Electrostatic sorting system, electrostatic sorting method, and learning device

Abstract

To provide an electrostatic sorting system capable of performing appropriate sorting.SOLUTION: An electrostatic sorting system of the present disclosure includes a charging unit that charges a sorting target material, a data acquisition unit that acquires feature information indicating a feature of the sorting target material, an inference unit that infers a sorting condition from the feature information acquired by the data acquisition unit using a learned model for inferring the sorting condition for sorting the sorting target material charged by the charging unit from the feature information, and an electrostatic sorting unit that electrostatically sorts the sorting target material charged by the charging unit on the basis of the sorting condition.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to an electrostatic separation system, an electrostatic separation method, and a learning device. [Background technology]

[0002] BACKGROUND ART Conventionally, there has been known an electrostatic sorting device that frictionally charges a plurality of types of materials to be sorted, and then sorts the materials in an electrostatic field.

[0003] The electrostatic separation device disclosed in Patent Document 1 detects when the charge amount of resin pieces (hereinafter, material to be separated) that have been charged before separation in an electrostatic field differs from the normal tendency of charge amount. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2023 / 187854 Summary of the Invention [Problem to be solved by the invention]

[0005] The results of electrostatic sorting can be affected by a variety of factors, such as the characteristics of the materials being sorted and the sorting conditions. By comprehensively considering these various factors, more appropriate sorting may be possible. [Means for solving the problem]

[0006] In order to solve the above problems, the electrostatic sorting system disclosed herein comprises a charging unit that charges the material to be sorted, a data acquisition unit that acquires characteristic information that indicates the characteristics of the material to be sorted, an inference unit that infers sorting conditions from the characteristic information acquired by the data acquisition unit using a trained model for inferring sorting conditions for sorting the material to be sorted that has been charged by the charging unit from the characteristic information, and an electrostatic sorting unit that electrostatically sorts the material to be sorted that has been charged by the charging unit based on the sorting conditions.

[0007] The electrostatic sorting method of the present disclosure includes a first step of charging the material to be sorted by a charging unit; The method includes a second step in which the data acquisition unit acquires characteristic information indicating the characteristics of the material to be sorted, a third step in which, after the first and second steps are completed, the inference unit infers sorting conditions from the characteristic information acquired by the data acquisition unit using a trained model for inferring sorting conditions for sorting the material to be sorted that has been charged by the charging unit from the characteristic information, and a fourth step in which, after the third step is completed, the electrostatic sorting unit electrostatically sorts the material to be sorted that has been charged by the charging unit based on the sorting conditions.

[0008] In addition, the learning device disclosed herein is an electrostatic sorting device having an electrostatic separation unit that has a charging unit that charges the material to be sorted, an electrostatic separation unit that has a plurality of electrodes to which a voltage is applied and that attracts the material to be sorted charged by the charging unit toward an electrode that corresponds to the polarity of the material to be sorted, and a partition unit that is provided below the electrostatic separation unit and that electrostatically sorts the material to be sorted that has been charged by the charging unit based on sorting conditions, and is equipped with: a data acquisition unit that acquires learning data including characteristic information that indicates the characteristics of the material to be sorted and the sorting results of the material to be sorted in the electrostatic sorting unit; and a model generation unit that uses the learning data to generate a trained model for inferring the sorting conditions from the characteristic information, and the sorting results include the probability that the material to be sorted has come into contact with the electrode in the electrostatic separation unit. [Effects of the Invention]

[0009] The electrostatic sorting system, electrostatic sorting method, and learning device according to the present disclosure have the effect of enabling more appropriate sorting. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a block diagram illustrating a configuration of an electrostatic separation system according to an embodiment of the present disclosure. [Figure 2] 1 is a configuration diagram of an electrostatic separation device and a detection device in an electrostatic separation system according to an embodiment of the present disclosure. [Figure 3] FIG. 1 is a configuration diagram of a learning device in an electrostatic separation system according to an embodiment of the present disclosure. [Figure 4] 10 is a flowchart illustrating processing by a learning device of an electrostatic separation system according to an embodiment of the present disclosure. [Figure 5] FIG. 1 is a configuration diagram of an inference device of an electrostatic separation system according to an embodiment of the present disclosure. [Figure 6] 10 is a flowchart illustrating processing by an inference device of an electrostatic separation system according to an embodiment of the present disclosure. [Figure 7] FIG. 10 is a configuration diagram of an electrostatic separation device and a detection device in an electrostatic separation system according to a modified example of an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. Note that the present disclosure is not limited to the following embodiments, and modifications or omissions are possible without departing from the spirit of the present disclosure. Furthermore, common elements in each drawing are designated by the same reference numerals, and redundant explanations will be omitted.

[0012] Embodiment Fig. 1 is a block diagram showing a configuration of an electrostatic separation system 100 according to an embodiment of the present disclosure. Fig. 2 is a configuration diagram of an electrostatic separation device 10 and a detection device 20 in the electrostatic separation system 100 according to an embodiment of the present disclosure. The configuration of the electrostatic separation system 100 will be described with reference to Figs. 1 and 2.

[0013] In addition, although the up and down directions of the electrostatic separation system 100 are defined in FIG. 2, they are defined for the purpose of explaining the embodiment and do not limit the arrangement and orientation of the devices and components of the present disclosure.

[0014] The electrostatic separation system 100 separates a group of resin pieces into individual types. The electrostatic separation system 100 includes an electrostatic separation device 10, a detection device 20, a learning device 30, a storage device 40, an inference device 50, a temperature and humidity adjustment device 60, and a control device 70. The electrostatic separation device 10, the detection device 20, the learning device 30, the storage device 40, the inference device 50, the temperature and humidity adjustment device 60, and the control device 70 each have a communication interface such as a network interface card (NIC) or a wireless communication module for connecting to a network. Examples of the network include a LAN, a WAN, the Internet, Bluetooth (registered trademark), a dedicated circuit, and infrared communication.

[0015] The electrostatic separation device 10 uses static electricity to separate a group of resin pieces containing multiple types of resin pieces. As shown in Figure 2, the electrostatic separation device 10 has a charging unit 110, a conveying unit 120, an electrostatic separation unit 130, and a storage unit 140.

[0016] The charging unit 110 charges the resin pieces contained in the resin piece group. Specifically, the charging unit 110 is a charging cylinder that agitates the resin piece group by rotating. In the charging unit 110, the resin pieces contained in the resin piece group rub against each other, and each resin piece is charged with a polarity and charge amount according to the frictional charging order.

[0017] The triboelectric order is a ranking in which materials that tend to become positively charged when different materials are rubbed together are placed at the top and materials that tend to become negatively charged are placed at the bottom. For example, the triboelectric order of plastics, from materials that tend to become positively charged to materials that tend to become negatively charged, is ABS (Acrylonitrile Butadiene Styrene), PS (Polystyrene), PP (Polypropylene), PET (Polyethylene Terephthalate), and PVC (Polyvinyl Chloride).

[0018] In the electrostatic sorting device 10 according to the embodiment, plastic pieces A and plastic pieces B are selected from among the resin pieces contained in the group of resin pieces. Plastic piece A is positively charged and plastic piece B is negatively charged by the charging unit 110. For example, plastic piece A is an ABS piece and plastic piece B is a PS piece. In FIG. 2, plastic piece A is shown in white and plastic piece B is shown in black.

[0019] The conveying unit 120 conveys the resin pieces charged by the charging unit 110 to the electrostatic separation unit 130. Specifically, the conveying unit 120 is a vibrating feeder that uses vibration to separate the resin pieces contained in the group of resin pieces charged by the charging unit 110, and moves them in a predetermined direction to cause them to fall.

[0020] The electrostatic sorting unit 130 electrostatically sorts the resin pieces transported from the transport unit 120. The electrostatic sorting unit 130 has an electrostatic separating unit 131 and a partition unit 135.

[0021] The electrostatic separator 131 includes an electrode unit 132 and a DC power supply 133. The electrode unit 132 is a pair of electrodes including a first electrode 132a and a second electrode 132b. The first electrode 132a and the second electrode 132b are disposed opposite to each other.

[0022] The DC power supply 133 applies a DC voltage between the first electrode 132 a and the second electrode 132 b, thereby generating an electrostatic field between the first electrode 132 a and the second electrode 132 b. The value of the DC voltage applied to the DC power supply 133 is controlled by a control device 70, which will be described later.

[0023] The resin pieces conveyed from conveying section 120 are attracted to first electrode 132a or second electrode 132b by electrostatic force depending on their charge state and fall. That is, the resin pieces fall to different positions depending on their charge state. In the embodiment, plastic piece A is positively charged and therefore is attracted to first electrode 132a and falls. Plastic piece B is negatively charged and therefore is attracted to second electrode 132b and falls.

[0024] The partition 135 is provided below the electrostatic separator 131 and to separate the storage section 140, which will be described later. The partition 135 is provided to appropriately store the resin pieces that drop from the electrostatic separator 131 in the storage section 140. In this embodiment, the partition 135 has a first partition 135a and a second partition 135b.

[0025] The first partitioning section 135a has a rotating shaft section 136 and a rotating section 137. The rotating shaft section 136 is fixed to the floor surface of the storage section 140. The rotating section 137 rotates around the rotating shaft section 136 in the direction of the arrow shown in FIG. 2. Similarly to the first partitioning section 135a, the second partitioning section 135b also has a rotating shaft section 138 and a rotating section 139 that rotates in the direction of the arrow. The rotation angles of the rotating section 137 and the rotating section 139 are controlled by the control device 70, which will be described later.

[0026] First partitioning unit 135a and second partitioning unit 135b can change the size or location of the space above storage unit 140 by rotating rotating unit 137 and rotating unit 139. In other words, first partitioning unit 135a and second partitioning unit 135b can change the correspondence between the falling position of the resin pieces contained in the resin piece group and storage unit 140 by rotating. Therefore, by changing the angle of rotating unit 137 or the angle of rotating unit 139, electrostatic separation system 100 can efficiently separate plastic pieces A and plastic pieces B from the resin piece group.

[0027] Because first partition 135a and second partition 135b are configured to rotate, they can be configured more simply than partitions that move horizontally. First partition 135a and second partition 135b also have the advantage of reducing friction with the surface on which they are installed. In this embodiment, partition 135 is configured to separate storage section 140, and rotation shafts 136 and 138 are fixed to the floor of storage section 140. Therefore, compared to a case where first partition 135a and second partition 135b move horizontally, they have the advantage of not being hindered from rotating by the stored material even when a large amount of resin pieces is stored in storage section 140.

[0028] The storage unit 140 is provided below the electrostatic separation unit 130 and stores the resin pieces separated by the electrostatic separation unit 130. The storage unit 140 has a first electrode side storage unit 141, a middle storage unit 142, and a second electrode side storage unit 143. The first electrode side storage unit 141 and the middle storage unit 142 are separated by a first partition 135a. The middle storage unit 142 and the second electrode side storage unit 143 are separated by a second partition 135b.

[0029] In this embodiment, first electrode side storage section 141 stores a large number of resin pieces attracted to first electrode 132a, and therefore contains a large number of plastic pieces A. Second electrode side storage section 143 stores a large number of resin pieces attracted to second electrode 132b, and therefore contains a large number of plastic pieces B. Middle storage section 142 contains a mixture of plastic pieces A and plastic pieces B.

[0030] The detection device 20 acquires pre-shaping characteristic information indicating the characteristics of the resin piece group, sorting status information indicating the sorting status of the resin piece group by the electrostatic sorting device 10, and environmental information which is information regarding the environment of the electrostatic separation unit 131. The detection device 20 has a pre-shaping characteristic information detection unit 21 that detects the pre-shaping characteristic information, a sorting status detection unit 22 that detects the sorting status information, and an environmental information detection unit 23 that detects the environmental information.

[0031] The pre-shaping characteristic information detection unit 21 is provided near the entrance of the charging unit 110 and near the exit of the charging unit 110. Specifically, the pre-shaping characteristic information detection unit 21 is a flow sensor and an infrared sensor provided near the entrance of the charging unit 110, and a charge amount sensor and a weight sensor provided near the exit of the charging unit 110. The pre-shaping characteristic information detection unit 21 uses the flow sensor to detect the amount of resin pieces fed into the charging unit 110 per unit time. The pre-shaping characteristic information detection unit 21 also uses the infrared sensor to detect the composition ratio of the resin pieces contained in the resin piece group. The pre-shaping characteristic information detection unit 21 also uses the charge amount sensor and the weight sensor to detect the charge amount and weight of the resin pieces contained in the resin piece group.

[0032] The pre-shaping characteristic information detection unit 21 transmits the pre-shaping characteristic information to the learning device 30 and the inference device 50. More specifically, the pre-shaping characteristic information detection unit 21 transmits information indicating the amount of resin piece group fed into the charging unit per unit time, information indicating the composition ratio of the resin pieces contained in the resin piece group, and information indicating the charge amount and weight of the resin pieces contained in the resin piece group.

[0033] Specifically, the sorting status detection unit 22 is a contact sensor provided in the electrode unit 132, and an infrared sensor provided in each of the first electrode-side storage unit 141 and the second electrode-side storage unit 143. The sorting status detection unit 22 uses the contact sensor to detect the number of times the resin pieces come into contact with the electrode unit 132. The sorting status detection unit 22 also uses the infrared sensor to detect substances contained in the storage items stored in the first electrode-side storage unit 141 and the second electrode-side storage unit 143. More specifically, the sorting status detection unit 22 detects the amount of plastic pieces A and the amount of plastic pieces B contained in the storage items stored in the first electrode-side storage unit 141 and the second electrode-side storage unit 143.

[0034] The sorting status detection unit 22 transmits to the learning device 30 information indicating the number of times that the resin pieces contained in the resin piece group come into contact with the electrode unit 132, and information indicating the substances contained in the storage materials stored in the first electrode side storage unit 141 and the second electrode side storage unit 143.

[0035] Specifically, the environmental information detection unit 23 is a hygrometer and a thermometer provided around the electrostatic separation unit 130. The environmental information detection unit 23 uses the hygrometer and the thermometer to detect the humidity and temperature around the electrostatic separation unit 130. The environmental information detection unit 23 transmits information indicating the humidity and temperature around the electrostatic separation unit 130 to the learning device 30 and the inference device 50. Note that the environmental information detection unit 23 may be provided in the electrostatic separation unit 130.

[0036] The learning device 30 generates a trained model for inferring the sorting conditions in the electrostatic sorting device 10, using the feature information, environmental information, and sorting result information as trained data. The feature information is information that has been shaped from the pre-shaping feature information detected by the detection device 20 so that it can be used as learning data for the learning device 30. Details of the feature information and the sorting result information will be described later. A detailed configuration of the learning device 30 will be described later. The sorting conditions are conditions for electrostatically sorting the material to be sorted. In the embodiment, the sorting conditions include the combination of the voltage value applied to the electrode unit 132 and the rotation angle of the rotating units 137 and 139 of the partition unit 135, as well as the temperature and humidity around the electrostatic sorting unit 130.

[0037] The storage device 40 stores the trained model generated by the learning device 30. The storage device 40 may also store information related to various processes in the electrostatic separation system 100 (such as information that serves as input to the processes, information generated during the processes, and information indicating the results of the processes). For example, the storage device 40 can store various pieces of information used in the process by the learning device 30 to generate a trained model and in the process by the inference device 50 to infer sorting conditions. Note that the storage device 40 does not have to be provided as a component of the electrostatic separation system 100, but may be provided as an external device accessible to the electrostatic separation system 100.

[0038] The inference device 50 infers the selection conditions from the feature information and the environmental information using the trained model stored in the storage device 40. The detailed configuration of the inference device 50 will be described later.

[0039] The temperature and humidity adjusting device 60 adjusts the temperature and humidity around the electrostatic separation unit 130. The temperature and humidity adjusting device 60 has a heating function, a cooling function, a dehumidifying function, and a humidifying function. The temperature and humidity adjusting device 60 is controlled by a control device 70, which will be described later.

[0040] The control device 70 receives the sorting conditions inferred by the inference device 50 and controls the electrostatic separation device 10 and the temperature and humidity adjustment device 60. Specifically, the control device 70 transmits to the electrostatic separation device 10 a voltage value to be applied to the electrode unit 132 and a control signal for controlling the rotation angle of the rotating unit 137 and the rotating unit 139 of the partition unit 135. The control device 70 also transmits a control signal for controlling the temperature and humidity to the temperature and humidity adjustment device 60. The control device 70 is, for example, a hardware configuration having a processor and a memory.

[0041] 3 is a configuration diagram of the learning device 30 in the electrostatic separation system 100 according to an embodiment of the present disclosure. A detailed configuration of the learning device 30 in the electrostatic separation system 100 will be described with reference to FIG.

[0042] The learning device 30 includes a data acquisition unit 31, a data shaping unit 32, and a model generation unit 33. The data acquisition unit 31 acquires pre-shaping feature information, sorting status information, and environmental information from the detection device 20.

[0043] The data shaping unit 32 shapes the data output from the data acquisition unit 31 as learning data. Specifically, the data shaping unit 32 calculates the specific charge of the resin pieces from information indicating the charge amount and weight of the resin pieces contained in the resin piece group, which is the pre-shaping characteristic information. In other words, the characteristic information used as learning data is the specific charge of the resin pieces contained in the resin piece group, the composition of the resin pieces contained in the resin piece group, and the amount of resin pieces fed into the charging unit 110 per unit time.

[0044] Furthermore, the data shaping unit 32 calculates a sorting result from the pre-shaping characteristic information and the sorting status information. In the embodiment, the sorting result is the probability that a resin piece contained in the resin piece group will come into contact with the electrode portion 132, and the purity and recovery rate of the resin pieces contained in the resin piece group. The data shaping unit 32 calculates the probability that a resin piece contained in the resin piece group will come into contact with the electrode portion 132 from the amount of resin pieces input and the number of times the resin pieces come into contact with the electrode portion 132.

[0045] In addition, the data shaping unit 32 calculates the purity of the resin pieces by determining the proportion of plastic pieces A among the stored materials stored in the first electrode side storage unit 141 and the proportion of plastic pieces B among the stored materials stored in the second electrode side storage unit 143.

[0046] Furthermore, the data shaping unit 32 calculates the amount of plastic piece A and the amount of plastic piece B contained in the resin piece group from the composition of the resin pieces contained in the resin piece group. The data shaping unit 32 calculates the recovery rate of the resin pieces by finding the ratio of the amount of plastic piece A stored in the first electrode side storage unit 141 to the amount of plastic piece A contained in the resin piece group, and the ratio of the amount of plastic piece B stored in the second electrode side storage unit 143 to the amount of plastic piece B contained in the resin piece group.

[0047] The data reforming unit 32 transmits the characteristic information, the environmental information, and the selection result information indicating the selection result to the model generation unit 33.

[0048] The model generation unit 33 learns sorting conditions based on training data created based on a combination of feature information, environmental information, and sorting result information received from the data reforming unit 32. That is, it generates a trained model that infers optimal sorting conditions from the feature information, environmental information, and sorting results of the electrostatic sorting system 100. Here, the training data is data in which the feature information, environmental information, and sorting results are associated with each other. The model generation unit 33 also transmits the generated trained model to the storage device 40.

[0049] Factors that affect the sorting of resin pieces contained in a resin piece group include the characteristics of the resin pieces and the environment surrounding the electrostatic separation unit 131. The electrostatic sorting system 100 is configured to infer sorting conditions from characteristic information and environmental information, thereby achieving the effect of being able to infer sorting conditions with higher accuracy than a configuration in which sorting conditions are inferred from characteristic information alone.

[0050] Furthermore, the learning device 30 uses the temperature and humidity around the electrostatic separation unit 130 as environmental information. If the temperature or humidity in the electrostatic separation unit 130 is too high, the resistance of the resin surface decreases, resulting in a decrease in separation efficiency. However, if the temperature and humidity in the electrostatic separation unit 130 are too low, the resin pieces contained in the resin piece group stick to the electrodes, causing back ionization discharge, resulting in a decrease in separation efficiency. Therefore, the temperature and humidity in the electrostatic separation unit 130 are correlated with separation efficiency, and therefore need to be controlled within an appropriate range.

[0051] Furthermore, the learning device 30 uses specific charge as characteristic information. In electrostatic sorting of resin pieces, the charge per mass of the resin pieces has a stronger correlation with the drop distribution than the charge amount of the resin pieces contained in the resin pieces. Therefore, by using specific charge as characteristic information, the electrostatic sorting system 100 has the effect of being able to infer sorting conditions with higher accuracy.

[0052] Furthermore, in addition to the specific charge, the learning device 30 uses the composition of the resin pieces contained in the resin piece group and the amount of resin pieces fed per unit time to the charging unit 110 as feature information. By using the feature information of multiple resin pieces, the electrostatic sorting system 100 has the effect of being able to infer sorting conditions with higher accuracy.

[0053] Furthermore, the learning device 30 uses the probability that a resin piece will come into contact with the electrode unit 132 as the sorting result information. If the charge amount of the resin pieces contained in the resin piece group is too large, the resin piece will collide with the electrode unit 132, thereby reducing the sorting accuracy. Therefore, the electrostatic sorting system 100 has the effect of including the probability that a resin piece contained in the resin piece group will come into contact with the electrode unit 132 as the sorting result information, thereby making it possible to infer the sorting conditions by taking into account the possibility that the sorting accuracy is reduced because the charge amount of the resin pieces is too large.

[0054] Furthermore, the learning device 30 uses the purity and recovery rate of the resin pieces contained in the resin piece group as the sorting result information. By using both the purity and recovery rate as the sorting result information, the electrostatic sorting system 100 has the effect of reducing the possibility that impurities other than the sorting target are stored in the storage section, and increasing the possibility that the sorting target material is stored in the desired storage section.

[0055] Furthermore, the learning device 30 infers, as a sorting condition, a combination of the voltage value applied to the electrode unit 132 and the rotation angle of the rotating unit 137 and rotating unit 139 of the partition unit 135. Both the voltage value applied to the electrode unit 132 and the rotation angle of the partition unit 135 affect the sorting results of the resin pieces. More specifically, the voltage value applied to the electrode unit 132 changes the distribution of the resin pieces as they fall after electrostatic separation. Furthermore, the rotation angle of the partition unit 135 changes the range in which the resin pieces that fall after electrostatic separation are stored in the first electrode side storage unit 141 or the second electrode side storage unit 143. Therefore, by changing the rotation angle of the partition unit 135 in accordance with the voltage value applied to the electrode unit 132, efficient sorting can be achieved. The electrostatic sorting system 100 has the effect of increasing sorting efficiency by outputting an appropriate combination of the voltage value applied to the electrode section 132 and the rotation angle of the rotating section 137 and rotating section 139 of the partition section 135.

[0056] Furthermore, the learning device 30 infers, as sorting conditions, the temperature and humidity around the electrostatic sorting unit 130. With this configuration, the electrostatic sorting system 100 can appropriately control the environment around the electrostatic sorting unit 130, thereby achieving the effect of improving sorting efficiency.

[0057] The learning algorithm used by the model generation unit 33 may be a known algorithm such as supervised learning, unsupervised learning, or reinforcement learning. As an example, a case where reinforcement learning is applied will be described. In reinforcement learning, an agent (subject of action) in a certain environment observes the current state (environmental parameters) and decides on an action to take. The environment changes dynamically depending on the agent's actions, and the agent is given a reward according to the environmental changes. The agent repeats this process and learns the course of action that will obtain the most reward through a series of actions. Q-learning and TD-learning are known as representative reinforcement learning methods. For example, in the case of Q-learning, a general update formula for the action value function Q(s, a) is expressed by Equation 1.

[0058]

number

[0059] In equation 1, s t represents the state of the environment at time t, and a t represents the action at time t. Action a t Therefore, the state is s t+1 Changes to r t+1 represents the reward that can be obtained depending on the change in state, γ represents the discount rate, and α represents the learning coefficient. Note that γ is in the range of 0<γ≦1, and α is in the range of 0<α≦1. When feature information and environmental information are used, behavior a t The selection condition is state s t and the state s at time t t Best Practices in a t Learn.

[0060] The update formula expressed by Equation 1 increases the action value Q if the action value Q of the action a with the highest Q value at time t+1 is greater than the action value Q of the action a executed at time t, and decreases the action value Q in the opposite case. In other words, the action value function Q(s, a) is updated so that the action value Q of the action a at time t approaches the best action value at time t+1. As a result, the best action value in a certain environment is propagated sequentially to the action value in the previous environment.

[0061] As described above, when generating a trained model by reinforcement learning, the model generation unit 33 includes a reward calculation unit 301 and a function update unit 302.

[0062] The reward calculation unit 301 calculates a reward based on the characteristic information, the environmental information, and the sorting result information. The reward calculation unit 301 calculates a reward r based on a sorting reward standard. For example, if the purity and recovery rate of the resin pieces contained in the resin piece group are high, or if the probability that the resin pieces contained in the resin piece group will come into contact with the electrode unit 132 is low, the reward r is increased (for example, a reward of "1"); on the other hand, if the purity and recovery rate of the resin pieces contained in the resin piece group are low, or if the probability that the resin pieces will come into contact with the electrode unit 132 is high, the reward r is decreased (for example, a reward of "-1");

[0063] The function update unit 302 updates the function for determining the selection condition according to the reward calculated by the reward calculation unit 301, and outputs the updated function to the storage device 40. For example, in the case of Q-learning, the action value function Q(s t ,a t ) is used as a function to calculate the selection conditions.

[0064] The model generation unit 33 repeatedly executes the above-described learning. The storage device 40 stores the action-value function Q(s t ,a t ), i.e., stores the trained model.

[0065] 4 is a flowchart showing the processing of the learning device 30 of the electrostatic separation system 100 according to an embodiment of the present disclosure. The processing executed by the learning device 30 will be described with reference to FIG.

[0066] In step S31, the data acquisition unit 31 acquires pre-shaping characteristic information, selection status information, and environmental information. Note that in step S31, the pre-shaping characteristic information, selection status information, and environmental information are acquired simultaneously, but it is sufficient that the pre-shaping characteristic information, selection status information, and environmental information are input in association with each other, and each piece of data may be acquired at a different time. In step S31, the processing ends when the data acquisition unit 31 acquires the pre-shaping characteristic information, selection status information, and environmental information.

[0067] Step S32 is performed after the processing of step S31. In step S32, the data shaping unit 32 shapes the data acquired in step S31. Specifically, the data shaping unit 32 calculates the specific charge of the resin pieces contained in the resin piece group as characteristic information from the pre-shaping characteristic information acquired in step S31. In addition, the data shaping unit 32 calculates sorting result information from the pre-sorting characteristic information and sorting status information acquired in step S31. In step S32, the processing ends when the data shaping unit 32 has shaped the data.

[0068] Step S33 is performed after the processing of step S32. In step S33, the model generation unit 33 calculates a reward based on the characteristic information, environmental information, and selection result information shaped in step S32. Specifically, the reward calculation unit 301 acquires the characteristic information, environmental information, and selection result information, and determines whether the criteria for increasing the reward are met based on predetermined selection reward criteria. In step S33, the model generation unit 33 determines whether the conditions are met, and the processing ends.

[0069] Step S34 is processed when it is determined in step S33 that the criteria for increasing the reward are met (step S33, Yes). In step S34, reward calculation unit 301 of model generation unit 33 increases the reward. In step S34, the processing ends when model generation unit 33 increases the reward.

[0070] Step S35 is processed when it is determined in step S33 that the criteria for increasing the reward are not met (step S33, No). In step S35, reward calculation unit 301 of model generation unit 33 reduces the reward. In step S35, the processing ends when model generation unit 33 reduces the reward.

[0071] Step S36 is performed after the processing of step S34 or after the processing of step S35. In step S36, the model generation unit 33 updates the action value function. Specifically, the function update unit 302 updates the action value function Q(s t ,a t ) is updated. In step S36, the model generation unit 33 updates the action-value function, and the process ends.

[0072] The learning device 30 repeatedly executes the above steps S31 to S36, and stores the generated action-value function Q(st, at) as a learned model.

[0073] 5 is a configuration diagram of the inference device 50 of the electrostatic separation system 100 according to an embodiment of the present disclosure. The configuration of the inference device 50 will be described with reference to FIG.

[0074] The inference device 50 includes a data acquisition unit 51, a data shaping unit 52, and an inference unit 53. The data acquisition unit 51 acquires pre-shaping characteristic information and environmental information. The data shaping unit 52 calculates the specific charge of the resin pieces from information indicating the charge amount and weight of the resin pieces included in the resin piece group, which is the pre-shaping characteristic information acquired by the data acquisition unit 51. The data shaping unit 52 also transmits the shaped data to the inference unit 53.

[0075] The inference unit 53 infers the selection conditions obtained using the trained model. That is, by inputting feature information and environmental information into this trained model, it is possible to output the selection conditions inferred from the feature information and environmental information. The inference unit 53 also transmits information indicating the inferred selection conditions to the control device 70.

[0076] 6 is a flowchart showing the processing of the inference device 50 of the electrostatic separation system 100 according to an embodiment of the present disclosure. The processing performed by the inference device 50 will be described with reference to FIG.

[0077] In step S51, the data acquisition unit 51 acquires pre-shaping characteristic information and environmental information. In step S51, when the data acquisition unit 51 acquires the pre-shaping characteristic information and environmental information, the process ends.

[0078] Step S52 is performed after the processing of step S51. In step S52, the data shaping unit 52 shapes the data acquired in step S51. Specifically, the data shaping unit 52 calculates the specific charge of the resin pieces contained in the resin piece group as characteristic information from the pre-shaping characteristic information acquired in step S51. In step S52, the processing ends when the data shaping unit 52 has shaped the data.

[0079] Step S53 is performed after the processing of step S52. In step S53, the inference unit 53 acquires a trained model from the storage device 40 and infers a selection condition by inputting the feature information and environmental information shaped in step S52 into the acquired trained model. In step S53, the processing ends when the inference unit 53 infers the selection condition.

[0080] Step S54 is performed after the processing of step S53. In step S54, the inference unit 53 transmits a signal indicating the selection conditions inferred in step S53 to the control device 70. In step S54, the processing ends when the inference unit 53 transmits the signal indicating the selection conditions. When step S54 ends, the inference processing of the inference device 50 ends.

[0081] The control device 70 uses the sorting conditions received in step S54 to control the electrostatic separation device 10 and the temperature and humidity adjustment device 60. With this configuration, the electrostatic separation system 100 has the effect of being able to perform separation according to the characteristics of the material to be sorted.

[0082] Variations of the embodiment. Next, an electrostatic separation system according to a modified example of the embodiment will be described. The electrostatic separation device according to the modified example of the embodiment is different from the electrostatic separation device 100 according to the embodiment in the position of the partition unit 155. The configuration of the electrostatic separation system 200, excluding the position of the partition unit, is the same as that of the electrostatic separation system 100 according to the embodiment, and therefore a description thereof will be omitted.

[0083] 7 is a configuration diagram of an electrostatic separation device 80 and a detection device 20 in an electrostatic separation system 200 according to a modified embodiment of the present disclosure. As shown in Fig. 7, the electrostatic separation device 80 includes a storage section divider 210. The storage section divider 210 includes a first storage section divider 211 that separates the first electrode side storage section 141 from the middle storage section 143, and a second storage section divider 212 that separates the second electrode side storage section 142 from the middle storage section 143.

[0084] The partition unit 155 also has a first partition unit 156 and a second partition unit 157. Like the partition unit 135 of the embodiment, the first partition unit 156 and the second partition unit 157 have a rotation shaft unit and a rotation unit, and rotate in the direction of the arrow in FIG. 7. The first partition unit 156 is located above the first storage unit partition 211, and the second partition unit 157 is located above the second storage unit partition 212. That is, in this modification of the embodiment, the partition unit 155 is provided below the electrostatic separator 131, but is not configured to separate the storage unit 140, but is provided on a member that separates the storage unit 140. The partition unit may be any unit that guides resin pieces that fall from the electrostatic separator 131.

[0085] In the embodiment, the configuration is such that plastic A and plastic B contained in a group of resin pieces are sorted, but this is not limiting. The configuration may be such that one type of plastic is sorted, or such that three or more types of plastic are sorted. Furthermore, the material to be sorted is not limited to resin pieces. For example, the configuration may be such that a specific object is sorted from materials to be sorted that include metals, or such that a specific object is sorted from materials to be sorted that include food.

[0086] In the embodiment, a case where reinforcement learning is applied to the learning algorithm used by the model generation unit 33 has been described, but the present invention is not limited to this. As for the learning algorithm, other than reinforcement learning, supervised learning, unsupervised learning, semi-supervised learning, or the like can also be applied.

[0087] Furthermore, the learning algorithm used in the model generation unit 33 may be deep learning, which learns to extract the features themselves, or machine learning may be performed according to other known methods, such as neural networks, genetic programming, inductive logic programming, and support vector machines.

[0088] The model generation unit 33 may also learn the sorting conditions using learning data acquired from multiple electrostatic separation systems 100. The model generation unit 33 may acquire learning data from multiple electrostatic separation systems 100 used in the same area, or may learn the sorting conditions using learning data collected from multiple electrostatic separation systems 100 operating independently in different areas. It is also possible to add or remove electrostatic separation systems from which learning data is collected during the process. Furthermore, the learning device 30 that has learned the sorting conditions for a certain electrostatic separation system 100 may be applied to another electrostatic separation system, and the sorting conditions for the other electrostatic separation system may be re-learned and updated.

[0089] Furthermore, in the electrostatic separation system 100 according to the embodiment, the detection device 20 is a separate device from the electrostatic separation device 10, but the detection device 20 may be configured to be built into the electrostatic separation device 10. Also, the detection device 20 may not be included as a component of the electrostatic separation system 100. The electrostatic separation system 100 may be configured in such a way that the learning device 30 and the inference device 50 can acquire the necessary information.

[0090] Furthermore, in the electrostatic separation system 100 according to the embodiment, the learning device 30, the storage device 40, and the inference device 50 are configured as separate devices, but this is not limiting. The learning device 30, the storage device 40, and the inference device 50 may be configured as a single device. The learning device 30, the storage device 40, or the inference device 50 may be built into the electrostatic separation device 10. Furthermore, the learning device 30 and the inference device 50 may be configured to exist on a cloud server.

[0091] Furthermore, the shape of the electrodes of the electrostatic separation system 100 according to the embodiment is not limited. The electrode unit 132 may be in a flat plate shape or a ring shape. Furthermore, in the electrostatic separation system 100 according to the embodiment, the electrodes are configured to face each other, but they do not have to face each other. Furthermore, the number of electrodes is not limited to two. The number of electrodes is not limited as long as the system is configured to include at least one positive electrode and one negative electrode.

[0092] Furthermore, although the electrostatic separation system 100 according to the embodiment has two partition units 135, this is not limiting. There may be only one partition, or there may be three or more partition units. Furthermore, although the partition unit 135 is configured to be fixed to the floor surface of the storage unit 140, this is not limiting. Both ends of the partition unit 135 may be fixed to the wall surface of the storage unit 140, or the partition unit 135 may be fixed by a member separate from the storage unit 140.

[0093] Furthermore, although the partition unit 135 of the electrostatic separation system 100 according to the embodiment has a rotating unit that rotates around a rotation axis, it may be configured to move in a manner other than rotation. For example, the partition unit 135 may be configured to move horizontally. Furthermore, when there are multiple partition units, only some of the partition units may be configured to rotate, while the other partition units may be configured to not move or to move horizontally.

[0094] Furthermore, in the embodiment, the data from the middle storage unit 142 in the storage unit 140 of the electrostatic separation system 100 is not used as learning data when generating a model, but the data from the middle storage unit 142 may be used as learning data. In this case, an infrared sensor detects substances contained in the material stored in the middle storage unit 142. Furthermore, although the storage unit 140 is configured to have three storage units, this is not limiting. The number of storage units in the storage unit 140 is not limited as long as it has two or more storage units.

[0095] Furthermore, the resin pieces stored in the middle storage section 142 in the storage section 140 of the electrostatic separation system 100 according to the embodiment may be configured to be returned to the charging section 110 again.

[0096] Furthermore, the sorting conditions of the electrostatic sorting system 100 according to the embodiment include a combination of the voltage value applied to the electrode unit 132 and the angle of the partition unit 135. However, the sorting conditions may be configured to include only the angle of the partition unit 135, or only the voltage value. The sorting conditions may also include other information. For example, the sorting conditions may include the rotation speed or rotation angle of the charging unit 110.

[0097] Furthermore, in the electrostatic separation system 100 according to the embodiment, the temperature and humidity adjustment device 60 is configured to adjust both the temperature and humidity around the electrostatic separation unit 130, but this is not limiting. It may be configured to adjust only one of the temperature or humidity. Alternatively, the electrostatic separation system may not be equipped with a temperature and humidity adjustment device.

[0098] Furthermore, the inference device 50 of the electrostatic sorting system 100 according to the embodiment is configured to infer sorting conditions from feature information and environmental information, but is not limited to this. The inference device 50 may be configured to infer sorting conditions from feature information only.

[0099] In addition, the electrostatic separation system 100 according to the embodiment is configured to use the specific charge calculated from the charge amount and weight of the resin pieces contained in the resin piece group as characteristic information, but it may also be configured to use the charge amount and weight as characteristic information.

[0100] Furthermore, the learning device 30 and the inference device 50 in the electrostatic sorting system 100 according to the embodiment are configured to simultaneously acquire pre-shaping characteristic information, sorting status information, and environmental information, but they do not have to be configured to acquire them simultaneously.

[0101] Furthermore, in the electrostatic separation system 100 according to the embodiment, the process of shaping the pre-shaping feature information into feature information is configured to be performed by the learning device 30 and the inference device 50, but this is not limited thereto. The data shaping process may be configured to be performed by the detection device 20 or an external device such as a calculation device. Furthermore, the learning device may use pre-shaping feature information that has not undergone data shaping as learning data.

[0102] Furthermore, the types of characteristic information or the types of sorting result information shown in the electrostatic sorting system 100 according to the embodiment are not limited to those described in the embodiment. The characteristic information or the sorting result information may be a part of the information described in the embodiment, or may include other information. For example, the sorting result information may include the speed of the resin pieces when they come into contact with the electrodes. [Explanation of symbols]

[0103] 10 electrostatic separation device, 20 detection device, 21 pre-shaping characteristic information detection unit, 22 sorting status detection unit, 23 environmental information detection unit, 30 learning device, 31 data acquisition unit, 32 data shaping unit, 33 model generation unit, 40 storage device, 50 inference device, 51 data acquisition unit, 52 data shaping unit, 53 inference unit, 60 temperature and humidity control device, 70 control device, 80 electrostatic separation device, 100 electrostatic separation system, 110 charging unit, 120 conveying unit, 130 electrostatic separation unit, 131 electrostatic separation unit, 132 electrode unit, 132a first electrode, 132b second electrode, 133 DC power supply, 135 partition unit, 135a first partition unit, 135b second partition unit, 136 rotating shaft unit, 137 rotating unit 138 Rotating shaft unit, 139 Rotating unit, 140 Storage unit, 141 First electrode side storage unit, 142 Middle storage unit, 143 Second electrode side storage unit, 155 Partition unit, 156 First partition unit, 157 Second partition unit, 200 Electrostatic separation system, 210 Storage unit partition, 211 First storage unit partition, 212 Second storage unit partition, 301 Reward calculation unit, 302 Function update unit.

Claims

1. a charging unit that charges the material to be sorted; a data acquisition unit that acquires characteristic information indicating characteristics of the material to be sorted; an inference unit that infers sorting conditions from the characteristic information acquired by the data acquisition unit using a trained model for inferring sorting conditions for sorting the material to be sorted that has been charged by the charging unit from the characteristic information; an electrostatic sorting unit that electrostatically sorts the materials to be sorted that have been charged by the charging unit based on the sorting conditions; An electrostatic separation system comprising:

2. The electrostatic sorting unit is an electrostatic separation unit having a plurality of electrodes, which attracts the materials to be sorted that have been charged by the charging unit to an electrode side corresponding to the polarity of the materials to be sorted; a partition portion provided below the electrostatic separator, The partition is provided to be movable or rotatable. The electrostatic separation system of claim 1 .

3. the partition unit includes a rotation shaft unit and a rotation unit that rotates around the rotation shaft unit, The selection conditions include a rotation angle of the rotating part.

3. The electrostatic separation system of claim 2.

4. The selection conditions include a combination of a voltage value applied to the electrode and a position or angle of the partition.

3. The electrostatic separation system of claim 2.

5. a humidity adjusting unit that adjusts the humidity of the electrostatic separation unit, or a temperature adjusting unit that adjusts the temperature of the electrostatic separation unit; and The sorting conditions include the humidity of the electrostatic separation unit or the temperature of the electrostatic separation unit.

3. The electrostatic separation system of claim 2.

6. The data acquisition unit further acquires information about the environment of the electrostatic separator or environmental information that is information about the environment around the electrostatic separator.

6. An electrostatic separation system according to any one of claims 2 to 5.

7. The environmental information includes information indicating the temperature of the electrostatic separator or information indicating the humidity of the electrostatic separator.

7. The electrostatic separation system of claim 6.

8. the data acquisition unit acquires the characteristic information of the material to be sorted that has been charged by the charging unit; The characteristic information includes information indicating the specific charge of objects contained in the material to be sorted.

8. An electrostatic separation system according to any one of claims 1 to 7.

9. The characteristic information includes information indicating the composition of the object contained in the material to be sorted, or information indicating the input amount of the material to be sorted input to the charging unit per unit time.

9. An electrostatic separation system according to any one of claims 1 to 8.

10. a first step in which the charging unit charges the material to be sorted; a second step in which a data acquisition unit acquires characteristic information indicating characteristics of the material to be sorted; a third step in which, after the first step and the second step are completed, an inference unit infers sorting conditions from the characteristic information acquired by the data acquisition unit using a trained model for inferring sorting conditions for sorting the material to be sorted that has been charged by the charging unit from the characteristic information; and a fourth step in which, after the third step is completed, an electrostatic separation unit electrostatically separates the materials to be separated that have been charged by the charging unit based on the separation conditions; An electrostatic separation method comprising:

11. a data acquisition unit for acquiring learning data including characteristic information indicating the characteristics of the material to be sorted and the sorting results of the material to be sorted in the electrostatic sorting unit, for an electrostatic sorting device having an electrifying unit that charges the material to be sorted, an electrostatic separating unit having a plurality of electrodes that attracts the material to be sorted charged in the electrifying unit toward an electrode corresponding to the polarity of the material to be sorted, and a partition unit provided below the electrostatic separating unit, and electrostatically sorting the material to be sorted that has been charged in the electrifying unit based on sorting conditions; a model generation unit that generates a trained model for inferring the selection condition from the feature information using the training data; Equipped with The sorting result includes a probability that the material to be sorted has come into contact with the electrode in the electrostatic separation unit. Learning device.

12. The electrostatic separation device includes a storage section that stores the materials to be separated electrostatically in the electrostatic separation section and has a first storage section and a second storage section that is different from the first storage section. and The sorting results include a purity indicating the proportion of a predetermined substance in the sorted material stored in the first storage unit or the second storage unit, and a recovery rate indicating the proportion of the amount of the predetermined substance contained in the sorted material stored in the first storage unit or the second storage unit to the amount of the predetermined substance contained in the sorted material. The learning device according to claim 11 .

Images