and obtain a first candidate speed set; S4. Generate a second set of candidate speeds using the DNN, input the second set of candidate speeds into the robot's digital twin model for simulation, and eliminate speeds that do not conform to actual working conditions from the second set of candidate speeds to obtain a third set of candidate speeds; S5. Use PSO to optimize the first candidate speed set and the third candidate speed set to obtain the updated speed; S6. Loop through steps S3-S5 based on the data from the optimization process until the loop termination condition is met, and use the updated speed as the speed at which the robot transports the object to be transported.
2. The method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector according to claim 1, wherein: The specific method of determining the fitness of speed is: S2-1. Calculate the friction force corresponding to the speed based on the robot's suction cup contact area and vacuum pressure difference. S2-2. Calculate the tangential force corresponding to the speed based on the speed of the robot and the weight of the object to be transported; S2-3. Calculate the centrifugal force corresponding to the speed based on the robot's arm length and speed; S2-4. Calculate the mechanical constraints corresponding to the speed based on the friction, tangential, and centrifugal forces corresponding to the speed; S2-5. Calculate the gas leakage constraint corresponding to the speed based on the pressure loss caused by gas leakage; S2-6. Calculate the speed adaptability based on the mechanical constraints and gas leakage constraints corresponding to the speed.
3. The method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector according to claim 2, wherein: The constraint condition of the law of mechanics is that the resultant force of tangential force and centrifugal force is less than or equal to friction force.
4. The method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector according to claim 3, wherein: The specific method of obtaining the first candidate speed set is: S3-1. Two parents use a simulated binary crossover operation to produce two offspring individuals; S3-2. Use the mutation operation of differential evolution on the generated offspring individuals to generate a first candidate speed set.
5. The method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector according to claim 3, wherein: The two offspring individuals generated are: c1=0.5·[(1+β)·p1+(1-β)·p2] c2=0.5·[(1-β)·p1+(1+β)·p2] Among them, p1 and p2 are parent individuals, c1 and c2 are offspring individuals, and β is a random variable that controls the deviation of the crossover process.
6. The method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector according to claim 3, wherein: The first candidate speed m in the first candidate speed set i for: m i =c1+F·(c2-c3) Where F is the scaling factor, c1, c2, and c3 are three randomly selected offspring individuals.
7. The method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector according to claim 1, wherein: The specific method of generating the second candidate speed set is: The vacuum pressure difference and the weight of the object to be transported are input into a DNN to obtain a second set of candidate velocities; the DNN is pre-trained.
8. The method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector according to claim 1, wherein: The specific method for calculating the first candidate speed set and the third candidate speed set is: S5-1. For each speed in the first candidate speed set and the third candidate speed set, take it as the initial position of a particle, calculate the fitness corresponding to the initial position, and use the initial position as the optimal position of the individual; S5-2 traverses the individual optimal positions of all particles and takes the individual optimal position with the largest fitness as the global optimal position; S5-3. Iteratively update the positions of all particles and calculate the fitness of the updated positions. Compare the fitness of the updated positions with the fitness of the optimal position. If the fitness of the updated particle position is greater than the fitness of the individual optimal position, the updated position is used as the individual optimal position. S5-4. Traverse the individual optimal positions of all particles after iteration, and take the individual optimal position with the largest fitness as the global optimal position; S5-6 repeats steps S5-3-S5-4 until the termination condition is reached, and outputs the global optimal position as the speed at which the robot transports the object to be transported.
9. The method for optimizing the dynamic speed of a robot using a suction cup tool as an end effector according to claim 1, wherein: The loop termination condition is: 50 consecutive generations meet the convergence condition or the calculation time exceeds 1 second; The convergence conditions are: (|Fitness(n)-Fitness(n-1)|) / Fitness(n-1)<0.001 Among them, Fitness(n) is the fitness of the speed obtained by the n-th cycle update, and Fitness(n-1) is the fitness of the speed obtained by the n-1-th cycle update.