Molding condition adjustment device, molding machine, molding condition adjustment method, and computer program
The molding condition adjustment device uses reinforcement learning to adjust molding conditions based on machine and product state data, eliminating the need for additional inspection devices and reducing complexity and costs.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- THE JAPAN STEEL WORKS LTD
- Filing Date
- 2022-07-29
- Publication Date
- 2026-07-30
AI Technical Summary
Existing molding condition adjustment methods require trial and error based on operator experience and necessitate additional inspection devices for reinforcement learning, increasing complexity and cost.
A molding condition adjustment device that utilizes reinforcement learning to adjust molding conditions without requiring a dedicated inspection device by acquiring output data from the molding machine and molded product state data through an external input device, learning the relationship between these data, and determining adjustment amounts.
Enables effective adjustment of molding conditions using reinforcement learning without the need for a dedicated inspection device, simplifying the process and reducing costs.
Smart Images
Figure 0007897736000001 
Figure 0007897736000002 
Figure 0007897736000003
Abstract
Description
Technical Field
[0006] , ,
[0005] , , ,
[0001] The present invention relates to a molding condition adjusting device, a molding machine, a molding condition adjusting method, and a computer program.
Background Art
[0002] When performing injection molding using an injection molding machine, first, a conditioning operation is performed to adjust the set values of various molding condition items to obtain the molding conditions. Examples of molding condition items include the temperature of the injection cylinder, injection pressure, injection speed, holding pressure switching position, and screw back position. The adjustment of these molding conditions is performed based on the operator's experience, and it is necessary to repeat trial and error to obtain appropriate molding conditions. The same applies when performing extrusion molding using an extruder.
[0003] Patent Document 1 discloses an injection molding machine system that adjusts the molding conditions of an injection molding machine by reinforcement learning.
Prior Art Documents
Patent Documents
[0004]
Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0005] When reinforcement learning the molding condition adjustment method, it is necessary to input reward data according to the state of the molded product into the learning device. In a factory where the molded product is visually inspected, it is necessary to additionally introduce a dedicated inspection device capable of detecting the state of the molded product. It is also necessary to additionally introduce an inspection device to obtain correct answer data in supervised learning.
[0006] The purpose of this disclosure is to provide a molding condition adjustment device, etc., that can adjust molding conditions set on a molding machine by reinforce learning a method for adjusting molding conditions set on the molding machine, without introducing an inspection device for detecting the state of the molded product. [Means for solving the problem]
[0007] A molding condition adjustment device relating to one aspect of the present disclosure is a molding condition adjustment device for adjusting the molding conditions of a molding machine, comprising: a first acquisition unit for acquiring output data obtained from the molding machine; a second acquisition unit for acquiring state data indicating the state of a molded product obtained from the molding machine; a learner for learning the relationship between the output data and the state data and the adjustment amount of the molding conditions, and determining the adjustment amount based on the acquired output data and the state data; and an external input device for receiving input operations from an observer of the molded product obtained from the molding machine, wherein the second acquisition unit acquires the state data indicating the state of the molded product via the external input device.
[0008] A molding machine relating to one aspect of this disclosure includes the above-mentioned molding condition adjustment device.
[0009] A molding condition adjustment method relating to one aspect of the present disclosure is a molding condition adjustment method that adjusts the molding conditions of a molding machine using a learning device, wherein output data obtained from the molding machine is acquired, state data indicating the state of the molded product is acquired via an external input device that accepts input operations from an observer of the molded product obtained from the molding machine is acquired, the relationship between the output data and the state data and the amount of adjustment of the molding conditions is learned by the learning device, and the amount of adjustment is determined using the acquired output data and state data.
[0010] A computer program relating to one aspect of this disclosure is a computer program that causes a computer having a learning device to perform a process to adjust the molding conditions of a molding machine, the program acquires output data obtained from the molding machine, acquires state data indicating the state of the molded product via an external input device that accepts input operations from an observer of the molded product obtained from the molding machine, causes the learning device to learn the relationship between the output data and the state data and the amount of adjustment of the molding conditions, and causes the computer to perform a process to determine the amount of adjustment based on the acquired output data and state data using the learning device. [Effects of the Invention]
[0011] According to the above, it is possible to adjust the molding conditions set on the molding machine by reinforcement learning, without introducing an inspection device to detect the condition of the molded product. [Brief explanation of the drawing]
[0012] [Figure 1] This is a schematic diagram illustrating an example of the configuration of an injection molding apparatus according to Embodiment 1. [Figure 2] This is a block diagram showing an example of the configuration of a molding condition adjustment device according to Embodiment 1. [Figure 3] This is a perspective view showing an example configuration of an external input device according to Embodiment 1. [Figure 4] This is a block diagram showing an example configuration of an external input device according to Embodiment 1. [Figure 5] This is a conceptual diagram showing the configuration of the reward table according to Embodiment 1. [Figure 6] This is a functional block diagram of an injection molding machine according to Embodiment 1. [Figure 7] This is a flowchart showing the processing procedure of the processor according to Embodiment 1. [Figure 8] This is a flowchart showing the processing procedure of the processor according to Embodiment 2. [Modes for carrying out the invention]
[0013] Specific examples of a molding condition adjustment apparatus, molding machine, molding condition adjustment method, and computer program according to embodiments of the present invention will be described below with reference to the drawings. At least some of the embodiments described below may be arbitrarily combined. However, the present invention is not limited to these examples and is intended to be shown in the claims, with all modifications in the sense and scope equivalent to the claims included.
[0014] Figure 1 is a schematic diagram showing an example of the configuration of an injection molding machine 101 according to Embodiment 1. The injection molding machine 101 according to Embodiment 1 comprises a mold clamping device 2 for clamping the mold 21, an injection device 3 for plasticizing and injecting the molding material, a control device 4, and an external input device 5. The mold clamping device 2 and the injection device 3 constitute the molding machine body 1. The external input device 5 is a device that receives input operations from an operator (observer) observing the state of the molded product and inputs state data indicating the state of the molded product obtained from the injection molding machine 101 to the control device 4. The control device 4 functions as a molding condition adjustment device according to Embodiment 1.
[0015] The clamping device 2 comprises a fixed platen 22 fixed on the bed 20, a clamping housing 23 slidably mounted on the bed 20, and a movable platen 24 that similarly slides on the bed 20. The fixed platen 22 and the clamping housing 23 are connected by multiple tie bars 25, 25, ... for example, four tie bars. The movable platen 24 is configured to slide freely between the fixed platen 22 and the clamping housing 23. A clamping mechanism 26 is provided between the clamping housing 23 and the movable platen 24.
[0016] The clamping mechanism 26 is composed of, for example, a toggle mechanism. Alternatively, the clamping mechanism 26 may be composed of a direct-pressure type clamping mechanism, i.e., a clamping cylinder. The fixed platen 22 and the movable platen 24 are each provided with a fixed mold 21a and a movable mold 21b, respectively, and when the clamping mechanism 26 is driven, the molds 21 are opened and closed.
[0017] The injection device 3 is provided on the base 30. The injection device 3 includes a heating cylinder 31 having a nozzle 31a at its tip, and a screw 32 rotatably arranged in the heating cylinder 31 in the circumferential and axial directions. The screw 32 is driven in the rotational and axial directions by a drive mechanism 33. The drive mechanism 33 is composed of a rotary motor that drives the screw 32 in the rotational direction, a motor, etc. that drives the screw 32 in the axial direction. Note that since the drive mechanism 33 shown in Fig. 1 is covered by a cover, its internal structure is not shown.
[0018] Near the rear end of the heating cylinder 31, a hopper 34 for injecting the molding material is provided. Further, the injection molding machine 101 includes a nozzle touch device 35 for moving the injection device 3 in the front-rear direction (left-right direction in Fig. 1). When the nozzle touch device 35 is driven, the injection device 3 advances so that the nozzle 31a of the heating cylinder 31 touches the contact portion of the fixed plate 22.
[0019] Fig. 2 is a block diagram showing a configuration example of the control device 4 according to Embodiment 1. The control device 4 is a computer that controls the operations of the mold clamping device 2 and the injection device 3, and includes a processor (reinforcement learning device) 41, a storage unit 42, a control signal output unit 43, a first acquisition unit 44, a second acquisition unit 45, a notification unit 46, and an operation panel 40 as its hardware configuration.
[0020] The control device 4 is a device for adjusting the molding conditions of the injection molding machine 101. Note that the control device 4 may be a server device connected to a network. Further, the control device 4 may be configured by a plurality of computers for distributed processing, may be realized by a plurality of virtual machines provided in one server, or may be realized using a cloud server.
[0021] The processor 41 includes arithmetic circuits such as a CPU (Central Processing Unit), multi-core CPU, GPU (Graphics Processing Unit), GPGPU (General-purpose computing on graphics processing units), TPU (Tensor Processing Unit), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and NPU (Neural Processing Unit), as well as internal storage devices such as ROM (Read Only Memory) and RAM (Random Access Memory), I / O terminals, a timing unit, and the like. The processor 41 performs the molding condition adjustment method according to this embodiment 1 by executing a computer program (program product) 42a stored in the storage unit 42 described later. Note that each functional unit of the control device 4 may be implemented in software, or some or all of them may be implemented in hardware.
[0022] The memory unit 42 is a non-volatile memory such as a hard disk, EEPROM (Electrically Erasable Programmable ROM), or flash memory. The memory unit 42 stores a computer program 42a that reinforces learning methods for adjusting molding conditions according to the state of the injection molding machine 101 and the molded product, and causes the computer to execute the molding condition adjustment process. The memory unit 42 stores various coefficients that characterize the reinforcement learning model. The memory unit 42 also stores a reward table 42b for calculating rewards for training the reinforcement learning model. Note that the computer program 42a may be configured to include the reward table 42b. Details of the reward table 42b will be described later.
[0023] The computer program 42a and reward table 42b according to this embodiment 1 may be recorded on a recording medium 49 in a manner that is computer-readable. The storage unit 42 stores the computer program 42a read from the recording medium 49 by the reading device. The recording medium 49 is a semiconductor memory such as flash memory. The recording medium 49 may also be an optical disc such as a CD (Compact Disc)-ROM, DVD (Digital Versatile Disc)-ROM, or BD (Blu-ray® Disc). Furthermore, the recording medium 49 may be a magnetic disc such as a flexible disk or hard disk, or a magneto-optical disc.
[0024] Furthermore, the computer program 42a and reward table 42b according to this embodiment 1 may be downloaded from an external server connected to the communication network and stored in the storage unit 42.
[0025] The control signal output unit 43 outputs control signals to the injection molding machine 101 to control the operation of the injection molding machine 101 in accordance with the control of the processor 41 based on the molding conditions.
[0026] The control panel 40 is an interface for setting molding conditions and other parameters of the injection molding machine 101 and for operating the injection molding machine 101. The control panel 40 includes a display panel and an operating device.
[0027] The display panel is a display device such as a liquid crystal display panel or an organic EL display panel, and, in accordance with the control of the processor 41, displays a reception screen for accepting the setting of molding conditions for the injection molding machine 101, or displays the status of the injection molding machine 101 and the implementation status of the molding condition adjustment method according to this embodiment 1.
[0028] The operating device is an input device for inputting and adjusting the molding conditions of the injection molding machine 101, and includes operating buttons, a touch panel, etc. The operating device provides the processor 41 with data indicating the received molding conditions.
[0029] The injection molding machine 101 is configured with set values that define molding conditions, such as the injection start position, mold resin temperature, nozzle temperature, cylinder temperature (heater temperature), hopper temperature, clamping force, injection speed, injection acceleration, injection peak pressure (injection pressure), and injection stroke.
[0030] Furthermore, the injection molding machine 101 is configured with setting values that define molding conditions, such as the resin pressure at the cylinder tip, the seating state of the anti-reverse ring, the holding pressure, the holding pressure switching speed, the holding pressure switching position, the holding pressure completion position, the cushion position, the metered back pressure, and the metered torque.
[0031] Furthermore, the injection molding machine 101 is configured with setting values that define molding conditions, such as metering completion position, screw retraction speed, cycle time, mold closing time, injection time, holding pressure time, metering time, and mold opening time. Additionally, the injection molding machine 101 is configured with setting values for cooling time, screw rotation speed, mold opening / closing speed, ejection speed, and number of ejections. The injection molding machine 101 then operates according to these setting values.
[0032] The first acquisition unit 44 is an input circuit that acquires operating status data (output data) indicating the operating status of the injection molding machine 101. The operating status data is, for example, data indicating the injection pressure in the injection molding machine 101 when the injection molding machine 101 performs molding, the servo motor current, the heater current of the heating cylinder 31, the heater temperature, the rotation speed of the servo motor or screw 32, etc. The operating status data is preferably time-series data such as the injection pressure mentioned above.
[0033] The clamping device 2 and the injection device 3 are equipped with one or more sensors 1a that detect physical quantities necessary to control the operation of the molding machine body 1, which are information indicating the operating status of the injection molding machine 101. The sensors 1a are connected to the first acquisition unit 44. Physical quantities include, for example, voltage, current, temperature, humidity, torque, pressure and other forces, speed of movable parts, acceleration, rotation angle, position, fluid flow rate, velocity, etc. Sensors 1a are, for example, current sensors, voltage sensors, temperature sensors, humidity sensors, torque sensors, pressure sensors, speed sensors, acceleration sensors, rotation angle sensors, positioning sensors, flow sensors, flow meters, etc.
[0034] Specifically, sensor 1a includes a load cell for detecting the injection pressure in the injection molding machine 101, a current sensor for detecting the current of the servo motor, a current sensor for detecting the heater current of the heating cylinder 31, a temperature sensor for detecting the heater temperature, and a rotational speed sensor for detecting the rotational speed of the servo motor or screw 32. Sensor 1a outputs a measurement signal indicating a physical quantity to the control device 4. The measurement signal output from sensor 1a is input to the first acquisition unit 44, and the first acquisition unit 44 acquires the measurement signal as operating status data.
[0035] The second acquisition unit 45 is an input circuit that acquires molded product status data indicating the state of the molded product. An external input device 5 is connected to the second output unit. The molded product status data includes, for example, data indicating that the molded product is in a good condition, data indicating that the molded product has shape defects, or data indicating that the molded product has surface defects. Shape defects of molded products include, for example, burrs, sink marks, voids (air bubbles), shorts, warping, gate residue, etc. Surface defects of molded products include, for example, silver streaks, weld lines, jetting, flow marks, cracks, crazing, etc.
[0036] The notification unit 46 is an output circuit that notifies that molded product status data indicating a defective state has been input by the external input device 5. The notification unit 46 is, for example, a light-emitting element such as an LED, or a speaker that outputs sound such as a buzzer sound. The notification unit 46 may also be a communication circuit that transmits information indicating that molded product status data indicating a defective state has been input to a communication terminal held by the operator. The notification unit 46 may be configured by appropriately combining a light-emitting element, a speaker, and a communication circuit. The display panel of the operation panel 40 may also be used as the notification unit 46. Although an example in which the control device 4 is equipped with a notification unit 46 has been described, the notification unit 46 may also be equipped with a notification unit 46 on the external input device 5.
[0037] Figure 3 is a perspective view showing an example configuration of the external input device 5 according to Embodiment 1, and Figure 4 is a block diagram showing an example configuration of the external input device 5 according to Embodiment 1. The external input device 5 comprises a housing 50 having an operating surface. The shape of the housing 50 is not particularly limited, but for example, it is approximately a rectangular parallelepiped shape. The operating surface of the housing 50 is provided with a plurality of operating parts 51, 52, 53, 54 corresponding to each of a plurality of different states of the molded product. The plurality of different states include a good state in which the molded product is a good product, and one or more defective states in which the molded product is a defective product. The defective state includes shape defects or surface defects. Shape defects of the molded product include, for example, burrs, sink marks, voids (air bubbles), shorts, warping, gate residue, etc. Surface defects of the molded product include, for example, silver streaks, weld lines, jetting, flow marks, cracks, crazing, etc. In this embodiment 1, for example, the operation unit 51 corresponds to a good state. The operation unit 52 corresponds to a burr defect state, the operation unit 53 corresponds to a short circuit defect state, and the operation unit 53 corresponds to a flowchart mark defect state.
[0038] The multiple operating sections 51, 52, 53, and 54 are push-button switches. The number of operating sections 51, 52, 53, and 54 is not particularly limited, but 3 to 6 is preferable. It is preferable that the surface area of the operating sections 51, 52, 53, and 54, that is, the surface area that the operator's fingers touch, is larger than the surface area of the operator's fingers. This can improve the operability of the external input device 5.
[0039] The multiple operating units 51, 52, 53, and 54 are of an alternate type that maintains their operating state. When one operating unit 51, 52, 53, or 54 is operated, the state in which that operating unit 51, 52, 53, or 54 is operated is maintained. The operating state of the operating units 51, 52, 53, and 54 may be maintained mechanically or software-wise.
[0040] The external input device 5 includes multiple light-emitting units 51a, 52a, 53a, and 54a, each corresponding to one of the multiple operation units 51, 52, 53, and 54, to indicate that the operation unit 51, 52, 53, and 54 has been operated. The multiple light-emitting units 51a, 52a, 53a, and 54a are, for example, LEDs and are provided on the corresponding multiple operation units 51, 52, 53, and 54. The multiple light-emitting units 51a, 52a, 53a, and 54a may be provided in the vicinity of each of the multiple operation units 51, 52, 53, and 54. The location of the multiple light-emitting units 51a, 52a, 53a, and 54a is not particularly limited as long as it can be shown that there is a one-to-one correspondence between the multiple light-emitting units 51a, 52a, 53a, and 54 and the multiple operation units 51, 52, 53, and 54.
[0041] As shown in Figure 4, the external input device 5 includes a microcontroller 5a. The microcontroller 5a is connected to a plurality of operation units 51, 52, 53, and 54, a plurality of light-emitting units 51a, 52a, 53a, and 54a, and an output terminal (status data output unit) 5b. One end of a cable 55 connecting the external input device 5 and the control device 4 (see Figure 2) is connected to the output terminal 5b. The microcontroller 5a can detect the operation status of each of the plurality of operation units 51, 52, 53, and 54 and recognize the operation status of each operation unit 51, 52, 53, and 54. The microcontroller 5a outputs molded product status data corresponding to the operated operation unit 51, 52, 53, and 54 to the control device 4 from the output terminal 5b. The molded product status data is, for example, data indicating that the molded product is in a good condition, data indicating that the molded product has a shape defect, or data indicating that the molded product has a surface defect. Furthermore, the microcontroller 5a can make each of the multiple light-emitting units 51a, 52a, 53a, and 54a light up by supplying current to each of them. Specifically, the microcontroller 5a makes the light-emitting units 51a, 52a, 53a, and 54a that correspond to the operation units 51, 52, 53, and 54 that have been operated by the operator, and the operation units 51, 52, 53, and 54 whose operation state is being maintained, light up.
[0042] Although the external input device 5 that outputs molded product status data to the control device 4 via a wired cable 55 has been described, it may also be configured to transmit molded product status data to the control device 4 via wireless communication.
[0043] Figure 5 is a conceptual diagram showing the configuration of the reward table 42b according to Embodiment 1. The reward table 42b stores, for example, statistical quantities indicating the state of the molded product and reward data corresponding to the state of the molded product in association with each other.
[0044] The statistics for the condition of molded products are obtained from molded product condition data that indicates the condition of the molded product. For example, these statistics include the percentage of good molded products and the percentage of defective products for multiple types of defects.
[0045] The statistical quantities indicating the state of the molded product described above are merely examples, and the method and format of expression are not particularly limited, as long as they can express whether or not the molded product is in a good state. The reward calculation unit 41b may store the molded product state data and the reward data in association.
[0046] Furthermore, the reward table 42b may also be a table that associates operating status features indicating the operating status of the injection molding machine 101, statistics indicating the state of the molded product, and reward data corresponding to the operating status and the state of the molded product. Operating condition features are characteristics obtained from operating condition data that shows the operating status of the injection molding machine 101. For example, operating condition features are data that expresses the operating status, such as the mean, maximum, and deviation of the time-series operating condition data. The above-mentioned characteristics of the operating conditions are merely examples, and the method and format of expression are not particularly limited, as long as they can express whether or not the injection molding machine 101 is in a good condition.
[0047] Figure 6 is a functional block diagram of the injection molding machine 101 according to Embodiment 1. The processor 41 of the control device 4 includes an observation unit 41a, a reward calculation unit 41b, an agent 41c, and an adjustment unit 41d as a reinforcement learner. Note that each functional unit of the control device 4 may be implemented in software, or some or all of them may be implemented in hardware.
[0048] The observation unit 41a acquires molding condition data indicating the molding conditions set in the injection molding machine 101. For example, if the storage unit 42 stores molding condition data, the observation unit 41a reads the molding condition data from the storage unit 42. The observation unit 41a also acquires operating status data and molded product status data from the first acquisition unit 44 and the second acquisition unit 45, respectively. The observation unit 41a outputs the acquired molding condition data, operating status data, and molded product status data to the agent 41c. The observation unit 41a also outputs the acquired molded product status data to the reward calculation unit 41b. In a configuration where reward data is calculated taking the operating status into account, the observation unit 41a outputs the molded product status data and operating status data to the reward calculation unit 41b.
[0049] The reward calculation unit 41b calculates reward data obtained by evaluating the currently set molding conditions based on the data observed by the observation unit 41a, particularly the molded product state data, and outputs the calculated reward data to agent 41c. Specifically, the reward calculation unit 41b calculates the probability of occurrence of each of several types of molded product states based on the molded product state data. Then, the reward calculation unit 41b obtains the reward data by reading the corresponding reward data from the reward table 42b using the calculated statistics indicating the molded product state as a key.
[0050] In a configuration where reward data is calculated taking into account the operating conditions, the reward calculation unit 41b calculates reward data obtained by evaluating the currently set molding conditions based on the operating conditions data and molded product condition data observed by the observation unit 41a, and outputs the calculated reward data to the agent 41c. Specifically, the reward calculation unit 41b calculates characteristic quantities of the operating conditions based on the operating conditions data. Then, the reward calculation unit 41b obtains the reward data by reading the corresponding reward data from the reward table 42b using the calculated characteristic quantities of the operating conditions and statistics indicating the state of the molded product as keys.
[0051] Although an example of obtaining reward data by referring to the reward table 42b has been described, the reward calculation unit 41b may be configured to calculate reward data by inputting operating status data and molded product status data into a predetermined calculation formula or function. For example, if the degree of defect in the molded product is large, the reward value will be small or negative. Also, if the characteristic quantity of the operating status falls outside the normal range, the reward value will be small or negative.
[0052] Agent 41c is, for example, a reinforcement learning model with a deep neural network such as DQN (Deep Q-Network), A3C, or D4PG, or a model-based reinforcement learning model such as PlaNet or SLAC. Below, we will describe an example in which Agent 41c is equipped with DQN.
[0053] Agent 41c uses DQN to determine an action a corresponding to the state s of the injection molding machine 101 indicated by the observed data. The state s includes molding condition data, operating status data, and molded product state data.
[0054] DQN is a neural network model that, given a state s represented by observed data as input, outputs the value of each of several actions a. The multiple actions a include items to be adjusted and adjustment amounts. Actions a with high value represent the adjustment items and adjustment amounts for setting appropriate conditions. Agent 41c selects an action a with high value, and the injection molding machine 101 transitions to another state as a result of the selected action a. After the state transition, agent 41c receives the reward calculated by the reward calculation unit 41b, and trains agent 41c to maximize the return, i.e., the cumulative reward.
[0055] More specifically, DQN has an input layer, an intermediate layer, and an output layer. The input layer comprises multiple nodes into which a state s, i.e., observed data, is input. The output layer comprises multiple nodes, each corresponding to one of several actions a, and outputting the value Q(s,a) of that action a in the input state s.
[0056] Based on the state s, action a, and the reward r obtained from that action, the DQN of agent 41c can be reinforced by adjusting various weight coefficients that characterize the DQN, using the value Q represented by the following equation (1) as training data for which the correct answer is. Q(s,a)←Q(s,a)+α(r+γmaxQ(snext,anext)-Q(s,a))...(1) however, s: state a:Action α: Learning rate r:Reward γ: discount rate maxQ(snext,anext): The maximum Q-value for the next possible action.
[0057] Figure 7 is a flowchart showing the processing procedure of the processor 41 according to Embodiment 1. The processor 41 receives initial settings for molding conditions via the operation panel 40 and sets the setting values for various items related to the received molding conditions (step S111). Then, the processor 41 controls the operation of the injection molding machine 101 according to the set molding conditions (step S112).
[0058] The injection molding performed in step S112 is a test shot for reinforcement learning in the reinforcement learner. The operator observes the molded product obtained from the test shot and confirms its condition. The operator then inputs the condition of the molded product by operating the control units 51, 52, 53, and 54 of the external input device 5. However, even if the operator operates the control units 51, 52, 53, and 54, the input content is not immediately finalized. As will be described later, the input content is finalized after a predetermined time has elapsed since the molding was performed.
[0059] Meanwhile, the processor 41 acquires operating status data of the molding machine body 1 via the first acquisition unit 44 (step S113).
[0060] The processor 41 measures the elapsed time since the completion of molding and determines whether a predetermined time has elapsed after the completion of molding (step S114). The time of completion of molding is not strictly defined and can be appropriately set as the time when molding is approximately complete. For example, the time when the ejector pin is driven and the molded product is pushed out of the mold 21 can be set as the time of completion of molding. The predetermined time is the time required for the operator to check the condition of the molded product and operate the external input device 5. The predetermined time is less than or equal to the molding cycle time.
[0061] If it is determined that a predetermined time has not elapsed after molding is complete (step S114: NO), the processor 41 returns to step S112. If it is determined that a predetermined time has elapsed after molding is complete (step S114: YES), the processor 41 acquires molded product status data from the external input device 5 (step S115). When a predetermined time has elapsed after molding is complete, the operating status of the external input device 5 is determined, and the molded product status data corresponding to the operating units 51, 52, 53, and 54 that are in the operating status at that time is input to the control device 4.
[0062] Next, the processor 41 stores the set molding condition data, acquired operating status data, and molded product status data in the storage unit 42 in association with each other (step S116).
[0063] Next, the processor 41 determines whether or not molded product status data indicating that the molded product is defective has been input (step S117). If molded product status data indicating that the molded product is defective has been input (step S117: YES), the processor 41 notifies the notification unit 46 that molded product status data indicating that the molded product is in a defective state has been input (step S118).
[0064] The operation units 51, 52, 53, and 54 of the external input device 5 maintain their operating state, and each time a predetermined amount of time has elapsed after molding is completed, molded product status data is automatically input to the operating unit 51, 52, 53, and 54 that is currently in operation. For example, if the operation unit 52 corresponding to a defective state where the molded product has burrs is operated, the state in which the operation unit 52 is operated is maintained. If the operator does not operate anything further, each time a predetermined amount of time has elapsed, the state corresponding to the operation unit 52, that is, molded product status data indicating that the molded product has burrs, will be input to the control device 4. As a result, there is a risk that molded product status data indicating a defective state will continue to be input unintentionally by the operator. However, in this embodiment 1, the notification unit 46 notifies the operator when molded product status data indicating a defective state has been input. Therefore, the operator can recognize that molded product status data indicating a defective state has been input, and the operator can be prevented from unintentionally inputting molded product status data indicating a defective state.
[0065] Furthermore, the notification unit 46 may be configured to notify the system that good condition data has been input for a molded product. Needless to say, the notification unit 46 distinguishes between good and bad conditions and notifies the system using different light emission patterns, different sounds, and different content. Furthermore, the system may be configured to transmit the input status of molded product condition data to a communication terminal held by the operator for notification.
[0066] If the process in step S118 is completed, or if no molded product status data indicating a defect has been entered (step S117: NO), the processor 41 determines whether a predetermined number of molded product status data, etc., necessary for reinforcement learning has been accumulated (step S119). If it is determined that a predetermined number of molded product status data, etc., has not been accumulated (step S119: NO), the processor 41 returns to step S112.
[0067] If it is determined that a predetermined number of molded product status data have been accumulated (step S119: YES), the processor 41 refers to the reward table 42b based on the molded product status data and calculates the reward data (step S120).
[0068] Next, the processor 41 performs reinforcement learning on agent 41c based on the observed data, which are molding condition data, operating status data, and molded product state data, and the reward data (step S121). For example, the DQN of agent 41c is reinforced by adjusting the various weight coefficients that characterize the DQN so that the value Q represented by the above equation (1) becomes the correct answer.
[0069] Next, the processor 41 determines whether or not to terminate reinforcement learning (step S122). For example, it determines whether or not to terminate reinforcement learning based on the probability of good and defective products occurring. The processor 41 may be configured to accept instructions from the operator to terminate reinforcement learning via the operation panel 40. If it is determined not to terminate reinforcement learning (step S122: NO), the processor 41 returns to step S112. If it is determined to terminate reinforcement learning (step S122: YES), the processor 41 terminates the reinforcement learning process.
[0070] According to the molding condition adjustment device, etc., of Embodiment 1 configured as described above, it is possible to adjust the molding conditions set on the injection molding machine 101 by reinforcement learning of the molding conditions of the injection molding machine 101 without introducing an inspection device that detects the state of the molded product. Specifically, the processor 41 or agent 41c can automatically adjust the molding conditions of the injection molding machine 101 by learning through reinforcement the relationship between the current molding conditions, operating status, and the state of the molded product and the appropriate amount of adjustment to the molding conditions.
[0071] The external input device 5 is equipped with multiple operating units 51, 52, 53, and 54 corresponding to various types of conditions of the molded product. Therefore, the operator can input molded product condition data simply by operating the corresponding operating unit 51, 52, 53, or 54. Specifically, the external input device 5 is equipped with an operating unit 51 corresponding to the good condition where the molded product is a good product, and multiple operating units 52, 53, and 54 corresponding to various types of defective conditions where the molded product is a defective product. By selectively operating these operating units 51, 52, 53, and 54, the operator can easily input molded product condition data indicating that the molded product is a good product, molded product condition data indicating that the molded product has shape defects or surface defects, etc., into the control device 4.
[0072] Since the operating sections 51, 52, 53, and 54 of the external input device 5 are of the alternate type, if molded products in the same state are being molded, the operator can input the molded product state data for that state without having to operate the external input device 5 each time.
[0073] Since the light-emitting units 51a, 52a, 53a, and 54a corresponding to the operating units 51, 52, 53, and 54 that are in operation are illuminated, the operator can visually recognize which operating unit 51, 52, 53, and 54 are currently in operation.
[0074] Furthermore, the input molded product state data is converted into reward data by the processor 41. The processor 41 can use the calculated reward data to reinforce the agent 41c that adjusts the molding conditions.
[0075] Furthermore, since the molded product status data corresponding to the operating units 51, 52, 53, and 54 that are in operation is input to the control device 4 each time a predetermined amount of time has elapsed after the molding is completed, the operator can input the molded product status data without having to operate the external input device 5 each time.
[0076] Since the notification unit 46 notifies the operator when molded product status data indicating that the molded product is in a defective state is entered, the operator can recognize that molded product status data indicating a defective state has been entered. Therefore, it is possible to prevent molded product status data indicating a defective state from being continuously entered against the operator's intention.
[0077] Although this embodiment mainly describes an example of adjusting the molding conditions of an injection molding apparatus, the molding condition adjustment method according to this embodiment may also be configured to adjust the molding conditions of an extruder.
[0078] In this embodiment 1, deep reinforcement learning using a neural network was described, but the molded product state may be adjusted using a reinforcement learning model equipped with a Q-value table or other known reinforcement learning models instead of DQN. Alternatively, the molded product state may be adjusted using a supervised learning model equipped with a neural network, SVM (Support Vector Machine), Bayesian network, or other known machine learning models.
[0079] Furthermore, although an example has been described in which molded product status data is automatically input to the control device 4 each time a predetermined amount of time has elapsed since molding, the external input device 5 may also be configured to include an input confirmation operation unit. In addition, the operation unit 54 may also be used as the input confirmation operation unit. The operator can input the confirmed molded product status data to the control device 4 by operating the operation units 51, 52, 53, and 54 corresponding to the state of the molded product and by operating the input confirmation operation unit.
[0080] Furthermore, although a configuration in which an external input device 5 is connected to the second acquisition unit 45 has been described, the second acquisition unit may also be configured to selectively connect a detection device for detecting the state of the molded product and the external input device 5. The detection device is, for example, a camera, a distance sensor, a weighing scale, etc., which detects the state of the molded product, measures physical quantities related to the state of the molded product, and outputs the physical quantity data obtained from the measurement to the control device 4. When the detection device is connected to the second acquisition unit 45, the processor 41 acquires the physical quantity data output from the detection device and calculates molded product state data indicating the state of the molded product based on the acquired physical quantity data. The processor 41 determines whether the molded product is good or defective by, for example, image recognition processing of image data obtained by imaging the molded product. The image recognition processing may be rule-based processing such as pattern matching, or it may be processing using a machine learning model.
[0081] The processor 41 may be configured to detect whether the device connected to the second acquisition unit 45 is an external input device 5 or a detection device. When the external input device 5 is connected to the second acquisition unit 45, the processor 41 acquires molded product state data using the process described in this embodiment and performs reinforcement learning. When the detection device is connected to the second acquisition unit 45, the processor 41 calculates molded product state data using, for example, the method described above and performs reinforcement learning. The processor 41 may also be configured to accept a selection of whether to use the external input device 5 or the detection device via the operation panel 40, and to perform reinforcement learning using the selected device.
[0082] Furthermore, the second acquisition unit 45 may be configured to allow connection of both the detection device and the external input device 5. The processor 41 calculates molded product state data using both the state data acquired via the external input device 5 and the physical quantity data obtained by measurement using the detection device. For example, the detection device is a camera, and the operator inputs molded product state data that cannot be recognized by the camera using the external input device 5. The processor 41 may, for example, adopt the molded product state data that indicates the worse condition from among the molded product state data obtained using the external input device 5 and the detection device.
[0083] Furthermore, if both the detection device and the external input device 5 are connected to the second acquisition unit 45, the system may be configured to allow specifying which data to prioritize. For example, the processor 41 of the control device 4 receives a request via the operation panel 40 or the like to determine which data from the detection device or the external input device 5 should be prioritized when evaluating the molding conditions. If it is specified that the data from the external input device 5 should be prioritized, the processor 41 may be configured to use the molded product status data from the external input device 5 for processing when the result obtained using the detection device is good and the result obtained using the external input device 5 is bad.
[0084] Furthermore, if both the detection device and the external input device 5 are connected to the second acquisition unit 45, the system may be configured to allow specifying which data to prioritize. For example, the processor 41 of the control device 4 receives weighting coefficients for the data obtained from the detection device and the external input device 5 via the operation panel 40 or the like. The processor 41 calculates molded product state data using the state data acquired via the external input device 5, the physical quantity data obtained by measurement using the detection device, and the weighting coefficients. For example, the processor 41 may be configured to calculate molded product state data and rewards as an overall evaluation by calculating a weighted average of the molded product state data obtained from each data using the weighting coefficients.
[0085] (Embodiment 2) The molding condition adjustment device, molding machine, molding condition adjustment method, and computer program 42a according to Embodiment 2 differ from Embodiment 1 in that the operating states of the operating units 51, 52, 53, and 54 of the external input device 5 are reset each time molded product state data is input to the control device 4 using the external input device 5. The other components of the molding condition adjustment device, etc. are the same as those of the molding condition adjustment device, etc. according to Embodiment 1, so the same reference numerals are used for the same parts, and detailed explanations are omitted.
[0086] Figure 8 is a flowchart showing the processing procedure of the processor 41 according to Embodiment 2. The processor 41 performs the same processing as steps S111 to S116 of Embodiment 1, acquiring and storing operating status data and molded product status data (steps S211 to S216). Next, the processor 41 resets the operating state of the external input device 5 (step S217). Specifically, the processor 41 controls the operating units 51, 52, 53, and 54 for inputting molded product status data indicating that the molded product is a good product to be in an operated state. For example, the processor 41 resets the operating state of the external input device 5 by outputting a reset signal to the external input device 5. The external input device 5 is equipped with input / output terminals that can input and output signals instead of an output terminal 5b. When a reset signal is input to the input / output terminals, the external input device 5 resets its operating state. In other words, the external input device 5 sets the operating unit 51 corresponding to the good state to an operated state.
[0087] With this configuration, the operating state is reset to a good state each time molding condition data is input to the control device 4 from the external input device 5. This prevents the operator from unintentionally continuing to input molded product status data indicating a defective state.
[0088] The processor 41 then performs the same processing as in steps S119 to S122 of Embodiment 1, and then completes the processing.
[0089] According to the molding condition adjustment device of Embodiment 2, the operating state is reset to a good state each time molding condition data is input to the control device 4, thus preventing the operator from unintentionally continuing to input molded product condition data indicating a defective state.
[0090] (Note 1) A molding condition adjustment device for adjusting the molding conditions of a molding machine, comprising: a first acquisition unit for acquiring output data obtained from the molding machine; a second acquisition unit for acquiring state data indicating the state of a molded product obtained from the molding machine; a learning device for learning the relationship between the output data and the state data and the adjustment amount of the molding conditions, and determining the adjustment amount based on the acquired output data and the state data; and an external input device for receiving input operations from an observer of the molded product obtained from the molding machine, wherein the second acquisition unit acquires the state data indicating the state of the molded product via the external input device. (Note 2) The molding condition adjustment device according to Note 1, wherein the external input device comprises a plurality of operating units corresponding to each of a plurality of states of the molded product, and a state data output unit that outputs state data indicating the state of the molded product corresponding to the operated operating unit. (Note 3) The molding condition adjustment device described in Note 2, wherein the multiple types of states include a good state in which the molded product is a good product and one or more types of defective states in which the molded product is a defective product. (Note 4) The molding condition adjustment device according to Note 2 or Note 3, comprising a plurality of light-emitting units corresponding to each of the plurality of operating units, which indicate that the operating unit has been operated. (Note 5) The molding condition adjustment device according to any one of Notes 2 to 4, wherein the plurality of operating parts are of the alternate type that maintains the operating state. (Note 6) The molding condition adjustment device according to any one of Notes 2 to 5, wherein the second acquisition unit acquires the state data corresponding to the operation unit that is in operation at the time the predetermined period of time has elapsed. (Note 7) A molding condition adjustment device according to any one of Notes 2 to 6, which includes a notification unit that provides a predetermined notification when it obtains the status data indicating that the molded product is a defective product. (Note 8) The molding condition adjustment device according to any one of Notes 2 to 7, wherein each time the second acquisition unit acquires the state data, the operation unit corresponding to the state in which the molded product is a good product is initialized to the state in which it has been operated. (Note 9) The molding condition adjustment device according to any one of Notes 1 to 8, wherein the output data includes operating status data indicating the operating status of the molding machine, and the status data includes data indicating a good condition in which the molded product is a good product, or data indicating a defective condition in which the molded product has shape defects or surface defects. (Note 10) The molding condition adjustment device according to any one of Notes 1 to 9, wherein the learning device comprises an agent that outputs an adjustment amount for the molding conditions when the acquired output data and the state data are input, and a reward calculation unit that calculates reward data based on the acquired state data, and the device causes the agent to perform reinforcement learning based on the acquired output data and the state data and the calculated reward data. (Note 11) A molding machine equipped with a molding condition adjustment device described in any one of Notes 1 to 10. [Explanation of symbols]
[0091] 1. Molding machine body 1a Sensor 2 Mold clamping device 3 Injection device 4. Control device 5. External Input Device 5a Microcontroller 5b Output terminal 40 Control Panel 41 processors 41a Observation Unit 41b Remuneration Calculation Department 41c Agent 41d Adjustment part 42 Storage section 42a Computer Programs 42b Reward Table 43 Control signal output section 46 Notification Department 49 Recording media 51,52,53,54 Operation section 51a, 52a, 53a, 54a Light-emitting part 55 Cables 101 Injection molding machine
Claims
1. A molding condition adjustment device for adjusting the molding conditions of a molding machine, A first acquisition unit acquires output data obtained from the molding machine, A second acquisition unit acquires state data indicating the state of the molded product obtained from the molding machine, A learning device that learns the relationship between the output data and the state data and the adjustment amount of the molding conditions, and determines the adjustment amount based on the acquired output data and state data, An external input device that receives input operations from an observer of the molded product obtained from the molding machine, and Equipped with, The aforementioned second acquisition unit is, The system is configured to acquire the state data indicating the state of the molded product via the external input device. The aforementioned learning device is When the acquired output data and the status data are input, an agent outputs the amount of adjustment for the molding conditions, A reward calculation unit that calculates reward data based on the acquired state data. Equipped with, The agent is reinforced based on the acquired output data and state data, and the calculated reward data. Molding condition adjustment device.
2. The external input device comprises a plurality of operating units corresponding to each of the multiple states of the molded product, A state data output unit that outputs state data indicating the state of the molded product corresponding to the operated unit, and A molding condition adjustment device according to claim 1, comprising:
3. The aforementioned multiple states include a good state in which the molded product is a good product, and one or more defective states in which the molded product is a defective product. The molding condition adjustment device according to claim 2.
4. The molding condition adjustment device according to claim 2 or claim 3, further comprising a plurality of light-emitting units corresponding to each of the plurality of operating units, which indicate that the operating unit has been operated.
5. The aforementioned plurality of operating units are of the alternate type, which maintains the operating state. A molding condition adjustment device according to claim 2 or claim 3.
6. The aforementioned second acquisition unit is, Each time a predetermined period of time has elapsed, the state data corresponding to the operating unit that is in the operating state at the time the predetermined period of time has elapsed is acquired. The molding condition adjustment device according to claim 5.
7. The system includes a notification unit that provides a predetermined notification when it obtains the status data indicating that the molded product is defective. The molding condition adjustment device according to claim 6.
8. Each time the second acquisition unit acquires the state data, the operation unit corresponding to the state in which the molded product is a good product is initialized to a state in which it has been operated. The molding condition adjustment device according to claim 6.
9. The output data includes operating status data indicating the operating status of the molding machine, The aforementioned status data includes data indicating that the molded product is in a good condition, or data indicating that the molded product has a shape defect or surface defect. A molding condition adjustment device according to claim 1 or claim 2.
10. A molding machine comprising a molding condition adjustment device according to claim 1 or claim 2.
11. A method for adjusting the molding conditions of a molding machine using a learning device that includes an agent that outputs an adjustment amount for the molding conditions of the molding machine when output data obtained from the molding machine and state data indicating the state of a molded product obtained from the molding machine are input, and a reward calculation unit that calculates reward data based on the acquired state data, The output data obtained from the molding machine is acquired, State data indicating the state of the molded product is acquired via an external input device that accepts input operations from an observer of the molded product obtained from the molding machine. Based on the acquired output data and state data, and the calculated reward data, the learning device is made to reinforce learn the relationship between the output data and state data and the adjustment amount of the molding conditions. Using the learning device, the adjustment amount is determined based on the acquired output data and state data. How to adjust molding conditions.
12. A computer program for causing a computer to perform a process to adjust the molding conditions of a molding machine, which includes a learning device having an agent that outputs an amount to adjust the molding conditions of a molding machine when output data obtained from a molding machine and state data indicating the state of a molded product obtained from the molding machine are input, and a reward calculation unit that calculates reward data based on the acquired state data, The output data obtained from the molding machine is acquired, State data indicating the state of the molded product is acquired via an external input device that accepts input operations from an observer of the molded product obtained from the molding machine. Based on the acquired output data and state data, and the calculated reward data, the learning device is made to reinforce learn the relationship between the output data and state data and the adjustment amount of the molding conditions. Using the learning device, the adjustment amount is determined based on the acquired output data and state data. A computer program that causes the computer to perform a process.