A precision intelligent control method for 3D printing concrete based on BP neural network

Abstract

The application discloses a kind of precision intelligent control methods of 3D printing concrete based on BP neural network, comprising the following steps;Step 1):establish the relationship between concrete rheology and the motor speed of 3D printing concrete extrusion system;Step 2):according to the relationship of step 1, the model of 3D printing concrete extrusion system rheology control system is established;Step 3):for the system of step 2, design BP-PID controller;Step 4):the parameters of BP-PID controller are optimized and set using PSO algorithm.The application combines BP neural network with PID controller, uses the adaptive learning ability of BP neural network to adjust PID control parameters online in real time, improves system stability.Improves the uniformity of printing material in the process of discharging, reduces the surface defects such as crack of printing finished product, improves the quality of concrete forming.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of 3D printing, in particular to a precision intelligent control method for 3D printing concrete based on BP neural network. BACKGROUND

[0002] 3D printing concrete technology is a technology that uses computer layer modeling and program commands to control the printing nozzle to lay materials layer by layer to build a free building structure. Compared with traditional building technology, 3D printing concrete technology has the characteristics of intelligentization, short construction period, zero-pollution construction, printable complex structure nodes, and low material cost.

[0003] However, the current popularization of 3D printing concrete still faces many problems, one of the main reasons being that the rheological properties of concrete materials in the mixing state have very sensitive time-varying development characteristics, which leads to very strict requirements for the control of the forming precision after the material is extruded. However, it is often difficult to achieve real-time control and rheological property feedback regulation of the material in the actual construction process, which directly leads to uneven appearance of the printed structure, initial defects such as cracks and interlayer weak layers, and serious weakening of the overall mechanical properties of the printed structure.

[0004] One of the existing methods for controlling the rheological properties of 3D printing concrete (publication number: CN115042295A) only considers the influence of temperature on the rheological properties of concrete during the printing process. Since the sensitivity of concrete materials to temperature is not particularly high, this method has little effect on the rheological properties of the printed material, reducing the effective working time of the printer and not improving the working efficiency. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a precision intelligent control method for 3D printing concrete based on BP neural network, to realize real-time monitoring and adjustment of the rheological properties of concrete during the printing process. The present application combines BP neural network with PID controller, uses the adaptive learning ability of BP neural network to adjust PID control parameters online in real time, improves the stability of the system, and uses the improved PSO algorithm to optimize the BP-PID self-learning process to ensure that the BP-PID system converges to the global optimal solution, improves the uniformity of the printed material during the feeding process, reduces surface defects such as cracks of the printed product, and improves the quality of concrete forming.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is:

[0007] A precision intelligent control method for 3D printing concrete based on BP neural network, comprising the following steps:

[0008] Step 1): Establish the relationship between the rheological property of concrete and the motor speed of the 3D printing concrete extrusion system;

[0009] Step 2): Establish a rheological property control model for the 3D printing concrete extrusion system based on the relationship in Step 1;

[0010] Step 3): Design a BP-PID controller for the rheological property control model established in Step 2;

[0011] Step 4): Optimize and set the parameters of the BP-PID controller using the PSO algorithm.

[0012] The step 1) establishes the relationship between the motor speed of the extrusion system and the rheological property of concrete;

[0013] Set the same batching ratio of concrete, take the motor speed as the variable, design a control experiment, measure the rheological property of concrete at the current motor speed during printing, and use the MATLAB data fitter to analyze and fit the results, to get the relationship between rheological property and speed x as follows:

[0014]

[0015] The step 2) establishes a rheological property control model for the 3D printing concrete extrusion system:

[0016] (1) The rheological property of concrete during printing should be maintained between 180-190, which makes the surface of the printed product uniform, and the given concrete fluidity threshold is taken as the input rin(k) of the 3D printing concrete rheological property control system;

[0017] (2) The sensor measures the size of the concrete fluidity in real time, and feeds back the output results to the system, and compares with the input signal to get the deviation signal e(k);

[0018] e(k)=rin(k)-rout(k)

[0019] (3) The deviation signal is transmitted to the controller to adjust the motor speed, and the general DC motor can be regarded as an inertial element, so the transfer function of the motor in the system is set as Where K is the motor transfer coefficient, T is the motor time constant; s is the function with s as the independent variable obtained after Laplace transform in control;

[0020] (4) Perform Laplace transform on the relationship obtained in step 1), and take the size of concrete fluidity as the output rout(k) of the system based on this relationship, and the control transfer function G2 is as follows:

[0021]

[0022] The step 3) BP-PID controller is specifically:

[0023] The incremental PID algorithm is used to realize the closed-loop control of the rheological property of the concrete, so that the fluidity and the motor speed are kept stable, and the three parameters k p , k i , and k d of the PID controller are adjusted online by the BP neural network; the BP neural network adjusts the three parameters according to the running state of the system, thereby indirectly adjusting the controlled object; the BP neural network has a self-learning feature, and the three control parameters of the PID can be controlled in real time through weight adjustment; the formula of the incremental PID algorithm is:

[0024] ΔU(k) = K o [e(k) - e(k-1)] + K i e(k) + K d {[e(k) - e(k-1)] - [e(k-1) - e(k-2)]}

[0025] That is, ΔU(k) = K p Δe(k) + K i e(k) + K d [Δe(k) - Δe(k-1)]

[0026] Where ΔU(k) is the output change, K p is the proportional coefficient, K i is the integral coefficient, K d is the differential coefficient, e(k) is the error value of the target and the actual value this time, e(k-1) is the error value of the target and the actual value last time, and e(k-2) is the error value of the target and the actual value two times ago.

[0027] After the PID controller is determined, the structure framework of the BP neural network is determined and applied to the PID controller.

[0028] (1) Selection of the number of hidden layer nodes:

[0029] The number of hidden layer nodes is selected according to the formula

[0030] Where m is the number of hidden layer nodes, n is the number of input layer nodes, and a is a constant between 1 and 10; the input layer is set to four neurons, which are the three parameters Δe(k), e(k), and Δe(k-1) of the incremental PID algorithm and the constant factor 1 for stabilizing the BP neural network; and the output layer is set to three output neurons.

[0031] (2) Selection of the activation function:

[0032] ​The tanh function is selected as the activation function of the hidden layer, and the unipolar Sigmoid function is selected as the activation function of the output layer to correspond to the three output non-negative PID parameters. The advantage of the Sigmoid function is that it can squeeze the input values in a large range into the output range of (0, 1). The tanh function is similar to the Sigmoid function, but it is a bipolar function, and its output value range is (-1, 1);

[0033] (3) Initial weight selection:

[0034] The initial weights of the hidden layer and the output layer are selected as random numbers close to 0, so as to speed up the training speed of the network at the beginning;

[0035] (4) Determination of learning rate:

[0036] The learning rate η is selected, and the selected learning rate is uniformly 0.2.

[0037] In the PID controller, the proportional element P reacts to the deviation instantaneously. Once the deviation occurs, the controller immediately produces a control action to change the control quantity in the direction of reducing the deviation. The integral element I can eliminate the static error, but it will also reduce the response speed of the system and increase the overshoot of the system. The role of the derivative element D is to prevent the change of the deviation, which controls according to the trend of the deviation. The introduction of the derivative action will help to reduce the overshoot, overcome the oscillation, and make the system stable. It speeds up the tracking speed of the system. After the 3D printing concrete device is started, the PID controller responds quickly, reduces the machine oscillation, and ensures the uniformity of the discharging system. However, since the 3D printing concrete process has nonlinear characteristics, the PID controller may be greatly disturbed, so the BP neural network is introduced to adjust the parameters of the PID controller in real time, which enhances the stability of the system and further improves the precision of the printing product.

[0038] The steps of step 4) are:

[0039] (1) BP network structure initialization, setting the learning rate, inertia coefficient initial parameters and change range of the BP network;

[0040] (2) PSO algorithm initialization; the BP neural network learning rate and inertia coefficient are used as the particles of PSO, the initial population number is set, the particle velocity and position vector are initialized, and the value of each particle center vector and the standardization constant is transmitted to the BP neural network control module for motor motion control;

[0041] (3) Particle update; the speed and position information of each particle are updated, and the P i of each particle and the P g of the whole colony are calculated.

[0042] (4) The particle swarm optimizes according to the neural network simulation step length, and the optimal value under the current step length is transmitted to the control module to control the movement of the motor in each cycle;

[0043] (5) The iteration is judged whether to terminate or not; the current state is judged whether to satisfy the iteration termination condition or not, if yes, the iteration is terminated, otherwise, it is transferred to step 3.

[0044] (6) The PSO iteration is terminated, and the output optimal learning rate and inertia coefficient are taken as the initial parameters of the BP-PID control system.

[0045] The beneficial effects of the present application are:

[0046] The present application aims at the time-varying problem of the rheological property of the concrete material in the printing process, and proposes a 3D printing concrete rheological property control algorithm, which detects the rheological property of the printing material in real time in the printing process, changes the motor speed and adjusts the rheological property of the printing material according to the data obtained through sensor feedback, aims at the nonlinearity problem of the 3D printing concrete process, combines the BP neural network with the PID controller, adjusts the PID control parameters in real time on line by using the adaptive learning ability of the BP neural network, and optimizes the BP-PID controller by using the PSO algorithm, improves the system stability and the uniformity of the printing material in the discharging process, makes the surface structure of the printing finished product more balanced, reduces the defects such as cracks, and thus improves the quality of the concrete forming. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 The method flow chart is designed for the rheological property control of the 3D printing concrete.

[0048] Figure 2 The system block diagram is designed for the rheological property control of the 3D printing concrete.

[0049] Figure 3 The simulation diagram is designed for the 3D printing concrete control system based on the BP-PID controller.

[0050] Figure 4 The flow chart is designed for the PSO-BP neural network control system. DETAILED DESCRIPTION

[0051] The present application will be further described in detail below with reference to the drawings.

[0052] Reference Figure 1 The specific steps of the present application are described in detail.

[0053] Step one: the relationship between the rheological property of the concrete and the motor speed of the printing system is established;

[0054] In step one: the method designs ten groups of control experiments, the material ratio of the concrete is fixed, the motor speed is changed, and the rheological property of the mixed concrete is measured respectively, and the data obtained is analyzed and fitted by using MATLAB;

[0055] Step two: establish a rheological property control model of 3D printing concrete extrusion system:

[0056] The step two specifically comprises the following steps:

[0057] (1) the threshold value of the rheological property of the concrete is taken as the input of the rheological property control of the 3D printing concrete, and the output is the actual rheological property of the concrete in the printing process;

[0058] (2) the transmission function of the motor is determined as The model of the rheological property control system is obtained by Laplace transformation of the relationship established in step one

[0059] Step three: design a BP-PID controller:

[0060] The step three: the designed BP-PID controller is applied to the model of the established control system, as shown in Figure 2 .

[0061] The calculation process of the BP neural network

[0062] (1) forward calculation process;

[0063] Let the input layer, the hidden layer and the output layer be represented by superscripts (1), (2) and (3) respectively, net be the input value of the current layer, O represent the output value of the current layer, and w be the weight matrix;

[0064] The output of the network input layer is:

[0065]

[0066] Wherein represents the output of node j;

[0067] The input and output of the network hidden layer are:

[0068]

[0069]

[0070] Wherein represents the input of node i;

[0071] The activation function of the hidden layer neuron is selected as the bipolar tanh function:

[0072]

[0073] The input and output of the output layer are:

[0074]

[0075]

[0076]

[0077] where K p is a proportional coefficient, K i is an integral coefficient, and K d is a differential coefficient;

[0078] The three nodes of the output layer correspond to K p , K i , and K D , respectively, which are three adjustable non-negative parameters. The activation function of the output layer is selected as a unipolar Sigmoid function:

[0079]

[0080] (2) Error backpropagation process;

[0081] Let the performance index function be:

[0082]

[0083] The network weight coefficient w is corrected using the gradient descent method, and the weight change is adjusted in the negative gradient direction of the network. For the performance index E k Given the learning rate η, the weight adjustment amount is:

[0084]

[0085] where:

[0086]

[0087] Then:

[0088]

[0089]

[0090]

[0091] The hidden layer weight adjustment amount is:

[0092]

[0093] where

[0094]

[0095] The output layer weight adjustment calculation formula is:

[0096]

[0097] where e(k) is the target and actual error value, Δe(k) is the target and actual error rate, and Δe(k-1) is the last target and actual error rate.

[0098] The design steps of the BP neural network PID are as follows:

[0099] (1) Set the network structure;

[0100] The BP neural network structure is determined to be 4 input layer nodes, 9 hidden layer nodes and 3 output layer nodes, the four input quantities are the parameters Δe(k), e(k), Δe(k-1) of the incremental PID algorithm and the constant 1 of the stable network, the hidden layer is the minimum number that meets the system performance, the output is the three parameters of the PID control, and the initial values of the input and output parameters of each layer are set to be 1.

[0101] (2) Network parameter setting;

[0102] The learning rate η is selected, and the initial weights in the hidden layer weight matrix w 2 (k) and the output layer weight matrix w (3) (k) are set to be random numbers between (-1, 1), the hidden layer activation function is set to be the bipolar tanh function, the output layer activation function is set to be the Sigmoid function, the input adjustment transfer function is set, the sampling period is set to be Ts, and the input expected value r(k) is set.

[0103] (3) Neural network training process;

[0104] The network sampling obtains the system expected value r(k) and the actual output value y(k) at the k time, the error e(k) = r(k)-y(k) at this time is calculated, and the input and output of each layer network are calculated according to the weight correction formula, the hidden layer weight matrix w 2 (k) and the output layer weight matrix w (3) (k) are adjusted, the learning number is increased by one, and the step is repeated until the sampling is completed. The simulation model is as follows: Figure 3 .

[0105] Step four: the parameters of the controller are optimized and set by using the PSO algorithm.

[0106] The step four includes the following steps:

[0107] (1) BP network structure initialization, set the learning rate, inertia coefficient initial parameters and variation range of the BP network;

[0108] (2) Initialize each parameter, such as the number of ants μ, pheromone importance factor α, pheromone evaporation factor β, maximum iteration number itermax, etc.

[0109] (3) Particle update, update the speed and position information of each particle, and calculate the P i of each particle and the P g of the entire colony.

[0110] (4) Determine whether the iteration is terminated. If the current state meets the iteration termination condition, output the optimal learning rate and inertia coefficient as the initial parameters of the BP-PID control system, otherwise go to step 3.

[0111] The specific implementation steps of step four are:

[0112] (1) Initialize the value of N particles and the update speed of particles. Each particle (pop = c1, c2, c3, c4, c5, b) is a vector containing the value of the center vector and the standardized constant, set the value range of the particle group, and the particles are randomly distributed and uniformly distributed within the respective value range. The initial update speed of the particle V(N, :) = 0.1rand(N, 6).

[0113] (2) The value of the center vector and the standardized constant of each particle is passed to the BP neural network control module for motor speed motion control. The error between the actual speed and the desired speed of the motor after the current step of the particle is run is used as the fitness of the particle.

[0114] (3) Sort the N fitness values, and the particle corresponding to the minimum fitness value is the global optimal particle g best , and the initialized particle is the individual optimal i best . The particle speed update and individual update formulas are respectively:

[0115] V(N, :) = ωV(N, :) + d1r1(i best (N, :) - pop(N, :)) + d2r2(g best -pop(N, :))

[0116] pop(N, :) = pop(N, :) + 0.3V(N, :)

[0117] The particle group optimizes according to the neural network simulation step size, and the optimal value at the current step is passed to the control module for motor speed control each time. The specific steps are as follows Figure 4 .

[0118] The method of the present application firstly experiments the relationship between the rheological property of the concrete material and the motor speed of the printer extrusion system, then carries out Laplace transformation on the obtained function relationship, and combines the motor speed of the extrusion system of the 3D printed concrete to establish a preliminary control model. Since the rheological property of the concrete material changes with the increase of time, which leads to the nonlinearity of the 3D printed concrete process, in view of this problem, the present application combines the BP neural network with the PID controller, uses the adaptive learning ability of the BP neural network to adjust the PID control parameters online in real time, improves the system stability, and in view of the problem that the BP-PID self-learning process is easy to fall into local minimum value, uses the improved PSO algorithm to optimize it, ensures that the BP-PID system converges to the global optimal solution.

[0119] The algorithm used in the present application is tested under the operation platform of Windows 10 using MATLAB R2022b development tool, and the specific software and hardware environment is as follows:

[0120] Table 1. Software and hardware environment parameter table

[0121]

[0122] The present application firstly obtains the relationship between the rheological property of the 3D printed concrete material and the motor speed through designing specific experiments, establishes a 3D printed concrete rheological property control system, designs a BP-PID controller for specific problems, adjusts the PID control parameters online in real time, improves the system stability, uses the PSO algorithm to optimize it, ensures that the BP-PID system converges to the global optimal solution, improves the uniformity of the printing material in the discharging process, reduces the surface defects such as cracks of the printed products, and improves the quality of the concrete forming.

Claims

1. A method for intelligent control of precision of 3D printing concrete based on BP neural network, characterized in that, It comprises the following steps: Step 1): Establish the relationship between the rheological property of concrete and the motor speed of the 3D printing concrete extrusion system; Step 2): Establish a rheological property control model for the 3D printing concrete extrusion system according to the relationship in step 1); Step 3): Design a BP-PID controller for the rheological property control model established in step 2); Step 4): Optimize and set the parameters of the BP-PID controller using the PSO algorithm; The step 1) establishes the relationship between the motor speed of the extrusion system and the rheological property of concrete; Set the same batching ratio of concrete, take the motor speed as the variable, design a control experiment, measure the rheological property of concrete at the current motor speed during printing, and use the MATLAB data fitter to analyze and fit the results, to obtain the relationship between the rheological property and the speed x as follows: The step 2) establishes a rheological property control model for the 3D printing concrete extrusion system: (1) The rheological property of concrete during printing should be maintained between 180-190 to make the surface of the printed product uniform, and the given concrete fluidity threshold is taken as the input rin(k) of the 3D printing concrete rheological property control system; (2) The sensor measures the size of the concrete fluidity in real time and feeds back the output results to the system, which are compared with the input signals to obtain the deviation signal e(k); e(k)=rin(k)-rout(k) (3) The deviation signal is transmitted to the controller to adjust the motor speed. Generally, the DC motor can be regarded as an inertial element, so the transfer function of the motor in the system is set as where K is the motor transfer coefficient, and T is the motor time constant. (4) Perform Laplace transform on the relationship obtained in step 1), and take the size of the concrete fluidity as the output rout(k) of the system based on this relationship, to obtain the control transfer function G2 as follows: The step 3) BP-PID controller is specifically: The incremental PID algorithm is used to realize the closed loop control of the concrete rheological property, so as to keep the fluidity and the motor rotating speed stable, and the three parameters k p , k i , k d of the PID controller are adjusted on line through the BP neural network; the BP neural network adjusts the three parameters according to the running state of the system, so as to indirectly adjust the controlled object; the BP neural network has the self-learning feature, and can realize the real time control of the three control parameters of the PID through the weight adjustment; the formula of the incremental PID algorithm is as follows: ΔU(k) = K p [e(k) - e(k-1)] + K i e(k) + K d {[e(k) - e(k-1)] -[e(k-1)-e(k-2)]} That is: ΔU(k) = K p Δe(k) + K i e(k) + K d [Δe(k) - Δe(k-1)] where ΔU(k) is the output change, K p is the proportional coefficient, K i is the integral coefficient, K d is the differential coefficient, e(k) is the error value of the target and the actual value this time, e(k-1) is the error value of the target and the actual value last time, and e(k-2) is the error value of the target and the actual value two times ago.

2. The precision intelligent control method for 3D printing concrete based on BP neural network according to claim 1, characterized in that, After determining the PID controller, determine the structure framework of the BP neural network applied to the PID controller; (1) Selection of the number of hidden layer nodes: According to the formula The number of nodes in the hidden layer is selected; Where m is the number of hidden layer nodes, n is the number of input layer nodes, a is a constant between 1-10, the input layer is set to 4 neurons, which are the three parameters Δe(k), e(k), Δe(k-1) of the incremental PID algorithm and the constant factor 1 for stabilizing the BP neural network, and the output layer is set to three output neurons; (2) Selection of activation function: The tanh function is selected as the activation function for the hidden layer, and the unipolar Sigmoid function is selected as the activation function for the output layer to correspond to the three non-negative PID parameters, and the Sigmoid function has the advantage of squeezing the input values in a large range into the output range of (0, 1), and the tanh function is similar to the Sigmoid function, but it is a bipolar function, and its output value range is (-1, 1); (3) Selection of initial weights: The initial weights of the hidden layer and the output layer are selected as random numbers close to 0 to speed up the training speed of the network at the beginning; (4) Determination of learning rate: Select the learning rate η. 3.The BP neural network-based precision intelligent control method for 3D printing concrete according to claim 1, characterized in that, The step 4) is: (1) Initialize the BP network structure, set the learning rate, initial parameters and variation range of the inertia coefficient of the BP network; (2) PSO algorithm initialization; the BP neural network learning rate and inertia coefficient are taken as the particles of PSO, the initial population number is set, the particle velocity and position vector are initialized, and the value of each particle center vector and standardization constant is transmitted to the BP neural network control module for motor motion control; (3) Particle update; update the velocity and position information of each particle, and calculate the P i of each particle and the P g of the entire cluster; (4) The particle swarm is optimized according to the neural network simulation step, and the optimal value under the current step is transmitted to the control module for motor motion control each time; (5) Determine whether the iteration is terminated; determine whether the current state meets the iteration termination condition, if yes, terminate the iteration, otherwise go to step 3; (6) PSO iteration termination, the output optimal learning rate and inertia coefficient are taken as the initial parameters of the BP-PID control system.

Images