Machine learning method, machine learning apparatus, computer program product, communication method, and mixing device

Abstract

The present invention relates to a machine learning method, obtaining a state variable including at least one first evaluation parameter related to performance evaluation of a mixture and at least one mixing condition; calculating a reward for a decision result of the at least one mixing condition based on the state variable; updating a function for deciding the at least one mixing condition based on the state variable based on the reward; and deciding a mixing condition that maximally obtains the reward by repeatedly updating the function, wherein the at least one first evaluation parameter includes at least one of a physical property and a shape feature related to the mixture.

Description

TECHNICAL FIELD

[0001] The present application relates to a technology of machine learning of mixing conditions of a mixing device. BACKGROUND

[0002] Patent Literature 1 discloses a technology of generating a machine learning model indicating whether mixing is normal or abnormal by machine learning of learning data of measured data of a state of a closed-type rubber mixing machine including instantaneous electric power value, temperature, rotational speed of a rotor, and position of upper and lower ram pressure (ram), and given teacher data indicating abnormality degree of mixing corresponding to the measured data, and then judging abnormality of mixing using the machine learning model.

[0003] However, Patent Literature 1 is a technology of judging presence or absence of abnormality of mixing using a machine learning model, and is not a technology of determining mixing conditions of a mixing device. For this reason, the technology of Patent Literature 1 cannot determine mixing conditions from which a proper mixed product can be obtained.

[0004] Up to the present time, proper mixing conditions are determined relying on years of experience of skilled technicians, and thus it is difficult to easily determine.

[0005] PRIOR ART DOCUMENTS

[0006] PATENT LITERATURE

[0007] Patent Literature 1: Japanese Patent Publication No. 2020-32676 SUMMARY

[0008] The present application is made to solve the above problems, and aims to provide a machine learning device or the like in which mixing conditions from which a proper mixed product can be obtained can be easily determined without relying on years of experience of skilled technicians.

[0009] An embodiment of the present application relates to a machine learning method for determining a mixing condition of a mixing device for mixing a polymer material to obtain a mixed product, the mixing device including a chamber into which the material is fed, a rotor of two or more shafts for mixing the material fed into the chamber, and a controller for controlling the rotor of two or more shafts, a mixing time of the material, and an operation step of the mixing device, the machine learning method including the steps of: obtaining state variables including at least one first evaluation parameter related to performance evaluation of the mixed product and at least one mixing condition; calculating a reward for a determination result of the at least one mixing condition based on the state variables; updating a function for determining the at least one mixing condition based on the state variables based on the reward; and determining a mixing condition that maximizes the reward by repeatedly updating the function.

[0010] Another embodiment of the present application relates to a machine learning device for determining a mixing condition of a mixing device for mixing a polymer material to obtain a mixed product, the mixing device including a chamber into which the material is fed, a rotor of two or more shafts for mixing the material fed into the chamber, and a controller for controlling the rotor of two or more shafts, a mixing time of the material, and an operation step of the mixing device, the machine learning device including: a state acquisition unit for obtaining state variables including at least one first evaluation parameter related to performance evaluation of the mixed product and at least one mixing condition; a reward calculation unit for calculating a reward for a determination result of the at least one mixing condition based on the state variables; an update unit for updating a function for determining the at least one mixing condition based on the state variables based on the reward; and a determination unit for determining a mixing condition that maximizes the reward by repeatedly updating the function.

[0011] Another embodiment of the present application relates to a machine learning program for a machine learning device that determines a mixing condition of a mixing device for mixing a polymer material to obtain a mixed product, the mixing device including: a chamber into which a material of the mixed product is fed; two or more rotors that mix the material fed into the chamber; and a controller that controls the two or more rotors, controls a mixing time of the material, and controls an operation step of the mixing device, the machine learning program causing a computer to function as: a state acquisition unit that acquires a state variable including at least one first evaluation parameter related to performance evaluation of the mixed product and at least one mixing condition; a reward calculation unit that calculates a reward for a determination result of the at least one mixing condition based on the state variable; an update unit that updates a function for determining the at least one mixing condition from the state variable based on the reward; and a determination unit that determines a mixing condition that maximizes the reward by repeatedly updating the function.

[0012] Another embodiment of the present application relates to a communication method for a machine learning device that determines a mixing condition of a mixing device for mixing a polymer material to obtain a mixed product, the mixing device including: a chamber into which a material of the mixed product is fed; two or more rotors that mix the material fed into the chamber; and a controller that controls the two or more rotors, controls a mixing time of the material, and controls an operation step of the mixing device, the communication method including: observing a state variable including at least one first evaluation parameter related to performance evaluation of the mixed product and at least one mixing condition; and transmitting the state variable to a network and receiving a machine-learned mixing condition, wherein the at least one first evaluation parameter includes at least one of a physical property and a shape feature of the mixed product.

[0013] Another embodiment of the present application relates to a mixing device for mixing a polymer material to obtain a mixed product, including: a chamber into which a material of the mixed product is fed; two or more rotors that mix the material fed into the chamber; a controller that controls the two or more rotors, controls a mixing time of the material, and controls an operation step of the mixing device; a state observation unit that acquires a state variable including at least one first evaluation parameter related to performance evaluation of the mixed product and at least one mixing condition; and a communication unit that transmits the state variable to a network and receives a machine-learned mixing condition, wherein the at least one first evaluation parameter includes at least one of a physical property and a shape feature of the mixed product.

[0014] According to the present application, appropriate mixing conditions can be easily determined without relying on years of experience of skilled technicians. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a front cross-sectional view of a mixing device to which the present embodiment relates.

[0016] Figure 2 is a diagram showing the overall configuration of a machine learning system that causes the mixing device to which the present embodiment relates to perform machine learning.

[0017] Figure 3 is a diagram showing one example of a mixing condition.

[0018] Figure 4 is a diagram showing one example of a mixing condition.

[0019] Figure 5 is a diagram showing one example of a mixing condition.

[0020] Figure 6 is a diagram showing one example of a mixing condition.

[0021] Figure 7 is a diagram showing one example of a first evaluation parameter.

[0022] Figure 8 is a diagram showing one example of a second evaluation parameter.

[0023] Figure 9 is a diagram showing Figure 2 is a flowchart showing one example of the processing of the machine learning system shown in

[0024] Figure 10 is a diagram showing the overall configuration of a machine learning system to which a modification of the present application relates. DETAILED DESCRIPTION

[0025] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. Note that the following embodiments are merely examples of the present application and do not limit the technical scope of the present application in any way.

[0026] Figure 1 is a front cross-sectional view of a mixing device 300 to which the present embodiment relates. As shown in FIG. 2, the mixing device 300 includes a mixing chamber 301, a first screw 302, a second screw 303, and a third screw 304. Figure 1As shown, the kneading device 300 includes a housing 3, a pair of rotors (one example of two or more rotors) 4, a door 5, a hopper 6, a weight 8, a cylinder 9, a piston 10, a piston rod 11, a motor 12, a speed reducer 13, a drive section 14, a controller 320, and an input section 350. The housing 3 has a chamber 2 inside. The pair of rotors 4 is disposed in the chamber 2. The hopper 6 is disposed to a feeding cylinder 7. The feeding cylinder 7 is vertically disposed above the housing 3 to supply the material dropped from the hopper 6 to the chamber 2. The weight 8 is freely movable up and down through the feeding cylinder 7.

[0027] The cylinder 9, which is constituted by a hydraulic cylinder or a pneumatic cylinder, for example, is connected to an upper portion of the feeding cylinder 7. The piston 10 is disposed in the cylinder 9. The piston 10 is connected to the weight 8 via the piston rod 11 that penetrates a lower lid side of the cylinder 9 in airtight. The weight 8 is lowered by the space in the upper portion of the cylinder 9 being pressurized. Thus, the weight 8 pushes the material supplied from the hopper 6 into the chamber 2. The material mainly includes a high molecular material such as resin or rubber. In addition, the material can include a filler (carbon, silica, or the like), a medicine, and an oil, and the like.

[0028] The housing 3 has a discharge port at a bottom portion. The door 5, which is also called a drop door, opens and closes the discharge port by power from an actuator (omitted from illustration). When the material is kneaded in the chamber 2, the door 5 blocks the discharge port. The door 5 discharges the kneaded material in the chamber 2 by opening the discharge port.

[0029] The pair of rotors 4 is disposed in parallel in a manner of abutting each other in a horizontal direction. Each rotor of the pair of rotors 4 rotates inwardly. Each rotor of the pair of rotors 4 has a plurality of kneading wings (omitted from illustration) on an outer peripheral surface thereof. There is a gap between a top of each kneading wing and the chamber 2. If the pair of rotors 4 rotates, a shearing force is applied to the material in the gap. Each kneading wing is twisted spirally around an axis of the pair of rotors 4. By the twisting, the material is pushed toward an axial direction of the pair of rotors 4 and thus flows in the axial direction.

[0030] The housing 3 has a passage (omitted from illustration) that extends in an axial direction of the pair of rotors 4 and allows a circulating medium to flow. The pair of rotors 4, the weight 8, and the door 5 each have a passage (omitted from illustration) that allows the circulating medium to flow.

[0031] The motor 12 rotates the pair of rotors 4 via the speed reducer 13. The speed reducer 13 is disposed between the motor 12 and the pair of rotors 4 to reduce a rotational speed of the pair of rotors 4. The drive section 14 is constituted by an air unit or a hydraulic unit. The drive section 14 lowers the weight 8 by pressurizing a space in an upper portion of the cylinder 9 and raises the weight 8 by depressurizing the space.

[0032] The controller 320 is responsible for the overall control of the mixing device 300. For example, the controller 320 controls the raising and lowering of the counterweight 8 by outputting a control signal for driving the driving section 14. Also, the controller 320 controls the temperature of the material in the chamber 2 by controlling a heater (not shown). Also, the controller 320 controls the rotation of the pair of rotors 4 by outputting a control signal for driving the motor 12. Also, the controller 320 controls the opening and closing of the door 5 by outputting a control signal to an actuator (not shown). The input section 350 is configured with an operating device such as a switch, for receiving an instruction from a user. The conveyor belt 15 conveys the material toward the hopper 6. In addition, the controller 320 is also responsible for the control of the pair of rotors 4, the control of the mixing time of the material, and the control of the action steps of the mixing device.

[0033] Figure 2 is a diagram showing the overall configuration of a machine learning system that causes the mixing device 300 involved in the embodiments to perform machine learning. The machine learning system includes a server 100, a communication device 200, and the mixing device 300. The server 100 and the communication device 200 are communicably connected to each other via a network NT1. The communication device 200 and the mixing device 300 are communicably connected to each other via a network NT2. The network NT1 is, for example, a wide area communication network such as the Internet. The network NT2 is, for example, a local area network. The server 100 is, for example, a cloud server constituted by one or more computers. The communication device 200 is, for example, a computer possessed by a user who uses the mixing device 300. The communication device 200 functions as a gateway that connects the mixing device 300 to the network NT1. The communication device 200 can be realized by installing dedicated application software on a computer possessed by the user himself or herself. Alternatively, the communication device 200 can be a dedicated device provided to the user by the manufacturer of the mixing device 300.

[0034] Hereinafter, the configuration of each device will be described in detail. The server 100 includes a processor 102 and a communication section 101. The processor 102 is a control device including a CPU or the like. The processor 102 includes a reward calculation section 110, an update section 120, a decision section 130, and a learning control section 140. Each module possessed by the processor 102 can be realized by causing the processor 102 to execute a machine learning program that causes the computer to function as the server 100 of the machine learning system, or can be realized by a dedicated circuit.

[0035] The reward calculation section 110 calculates a reward for the decision result for at least one mixing condition, based on the state variable observed by the state observation section 321.

[0036] The update unit 120 updates a function for determining the mixing condition based on the state variable observed by the state observation unit 321, based on the reward calculated by the reward calculation unit 110. As the function, an action value function described later can be adopted.

[0037] The determination unit 130 determines the mixing condition that can obtain the most reward, by repeatedly performing the update of the function while changing at least one of the mixing conditions.

[0038] The learning control unit 140 is responsible for the overall control of the machine learning. The machine learning system of the present embodiment learns the mixing condition by reinforcement learning. Reinforcement learning is a machine learning method in which an agent (action subject) selects a certain action based on the state of an environment, and the agent learns to select a better action by causing the environment to change based on the selected action and giving the agent a reward accompanying the change in the environment. As the reinforcement learning, Q-learning and TD-learning can be adopted. In the following description, Q-learning is described as an example. In the present embodiment, the reward calculation unit 110, the update unit 120, the determination unit 130, the learning control unit 140, and the state observation unit 321 described later correspond to the agent. In the present embodiment, the communication unit 101 is one example of a state acquisition unit that acquires the state variable.

[0039] The communication unit 101 is configured by a communication circuit that connects the server 100 to the network NT1. The communication unit 101 receives the state variable observed by the state observation unit 321 via the communication device 200. The communication unit 101 transmits the mixing condition determined by the determination unit 130 to the mixing device 300 via the communication device 200. The communication unit 101 transmits the processing instruction determined by the learning control unit 140 to the mixing device 300.

[0040] The communication device 200 includes a transmitter 201 and a receiver 202. The transmitter 201 transmits the state variable transmitted from the mixing device 300 to the server 100, and transmits the mixing condition transmitted from the server 100 to the mixing device 300. The receiver 202 receives the state variable transmitted from the mixing device 300, and receives the mixing condition transmitted from the server 100.

[0041] The mixing device 300 includes a communication unit 310, a controller 320, a storage 330, a sensing unit 340, and an input unit 350.

[0042] The communication unit 310 is a communication circuit for connecting the mixing device 300 to the network NT2. The communication unit 310 transmits the state variable observed by the state observation unit 321 to the server 100. The communication unit 310 receives the mixing condition determined by the determination unit 130 of the server 100. The communication unit 310 receives the processing instruction described later determined by the learning control unit 140.

[0043] The controller 320 is a computer including a CPU or the like. The controller 320 includes a state observation section 321 and a kneading execution section 322. The communication section 310 transmits the state variable acquired by the state observation section 321 to the server 100. Each module possessed by the controller 320 can be realized, for example, by causing the CPU to execute a machine learning program that causes the kneading device 300 to function as a machine learning system.

[0044] The state observation section 321 observes the state variable including the first evaluation parameter related to the performance evaluation of the kneaded material and at least one of the kneading conditions after the kneading ends. Here, the kneading conditions are the measurement values of the sensing section 340 and the set values of the kneading execution section 322. Also, the kneading conditions are also the first evaluation parameter, the measurement values of the sensing section 340, and the like. In addition, the state observation section 321 can also acquire the second evaluation parameter related to the stability of the operation of the kneading device 300.

[0045] Figures 3 to 6 is a schematic view showing one example of the kneading conditions. The kneading conditions are roughly classified into the categories. In the categories, at least one of the first parameter related to the material, the second parameter related to the rotor control, the third parameter related to the operation step, the fourth parameter related to the counterweight operation, the fifth parameter related to the temperature adjustment, and the sixth parameter related to the equipment specifications is included.

[0046] The first parameter includes at least one of the kind of the input material including the mixing amount, the weight of the material, the specific gravity of the material, the order of input, and the filling rate. The kind of the input material also includes the kind of the component of the material. The mixing amount includes the amount of each kind of the component of the material or the proportion of the amount of each kind. As the kind of the component, for example, rubber or resin, carbon fiber, oil, and medicine, and the like are included. The weight of the material includes the weight of the component of the material. The specific gravity of the material indicates the specific gravity of each kind of the component of the input material. The order of input is the order of the component of the input material. The filling rate is the proportion (volume rate) of the volume of the material input to the chamber 2 with respect to the volume of the chamber 2.

[0047] The second parameter includes at least one of the rotor rotation speed, the rotor phase, and the rotor speed ratio. The rotor rotation speed is the rotation speed of the pair of rotors 4 per unit time at each operation step.

[0048] The mixing device 300 performs the step process of the multiple material feeding from the hopper 6 until the mixing product is discharged from the door 5. Hereinafter, the period from the material feeding to the next material feeding or the material cleaning in one step process will be referred to as a step unit. That is, one step process is divided into one or more step units, such as the first step unit (1st), the second step unit (2nd). Further, each step unit is also divided into a material feeding process and a mixing process or a cleaning process. Each of these material feeding process, mixing process, and cleaning process is collectively referred to as an operation step (one example of a step).

[0049] The material feeding time indicates when the material feeding process is performed. In the material feeding process, the material is fed from the hopper 6 and pushed into the chamber 2 by the counterweight 8. The mixing time indicates when the mixing process is performed. In the mixing process, the material is mixed by the pair of rotors 4 in a state where the counterweight 8 is lowered. The cleaning time indicates when the cleaning process is performed. In the step units after the second, the cleaning process can be performed instead of the material feeding process as appropriate. In the step units after the second, the counterweight 8 is raised when the material feeding process or the cleaning process is started. In the step units after the second, the counterweight is lowered when the mixing process is started.

[0050] For example, "1st. material feeding time" indicates the rotor speed in the material feeding process in the first step unit, and "1st. mixing time" indicates the rotor speed in the mixing process in the first step unit. Further, "2nd. material feeding time or cleaning time" indicates the rotor speed in the material feeding process in the second step unit or the rotor speed in the cleaning process in the second step unit. The same applies to the step units after the third. Here, the step units up to the third are indicated, but the step units after the fourth can also be indicated.

[0051] The rotor phase is the setting angle of the pair of rotors 4. For example, the pair of rotors 4 is set with a phase offset of 90 degrees, 180 degrees, and the like. The rotor speed ratio is the difference in the rotor speed per unit time of the pair of rotors 4.

[0052] Referring to Figure 4 The third parameter includes at least one of the mixing time of each step, the material feeding time, the step advancing condition, the cumulative mixing time, and the cumulative electric quantity. The mixing time of each step is the time required for the mixing process of each step unit. For example, "1st mixing step" indicates the time of the mixing process in the first step unit. Here, the step units up to the third are indicated, but the step units after the fourth can also be indicated. In addition, the mixing time of each step can also be constituted by the mixing time of at least one of the step units among the mixing times of a plurality of step units. For example, the mixing time of each step can also be constituted by only the mixing time of "1st mixing step".

[0053] The material input time is the time of the material input process of each step unit. For example, the "1st input step" is the time of the material input process of the first step unit. In addition, the material input time can also be constituted by the material input time of at least one of the step units among the material input times of a plurality of step units. For example, the material input time can also be constituted only by the kneading time of the "1st input step".

[0054] The step advancement condition is a condition for advancing to the next action step from each action step. The step advancement condition includes at least one of the kneading time, the temperature of the material, the temperature of temperature-keeping kneading, the instantaneous electric power, the cumulative electric power, the instantaneous electric current, the torque, and the temperature at the time of discharge of the material. The kneading time indicates a prescribed kneading time for advancing to the next action step. If the kneading time reaches the prescribed kneading time, the kneading process ends and advances to the next action step. The temperature of the material indicates a prescribed temperature of the material for advancing to the next action step. If the temperature of the material reaches the prescribed temperature, the kneading process ends and advances to the next action step. The temperature of temperature-keeping kneading is a set temperature when the temperature is kept constant to perform kneading in the kneading process. The instantaneous electric power indicates a prescribed instantaneous electric power of the motor 12 for advancing to the next action step. If the instantaneous electric power of the motor 12 reaches the prescribed instantaneous electric power, the kneading process ends and advances to the next action step. The cumulative electric power indicates a prescribed cumulative electric power of the motor 12 for advancing to the next action step. If the cumulative electric power of the motor 12 reaches the prescribed cumulative electric power, the kneading process ends and advances to the next action step. The instantaneous electric current indicates a prescribed instantaneous electric current of the motor 12 for advancing to the next action step. If the instantaneous electric current of the motor 12 reaches the prescribed instantaneous electric current, the kneading process ends and advances to the next action step. The torque indicates a prescribed torque of the motor 12 for advancing to the next action step. If the torque of the motor 12 reaches the prescribed torque, the kneading process ends and advances to the next action step. The temperature at the time of discharge of the material is a prescribed temperature at the time of discharging the material. In addition, the step advancement condition can also be constituted by at least one of the conditions for advancing to the next step unit among a plurality of conditions. For example, the step advancement condition can also be constituted by at least one of the conditions among the conditions of the kneading time, the temperature of the material, the temperature of temperature-keeping kneading, the instantaneous electric power, the cumulative electric power, the instantaneous electric current, the torque, and the temperature at the time of discharge of the material.

[0055] The cumulative kneading time is the cumulative kneading time of one batch. The cumulative electric amount is the cumulative electric amount spent for one batch.

[0056] Referring toFigure 5 The fourth parameter includes at least one of the following: counterweight cylinder pressure, counterweight position, and counterweight speed. Counterweight cylinder pressure is the pressure at which the counterweight 8 pushes the material into the interior of chamber 2 during the mixing process of each step unit. For example, "1st mixing step" indicates the counterweight cylinder pressure during the mixing process of the first step unit.

[0057] The counterweight position is the position of the counterweight 8 when it is pressing the material. The counterweight speed is the speed of the counterweight 8 when it is pressing the material.

[0058] The fifth parameter includes at least one of the following: outside air temperature, equipment component temperature, and circulating medium temperature. Outside air temperature is the ambient temperature at the time of the processing step. Equipment component temperature is the temperature of each part of the equipment during the processing step. For example, the equipment component temperature is... Figure 1 The components shown include the supply cylinder 7, cylinder 9, motor 12, reducer 13, housing 3, a pair of rotors 4, and door 5.

[0059] The circulating medium temperature includes at least one of the following: inlet chamber, outlet chamber, inlet rotor, outlet rotor, inlet counterweight, outlet counterweight, door, and outlet. The inlet chamber is the temperature of the circulating medium entering the housing 3. The outlet chamber is the temperature of the circulating medium exiting the housing 3. The inlet rotor is the temperature of the circulating medium entering a pair of rotors 4. The outlet rotor is the temperature of the circulating medium exiting the pair of rotors 4. The inlet counterweight is the temperature of the circulating medium entering the counterweight 8. The outlet counterweight is the temperature of the circulating medium exiting the counterweight 8. The door door is the temperature of the circulating medium entering the door 5. The outlet door is the temperature of the circulating medium exiting the door 5. The circulating medium temperature affects the properties of the compound. Furthermore, the circulating medium temperature can be used to calculate the thermal history of the compounded material. The circulating medium can be, for example, cooling water, steam, or hot oil. The circulating medium heats and / or cools the housing 3, the pair of rotors 4, the counterweight 8, and the door 5.

[0060] Reference Figure 6 The sixth parameter includes at least one of the following: rotor type, surface treatment, door top shape, and counterweight shape. Rotor type refers to the type of a pair of rotors 4 according to their shape. Surface treatment refers to the type of hardening treatment applied to the chamber 2. Surface treatments include thickened plating, etc. Surface treatment can improve the wear resistance and corrosion resistance of the chamber 2. Door top shape refers to the type of door 5 according to its shape. Counterweight shape refers to the type of door 5 (which should be a counterweight) according to its shape.

[0061] The above is an example of the mixing conditions. Further, the parameters particularly important among the above mixing conditions are described below. For example, "mixing amount" and "filling rate" in the first parameter. For example, "rotor rotation speed" at the time of mixing in the second parameter. For example, "mixing time of each step", "step advancing condition", "cumulative mixing time", and "cumulative electric amount" in the third parameter. For example, "counterweight cylinder pressure" in the fourth parameter. For example, "circulating medium temperature" in the fifth parameter.

[0062] Next, the first evaluation parameter is described. Figure 7 is a schematic view showing an example of the first evaluation parameter. The first evaluation parameter is roughly classified into a category. The category includes at least one of physical properties and shape characteristics. The physical properties include at least one of mooney viscosity, vulcanization properties, Payne effect, dispersion of additives, dynamic viscoelasticity (Tan δ), hardness, tension stress, elongation, tension strength, discharge actual weight, discharge actual temperature, wear characteristics, bending strength, impact strength, breaking strength, modulus of elasticity, capillary viscosity, flow characteristics, and rubber mixing rotation speed (surface renewal amount). The shape characteristics include at least one of remaining particles and surface properties.

[0063] Next, the second evaluation parameter is described. Figure 8 is a schematic view showing an example of the second evaluation parameter. The second evaluation parameter includes at least one of a category of a state of a counterweight, a state of a bearing, a state of a hydraulic pressure, and a state of a rotor. The state of the counterweight indicates a state of the counterweight. The state of the counterweight includes a counterweight pressure variation. The counterweight pressure variation indicates a variation of a thrust pressure of the counterweight 8 at the time of mixing. The thrust pressure varies according to a push of the material. The state of the bearing indicates a state of the bearing possessed by the pair of rotors 4. The state of the bearing includes at least one of a thrust load and a radial load. The thrust load is an axial load applied to the bearing of the pair of rotors 4 at the time of mixing. The radial load is a radial load applied to the bearing of the pair of rotors 4 at the time of mixing.

[0064] The state-hydraulic pressure indicates a state of a hydraulic pressure of working oil in which the hydraulic device of the mixing apparatus 300 operates. The state-hydraulic pressure includes at least one of a mixer circuit pressure, a counterweight circuit pressure, and an oil cleanliness. The mixer circuit pressure is a pressure of working oil supplied to an actuator for driving the door 5. The counterweight circuit pressure is a pressure of working oil supplied to an actuator for moving the counterweight 8 up and down. The oil cleanliness is a cleanliness of the working oil.

[0065] The state-rotor indicates a state of the rotor 4. The state-rotor includes at least one of an instantaneous electric power, a cumulative electric power, an instantaneous electric current, and a torque. The instantaneous electric power is an instantaneous electric power of the motor 12. The cumulative electric power is a cumulative electric power of the motor 12. The instantaneous electric current is an instantaneous electric current of the motor 12. The torque is a torque of the motor 12.

[0066] Among the first evaluation parameters and the second evaluation parameters, parameters that are particularly important are, for example, Mooney viscosity, vulcanization characteristics, Payne effect, dispersion of additives, dynamic viscoelasticity, hardness, discharge actual weight, and discharge actual temperature.

[0067] Referring back to Figure 2 , the mixing execution section 322 controls a mixing process performed by the mixing apparatus 300. For example, the mixing execution section 322 performs control of the rising and lowering of the counterweight 8, control of the pressure of the counterweight cylinder, control of the heater, control of the motor 12, control of the opening and closing of the door 5, and the like, as described with regard to the controller 320.

[0068] The memory 330 is, for example, a nonvolatile storage device, and stores, for example, an optimally determined mixing condition.

[0069] The sensing section 340 is various sensors for measuring Figures 3 to 6 the mixing condition shown in FIG. 1, Figure 7 the first evaluation parameter shown in FIG. 2, and Figure 8 the second evaluation parameter shown in FIG. 3. Specifically, the sensing section 340 includes a sensor for detecting the rotor rotation speed, a timer for measuring the mixing time and the material charging time, and the like, a sensor for measuring the temperature of the material or the temperature of the circulating medium, a sensor for measuring the current, the voltage, and the electric power supplied to the motor 12, a sensor for measuring the torque of the motor 12, a sensor for measuring the pressure of the counterweight 8, an outside air temperature sensor, a sensor for measuring the position of the counterweight 8, and a sensor for measuring the speed of the counterweight 8, and the like. Furthermore, the sensing section 340 includes a sensor for measuring the weight of the material and the mixed product, a sensor for measuring the load in the thrust direction and the radial direction applied to the bearings of the pair of rotors 4, and the like.

[0070] The input unit 350 is an input device such as a keyboard and mouse. The input unit 350 allows input from users, for example... Figure 6 The sixth parameter shown contains various data. Furthermore, the input unit 350 can receive data, for example, through user input. Figure 7 The measured value of the first evaluation parameter is shown. Furthermore, the input unit 350, for example, inputs... Figure 8 The data shown includes various parameters such as oil cleanliness.

[0071] Figure 9 It means Figure 2 The flowchart illustrates an example of the processing of a machine learning system. In step S1, the learning control unit 140 acquires input values ​​of the mixing conditions input by the user via the input unit 350. The input values ​​acquired here contain... Figure 3 The mixing quantities shown include the type of material, material weight, material specific gravity, input sequence, filling rate, rotor phase, and rotor speed ratio. Figure 6 The rotor type, surface treatment, gate top shape, counterweight shape, and... Figure 5 The outside air temperature, etc., are shown.

[0072] In step S2, the learning control unit 140 determines at least one mixing condition and a set value for each mixing condition. Here, the mixing condition to be set is... Figures 3 to 6 At least one of the listed mixing conditions has a settable value. Examples of mixing conditions with settable values ​​include... Figures 3 to 6 The mixing conditions illustrated are those other than those obtained as input values ​​in step S1. Here, the determined mixing condition settings are equivalent to actions in reinforcement learning.

[0073] Specifically, the learning control unit 140 randomly selects a setting value for each mixing condition that is the setting target. Here, the setting value is randomly selected from a predetermined range for each mixing condition. For example, the ε-greedy method can be used to select the setting value for the mixing condition.

[0074] In step S3, the learning control unit 140 sends a mixing execution command to the mixing apparatus 300, causing the mixing apparatus 300 to begin the mixing process. If the mixing execution command is received via the communication unit 310, the mixing execution unit 322 sets the mixing conditions according to the mixing execution command, thus initiating the mixing process. The mixing execution command includes input values ​​for the mixing conditions set in step S1 and set values ​​for the mixing conditions determined in step S2.

[0075] If the mixing process is complete, the state observation unit 321 observes the state variables (step S4). Specifically, the state observation unit 321 will... Figure 7 as well as Figure 8the first evaluation parameter and the second evaluation parameter Figures 3 to 6 Among the mixing conditions shown, the mixing conditions that are observation targets are acquired as state variables. The state observation section 321 can acquire the measured values of the various measuring instruments input to the input section 350 and the measured values measured by the sensing section 340 as the mixing conditions, the first evaluation parameter, and the second evaluation parameter. In addition, the first evaluation parameter and the second evaluation parameter can be acquired by causing the mixing device 300 to communicate with the various measuring instruments. Furthermore, the mixing conditions that are observation targets can adopt Figures 3 to 6 the mixing conditions that are predetermined among the mixing conditions shown. Furthermore, the state observation section 321 transmits the acquired state variables to the server 100 via the communication section 310.

[0076] In step S5, the decision section 130 evaluates the first evaluation parameter and the second evaluation parameter. Here, the decision section 130 evaluates the first evaluation parameter and the second evaluation parameter by judging whether the evaluation parameters that become evaluation targets (hereinafter, referred to as target evaluation parameters) among the first evaluation parameter and the second evaluation parameter acquired in step S4 reach a prescribed reference value. The target evaluation parameters are Figure 7 and Figure 8 one or more of the first evaluation parameter and the second evaluation parameter. In the case where the target evaluation parameters are plural, the reference value becomes a plurality of reference values that exist in correspondence with the respective target evaluation parameters. The reference value can adopt, for example, a predetermined value that indicates that the target evaluation parameter reaches a certain reference.

[0077] The reference value can also be a value that includes an upper limit value and a lower limit value. In this case, when the target evaluation parameter enters a range of the upper limit value and the lower limit value, it is judged that the reference value has been reached. The reference value can also be one value. In this case, when the target evaluation parameter exceeds the reference value or is lower than the reference value, it is judged that a certain reference is satisfied.

[0078] The decision section 130, in the case where it is judged that the target evaluation parameter reaches the reference value (YES in step S6), outputs the mixing conditions set in step S2 as the final mixing conditions (step S7). On the other hand, the decision section 130, in the case where it is judged that the target evaluation parameter does not reach the reference value (NO in step S6), causes the process to proceed to step S8. In addition, in the case where the target evaluation parameters are plural, the decision section 130 can judge YES in step S6 in the case where all of the target evaluation parameters reach the reference value.

[0079] At step S8, the reward calculation section 110 determines whether the object evaluation parameter is close to the reference value. In the case where the object evaluation parameter is close to the reference value (YES at step S8), the reward calculation section 110 increases the reward to the agent (step S9). On the other hand, in the case where the object evaluation parameter is not close to the reference value (NO at step S8), the reward calculation section 110 decreases the reward to the agent (step S10). In this case, the reward calculation section 110 can increase or decrease the reward by a predetermined increase or decrease value of the reward. In addition, in the case where the object evaluation parameters are plural, the reward calculation section 110 can perform the determination of step S8 for each object evaluation parameter. In this case, the reward calculation section 110 can increase or decrease the reward for each object evaluation parameter based on the determination result of step S8. Also, the increase or decrease value of the reward can take different values depending on the object evaluation parameter. For example, the increase or decrease value of the reward for the above-mentioned important evaluation parameter among the first evaluation parameter and the second evaluation parameter can be set to be larger than that of the other evaluation parameters.

[0080] At step Sll, the update section 120 updates the action value function using the reward given to the agent. The Q-learning employed in the present embodiment is a method of learning a value, Q value (Q(s, a), when an action a is selected in a certain environment state s. In addition, the environment state s t corresponds to the state variable of the above-described flow. Also, in the Q-learning, an action a is selected in which Q(s, a) is the highest in a certain environment state s. In the Q-learning, by repeating the trial, various actions a in a certain environment state s are acquired, and the correct Q(s, a) is learned using the reward at that time. The action value function Q(s t , a t ) is updated by the following formula (1).

[0081]

[0082] Here, s t , a t respectively indicate the environment state and the action at time t. By the action a t , the environment state changes to s t+1 , and the reward r t+1 is calculated based on the change of the environment state. Also, the term with max is a term in which the Q value (Q(s t+1 , a) is multiplied by the term of γ in which the action a is selected in which the value known at that time is the highest in the environment state s t+1 . Here, γ is a discount rate, and takes a value of 0 < γ ≤ 1 (typically, 0.9 to 0.99). α is a learning coefficient, and takes a value of 0 < α ≤ 1 (typically, about 0.1).

[0083] The updated formula, if γ·maxQ(s) t+1 a) is greater than the Q value of action a in state s, i.e., Q(s) t a t ), which makes Q(s) t a t ) increases, where γ·maxQ(s t+1 a) is based on the choice made by taking action a in the next environmental state s. t+1 The optimal Q-value under the given conditions. On the other hand, the updated formula, if γ·maxQ(s) t+1 a) less than Q(s) t a t ), which makes Q(s) t a t Decrease. That is, make the state s decrease. t The value of a certain action 'a' is close to the value of that action in the next state 's'. t+1 The value of the best action. Therefore, the optimal mixing conditions are determined.

[0084] If the processing in step S11 is completed, the process returns to step S2, where the setting value of the mixing conditions is changed, and the action value function is updated in the same way. Although the update unit 120 updates the action value function, the present invention is not limited to this and can also update the action value table.

[0085] Q(s, a) can also be stored in tabular form for the values ​​of all pairs (s, a) of states and actions. Alternatively, Q(s, a) can be represented by an approximation function that approximates the values ​​of all pairs (s, a) of states and actions. This approximation function can also be constructed using a multi-layered neural network. In this case, the neural network can learn in real time from the data obtained by actually operating the mixing device 300, and perform online learning to reflect the learned data in the next action. Thus, deep reinforcement learning is achieved.

[0086] To date, research has been conducted on mixing conditions, as shown below: obtaining a good mixture by changing the mixing conditions in a mixing unit. To obtain good mixing conditions, it is necessary to determine the correlation between the first evaluation parameter, the second evaluation parameter, and the mixing conditions. However, as... Figures 3 to 6 As shown, due to the vast variety of mixing conditions, a large number of physical models are needed to define such correlations, and describing these correlations using physical models is also difficult. Furthermore, constructing such physical models requires manually identifying which parameter affects the evaluation of which evaluation parameter, making the process extremely challenging.

[0087] According to the present embodiment, the first to sixth parameters, the first evaluation parameter, and the second evaluation parameter described above are observed as state variables. Further, a reward for a decision result of the mixing condition is calculated based on the observed state variables, the action value function for deciding the mixing condition from the state variables is updated based on the calculated reward, and the mixing condition from which the most reward can be obtained is learned by repeating the update. Thus, the present embodiment decides the mixing condition by machine learning without using the physical model described above. As a result, the present embodiment can easily decide the appropriate mixing condition without depending on the experience of skilled technicians over many years.

[0088] In addition, the present application can adopt the following modification.

[0089] (1) Figure 10 is a whole configuration diagram of the machine learning system to which the modification of the present application relates. The machine learning system to which the modification relates is configured only with the mixing device 300A. The mixing device 300A includes the controller 320A, the input section 391, and the sensing section 392. The controller 320A includes the machine learning section 370 and the mixing section 380. The machine learning section 370 includes the reward calculation section 371, the update section 372, the decision section 373, and the learning control section 374. The reward calculation section 371 to the learning control section 374 are respectively the same as the reward calculation section 110 to the learning control section 140 illustrated in Figure 2 . The mixing section 380 includes the state observation section 381 and the mixing execution section 382. The state observation section 381 and the mixing execution section 382 are respectively the same as the state observation section 321 and the mixing execution section 322 illustrated in Figure 2 . The input section 391 and the sensing section 392 are respectively the same as the input section 350 and the sensing section 340 illustrated in Figure 2 . In the modification, the state observation section 381 is one example of the state acquisition section that acquires state information.

[0090] Thus, according to the machine learning system to which the modification relates, the optimal mixing condition can be learned only with the mixing device 300A.

[0091] (2) The flowchart described above observes the state variables after the end of the processing, but this is only one example, and the state variables can be observed multiple times during one processing. For example, in a case where the state variables are constituted only by parameters that can be measured instantaneously, multiple state variables can be observed during one processing. Thereby, the learning time can be shortened.

[0092] Summary of the Embodiment

[0093] Another embodiment of the present application relates to a machine learning method for determining a mixing condition of a mixing device for mixing a polymer material to obtain a mixed product, the mixing device including a chamber into which the material is fed, a rotor of two or more shafts for mixing the material fed into the chamber, and a controller for controlling the rotor of two or more shafts, a mixing time of the material, and an operation step of the mixing device, the machine learning method including the steps of: acquiring a state variable including at least one first evaluation parameter related to performance evaluation of the mixed product and at least one mixing condition; calculating a reward for a determination result of the at least one mixing condition based on the state variable; updating a function for determining the at least one mixing condition based on the state variable based on the reward; and determining a mixing condition that maximally obtains the reward by repeatedly updating the function, wherein the at least one first evaluation parameter includes at least one of a physical property and a shape feature of the mixed product.

[0094] According to this embodiment, at least one mixing condition is acquired as a state variable. Also, a first evaluation parameter including at least one of a physical property and a shape feature of the mixed product is acquired as a state variable.

[0095] Then, a reward for a determination result of the mixing condition is calculated based on the acquired state variable, a function for determining the mixing condition based on the state variable is updated based on the calculated reward, and a mixing condition that maximally obtains the reward is learned by repeatedly performing the update. Therefore, this configuration can easily determine an appropriate mixing condition without depending on years of experience of skilled technicians.

[0096] In the machine learning method, the at least one mixing condition can be at least one of a first parameter related to the material and a second parameter related to control of the rotor.

[0097] According to this configuration, since at least one of a first parameter related to the material of the mixed product and a second parameter related to control of the rotor is acquired as a state variable, a more appropriate mixing condition can be determined considering the first parameter and the second parameter.

[0098] In the machine learning method, the at least one mixing condition can be at least one of a first parameter related to the material, a second parameter related to control of the rotor, and a third parameter related to an operation step, the third parameter being at least one of a mixing time of at least one of the steps, a material feeding time of at least one of the steps, at least one of conditions for moving to the next step, a cumulative mixing time, and a cumulative electric power.

[0099] According to this configuration, since the at least one mixing condition of the first parameter related to the material of the mixture, the second parameter related to the control of the rotor, and the third parameter related to the operation step is acquired as a state variable, a more appropriate mixing condition can be determined considering the first parameter, the second parameter, and the third parameter.

[0100] In the machine learning method, preferably, the mixing device further includes a counterweight, and the at least one mixing condition includes a fourth parameter related to operation of the counterweight.

[0101] According to this configuration, since the fourth parameter related to the counterweight is also acquired as a state variable, a more appropriate mixing condition can be determined.

[0102] In the machine learning method, preferably, the mixing device further includes a temperature adjustment mechanism, and the at least one mixing condition includes a fifth parameter related to temperature adjustment.

[0103] According to this configuration, since the fifth parameter related to the temperature adjustment is also acquired as a state variable, a more appropriate mixing condition can be determined.

[0104] In the machine learning method, preferably, the first parameter includes at least one of a mixing amount of the material, a feeding order of components of the material, and a filling rate of the material in the chamber.

[0105] According to this configuration, since at least one of the mixing amount of the material, the feeding order of components of the material, and the filling rate of the material in the chamber is adopted as the first parameter, a more appropriate mixing condition can be determined.

[0106] In the machine learning method, preferably, the second parameter includes at least one of a rotational speed of the two or more rotors, a phase of the two or more rotors, and a speed ratio of the rotors.

[0107] According to this configuration, since at least one of the rotational speed of the two or more rotors, the phase of the two or more rotors, and the speed ratio of the rotors is adopted as the second parameter, a more appropriate mixing condition can be determined.

[0108] In the machine learning method, preferably, the conditions for entering the next step in the third parameter include at least one of the mixing time for entering the next step, the temperature of the material, the temperature of the material that should be maintained at each step, the instantaneous power of the motor for driving the rotors of the two or more axes, the cumulative power of the motor, the instantaneous current of the motor, the torque of the motor, and the discharge temperature of the material.

[0109] According to this configuration, since at least one of the mixing time at each step, the temperature of the material that should be maintained at each step, the instantaneous power of the motor for driving the rotors of the two or more axes, the cumulative power of the motor, the instantaneous current of the motor, the torque of the motor, and the discharge temperature of the material is adopted as the third parameter as the condition for entering the next step, a more appropriate mixing condition can be determined.

[0110] In the machine learning method, preferably, the fourth parameter includes at least one of the push pressure of the counterweight when the material is pushed into the chamber, the position of the counterweight, and the speed of the counterweight.

[0111] According to this configuration, since at least one of the push pressure of the counterweight when the material is pushed into the chamber, the position of the counterweight, and the speed of the counterweight is adopted as the fourth parameter, a more appropriate mixing condition can be determined.

[0112] In the machine learning method, preferably, the fifth parameter includes at least one of the temperature of the circulating medium entering the chamber, the temperature of the circulating medium coming out of the chamber, the temperature of the circulating medium entering the rotors of the two or more axes, the temperature of the circulating medium coming out of the rotors of the two or more axes, the temperature of the circulating medium entering the door for discharging the material, and the temperature of the circulating medium coming out of the door.

[0113] According to this configuration, since at least one of the temperature of the circulating medium entering the chamber, the temperature of the circulating medium coming out of the chamber, the temperature of the circulating medium entering the rotors of the two or more axes, the temperature of the circulating medium coming out of the rotors of the two or more axes, the temperature of the circulating medium entering the door, and the temperature of the circulating medium coming out of the door is adopted as the fifth parameter, a more appropriate mixing condition can be determined.

[0114] In the machine learning method, preferably, the state variable further includes a second evaluation parameter related to the stability of the operation of the mixing device.

[0115] According to this configuration, since the parameter related to the stability of the operation is included in the state variable, a mixing condition from which a suitable mixed material can be obtained can be obtained while seeking the stability of the operation of the mixing device.

[0116] In the machine learning method, preferably, the physical property includes at least one of Mooney viscosity, cure characteristics, Payne effect, dispersion of additives, dynamic viscoelasticity, hardness, weight of the compound, and temperature of the compound.

[0117] According to this configuration, since at least one of Mooney viscosity, cure characteristics, Payne effect, dispersion of additives, dynamic viscoelasticity, hardness, weight of the compound, and temperature of the compound is adopted as the physical property, the mixing conditions that can obtain the compound satisfying the physical property can be easily obtained.

[0118] In the machine learning method, preferably, the function is updated in real time using deep reinforcement learning.

[0119] According to this configuration, since the function is updated in real time using deep reinforcement learning, the update of the function can be performed correctly and quickly.

[0120] In the machine learning method, preferably, when the at least one first evaluation parameter approaches a predetermined reference value corresponding to each first evaluation parameter, the reward is increased, and when the at least one first evaluation parameter does not approach the reference value corresponding to each first evaluation parameter, the reward is decreased, in calculating the reward.

[0121] According to this configuration, since the reward is increased as the first evaluation parameter approaches the reference value, the first evaluation parameter can be quickly brought to the reference value.

[0122] In addition, in the present application, each process of the machine learning method can be installed to a machine learning device or circulated as a machine learning program. The machine learning device can be configured with a server or a mixing device.

[0123] Another embodiment of the present application relates to a communication method for machine learning of mixing conditions of a mixing device for mixing a polymer material to obtain a compound, the mixing device including: a chamber into which a material for the compound is fed; a two-shaft or more rotor that mixes the material fed into the chamber; and a controller that controls the two-shaft or more rotor, controls a mixing time of the material, and controls an operation step of the mixing device, the communication method including: observing a state variable including at least one first evaluation parameter related to performance evaluation of the compound and at least one mixing condition; and transmitting the state variable to a network, and receiving a mixing condition after machine learning, wherein the at least one first evaluation parameter includes at least one of a physical property and a shape characteristic of the compound.

[0124] According to this configuration, it is possible to provide necessary information required when machine learning of a mixing condition is performed. Such a communication method can be installed to a mixing device.

Claims

1. A machine learning method of causing a machine learning device to determine a mixing condition of a mixing device for mixing a polymer material to obtain a mixed product, wherein The mixing device is provided with: a chamber into which a material is fed; two or more rotors that mix the material fed into the chamber; and a controller that controls the two or more rotors, the mixing time of the material, and the operation steps of the mixing device, characterized in that the machine learning method includes the steps of: acquiring state variables including at least one first evaluation parameter related to performance evaluation of the mixture and at least one mixing condition; calculating a reward for a decision result of the at least one mixing condition based on the state variables; updating a function for deciding the at least one mixing condition from the state variables based on the reward; and deciding a mixing condition that maximally obtains the reward by repeatedly updating the function, wherein the at least one first evaluation parameter includes at least one of a physical property and a shape feature of the mixture, the at least one mixing condition is at least one of a first parameter related to the material, a second parameter related to control of the rotors, and a third parameter related to an operation step, the third parameter being a parameter selected from at least one of a mixing time of at least one step, a material feeding time of at least one step, at least one condition for entering a next step, a cumulative mixing time, and a cumulative electric power among a plurality of steps.

2. The machine learning method according to claim 1, characterized in that the mixing device further includes a counterweight, the at least one mixing condition includes a fourth parameter related to operation of the counterweight.

3. The machine learning method according to claim 1, characterized in that the mixing device further includes a temperature adjustment mechanism, the at least one mixing condition includes a fifth parameter related to temperature adjustment.

4. The machine learning method according to claim 1, characterized in that the first parameter includes at least one of a mixing amount of the material, a feeding order of components of the material, and a filling rate of the material in the chamber.

5. The machine learning method according to claim 1, characterized in that the second parameter includes at least one of a rotational speed of the two or more rotors, a phase of the two or more rotors, and a speed ratio of the rotors.

6. The machine learning method according to claim 1, characterized in that the condition for entering a next step in the third parameter includes at least one of a mixing time for entering a next step, a temperature of the material, a temperature of the material that should be maintained at each step, an instantaneous electric power of a motor for driving the two or more rotors, a cumulative electric power of the motor, an instantaneous current of the motor, a torque of the motor, and a discharge temperature of the material.

7. The machine learning method according to claim 2, characterized in that the fourth parameter includes at least one of a pressing pressure of the counterweight when the material is pushed into the chamber, a position of the counterweight, and a speed of the counterweight.

8. The machine learning method according to claim 3, wherein the fifth parameter includes at least one of a temperature of the circulating medium entering the chamber, a temperature of the circulating medium exiting the chamber, a temperature of the circulating medium entering the two or more rotors, a temperature of the circulating medium exiting the two or more rotors, a temperature of the circulating medium entering a door through which the material is discharged, and a temperature of the circulating medium exiting the door.

9. The machine learning method according to claim 1, wherein the state variable further includes a second evaluation parameter related to stability of the operation of the mixing device.

10. The machine learning method according to claim 1, wherein the physical property includes at least one of Mooney viscosity, cure characteristics, Payne effect, dispersion of additives, dynamic viscoelasticity, hardness, weight of the mixed material, and temperature of the mixed material.

11. The machine learning method according to claim 1, wherein the function is updated in real time using deep reinforcement learning.

12. The machine learning method according to any one of claims 1 to 11, wherein when the at least one first evaluation parameter approaches a prescribed reference value corresponding to each first evaluation parameter, the reward is increased, and when the at least one first evaluation parameter does not approach the reference value corresponding to each first evaluation parameter, the reward is decreased, in calculating the reward.

13. A machine learning device that determines a mixing condition of a mixing device for mixing a polymer material to obtain a mixed material, the mixing device comprising: a chamber into which the material is fed; two or more rotors that mix the material fed into the chamber; and a controller that controls the two or more rotors, controls a mixing time of the material, and controls an operation step of the mixing device, the machine learning device comprising: a state acquisition unit that acquires a state variable including at least one first evaluation parameter related to performance evaluation of the mixed material and at least one mixing condition; a reward calculation unit that calculates a reward of a determination result of the at least one mixing condition based on the state variable; an update unit that updates a function for determining the at least one mixing condition from the state variable based on the reward; and a determination unit that determines a mixing condition that maximizes the reward by repeatedly updating the function, wherein the at least one first evaluation parameter includes at least one of a physical property and a shape characteristic of the mixed material, the at least one mixing condition is at least one of a first parameter related to the material, a second parameter related to control of the rotors, and a third parameter related to an operation step, the third parameter being a parameter selected from at least one of a mixing time of at least one step among a plurality of steps, a material feeding time of at least one step among the steps, at least one condition for a condition for entering a next step, a cumulative mixing time, and a cumulative electric power. ​ ​ ​ characterized in that ​ ​ ​ ​ ​ ​ ​ 14. A computer program product of a machine learning device that determines a mixing condition of a mixing device for mixing a polymer material to obtain a mixed product, the mixing device comprising: a chamber into which a material to be mixed into the mixed product is fed; two or more rotors that mix the material fed into the chamber; and a controller that controls the two or more rotors, a mixing time of the material, and an operation step of the mixing device, the computer program product causing a computer to function as: a state acquisition unit that acquires a state variable including at least one first evaluation parameter related to performance evaluation of the mixed product and at least one mixing condition; a reward calculation unit that calculates a reward for a determination result of the at least one mixing condition based on the state variable; an update unit that updates a function for determining the at least one mixing condition from the state variable based on the reward; and a determination unit that determines a mixing condition that maximizes the reward by repeatedly updating the function, wherein the at least one first evaluation parameter includes at least one of a physical property and a shape feature of the mixed product, the at least one mixing condition is at least one of a first parameter related to the material, a second parameter related to control of the rotors, and a third parameter related to the operation step, the third parameter being a parameter selected from at least one of a mixing time of at least one of the steps, a material feeding time of at least one of the steps, at least one of conditions for moving to a next step, a cumulative mixing time, and a cumulative power amount.

15. A communication method of machine learning a mixing condition of a mixing device for mixing a polymer material to obtain a mixed product, the mixing device comprising: a chamber into which a material to be mixed into the mixed product is fed; two or more rotors that mix the material fed into the chamber; and a controller that controls the two or more rotors, a mixing time of the material, and an operation step of the mixing device, the communication method comprising: observing a state variable including at least one first evaluation parameter related to performance evaluation of the mixed product and at least one mixing condition; and transmitting the state variable to a network, and receiving a machine-learned mixing condition, the machine-learned mixing condition being obtained by calculating a reward for a determination result of the at least one mixing condition based on the state variable, updating a function for determining the at least one mixing condition from the state variable based on the reward, and determining a mixing condition that maximizes the reward by repeatedly updating the function, wherein the at least one first evaluation parameter includes at least one of a physical property and a shape feature of the mixed product. ​ ​ characterized in that ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ characterized in that ​ ​ ​ ​ The at least one mixing condition is at least one of a first parameter related to the material, a second parameter related to the control of the rotor, and a third parameter related to an action step, the third parameter being at least one selected from a mixing time of at least one of the steps, a material feeding time of at least one of the steps, at least one of the conditions for entering the next step, a cumulative mixing time, and a cumulative power amount.

16. A kneader that kneads a polymer material to obtain a kneaded product, characterized by comprising Comprises: a chamber into which a material to be mixed is fed; a two- or more shaft rotor that mixes the material fed into the chamber; a controller that controls the two- or more shaft rotor, controls the mixing time of the material, and controls the action steps of the mixing device; a state observation unit that acquires state variables including at least one first evaluation parameter related to the performance evaluation of the mixed material and at least one mixing condition; and a communication unit that transmits the state variables to a network, receives a machine-learned mixing condition, the machine-learned mixing condition being a result of determining a reward for the at least one mixing condition based on the state variables, updating a function for determining the at least one mixing condition based on the state variables based on the reward, and determining a mixing condition that maximally obtains the reward by repeatedly updating the function, the at least one first evaluation parameter including at least one of a physical property and a shape feature of the mixed material, the at least one mixing condition is at least one of a first parameter related to the material, a second parameter related to the control of the rotor, and a third parameter related to an action step, the third parameter being at least one selected from a mixing time of at least one of the steps, a material feeding time of at least one of the steps, at least one of the conditions for entering the next step, a cumulative mixing time, and a cumulative power amount.

Images