CN107697660A - Screw material disperser control method based on machine learning - Google Patents

Abstract

The invention discloses the screw material disperser control method based on machine learning, first Dynamic Recurrent Elman neutral nets are established in the controller, by blanking bin material position, falling down error, blanking rate, the helical blade diameter of material density and auger conveyor, 7 input quantities of pitch and screw rod maximum (top) speed are mapped as the material air weighting in screw material disperser blanking process, after gradient descent method off-line training network, processing module carries out closing control in advance by output module according to the air weighting predicted value that neutral net exports to auger conveyor in On-line Control blanking process.The present invention is modeled using nonlinear network to blanking process, network after training can carry out Accurate Prediction to the air weighting under different blanking states, produce so as to direct accurate blanking and suitable for small lot, and blanking efficiency is improved because screw rod can keep high running speed.

Description

Control method of screw material batching machine based on machine learning

technical field

The invention relates to the field of quantitative cutting, in particular to a machine learning-based control method for a screw-type material batching machine.

Background technique

In industrial and agricultural manufacturing and commodity packaging, there are a large number of powder materials, such as coal powder and other iron-making raw materials, polypropylene, polystyrene, polyvinyl chloride, light methyl cellulose, polypropylene nitrile, epoxy resin powder coating Chemical raw materials such as quartz sand, cement and other building materials, household chemical products such as washing powder, agricultural products such as millet and soybeans, grains and beans, or powder, slag, granular processed food, agricultural production materials such as feed, chemical fertilizers, pesticides, and powder Granular health care products, Chinese and Western medicines, condiments, etc. all require automatic quantitative packaging or ingredient manufacturing.

At present, many enterprises in our country still use manual quantitative batching or packaging. On the one hand, the labor intensity is high, the speed is slow, and the economic benefit is poor; It is easy to cause harm to the human body. Therefore, for production enterprises, there is an urgent need to provide inexpensive multi-component automatic quantitative feeding equipment or devices with high speed and accuracy to meet the quantitative packaging or batching manufacturing requirements of a large number of materials.

At present, there are two commonly used methods for automatic quantitative feeding and batching devices for powder materials at home and abroad, volumetric and weighing. Volumetric quantitative filling or feeding is performed according to the volume of the material. The quantitative feeding is fast, but the quality of the quantitative material is changed by the change of the material density. In order to improve the feeding accuracy, various adjustment methods have appeared, such as the Chinese patent application number 201320001933.3, which adopts frequency conversion speed regulation for the screw, and gradually slows down the feeding speed when it is close to the target value, reducing the drop in the air; the application number is 201310234280.8 Chinese patent, in the three-speed frequency conversion feeding process of soda ash packaging machine, large and small screws are used to feed in multiple stages; the Chinese patent with application number 200920248298.2 considers that it is difficult to control the quantitative during fast feeding, and it is reduced by the method of fast first and then slow The impact of the feed drop; the final value of the blanking of these non-weighing schemes can only be close to the expected value, and the accuracy is not high.

The weighing type quantification is based on the quality of the material to measure, fill or feed. It needs to be weighed continuously during the feeding process, and the feeding amount is controlled according to the feedback of the weighing result. Because the weighing is greatly affected by the impact of the feeding and the lagging material in the air, the components are lowered. Material speed and accuracy are faced with many difficulties. In order to compensate for the interference of airborne materials on the measurement accuracy, many schemes adopt the technology of closing the valve in advance. For example, the Chinese patent application number 201410230888.8 divides the ingredient weighing process into three stages, and uses iterative learning control method to calculate in the last stage Turn off the advance control amount.

Compared with the indirect linear iterative prediction in iterative learning control, if a nonlinear mapping can be constructed through the analysis of various factors affecting the amount of material dropped in the air during the blanking process, the blanking process can be described more intuitively and Accurate and straightforward predictions of material air volumes are based on this mapping.

Contents of the invention

The simple screw feeder belongs to the category of volumetric quantitative filling. Volumetric quantitative filling is based on volume to measure the quantity of filling materials. It has a simple structure and low cost, but the stability and accuracy of quantitative filling speed depend on the stability of the apparent specific gravity of the material. The physical and chemical properties such as the looseness of the material, the uniformity of the particles, and the hygroscopicity have a great influence. Since the ordinary volumetric type is essentially a conversion type, it is impossible to grasp the exact quality of the blanking like the weighing type. Although a combined weighing scheme appeared later, because there is no prediction of the empty volume, it can only rely on the extremely low final stage of blanking. The feeding speed to ensure the accuracy.

For this reason, the present invention combines the dynamic weighing detection with the screw feeder to increase the feeding speed. However, in weighing blanking, it is necessary to estimate the amount of air and its impact. The amount of material falling in the air during the feeding process is the amount of air, which is affected by many factors, such as the closing speed of the conveying device, the size of the drop between the feeding port and the material surface of the scale bucket, the flow rate of the falling material, etc., so the feeding and conveying should be closed in advance The timing of the device is difficult to determine in one go by offline experiments.

According to the in-depth test and analysis of the blanking process, it is found that the most important factors affecting the air volume of the screw material batching machine include: the material level of the feeding bin, the drop in the air, the blanking rate, the material density, and the diameter and pitch of the screw blade of the screw conveyor. and the maximum screw speed. The air volume is a complex nonlinear mapping of these physical quantities. In order to predict the air volume and then perform accurate feeding by closing the screw conveyor in advance, it is necessary to identify and express the mapping relationship.

The method of identifying the system and modifying the parameters based on the linear system theory can be well applied to the linear system, but it cannot be applied to the complex nonlinear system. Artificial neural network is a network formed by the extensive interconnection of a large number of processing units. It has large-scale parallel simulation processing capabilities and strong self-adaptation, self-organization, and self-learning capabilities. It is generally valued in system modeling, identification, and control. Its nonlinear transformation characteristics provide an effective method for system identification, especially the identification of nonlinear systems.

At present, the most widely used in nonlinear system identification is the multi-layer forward network. The multi-layer forward network has the ability to approximate any continuous nonlinear function, but this kind of network structure is generally static. From the analysis of the material falling process, it can be seen that It is found that since the material level of the lower silo and the drop in the air are gradually changing, there is also a close relationship between the amount of air in the two consecutive sampling periods. For this reason, the present invention uses a dynamic recurrent neural network to model the system. Different from the static feedforward neural network, the dynamic recurrent network has the function of mapping dynamic features by storing the internal state, so that the system has the ability to adapt to time-varying characteristics, and is more suitable for the identification of nonlinear dynamic systems. The present invention is based on the dynamic recursive Elman neural network, and compares the relationship between the air volume and the material level c of the lower hopper, the air drop h, the blanking rate d, the material density ρ, the screw blade diameter D of the screw conveyor, the screw pitch S, and the maximum screw speed v _R Identify the mapping relationship between them, and detect and dynamically adjust the distribution of materials in the feeding bin during the feeding process, so that the trained neural network can directly predict the amount of air in different states, so as to achieve high-precision cutting .

The technical solution of the present invention is to provide a machine learning-based control method for a screw type material batching machine, comprising the following steps:

S1, set up a neural network module: the neural network module adopts a dynamic recursive Elman neural network, and its input layer receives the material level of the lower silo, the drop in the air, the blanking rate, the material density and the screw blade diameter of the screw conveyor from the processing module respectively. There are 7 input quantities of screw pitch and maximum screw speed, and the output of the output layer is transmitted to the iterative learning module and the processing module through the first connection array and the second connection array respectively;

S2. Obtain training samples: use the screw-type material batching machine to repeat the feeding. After each feeding starts, when the material forms a continuous material flow from the screw conveyor at the bottom of the feeding bin to the metering hopper, continue feeding for a period of time. When the screw conveyor is closed, the initial weight reading W of the weighing module is read in real time and the value of the input amount is obtained by the processing module. After the material falls, the weight reading WD of the weighing module is read, and the time when the screw conveyor is closed The amount of air in the state is A=WD-W, with A as the actual value of the amount of air in the sample output;

S3. Offline training neural network: based on the obtained training samples, the iterative learning module uses the gradient descent method to iteratively adjust the neural network according to the actual value of the material air volume input by the processing module and the neural network through the first connection array and the network output value respectively. connection weight;

S4. On-line blanking control:

The signal acquisition module collects the sensing signals of the material level of the lower hopper, the material level of the metering hopper, and the weight of the falling material in real time through the sensor of the position sensor in the lower hopper, the sensor of the bucket position in the metering hopper, and the weighing module carrying the metering hopper, and transmits them to the The processing module performs data processing and analysis, and obtains the material level of the feeding bin, the drop in the air, and the feeding rate;

The trained neural network is used to predict the air volume to obtain the predicted value yA and send it to the processing module. After processing and analysis by the processing module, the closing time of the screw conveyor at the bottom opening of the lower silo is controlled by the output module.

As preferably, in the online blanking control process, assuming that the primary blanking amount of the current component is Ws, when starting blanking, the controller obtains the initial weight of the weighing hopper as G by reading the sensing value of the weighing module. ; Afterwards, the controller continuously reads the sensing value of the weighing module, and when the value reaches (G0+Ws-yA), the screw conveyor is closed.

Preferably, in the process of obtaining the training samples, the training samples cover enough blanking states, and the moment of closing the screw conveyor each time can be set as a random value after the moment when the initial weight reading of the weighing module is a certain value .

As a preference, during the control process of online blanking, in addition to the predicted value of the amount in the air, the current cumulative blanking error must also be compensated, that is, when it is detected that the weight of the weighing hopper reaches (G0+Ws-yA-E), the screw will be closed Conveyor, where E is the current cumulative feeding error of this component.

Preferably, the output module is also connected to the discharge valve at the bottom opening of the weighing hopper, and controls the opening and closing of the discharge valve according to the instructions of the processing module;

The output module is also connected to the rotatable base of the position sensor in the lower bin, and controls the operation of the base according to the instructions of the processing module;

The output module is also connected to the vibrating rod installed on the frame near the side wall of the lower bin, and controls the start, stop and operation of the vibrating rod according to the instructions of the processing module;

The signal acquisition module also collects the material level in the mixing hopper through a material level sensor located in the mixing hopper below the blanking valve, and the output module is also respectively connected to the push plate at the bottom of the mixing hopper and installed in the mixing hopper. The mixer, and control the start and stop of the push plate and the mixer according to the instructions of the processing module.

Preferably, the output module is also connected to the feed pump connected in series between the storage bin and the lower feed bin, and controls the start, stop and operation of the feed pump according to the instructions of the processing module, wherein the speed of the feed pump is as follows: Take control:

Among them, V _{into 0} is a set maximum feeding speed, l is the current material level of the lower silo, L _M and L _m are the preset highest and lowest material levels of the lower silo respectively.

As preferably, the model of the neural network is:

x c _k (t) = x _k (t-mod (k, q)-1),

Among them, mod is the remainder function, the f() function is taken as the sigmoid function; xc _k (t) is the output of the receiving layer, x _j (t) is the output of the hidden layer, u _i (t-1) and y(t) are the input layer input and output layer output respectively, ω _j , ω _jk and ω _ji are the connection weights from the hidden layer to the output layer, the connection weights from the successor layer to the hidden layer, and the connection from the input layer to the hidden layer Weights, θ and θ _j are the thresholds of the output layer and the hidden layer respectively; k=1, 2...m, q is the selected regression delay scale, which is optimized according to the sampling period and feeding rate, if optional q=5; j==1, 2...m, i=1, 2...7, the number m of hidden layer and receiving layer nodes can be selected from 11 to 20, for example, 16 is preferred.

Preferably, there are multiple neural network modules, and each neural network module corresponds to a screw conveyor of the batching machine.

As a preference, the output module controls the running speed of the screw conveyor in the following manner:

A. Start from the stop state with μ·a _max acceleration, and keep the speed constant when the speed reaches λ·v _R ;

B. When the closing time is up, decelerate at the acceleration of μ·a _max until it stops;

Among them, a _max is the rated maximum acceleration of the screw of the screw conveyor, v _R is the maximum speed, μ is the acceleration coefficient between (0.5-0.9), and λ is the speed coefficient between (0.85-1.0);

The closing time means that the weight of the material that is currently read from the weighing module is equal to:

Among them, Ws and Wa are the predicted value of the current material's one-time feeding amount and air volume respectively, d is the feeding rate of the screw conveyor when the screw runs at the maximum speed, t _s is the length of time for deceleration and stop: t _s = λ·v _R /μ·a _max .

Compared with the prior art, the solution of the present invention has the following advantages: the present invention uses a nonlinear network to construct and model the mapping relationship between the influencing factors of the blanking process and the amount of space in the air, and the network after offline training can be used for different The air volume in the blanking state can be accurately predicted, so that the continuous blanking control can be directly carried out according to the predicted value in the online application, and the blanking error fluctuation in the online iterative learning can be avoided. The control of material accumulation error reduces the total error of batch blanking, and compared with the general screw type blanking device, the screw conveyor can maintain a higher operating speed, which improves the blanking efficiency.

Description of drawings

Figure 1 is a structural diagram of the controller of a screw-type material batching machine based on machine learning;

Fig. 2 is a schematic diagram of Elman neural network structure;

Fig. 3 is a structural diagram of the screw type material batching machine;

Fig. 4 is the external structure diagram of the screw type material batching machine;

Fig. 5 is a schematic diagram of the material falling process;

Fig. 6 is a schematic diagram of the partial structure of the storage bin and the lower bin;

Fig. 7 is a schematic diagram of the structure of the vibrating rod;

Fig. 8 is a schematic diagram of material flow laminar flow in the lower silo;

Fig. 9 is a schematic diagram of the distribution of multi-component materials in the metering hopper;

Figure 10 is a diagram of weighing changes during the material falling process.

Among them: 1. Lower material bin 2. Screw conveyor 3. Measuring hopper 4. Weighing module 5. Blanking valve 6. Mixing hopper 7. Push plate 8. Delivery pipe 9. Controller 10. Storage bin 11. Feed pump 12, bin level sensor 13, bucket level sensor 14, material level sensor 15, feed pipe 16, material nozzle 17, small hole 18, vibrating rod 19, material level surface 20, stop point 21, scanning line 22, Mixer 23, feeding pipe

30. Rack 91, signal acquisition module 92, processing module 93, neural network module 94, iterative learning module 95, storage module 96, first connection array 97, second connection array 98, output module

101, pumping plate 181, pillar 182, pan platform 183, vibrator 184, vibrating rod 185, particle protrusion 186, vibrating rod track

201, screw box 202, conveying screw 203, connector 204, motor

detailed description

Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention is not limited to these embodiments. The present invention covers any alternatives, modifications, equivalent methods and schemes made within the spirit and scope of the present invention.

In order to provide the public with a thorough understanding of the present invention, specific details are set forth in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details.

In the following paragraphs the invention is described more specifically by way of example with reference to the accompanying drawings. It should be noted that all the drawings are in simplified form and use inaccurate scales, which are only used to facilitate and clearly assist the purpose of illustrating the embodiments of the present invention.

As shown in Figure 1, the present invention adopts the screw-type material batching machine controller based on machine learning, and described controller comprises signal acquisition module 91, processing module 92, neural network module 93, iterative learning module 94, storage module 95, the first A connection matrix 96 , a second connection matrix 97 and an output module 98 . Wherein, the neural network module 93 adopts an Elman neural network, and the storage module 95 is a memory for storing data.

As shown in Figure 2, the Elman neural network used has a recursive structure. Compared with the BP neural network, the Elman neural network includes a receiving layer in addition to the input layer, hidden layer and output layer. The feedback connection makes it possible to express the time delay between input and output and the timing characteristics of parameters, so that the network has a memory function. Referring to Figure 2, the input layer of the established neural network has 7 units, and the number m of nodes in the hidden layer and the receiving layer can be selected between 11 and 20. If 16 is selected, the output layer has only one unit.

The model of the neural network is:

x c _k (t) = x _k (t-mod (k, q)-1) (1),

Among them, mod is the remainder function, the f() function is taken as the sigmoid function; xc _k (t) is the output of the receiving layer, x _j (t) is the output of the hidden layer, u _i (t-1) and y(t) are the input layer input and output layer output respectively, ω _j , ω _jk and ω _ji are the connection weights from the hidden layer to the output layer, the connection weights from the successor layer to the hidden layer, and the connection from the input layer to the hidden layer Weights, θ and θ _j are the thresholds of the output layer and the hidden layer respectively; k=1, 2...m, q is the selected regression delay scale, which is optimized according to the sampling period and feeding rate, if optional q=5; j==1, 2...m, i=1, 2...7.

Referring to Fig. 3 and Fig. 4, the machine learning-based control method of the screw-type material batching machine of the present invention is used to accurately control the feeding of the screw-type material batching machine. The screw-type material batching machine includes a feeding bin 1 , a screw conveyor 2 , a weighing hopper 3 , a weighing module 4 , a blanking valve 5 , a mixing hopper 6 , and a controller 9 . The screw-type material batching machine is used for feeding and batching various component materials, and each component of the material has a set of feeding bins 1 and screw conveyors 2 corresponding to each other.

There is a pumping plate 101 at the bottom of the feeding bin 1, and the pumping board is opened during feeding, and the material flows out from the opening at the bottom of the feeding bin. Screw conveyor 2 comprises screw case 201, conveying screw rod 202, connector 203 and motor 204, and the shell of motor 204 links to each other with screw case 201 by connector 203, and the conveying screw rod 202 that is positioned at the screw case 201 passes through axle sleeve and motor 204 The shafts are connected; the upper surface of the screw box 201 is opposite to the opening at the bottom of the lower bin 1. There is a feed inlet, and the opposite end of the screw box 201 is connected to the feeding pipe 23 opposite to the motor, and the feeding pipe 23 is fixed on the frame 30.

As shown in Fig. 3 and Fig. 4, when unloading, the pumping plate 101 is controlled to open, and the material falls into the screw box 201 of the screw conveyor 2 from the unloading bin 1, and the motor 204 is started at the same time, and the conveying screw 202 rotates together with the motor. The material is conveyed to the feeding pipe 23 at the end, and falls into the metering hopper 3 below from the feeding pipe 23 .

The frame 30 serves as the frame of the equipment and is used to fix and support other components. The weighing module 4 is fixed on the frame 30 , and the weighing bucket 3 is pressed on the weighing module 4 in a movable manner. The bottom of the weighing bucket 3 has an opening, and the opening and closing of the opening are controlled by the blanking valve 5 . The metering hopper 4 is located at the bottom of the feeding pipe 23 , and a plurality of screw conveyors 2 are radially distributed relative to the center of the feeding pipe 23 and the metering hopper 4 .

The mixing hopper 6 is located below the discharge valve 5, and has a push plate 7 at its bottom, and a delivery pipe 8 is connected below the push plate, and the latter transports the multi-component mixture to packaging bags or production equipment. A material level sensor 14 is installed on the side wall of the mixing hopper 6, and a mixer 22 is also arranged inside it, and the capacity of the mixing hopper 6 is several such as 15 times of the measuring hopper 3.

As shown in FIG. 1 and FIG. 4 , the controller 9 may also include an input module, such as a touch screen. During operation, the formula of multi-component materials and other parameters can be set through the man-machine interface of the touch screen, and the switch between the first connection array and the second connection array can also be realized through the touch screen operation, so that the controller works in offline training or online prediction mode.

As shown in Figure 4 and Figure 5, the controller can dynamically read the current reading of the weighing module in real time, but during the feeding process, the reading it reads is not the actual weight of the material falling into the weighing hopper, but includes The impact of the material; and after the motor is turned off, the material between the screw conveyor and the weighing hopper will continue to fall into the weighing hopper. Therefore, it is generally used to deduct this part of the air volume, that is, to close the screw conveyor when the weighing measurement is less than a certain value of the target weight to control the feeding and batching. But in practice, how to accurately obtain the size of the air volume is a difficult problem that must be solved.

Figure 5 shows the change of the impact of the material level drop and the falling speed on the weighing hopper during the falling process of the material. The material falls from the screw conveyor 2 at the initial velocity v0, and the distance between the outlet of the screw conveyor 2 and the bottom of the weighing hopper 3 is H. As the material level _h2 in the metering hopper increases, the air drop _h1 will become smaller.

The mass equivalent change of the material in the weighing hopper can be expressed by the following formula:

Among them, at time t, dm is the dropping mass per unit time (g/s) at the outlet of the screw conveyor, v ₀ is the initial velocity when the material falls, and the velocity of the Δm material when it falls into the weighing hopper changes from Velocity v ₁ becomes 0.

It can be seen from formula (4) that with the change of the air drop _h1 , the impact of the material on the weighing bucket also changes. Therefore, the weight of the weighing bucket changes with time.

Through repeated experimental tests and analysis on the blanking process, it is concluded that for the screw type material batching machine, the most important factors affecting the amount of air in the blanking process include: the material level c of the blanking bin, the drop in the air h, the blanking rate d, The material density ρ and the screw blade diameter D of the screw conveyor, the screw pitch S and the maximum screw speed v _R . Aerial quantities are complex nonlinear mappings of these physical quantities. In order to accurately obtain the predicted value of the material air volume in different states, so as to accurately discharge the material by closing the screw conveyor in advance, it is necessary to identify and express the mapping relationship.

Based on the complex non-linear characteristics of the mapping, and considering the close relationship between the air volume in two consecutive sampling periods, the present invention uses a dynamic recursive Elman neural network to model the air volume and the material level c of the lower hopper, the air volume Identify the mapping relationship between the drop h, the blanking rate d, the material density ρ, the screw blade diameter D of the screw conveyor, the screw pitch S and the maximum screw speed v _R.

As shown in Figure 1 and Figure 2, the established neural network, the input layer includes 7 nodes, in which the material density ρ, the screw blade diameter D of the screw conveyor, the screw pitch S and the maximum screw speed are definite values, which can be input through the touch screen to the controller; the other three quantities need to be collected dynamically and in real time through the signal acquisition module. By introducing multiple nodes of the receiving layer with different delay regressions in the network, the network structure is more suitable for the blanking process, so that the network training converges faster.

As shown in Fig. 1 and Fig. 2, when training the neural network offline, the iterative learning module adjusts the connection weight of the neural network according to the actual value of the material air volume input by the processing module and the neural network and the network output value respectively through the first connection matrix. value.

In order to obtain training samples, after the start of feeding, when the material forms a continuous material flow from the screw conveyor at the bottom of the feeding bin to the weighing hopper, continue feeding for a period of time, and read the weighing module in real time when the screw conveyor is turned off For the weight reading W, wait for the material to fall and read the weight reading WD of the weighing module, then the air volume at the moment when the screw conveyor is closed is A=WD-W, this value is the actual value of the sample output value y, that is, the expected value y _d .

The neural network training adopts the gradient descent method, and the weight and threshold adjustment methods during training are as follows.

Assuming there are a total of P training samples, let the error function be:

Then the adjustment formula of the connection weights from the hidden layer to the output layer is as follows:

ω _j (t+1)=ω _j (t)+Δω _j (t+1) (6),

in,

δ _y =-(y _d -y)·y·(1-y) (8),

The adjustment formula of the output layer threshold is:

θ(t+1)=θ(t)+Δθ(t+1) (9),

in,

Similarly, the adjustment formula for the connection weights from the input layer to the hidden layer is:

_ωji (t+1)＝ _ωji (t)+ _Δωji (t+1) (11),

in,

δ _j = δ _y · ω _j · x _j (t) · (1-x _j (t)) (13),

The adjustment formula of hidden layer threshold is:

θ _j (t+1)=θ _j (t)+Δθ _j (t+1) (14),

in,

Regardless of the dependence of xc _k (t) on the connection weight ω _jk , the adjustment formula of the connection weight from the receiving layer to the hidden layer is:

ω _jk (t+1)=ω _jk (t)+Δω _jk (t+1) (16),

in,

The initial value range of each weight is taken as (-0.1, 0.1) interval, and the learning rate η is a decimal less than 1, which can be adjusted dynamically by using a fixed rate or according to the total error of the current network output.

The training end condition can be set as the total error or its change is less than a set value or the number of training times reaches a certain amount.

Preferably, in order to make the training samples cover more situations, the moment of closing the screw conveyor each time can be set as a random value after the moment when the screw conveyor is started or the weight reading of the weighing module reaches a certain value.

Before network training, perform normalized preprocessing on 7 inputs and 1 output:

r'=rr _min /r _max -r _min (18),

Among them, r is the unprocessed physical quantity, r' is the normalized physical quantity, r _max and r _min are the maximum and minimum values of the sample data set, respectively.

When calculating the predicted value of the air volume, the following formula is used to convert the network output volume back to the air volume value:

r=r _min +r'·(r _max -r _min ) (19).

When the material is controlled online, the first connection array is disconnected, and the neural network predicts the amount yA in the air and outputs it to the processing module through the second connection array. Conveyor for shutdown control:

Assuming that the one-time feeding amount of the current component is Ws, when starting feeding, the controller obtains the initial weight of the weighing hopper as G0 by reading the sensor value of the weighing module; then, the controller continuously reads the weighing module’s Sensing value, when the value reaches (G0+Ws-yA), close the screw conveyor.

As a preference, in addition to the predicted value of the amount in the air, the current accumulated feeding error should be compensated, that is, when the weight of the weighing hopper is detected to reach (G0+Ws-yA-E), the screw conveyor will be closed, where E is the component The current accumulative blanking error.

As preferably, the controller controls the operating speed of the screw conveyor in the following manner:

A. Start from the stop state with μ·a _max acceleration, and keep the speed constant when the speed reaches λ·v _R ;

B. When the closing time is up, decelerate at the acceleration of μ·a _max until it stops;

Among them, a _max is the rated maximum acceleration of the screw of the screw conveyor, v _R is the maximum speed, μ is the acceleration coefficient between (0.5-0.9), and λ is the speed coefficient between (0.85-1.0);

The closing time means that the weight of the material that is currently read from the weighing module is equal to:

Among them, Ws and Wa are the predicted value of one-time feeding amount and empty amount of the current material respectively, d is the feeding rate of the screw conveyor when the screw runs at the maximum speed, t _s is the length of deceleration and stop time, t _s = λ·v _R /μ·a _max .

This same speed control is used for the screw conveyor both when taking the training samples offline and when applying the neural network online. Among the 7 input quantities of the training samples, the material density, the screw blade diameter, the screw pitch and the maximum screw speed of the screw conveyor should be set in advance. The signal acquisition module collects the sensing signals of the material level of the lower hopper, the material level of the metering hopper, and the weight of the falling material in real time through the sensor of the position sensor in the lower hopper, the sensor of the bucket position in the metering hopper, and the weighing module carrying the metering hopper, and transmits them to the The processing module performs data preprocessing, and then inputs it to the neural network. The output value of the neural network and the expected output value preprocessed by the processing module are sent to the iterative learning module through the first connection matrix, and the iterative learning module will adjust the output value according to the gradient descent method. The weights of are passed back to the neural network.

The bucket position sensor can be set similarly to the storage position sensor. Both the bin position sensor and the bucket position sensor can use distance sensors to detect the material level height of the material in the lower bin and the weighing hopper respectively, wherein the bucket level signal of the weighing hopper is processed by the processing module and converted into the material drop in the air.

By collecting the signals of the weighing module periodically, the processing module can calculate the material falling mass equivalent per unit time, that is, the material falling rate.

It can also be analyzed from formula (4) that the mass equivalent of the material dropped per unit time, that is, the detected material falling rate, is also affected by the shape distribution of the material in the lower bin 1.

There are two main types of particulate matter flowing out from the lower silo under the action of gravity: overall flow and central flow. In the flow pattern of the overall flow, the entire particle layer in the silo can flow out roughly evenly, and basically every particle is moving; while in the flow pattern of the central flow, some particles are stationary, and there is a gap between the flowing and stationary particles. flow channel boundaries. The overall feeding rate of the bulk flow is larger than that of the central flow, and the fluctuation of the feeding rate is small and the flow is stable. In the actual production process, the material in the warehouse may have a flow pattern of central flow, so that when the material port starts to discharge, the material will be solidified into a plate due to the compaction stress generated by the pressure of the warehouse.

As shown in Figures 6 to 8, in order to reduce the fluctuation range of the blanking rate in the lower hopper, so as to better perform system modeling and air volume prediction, the material batching machine applied to the present invention adopts a distance sensor type silo The sensor and the rotatable vibrating rod detect and adjust the material accumulation form in the lower material bin, so that the formation and collapse of the dynamic material arch alternately appear above the lower material opening, ensuring the stability of the material density and the material discharge form.

As shown in Figure 6, the lower bin 1 continuously discharges the material, and when the material level in the bin drops to a certain value, it needs to be replenished. For this reason, a storage bin 10 is set above the lower bin 1 , and the material in the storage bin 10 enters the lower bin 1 through a feed pump 11 and a feed pipe 15 . In order to make the material particles evenly fed, a material nozzle 16 is provided at the outlet of the end of the feed pipe 15. The surface of the material nozzle 16 is spherical and has circular small holes 17 distributed on the surface. The diameter of the small holes depends on the particle size of the material. Make an optimization. Feed pump 11 adopts screw conveyor, and its action is controlled by controller.

During the feeding process of the feeding bin 1, with the reduction of the material level surface 19, the feed pump 11 acts under the control of the controller, so that the material level of the top surface of the material in the feeding bin remains near the preset value, and its rotating speed Control as follows:

Among them, V _{into 0} is a set maximum feeding speed, l is the current material level of the lower silo, L _M and L _m are the preset highest and lowest material levels near _LT respectively.

The two figures in Fig. 6 are observed from the side view and the top view direction of the lower silo 1 respectively, as shown in Fig. 6a and 6b, a position sensor 12 is installed on a vertex near the center of the frame of the lower hopper 1, which has a The rotating base can be pitched and rotated, so that the warehouse level sensor can detect materials in the direction of different docking points 20, and each docking point 20 forms a scanning line 21 close to concentric circles, thereby judging the distribution of the material discharge surface 19.

As shown in Figure 7, the controller improves the distribution of materials through a vibrating rod 18 installed on the side wall of the lower bin 1 by the batching machine. The vibrating bar 18 comprises a pillar 181, a cloud platform 182, a vibrator 183, and a vibrating bar 184 which are connected successively. A spring buffer is arranged at the bottom of the vibrator 183, and particle protrusions 185 are distributed on the vibrating bar 184 surface. Rotate so that the vibrating rod 184 makes a curved movement in the lower bin 1.

During the feeding process, the controller judges the distribution of materials in the feeding bin through the detection of the bin level sensor and the tracking of the feeding rate per unit time, so that the material level surface in the feeding bin remains approximately parabolic. As shown in Figures 6 to 8, when the materials are evenly distributed, the distance values of the materials detected by the position sensor at different orientations are approximately concentrated in a small range after the geometric transformation of the detection ray and the vertical inclination angle. The detected distance value exceeds this range when the material is partially compacted or a stable material arch occurs. At the same time, the feeding rate of each feeding bin is tracked in real time through the weighing module. When the distance sensor detects the above-mentioned abnormal state or finds that the fluctuation of the discharge amount per unit time exceeds the set threshold such as 5%, the controller commands the vibrating rod to move. Through the operation of the pan/tilt, the vibrating rod starts from the starting point and passes through the high material level area. Go to the low material level area, do serpentine stirring, the vibration rod track 126 at the end of the vibration rod 124 in the lower bin 1 is shown in Figure 7; at the same time, the vibrator starts to vibrate, and the particle protrusions on the vibration rod drive the surrounding particles. Particles, so as to break the occasional compaction or material arch, so that the material distribution can be restored to uniformity. Through the dynamic detection and control of material distribution, the fluctuation of material density is reduced, so as to ensure the stability of filling amount per unit time. While the vibrating rod is moving, the feeding is suspended and the pumping plate is closed.

As shown in Figure 8, the present invention greatly weakens the effect of the compaction force produced by the impact of charging through the detection and action of the distance sensor and the vibrating rod, effectively preventing the particle size segregation of the materials in the bin, and making the lower bin The activation of the material inside improves the distribution of the material.

The controller acts as the control center of the whole screw material batching machine. During the feeding process of multi-component materials, the controller controls the action of each screw conveyor in turn. The bucket falls into the mixing hopper.

After completing multiple batches of material, the controller reads the status of the material level sensor in the mixing hopper, and if it detects that the material level exceeds the set threshold, it controls the mixer to rotate and stir through the output module to mix various materials evenly Finally, under the control of the controller, the push plate is opened, and the mixed material is output from the delivery pipe.

Figures 9 to 10 supplement the records of the multi-component material unloading process, in which Figure 9 shows the schematic diagram of the material distribution in the weighing hopper when the four components are unloaded; Figure 10 shows the weighing module during a 900ms long material falling process The change curve of the readings, where the abscissa is the delay time after closing the screw conveyor. It can be seen that due to the impact force, the weighing reading will overshoot and then return to the actual weight; and the material in the air will completely fall into the weighing hopper after the screw conveyor is closed for about 700ms, and the reading of the weighing module will tend to be stable. .

Applying the control method of the present invention to carry out feeding control, the feeding behavior of each screw conveyor is first modeled separately off-line, and the feeding of each component material is carried out separately in the process of collecting samples, so that all materials can be recovered without causing waste. Periodically collect the material level of the feeding bin, the drop in the air, and the blanking rate during the actual feeding, and can predict the current amount in the air in real time. Therefore, starting from the first batch, the feeding can be done accurately and other methods such as online iteration can be avoided. Blanking error fluctuations in learning scenarios.

The implementation methods described above do not constitute a limitation to the scope of protection of the technical solution. Any modifications, equivalent replacements and improvements made within the spirit and principles of the above implementation methods shall be included in the protection scope of the technical solution.

Claims (9)

1. A screw type material dosing machine control method based on machine learning comprises the following steps:

s1, establishing a neural network module: the neural network module adopts a dynamic recursive Elman neural network, an input layer of the neural network module respectively receives 7 input quantities of material level, air fall, blanking rate, material density of a blanking bin, the diameter and the screw pitch of a helical blade of a screw conveyor and the maximum rotating speed of a screw from the processing module, and the output quantity of an output layer is respectively transmitted to the iterative learning module and the processing module through a first connecting array and a second connecting array;

s2, obtaining a training sample: repeatedly feeding materials by using a screw type material dosing machine, continuously feeding materials for a period of time when continuous material flow is formed between a screw conveyor at the bottom of a feeding bin and a weighing hopper after each feeding is started, reading an initial weight reading W of a weighing module in real time when the screw conveyor is closed, acquiring a value of an input quantity by a processing module, reading a weight reading WD of the weighing module after the materials fall, and taking the empty quantity A as an empty quantity actual value of a sample output quantity when the screw conveyor is closed;

s3, off-line training of the neural network: based on the obtained training sample, the iterative learning module iteratively adjusts the connection weight of the neural network by adopting a gradient descent method according to the actual value of the material empty space and the network output value which are respectively input by the processing module and the neural network through the first connection array;

s4, online blanking control:

the signal acquisition module acquires sensing signals of the material level of the blanking bin, the material level of the measuring hopper and the weight of falling materials in real time through a bin level sensor in the blanking bin, a hopper level sensor in the measuring hopper and a weighing module bearing the measuring hopper respectively and transmits the sensing signals to the processing module for data processing and analysis to obtain the material level of the blanking bin, the air drop and the blanking rate;

and predicting the air quantity by using the trained neural network to obtain a predicted value yA, transmitting the predicted value yA to a processing module, and controlling the closing time of the screw conveyor at the opening at the bottom of the blanking bin through an output module after the predicted value yA is processed and analyzed by the processing module.

2. The machine learning-based screw material dispensing machine control method of claim 1, wherein:

in the online blanking control process, assuming that the primary blanking amount of the current component is Ws, when blanking is started, the controller obtains the initial weight of the weighing hopper G0 by reading the sensing value of the weighing module; the controller then constantly reads the sensor value of the weighing module and closes the screw conveyor when this value reaches (G0+ Ws-yA).

3. The machine learning-based screw material dispensing machine control method of claim 1, wherein:

in the process of obtaining the training sample, the training sample covers enough blanking states, and the moment when the screw conveyor is closed each time can be set to be a random value after the moment when the initial weight reading of the weighing module is a certain determined value.

4. The machine learning-based screw material dispensing machine control method of claim 1, wherein:

in the on-line feeding control process, in addition to the predicted empty space amount value, the current accumulated feeding error is compensated, namely when the weight of the weighing hopper is detected to reach (G0+ Ws-yA-E), the screw conveyor is closed, wherein E is the current accumulated feeding error of the component.

5. The machine learning-based screw material dispensing machine control method of claim 1, wherein:

the output module is also connected to a blanking valve at the opening at the bottom of the metering hopper and controls the opening and closing of the blanking valve according to the instruction of the processing module;

the output module is also connected to a rotatable base of a bin level sensor in the blanking bin and controls the base to operate according to the instruction of the processing module;

the output module is also connected to a vibrating rod arranged on the side wall of the machine frame close to the blanking bin, and controls the start-stop and operation of the vibrating rod according to the instruction of the processing module;

the signal acquisition module is also used for acquiring the material level in the mixing hopper through a material level sensor positioned in the mixing hopper below the blanking valve, the output module is also respectively connected to the push plate at the bottom of the mixing hopper and the mixer installed in the mixing hopper, and the start and stop of the push plate and the mixer are respectively controlled according to the instruction of the processing module.

6. The machine learning-based screw material dispensing machine control method of claim 1, wherein:

the output module is also connected to a feeding pump which is connected between the storage bin and the discharging bin in series, and controls the start, stop and operation of the feeding pump according to the instruction of the processing module, wherein the rotating speed of the feeding pump is controlled according to the following formula:

wherein, V_{0 feeding}A set maximum feeding speed, L is the current material level of the feed bin, L_MAnd L_mRespectively the preset highest and lowest feeding bin material positions.

7. The machine learning-based screw material dosing machine control method of any one of claims 1-6, wherein the neural network model is:

xc_k(t)＝x_k(t-mod(k，q)-1)，

<mrow> <msub> <mi>x</mi> <mi>j</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>f</mi> <mrow> <mo>(</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <msub> <mi>&amp;omega;</mi> <mrow> <mi>j</mi> <mi>k</mi> </mrow> </msub> <mo>&amp;CenterDot;</mo> <msub> <mi>xc</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>+</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mn>5</mn> </munderover> <msub> <mi>&amp;omega;</mi> <mrow> <mi>j</mi> <mi>i</mi> </mrow> </msub> <mo>&amp;CenterDot;</mo> <msub> <mi>u</mi> <mi>i</mi> </msub> <mo>(</mo> <mi>t</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>&amp;theta;</mi> <mi>j</mi> </msub> <mo>)</mo> <mo>,</mo> </mrow>

<mrow> <mi>y</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>f</mi> <mrow> <mo>(</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <msub> <mi>&amp;omega;</mi> <mi>j</mi> </msub> <mo>&amp;CenterDot;</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>-</mo> <mi>&amp;theta;</mi> <mo>)</mo> </mrow> <mo>,</mo> </mrow>

wherein mod is a remainder function, and f () is a sigmoid function; xc_k(t) is the carry layer output, x_j(t) is the hidden layer output, u_i(t-1) and y (t) are input layer input and output layer output, ω_j、ω_jkAnd ω_jiRespectively, the connection weight from the hidden layer to the output layer, the connection weight from the accepting layer to the hidden layer and the connection weight from the input layer to the hidden layer, theta and theta_jOutput layer and hidden layer thresholds, respectively; k is 1, 2 … m, q is the selected regression delay scale, and is preferred according to the sampling period and the blanking rate, and if the selectable q is 5; j is 1, 2 … m, i is 1, 2 … 7, and the number m of hidden layer and receiving layer nodes can be selected from 11 to 20, such as 16.

8. The machine learning-based screw material dispensing machine control method of claim 7, wherein: the neural network module has a plurality of, and every neural network module corresponds a screw conveyer of proportioning machine.

9. The machine learning-based screw material dispensing machine control method of claim 1, wherein the output module controls the operating speed of the screw conveyor as follows:

A. from a stopped state at a rate of mu a_maxStarting at acceleration, when the speed reaches lambda.v_RKeeping the speed unchanged;

B. when the closing time is up, the value is expressed in mu a_maxThe acceleration starts to decelerate until stopping;

wherein, a_maxRated maximum acceleration, v, of the screw conveyor_RThe maximum speed is mu is an acceleration coefficient between 0.5 and 0.9, and the lambda is a speed coefficient between 0.85 and 1.0;

the closing time means that the current baiting weight read from the weighing module is equal to:

wherein Ws and Wa are respectively the predicted values of the current one-time material discharge amount and the empty space amount, d is the discharge rate of the screw conveyer when the screw rotates at the maximum speed, and t is the discharge rate of the screw conveyer when the screw rotates at the maximum speed_sFor the length of deceleration stop time, t_s＝λ·v_R/μ·a_max。