Manufacturing system, manufacturing method, and manufacturing program

Abstract

This manufacturing system comprises: a setting unit that sets a candidate action parameter indicating action on a workpiece, said candidate action parameters indicating at least one from among a position of the action, a magnitude of the action, and a direction of the action; an inference unit that inputs a candidate action parameter into an inference engine to infer a result of processing the workpiece according to the candidate action parameter, the inference engine being configured to accept input of a candidate action parameter to infer the effect of the action on the workpiece; a determination unit that determines an actual action parameter on the basis of the result of processing; and an execution unit that applies action to the workpiece according to the actual action parameter.

Description

Manufacturing system, manufacturing method, and manufacturing program

[0001] One aspect of the present disclosure relates to a manufacturing system, a manufacturing method, and a manufacturing program.

[0002] As an example of a method for manufacturing a workpiece, Patent Document 1 describes a straightening device for a linear workpiece. This straightening device comprises a pulling mechanism that applies a pulling force in the longitudinal direction of the workpiece while clamping both ends of the workpiece of a predetermined length, and a pressing mechanism equipped with a straightening ram that applies a pressure force to the clamped workpiece in the straightening direction.

[0003] JP 2018-130731 A

[0004] There is a demand for a mechanism that can automate the manufacturing of workpieces with high precision.

[0005] A manufacturing system according to one aspect of the present disclosure includes a setting unit that sets candidate action parameters indicating an action on a workpiece, the candidate action parameters indicating at least one of the position of the action, the magnitude of the action, and the direction of the action; an inference unit that inputs the candidate action parameters to an inference device that accepts input of the candidate action parameters and infers the effect of the action on the workpiece, and infers the processing results of the workpiece using the candidate action parameters; a determination unit that determines actual action parameters based on the processing results; and an execution unit that applies an action to the workpiece using the actual action parameters.

[0006] According to one aspect of the present disclosure, the manufacturing of a workpiece can be automated with high precision.

[0007] FIG. 1 is a diagram illustrating an example of the functional configuration of a manufacturing system. FIG. 2 is a diagram illustrating an example of the hardware configuration of a computer used for the manufacturing system. FIG. 3 is a diagram illustrating an example of the configuration of an inference unit. FIG. 4 is a diagram illustrating another example of the configuration of the inference unit. FIG. 5 is a flowchart illustrating an example of processing executed by the manufacturing system. FIG. 6 is a diagram illustrating another example of the functional configuration of the manufacturing system. FIG. 7 is a diagram illustrating another example of the configuration of the inference unit. FIG. 8 is a flowchart illustrating another example of processing executed by the manufacturing system.

[0008] Various examples of the present disclosure will be described in detail below with reference to the accompanying drawings. In the description of the drawings, the same or equivalent elements are designated by the same reference numerals, and redundant description will be omitted.

[0009] [System Overview] The manufacturing system according to the present disclosure is a control system that processes a workpiece using action parameters that indicate an action on the workpiece, thereby manufacturing the workpiece. The action applied to the workpiece may be an action that changes the physical state of the workpiece, in which case the manufacturing system applies the action to the workpiece so as to change the current physical state of the workpiece. The physical state of the workpiece may be a visually recognizable state such as its shape (external shape), or may be an internal state that is not visually recognizable such as its internal temperature. The change in the physical state may be a visually recognizable change or a visually imperceptible change. The action may be applied to a portion of the workpiece, or may be applied to the entire workpiece.

[0010] For example, the manufacturing system determines action parameters for correcting a workpiece that is at least partially distorted, strikes the workpiece based on the action parameters to correct the workpiece, and provides the corrected workpiece as a finished product or intermediate product. In this example, the physical state of the workpiece is the shape of the workpiece, and the action is striking the workpiece.

[0011] There are still areas where the application of actions to workpieces is performed manually or where automation is insufficient. Taking a metal plate as an example, the correction work of hammering the metal plate to flatten it is currently performed manually. Depending on the type of metal plate or the quality target, the worker may have to hammer the metal plate multiple times to attempt correction, which increases the work time and burden on the worker. According to the manufacturing system disclosed herein, action parameters are automatically determined and actions are automatically applied to the actual workpiece based on the action parameters, making it possible to automate the manufacturing of workpieces with high precision.

[0012] [System Configuration] Figure 1 is a diagram showing the functional configuration of a manufacturing system according to an example. The manufacturing system 1 shown in Figure 1 applies an action to an actual workpiece 9 so that the workpiece 9 becomes flat. Therefore, in this example, the action applied to the workpiece 9 is an action that changes the shape of the workpiece 9, and the manufacturing system 1 applies an action to the workpiece 9 so that the shape of the workpiece 9 changes. In the following, it is assumed that the workpiece 9 is a metal plate. The manufacturing system 1 includes, as functional components, a measurement unit 11, a setting unit 12, an inference unit 13, a determination unit 14, a loop unit 15, an execution unit 16, and a transfer robot 17.

[0013] The measurement unit 11 is a functional module that measures the physical state of the workpiece 9. For example, the measurement unit 11 includes one or more sensors for measuring the shape of the workpiece 9.

[0014] The setting unit 12 is a functional module that sets candidate action parameters that indicate an action on the workpiece 9. The candidate action parameters are candidates for actual action parameters for actually applying an action to the workpiece 9. The candidate action parameters indicate at least one of the position of the action, the magnitude of the action, and the direction of the action. The position of the action refers to the position where the action is applied to the workpiece. The magnitude of the action refers to the magnitude of the action that is to be applied to the workpiece. The direction of the action refers to the direction in which the action is applied to the workpiece.

[0015] The inference unit 13 is a functional module that inputs candidate action parameters to the inference device 20 and infers the processing results of the workpiece 9 based on the candidate action parameters. The inference device 20 is a computational model that receives the input of candidate action parameters and infers the influence of the action indicated by the candidate action parameters on the workpiece 9. The inference device 20 is, for example, a neural network.

[0016] The determining unit 14 is a functional module that determines actual effect parameters based on the processing results. Corresponding to the candidate effect parameters, the actual effect parameters indicate at least one of the position of the effect, the magnitude of the effect, and the direction of the effect.

[0017] The loop unit 15 is a functional module that manages the repetition of a series of processes performed by the setting unit 12, the inference unit 13, and the decision unit .

[0018] The execution unit 16 is a functional module that applies an action to the workpiece 9 according to the actual action parameters. For example, the execution unit 16 applies an action to the workpiece 9 according to the actual action parameters so as to change the current physical state of the workpiece 9. In the example shown in FIG. 1 , the execution unit 16 includes a cylinder 16a for hitting the workpiece 9. In one example, the cylinder 16a is movable horizontally along the top surface of the workpiece 9 and is also movable toward or away from the top surface of the workpiece 9. The execution unit 16 may further include a mechanism for adjusting the stroke width or angle at which the top surface of the workpiece 9 is hit. The workpiece 9 can be deformed by the cylinder 16a hitting the top surface of the workpiece 9.

[0019] The transfer robot 17 is a functional module that transfers the workpiece 9 between the execution unit 16 and the measurement unit 11. The transfer robot 17 includes a controller that controls the transfer and a robot main body that actually transfers the workpiece 9.

[0020] The manufacturing system 1 can be realized using any type of computer. The computer may be a general-purpose computer such as a personal computer or a business server, or may be incorporated into a dedicated device that executes a specific process.

[0021] 2 is a diagram showing an example of the hardware configuration of the computer 100 used for the manufacturing system 1. In this example, the computer 100 includes a main body 110, a monitor 120, and an input device .

[0022] The main body 110 is a device having a circuit 160. The circuit 160 has a processor 161, a memory 162, a storage 163, an input / output port 164, and a communication port 165. The number of each hardware component may be one or more. The storage 163 records programs for configuring each functional module of the main body 110. The storage 163 is a computer-readable recording medium such as a hard disk, a non-volatile semiconductor memory, a magnetic disk, or an optical disk. The memory 162 temporarily stores programs loaded from the storage 163, calculation results of the processor 161, and the like. The processor 161 configures each functional module by executing programs in cooperation with the memory 162. The input / output port 164 inputs and outputs electrical signals to and from the monitor 120 or the input device 130 in response to instructions from the processor 161. The communication port 165 performs data communication with other devices via a communication network N in response to instructions from the processor 161.

[0023] The monitor 120 is a device for displaying information output from the main body 110. For example, the monitor 120 is a device capable of displaying graphics, such as a liquid crystal panel.

[0024] The input device 130 is a device for inputting information to the main body 110. Examples of the input device 130 include operation interfaces such as a keypad, a mouse, and an operation controller.

[0025] The monitor 120 and the input device 130 may be integrated as a touch panel. For example, the main body 110, the monitor 120, and the input device 130 may be integrated as a tablet computer.

[0026] Each functional module of the manufacturing system 1 is realized by loading a manufacturing program onto the processor 161 or memory 162 and having the processor 161 execute the program. The manufacturing program includes code for realizing each functional module of the manufacturing system 1. The processor 161 operates the input / output port 164 and the communication port 165 in accordance with the manufacturing program, and reads and writes data from and to the memory 162 or the storage 163.

[0027] The manufacturing program may be provided in a state recorded on a non-transitory recording medium such as a CD-ROM, a DVD-ROM, a semiconductor memory, etc. Alternatively, the manufacturing program may be provided via a communication network as a data signal superimposed on a carrier wave.

[0028] [Inference Unit and Inference Unit] In one example, the inference unit 20 used by the inference unit 13 is generated by a predetermined learning device and provided to the manufacturing system 1. The learning device accesses a predetermined database storing training data and performs machine learning based on the training data to generate the inference unit 20. Each data record of the training data represents a combination of data input to a machine learning model and ground truth data to be compared with an inference result output from the machine learning model. The training data may be generated using at least one of actual measurement data obtained by past actual processing of one or more sample works and virtual data obtained by computer simulation. The actual measurement data may be obtained manually or automatically by a control system such as the manufacturing system 1. For one data record of the training data, the learning device provides input data to a predetermined machine learning model and obtains an inference result output from the machine learning model. The learning device performs backpropagation based on the error between the inference result and the ground truth data to update a set of parameters in the machine learning model. The learning device performs its learning using each data record until a given termination condition is met, thereby generating a reasoner 20 .

[0029] Various examples of the inference unit 13 will be described with reference to Figures 3 and 4. Figures 3 and 4 are both diagrams showing examples of the configuration of the inference unit 13.

[0030] FIG. 3 shows an inference unit 13A, which is an example of the inference unit 13. The inference unit 13A uses an inference unit 20A, which is an example of the inference unit 20, and includes an evaluation unit 18. The inference unit 20A receives input of the physical state of the workpiece 9 and input of a candidate action parameter, and infers the next physical state, which is the physical state of the workpiece 9 affected by the action indicated by the candidate action parameter, as the effect of the action on the workpiece 9. Any of these physical states may be the shape of the workpiece 9. The shape may be expressed by the position (height or depth) in the thickness direction of the workpiece 9 in each of multiple small regions obtained by dividing the top surface of the workpiece 9 into a mesh. The evaluation unit 18 is a functional module that evaluates the next physical state inferred by the inference unit 20A and calculates an evaluation value. The evaluation value is an index that indicates how close the state of the workpiece is to the quality target for the workpiece. The higher the evaluation value, the more desirable the next physical state is. The evaluation value may be expressed as a continuous value. As another example, the evaluation value may be expressed as a binary value indicating whether the evaluation of the next physical state of the work 9 is higher or lower than the evaluation of the previous physical state, which is the physical state of the work 9 before it is affected by the action.

[0031] Each data record of the training data for generating the inference unit 20A by the above-described machine learning represents a combination of a pre-state, which is the physical state of the sample work before an action is applied, an action parameter indicating the action, and a post-state, which is the physical state of the sample work after the action is applied. The post-state is used as the correct answer in the machine learning.

[0032] The inference unit 13A inputs the physical state of the workpiece 9 and the candidate action parameters to the inference device 20A. The inference device 20A infers the next physical state of the workpiece 9 as the effect of the action on the workpiece 9 indicated by the candidate action parameters. The evaluation unit 18 evaluates the inferred next physical state and calculates an evaluation value. For example, the evaluation unit 18 may calculate the evaluation value based on the difference between the quality target of the workpiece 9 and the inferred next physical state. The inference unit 13A acquires the evaluation value as the processing result. Through this series of procedures, the inference unit 13A infers the processing result of the workpiece 9 using the candidate action parameters.

[0033] FIG. 4 shows an inference unit 13B, which is another example of the inference unit 13. The inference unit 13B uses an inference unit 20B, which is another example of the inference device 20. The inference device 20B receives input of the physical state of the workpiece 9 and input of a candidate action parameter, and infers, as the effect of the action on the workpiece 9, an evaluation value of the next physical state, which is the physical state of the workpiece 9 affected by the action indicated by the candidate action parameter. As in the example of FIG. 3, the physical state may be the shape of the workpiece 9, and the shape may be represented by the position in the thickness direction of the workpiece 9 in each of a plurality of small regions obtained by dividing the top surface of the workpiece 9 into a mesh-like pattern. Because the inference device 20B directly infers the evaluation value, the inference unit 13B does not include an evaluation unit 18.

[0034] Each data record of the training data for generating the inference unit 20B by the above-described machine learning represents a combination of a pre-state, which is the physical state of the sample work before an action is applied, an action parameter indicating the action, and an evaluation value of a post-state, which is the physical state of the sample work after the action is applied. The evaluation value is used as a correct answer in the machine learning.

[0035] The inference unit 13B inputs the physical state of the workpiece 9 and the candidate action parameters to the inference device 20B. The inference device 20B infers an evaluation value of the next physical state of the workpiece 9 as the effect of the action on the workpiece 9 indicated by the candidate action parameters. The inference unit 13B acquires the evaluation value as the processing result. Through this series of procedures, the inference unit 13B infers the processing result of the workpiece 9 using the candidate action parameters.

[0036] At least one of the inference unit 20A and the inference unit 20B may include an encoder for compressing the physical state (e.g., shape) of the workpiece 9 into a low dimension. As described above, when the shape of the workpiece 9 is represented by the position in the thickness direction of the workpiece 9 in each of a plurality of small regions obtained by dividing the top surface of the workpiece 9 into a mesh, the encoder compresses the plurality of small regions into a low dimension. The inference unit 20A including the encoder infers the next physical state based on the compressed physical state and the candidate action parameters. The inference unit 20B including the encoder infers an evaluation value based on the compressed physical state and the candidate action parameters.

[0037] [Manufacturing Method] As an example of a manufacturing method according to the present disclosure, an example of processing executed by the manufacturing system 1 will be described with reference to Fig. 5. Fig. 5 is a flowchart showing the example as processing flow S1. That is, the manufacturing system 1 executes processing flow S1. In the following description, it is assumed that the physical state of the workpiece 9 is the shape of the workpiece 9, and that the action applied to the workpiece 9 is an action that changes the shape of the workpiece 9.

[0038] In step S11, the measurement unit 11 measures the current physical state of the workpiece 9. In one example, the transfer robot 17 transfers the workpiece 9 to the measurement unit 11, and the measurement unit 11 measures the shape of the transferred workpiece 9 as the physical state. The measurement result may indicate distortion of the workpiece 9.

[0039] In step S12, the measurement unit 11 determines whether the workpiece 9 satisfies the quality target. For example, the evaluation unit 18 determines that the workpiece 9 satisfies the quality target when the difference between the measured shape (physical state) and the quality target is less than a predetermined threshold, and determines that the workpiece 9 does not satisfy the quality target when the difference is equal to or greater than the threshold.

[0040] If the workpiece 9 satisfies the quality target (YES in step S12), the manufacturing system 1 ends the process. This means that the manufacturing of the workpiece 9 by the manufacturing system 1 is complete. The manufactured workpiece 9 may be transported to the next process, or may be shipped as a finished product or an intermediate product.

[0041] On the other hand, if the workpiece 9 does not satisfy the quality target (NO in step S12), the process proceeds to step S13. In this case, the transport robot 17 transports the workpiece 9 to the execution unit 16.

[0042] In step S13, the setting unit 12 sets one or more candidate action parameters. For example, the setting unit 12 sets a plurality of candidate action parameters that differ from each other in at least one of the position, magnitude, and direction of the action.

[0043] In step S14, the inference unit 13 infers, for each of one or more candidate action parameters, the processing result of the workpiece 9 according to the candidate action parameter. For example, for each of one or more candidate action parameters, the inference unit 13 inputs the candidate action parameter and the measured current physical state to the inference device 20 to infer the processing result. The inference unit 13 infers the evaluation value of the next physical state of the workpiece 9 as the processing result.

[0044] When the inference unit 13A and the inference device 20A are used, the inference unit 13A inputs candidate action parameters and the current physical state to the inference device 20A. The inference device 20A receives the input data and infers the next physical state of the workpiece 9. The evaluation unit 18 evaluates the inferred next physical state and calculates an evaluation value. The inference unit 13A obtains the evaluation value as a processing result.

[0045] When the inference unit 13B and the inference device 20B are used, the inference unit 13B inputs candidate action parameters and the current physical state to the inference device 20B. The inference device 20B receives the input data and infers an evaluation value of the next physical state of the workpiece 9. The inference unit 13B obtains the evaluation value as a processing result.

[0046] In step S15, the determination unit 14 determines an actual action parameter based on one or more processing results obtained from one or more candidate action parameters. If the inference unit 13 has inferred a processing result for each of a plurality of candidate action parameters, the determination unit 14 determines one of the plurality of candidate action parameters as the actual action parameter based on the plurality of processing results.

[0047] When the inference unit 13A or the inference unit 13B is used, the determination unit 14 determines an actual action parameter from a plurality of candidate action parameters based on a plurality of pairs of candidate action parameters and evaluation values. When the evaluation values ​​are expressed by continuous values, the determination unit 14 may determine the candidate action parameter corresponding to the highest evaluation value as the actual action parameter. When the evaluation values ​​are expressed by binary values, the determination unit 14 may determine one of the candidate action parameters for which the evaluation of the next physical state is higher than the evaluation of the previous physical state as the actual action parameter. For example, the determination unit 14 may determine the candidate action parameter corresponding to the highest position in the thickness direction of the workpiece 9 as the actual action parameter.

[0048] When the inference unit 13 infers a processing result for a single candidate action parameter, the determination unit 14 may determine that candidate action parameter as the actual action parameter as is.

[0049] In step S16, the loop unit 15 determines whether a predetermined number of actual action parameters have been obtained. The predetermined number may be 1 or a value equal to or greater than 1. That is, the manufacturing system 1 may determine a single actual action parameter, or may determine a set of two or more actual action parameters that indicate two or more actions to be applied to the workpiece 9 in order (i.e., sequentially).

[0050] If the predetermined number of actual action parameters have not yet been obtained (NO in step S16), the processing returns to step S13. The loop unit 15 executes a series of repeated processes by the setting unit 12, the inference unit 13, and the determination unit 14 to determine a set of two or more actual action parameters. To achieve this repetition, the inference unit 13 infers the physical state of the workpiece 9 affected by the action, i.e., the next physical state, as at least a part of the processing result. When the inference unit 13A and the inference unit 20A are used, the inference unit 13A can acquire the next physical state inferred by the inference unit 20A as the processing result in addition to the evaluation value.

[0051] In the repeated step S13, the setting unit 12 sets one or more next candidate action parameters. The setting unit 12 may set the next candidate action parameters to be different from the previously determined actual action parameters. In the repeated step S14, the inference unit 13 inputs one or more next candidate action parameters and the previously inferred physical state (next physical state) of the workpiece 9 to the inference device 20, and infers a processing result for each of the one or more next candidate action parameters. In the repeated step S15, the determination unit 14 determines a next actual action parameter based on the one or more inferred processing results. The next actual action parameter is an action parameter to be used next to the previously determined actual action parameter.

[0052] If a predetermined number of actual action parameters have been obtained (YES in step S16), the process proceeds to step S17.

[0053] As shown by steps S13 to S16, in each iteration, the setting unit 12 sets candidate action parameters, the inference unit 13 inputs the candidate action parameters and the previously inferred physical state of the workpiece to an inference unit to infer processing results for the candidate action parameters, and the determination unit 14 determines actual action parameters based on the processing results, resulting in two or more sets of actual action parameters.

[0054] In step S17, the execution unit 16 executes processing on the workpiece 9 based on the determined actual action parameters. That is, the execution unit 16 applies an action to the workpiece 9 using the actual action parameters. When two or more sets of actual action parameters are obtained, the execution unit 16 applies two or more actions to the workpiece in sequence using the sets. For example, the execution unit 16 adjusts one of the position and angle of the cylinder 16a relative to the workpiece 9 based on the actual action parameters, and hits the workpiece 9 with the cylinder 16a with a set force. The force with which the workpiece 9 is hit is adjusted, for example, by the stroke width of the cylinder 16a. When two or more sets of actual action parameters are obtained, the execution unit 16 hits the workpiece 9 with the cylinder 16a multiple times while changing at least one of the position, angle, and force as necessary.

[0055] After step S17, the process returns to step S11. In this case, the transport robot 17 transports the workpiece 9 to the measurement unit 11. In the repeated step S11, the measurement unit 11 measures the current physical state of the workpiece 9, i.e., the physical state of the workpiece 9 after the action has been applied. In the repeated step S12, the measurement unit 11 again determines whether the workpiece 9 satisfies the quality target. If the workpiece 9 satisfies the quality target (YES in step S12), the manufacturing system 1 ends the process. If the workpiece 9 still does not satisfy the quality target (NO in step S12), the process from step S13 onwards is executed again. In step S13, the setting unit 12 sets one or more new candidate action parameters. In step S14, the inference unit 13 identifies the measured physical state as the current physical state, and for each of the one or more new candidate action parameters, the new candidate action parameter and the identified current physical state are input to the inference device 20 to infer a new processing result for the workpiece 9. In step S15, the determination unit 14 determines new actual action parameters based on one or more new processing results. As shown in step S16, the manufacturing system determines a predetermined number of new actual action parameters. In step S17, the execution unit 16 applies one or more new actions to the workpiece 9 using the one or more new actual action parameters.

[0056] As described above, the manufacturing system 1 uses the inference unit 20 to infer the next physical state of the workpiece 9 from the current physical state of the workpiece 9 and the candidate action parameters, and calculates an evaluation value for the next physical state. Alternatively, the manufacturing system 1 uses the inference unit 20 to directly infer an evaluation value from the current physical state of the workpiece 9 and the candidate action parameters. Then, based on the evaluation value, the manufacturing system 1 determines actual action parameters that indicate the next action to be applied to the workpiece. This series of processes enables the automation of the manufacturing of the workpiece 9 with high precision.

[0057] [Modifications] The technology according to the present disclosure has been described in detail above based on various examples. However, the present disclosure is not limited to the above examples. The technology according to the present disclosure can be modified in various ways without departing from the spirit of the present disclosure.

[0058] Fig. 6 is a diagram showing the functional configuration of a manufacturing system according to another example. The manufacturing system 1X shown in Fig. 6 differs from the manufacturing system 1 in that it does not include the setting unit 12, the determination unit 14, and the loop unit 15, and includes an inference unit 13X and an inference unit 20X instead of the inference unit 13 and the inference unit 20. The following will particularly describe these differences.

[0059] The inference unit 13X is a functional module that inputs the physical state of the workpiece 9 into the inference unit 20X and infers action parameters that correspond to the physical state and indicate the action exerted on the workpiece 9 as actual action parameters. The inference unit 20X is a computational model that receives input of the physical state of the workpiece 9 and infers the action parameters. The inference unit 20X is, for example, a neural network. The physical state of the workpiece 9 input to the inference unit 20X is, for example, the current physical state of the workpiece 9. As in the example of the manufacturing system 1, the physical state may be the shape of the workpiece 9, and the shape may be represented by the position in the thickness direction of the workpiece 9 in each of multiple small regions obtained by dividing the top surface of the workpiece 9 into a mesh. The inferred actual action parameter indicates at least one of the position of the action, the magnitude of the action, and the direction of the action.

[0060] FIG. 7 is a diagram showing the configuration of the inference unit 13X. Each data record of the training data for generating the inference unit 20X by the above-described machine learning indicates a combination of a previous state, which is the physical state of the sample work before an action is applied, and an action parameter applied to the sample work in that previous state. The action parameter is used as a correct answer in machine learning. As described above, the training data may also be generated using at least one of actual measurement data and virtual data. The inference unit 13X inputs the physical state of the work 9 to the inference unit 20X. The inference unit 20X infers the actual action parameter based on that physical state.

[0061] In one example, the inference unit 20X may directly infer the actual action parameters. Alternatively, the inference unit 20X may infer evaluation values ​​for each of a plurality of small regions obtained by dividing the top surface of the workpiece 9 into a mesh pattern, generate a matrix of evaluation values, and infer actual action parameters that indicate at least the position corresponding to the small region with the highest evaluation value. In other words, the inference unit 20X may indirectly infer the actual action parameters. In either case, the inference unit 13X acquires the inferred actual action parameters.

[0062] An example of processing executed by the manufacturing system 1X will be described with reference to Fig. 8. Fig. 8 is a flowchart showing this example as processing flow S2. That is, the manufacturing system 1X executes processing flow S2. As with processing flow S1, in the following description, the physical state of the workpiece 9 is the shape of the workpiece 9, and the action applied to the workpiece 9 is an action that changes the shape of the workpiece 9.

[0063] In step S21, the measurement unit 11 measures the current physical state of the workpiece 9. Step S21 is similar to step S11 described above.

[0064] In step S22, the measurement unit 11 determines whether the workpiece 9 satisfies the quality target. Step S22 is the same as step S12 described above. If the workpiece 9 satisfies the quality target (YES in step S22), the manufacturing system 1X ends the processing. On the other hand, if the workpiece 9 does not satisfy the quality target (NO in step S22), the processing proceeds to step S23. In this case, the transport robot 17 transports the workpiece 9 to the execution unit 16.

[0065] In step S23, the inference unit 13X infers actual action parameters for the current physical state of the workpiece 9. The inference unit 13X inputs the measured current physical state to the inference device 20X. The inference device 20X accepts the physical state and infers actual action parameters. The inference unit 13X acquires the actual action parameters.

[0066] In step S24, the execution unit 16 executes processing on the workpiece 9 based on the estimated actual action parameters. Step S24 is similar to step S17 described above.

[0067] After step S24, the process returns to step S21. In this case, the transport robot 17 transports the workpiece 9 to the measurement unit 11. In the repeated step S21, the measurement unit 11 measures the current physical state of the workpiece 9, i.e., the physical state of the workpiece 9 after the action has been applied. In the repeated step S22, the measurement unit 11 again determines whether the workpiece 9 satisfies the quality target. If the workpiece 9 satisfies the quality target (YES in step S22), the manufacturing system 1X terminates the process. If the workpiece 9 still does not satisfy the quality target (NO in step S22), the manufacturing system 1X again executes the processes from step S23 onwards. In step S23, the inference unit 13X infers new actual action parameters for the measured current physical state. In step S24, the execution unit 16 applies a new action to the workpiece 9 using the new actual action parameters.

[0068] The inference device may accept input of candidate action parameters without accepting input of the physical state of the workpiece, and infer the effect of the action on the workpiece. As in the above example, the inferred effect may be an evaluation value or the next physical state of the workpiece. Each data record of the training data for generating the inference device by the above machine learning indicates a combination of an action parameter indicating an action and the physical state of the sample workpiece after the action is applied or an evaluation value of the physical state. The physical state or evaluation value is used as the correct answer in the machine learning. In this variant, the inference unit inputs candidate action parameters to the inference device, and infers the processing result of the workpiece using the candidate action parameters.

[0069] The manufacturing system does not have to include at least one of the measuring unit and the transport robot. At least one of the measuring unit and the transport robot may be a component of a control system different from the manufacturing system. Alternatively, the work performed by at least one of the measuring unit and the transport robot may be performed manually.

[0070] The manufacturing system does not need to include a loop unit. In this case, in response to the determination of one actual action parameter, the execution unit applies an action to the workpiece according to the actual action parameter.

[0071] In process flow S1, the manufacturing system 1 may end the process after executing step S17. In process flow S2, the manufacturing system 1X may end the process after executing step S24. As in these examples, the manufacturing system may execute a process for one workpiece only once, which applies one or more actions to the workpiece based on one or more actual action parameters.

[0072] When the inference unit 13A and the inference device 20A are used, the manufacturing system 1 may further include an update unit for updating the inference device 20A. The update unit generates a new data record indicating a combination of the physical state of the workpiece 9 input to the inference device 20A, the determined actual action parameters, and the physical state of the workpiece measured by the measurement unit 11. Compared with the example of FIG. 3 , the new data record indicates a combination of a pre-state, which is the physical state of the workpiece 9 before an action is applied, an action parameter indicating the action, and a post-state, which is the physical state of the workpiece 9 after the action is applied. The manufacturing system 1 stores the new data record in a predetermined database as at least a part of the training data. The update unit performs machine learning using training data including one or more new data records to generate a new inference device 20A. The update unit replaces the existing inference device 20A with the generated inference device 20A, thereby updating the inference device 20A. This update can improve the inference accuracy of the inference device 20A.

[0073] While the above example illustrates a straightening operation as an operation performed by the manufacturing system, the manufacturing system may be used in other situations. For example, the workpiece may be a weld bead, and the operation may be polishing to smooth the weld bead. In this case, the manufacturing system sets actual operation parameters for polishing. Alternatively, the operation may be spot welding, and the manufacturing system sets actual operation parameters for spot welding the workpiece. As described above, setting the actual operation parameters may involve determining or inferring the actual operation parameters.

[0074] In the above example, both the candidate action parameter and the actual action parameter indicate at least one of the position of the action, the magnitude of the action, and the direction of the action, but the action parameter may also include other indicators such as the number of times the action is applied to the workpiece.

[0075] The hardware configuration of the system is not limited to a configuration in which each functional module is realized by executing a program. For example, at least some of the functional modules may be configured by logic circuits specialized for the functions, or may be configured by an ASIC (Application Specific Integrated Circuit) that integrates the logic circuits.

[0076] The processing steps of the method executed by at least one processor are not limited to the above examples. For example, some of the steps or processes described above may be omitted, or the steps may be executed in a different order. Furthermore, any two or more of the steps described above may be combined, or some of the steps may be modified or deleted. Alternatively, other steps may be executed in addition to the steps described above.

[0077] When comparing the magnitude of two numbers within a computer system or computer, either of the two criteria "greater than or equal to" and "greater than" can be used, or either of the two criteria "less than or equal to" and "under".

[0078] [Appendix A] As can be seen from the various examples above, the present disclosure includes the following aspects.

[0079] (Appendix A1) A manufacturing system comprising: a setting unit that sets candidate action parameters indicating an action on a workpiece, the candidate action parameters indicating at least one of the position, magnitude, and direction of the action; an inference unit that inputs the candidate action parameters to an inference unit that receives input of the candidate action parameters and infers the effect of the action on the workpiece, and infers a processing result of the workpiece according to the candidate action parameters; a determination unit that determines actual action parameters based on the processing result; and an execution unit that applies the action to the workpiece according to the actual action parameters. According to Appendix A1, the processing result of the workpiece due to the action indicated by the candidate action parameters is inferred using the inference unit, and the actual action parameters are determined based on the processing result. Since the actual action parameters for actually applying the action to the workpiece are automatically determined using the inference unit, the manufacturing of the workpiece can be automated with high precision. Furthermore, even when the action parameters indicate a combination of two or more of the position, magnitude, and direction of the action, the actual action parameters can be appropriately determined.

[0080] (Appendix A2) The manufacturing system according to Appendix A1, wherein the inference unit further receives input of a physical state of the workpiece to infer the effect, and the inference unit inputs the candidate action parameters and the current physical state of the workpiece to the inference unit to infer the processing result. According to Appendix A2, the physical state of the workpiece is further input to the inference unit in addition to the candidate action parameters to infer the processing result of the workpiece, so that the processing result can be inferred with higher accuracy.

[0081] (Appendix A3) The manufacturing system according to Appendix A2, wherein the action is an action that changes a physical state of the workpiece, and the execution unit applies the action to the workpiece according to the actual action parameters so as to change the current physical state of the workpiece. According to Appendix A3, the physical state of the actual workpiece can be changed with high precision.

[0082] (Appendix A4) The manufacturing system according to Appendix A2 or A3, wherein the setting unit sets a plurality of the candidate action parameters, the inference unit inputs each of the plurality of candidate action parameters to the inference device and infers the processing result for each of the plurality of candidate action parameters, and the determination unit determines the actual action parameter from the plurality of candidate action parameters based on the plurality of processing results. According to Appendix A4, a processing result is inferred for each of the plurality of candidate action parameters, and an actual action parameter is determined from the plurality of candidate action parameters based on these processing results. By performing inference on a plurality of candidate action parameters, it is possible to more reliably determine an appropriate actual action parameter and to impart a more appropriate action to the workpiece.

[0083] (Appendix A5) The manufacturing system according to Appendix A4, wherein the inference unit infers, as the processing result, an evaluation value of the next physical state of the workpiece, which is the physical state of the workpiece affected by the action, and the determination unit determines the actual action parameter from the plurality of candidate action parameters based on a plurality of pairs of the candidate action parameters and the evaluation values. According to Appendix A5, the actual action parameter is determined based on the evaluation value of the next physical state of the workpiece, so that the action can be applied to the workpiece in accordance with a predetermined evaluation criterion. As a result, the quality of the workpiece can be managed based on the evaluation criterion.

[0084] (Appendix A6) The manufacturing system according to Appendix A5, wherein the inference device infers the evaluation value as the influence, and the inference unit acquires the evaluation value inferred by the inference device as the processing result. According to Appendix A6, since the inference device directly evaluates the physical state of the workpiece, it is possible to incorporate the subjective evaluation of an operator, such as an expert, into the inference device. As a result, it is possible to manufacture workpieces with high precision at the same level as an operator.

[0085] (Appendix A7) The manufacturing system according to Appendix A5, wherein the inference device infers the next physical state as the effect, and the inference unit includes an evaluation unit that evaluates the next physical state inferred by the inference device to calculate the evaluation value, and the evaluation value calculated by the evaluation unit is acquired as the processing result. According to Appendix A7, the inference of the next physical state by the inference device and the evaluation of the next physical state are executed separately. Because the inference device infers a physical state, which is objective data, it becomes easier to obtain training data for generating the inference device and to verify the inference.

[0086] (Appendix A8) The manufacturing system according to any one of Appendices A2 to A7, further comprising a loop unit that executes a series of repetitive processes by the setting unit, the inference unit, and the determination unit to determine a set of two or more actual action parameters indicating the two or more actions to be applied to the workpiece in sequence, wherein the inference unit infers a physical state of the workpiece affected by the action as at least a part of the processing result, the setting unit sets the candidate action parameter in each repetition, the inference unit inputs the candidate action parameter and the previously inferred physical state of the workpiece to the inference device in each repetition to infer the processing result for the candidate action parameter, the determination unit determines the actual action parameter based on the processing result in each repetition, and the execution unit applies the two or more actions to the workpiece in sequence using the set of two or more actual action parameters obtained by the repetition. According to Appendix A8, after actual action parameters for each of the multiple actions to be applied in sequence (i.e., sequentially) are determined, the multiple actions are applied to the workpiece. Therefore, it becomes possible to operate the manufacturing system more efficiently, and in turn, it is possible to improve the manufacturing efficiency of the workpieces.

[0087] (Appendix A9) The manufacturing system according to Appendix A8, further comprising a measurement unit that measures the physical state of the workpiece to which the two or more actions have been applied, wherein the setting unit sets a new candidate action parameter, the inference unit identifies the measured physical state as the current physical state, and inputs the new candidate action parameter and the identified current physical state into the inference device to infer a new processing result of the workpiece, the determination unit determines a new actual action parameter based on the new processing result, and the execution unit applies a new action to the workpiece using the new actual action parameter. According to Appendix A9, the physical state of the workpiece to which multiple actions have been applied is measured, and new actual action parameters are determined based on the measurement results. Therefore, even if an error occurs in a previous process based on multiple actual action parameters, the error can be corrected to manufacture the workpiece with high precision.

[0088] (Appendix A10) The manufacturing system according to Appendix A9 further includes a robot that transports the workpiece between the execution unit and the measurement unit. According to Appendix A10, because the robot transports the workpiece, general-purpose devices can be used as the measurement unit and execution unit, rather than dedicated machines. This reduces the cost of the manufacturing equipment. While manufacturing efficiency is affected by the number of transports performed by the robot, by sequentially performing multiple actions before transporting, as shown in Appendix A9, it is possible to achieve both improved manufacturing efficiency and reduced manufacturing equipment costs.

[0089] (Appendix A11) A manufacturing system comprising: an inference unit that inputs the current physical state of a workpiece into an inference unit that receives input of the physical state of the workpiece and infers action parameters indicating an action to be applied to the workpiece, and infers the action parameters corresponding to the current physical state as actual action parameters, wherein the action parameters indicate at least one of the position of the action, the magnitude of the action, and the direction of the action; and an execution unit that applies the action to the workpiece using the actual action parameters inferred from the current physical state. According to Appendix A11, the inference unit directly infers the actual action parameters from the current physical state of the workpiece. This mechanism enables the automation of workpiece manufacturing with high precision. Furthermore, the processing procedures up to setting the actual action parameters can be simplified.

[0090] (Appendix A12) A manufacturing method executed by a manufacturing system having at least one processor, comprising: a step of setting candidate action parameters indicating an action on a workpiece, the candidate action parameters indicating at least one of the position, the magnitude, and the direction of the action; a step of inputting the candidate action parameters to an inference device that receives input of the candidate action parameters and infers the influence of the action on the workpiece, and inferring a processing result of the workpiece according to the candidate action parameters; a step of determining actual action parameters based on the processing result; and a step of applying the action to the workpiece according to the actual action parameters. According to Appendix A12, it is possible to obtain the same technical effect as in Appendix A1.

[0091] (Appendix A13) A manufacturing program that causes a computer to execute the following steps: a step of setting a candidate action parameter that indicates an action on a workpiece, the candidate action parameter indicating at least one of the position of the action, the magnitude of the action, and the direction of the action; a step of inputting the candidate action parameter into an inference device that receives input of the candidate action parameter and infers the influence of the action on the workpiece, and inferring a processing result of the workpiece according to the candidate action parameter; a step of determining an actual action parameter based on the processing result; and a step of applying the action to the workpiece according to the actual action parameter. According to Appendix A13, a technical effect similar to that of Appendix A1 can be obtained.

[0092] [Appendix B] As can be seen from the various examples above, the present disclosure also includes the following aspects.

[0093] (Appendix B1) A workpiece manufacturing device comprising: a setting unit that sets multiple action parameters representing at least one of the position on the workpiece and the magnitude of the action when an action that causes a physical change in the workpiece is applied to the workpiece; an inference unit that inputs the multiple action parameters set by the setting unit to an inference device that infers a physical change in the workpiece upon receiving input of at least the action parameters, and infers the physical change in the workpiece; a determination unit that determines one action parameter from the multiple action parameters based on the inferred change; and an execution unit that applies an action based on the one action parameter determined by the determination unit to the workpiece. According to Appendix B1, for each of the multiple action parameters, it is inferred what change may occur in the workpiece if an action is applied based on the action parameter, and the action parameter to be actually applied is determined based on the inference result. This allows an action to be applied according to the quality of the workpiece to be manufactured, enabling the automation of workpiece manufacturing that requires high precision, which has been difficult to achieve until now. The inference device learns changes that may occur in the workpiece and predicts physical phenomena. Therefore, the influence of bias in the learning data can be reduced compared to directly inferring action parameters.

[0094] (Appendix B2) The workpiece manufacturing device according to Appendix B1, wherein the inference unit further inputs the physical state of the workpiece to the inference device, and infers, as the inferred change, a physical state after the input physical state is affected by the input action parameter from the input physical state. According to Appendix B2, since the physical state is inferred, it becomes easier to verify the inference result.

[0095] (Appendix B3) The workpiece manufacturing apparatus according to Appendix B2, further comprising an evaluation unit that evaluates the physical state inferred by the inference unit, wherein the determination unit determines one action parameter from the plurality of action parameters based on the evaluation result by the evaluation unit for the inferred physical state. According to Appendix B3, the evaluation unit performs an evaluation and the action parameter is determined according to the result. Therefore, the quality of the workpiece can be controlled by controlling the evaluation method of the evaluation unit.

[0096] (Appendix B4) The workpiece manufacturing apparatus according to Appendix B2 or B3, further comprising a loop unit that repeats multiple times inputting the physical state previously inferred by the inference unit into the inference device together with the action parameter different from that used in the previous inference, wherein the determination unit determines the one action parameter for each of multiple actions to be performed consecutively based on the inference results by the inference unit in the multiple repetitions, and the execution unit, after determining the one action parameter for each of the multiple actions to be performed consecutively, applies each of the multiple actions to the workpiece based on the one action parameter for each of the multiple actions. According to Appendix B4, after determining the action parameter for each of the multiple actions to be performed sequentially, the multiple sequential actions can be applied to the workpiece. Therefore, manufacturing efficiency can be improved.

[0097] (Appendix B5) The workpiece manufacturing apparatus according to Appendix B4 further comprises a measurement unit that measures the physical state of the workpiece after the execution unit applies each of the multiple actions to be performed consecutively to the workpiece, wherein the inference unit performs inference using the new physical state measured by the measurement unit when applying a new action to the workpiece, and the determination unit determines action parameters of the new action to be applied to the workpiece based on the inference result by the inference unit. According to Appendix B5, the inference error by the inference unit can be corrected by confirming the results of applying the multiple sequential actions through measurement and re-determining the action parameters. Additionally, the action parameters of the sequential actions performed and the resulting measurement results can be accumulated as learning data for improving the inference accuracy of the inference device.

[0098] (Appendix B6) The workpiece manufacturing apparatus according to Appendix B5, further comprising a robot that transports the workpiece between the execution unit and the measurement unit. According to Appendix B6, because the robot transports the workpiece, general-purpose devices can be used as the measurement unit and execution unit, rather than dedicated machines. Therefore, the cost of the manufacturing apparatus can be reduced. While manufacturing efficiency is affected by the number of transports by the robot, as shown in Appendix B5, it is sufficient to perform multiple actions sequentially before transporting, thereby achieving both improved manufacturing efficiency and reduced manufacturing apparatus costs.

[0099] (Appendix B7) A method for manufacturing a workpiece, comprising the steps of: setting a plurality of action parameters representing at least one of a position on the workpiece and the magnitude of an action when an action that causes a physical change in the workpiece is applied to the workpiece; inferring a physical change in the workpiece by inputting the set plurality of action parameters into an inference device that infers a physical change in the workpiece upon receiving input of at least the action parameters; determining one action parameter from the plurality of action parameters based on the inferred change; and applying an action according to the determined one action parameter to the workpiece. According to Appendix B7, it is possible to obtain the same technical effect as in Appendix B1.

[0100] 1, 1X... manufacturing system, 9... work, 11... measurement unit, 12... setting unit, 13, 13A, 13B, 13X... inference unit, 14... decision unit, 15... loop unit, 16... execution unit, 18... evaluation unit, 20, 20A, 20B, 20X... inference device

Claims

1. A manufacturing system comprising: a setting unit that sets candidate action parameters indicating an action on a workpiece, the candidate action parameters indicating at least one of the position of the action, the magnitude of the action, and the direction of the action; an inference unit that inputs the candidate action parameters to an inference device that accepts input of the candidate action parameters and infers the effect of the action on the workpiece, and infers a processing result of the workpiece using the candidate action parameters; a determination unit that determines actual action parameters based on the processing result; and an execution unit that applies the action to the workpiece using the actual action parameters.

2. The manufacturing system of claim 1, wherein the inference unit further accepts input of the physical state of the work to infer the effect, and the inference unit inputs the candidate action parameters and the current physical state of the work to the inference unit to infer the processing result.

3. The manufacturing system of claim 2, wherein the action is an action that changes the physical state of the work, and the execution unit imparts the action to the work according to the actual action parameters so as to change the current physical state of the work.

4. A manufacturing system as described in claim 2 or 3, wherein the setting unit sets a plurality of the candidate action parameters, the inference unit inputs each of the plurality of candidate action parameters to the inference device and infers the processing result for each of the plurality of candidate action parameters, and the determination unit determines the actual action parameter from the plurality of candidate action parameters based on the plurality of the processing results.

5. The manufacturing system described in claim 4, wherein the inference unit infers, as the processing result, an evaluation value of the next physical state of the work, which is the physical state of the work affected by the action, and the determination unit determines the actual action parameter from the multiple candidate action parameters based on multiple pairs of the candidate action parameters and the evaluation values.

6. The manufacturing system according to claim 5, wherein the inference unit infers the evaluation value as the effect, and the inference unit obtains the evaluation value inferred by the inference unit as the processing result.

7. The manufacturing system of claim 5, wherein the inference unit infers the next physical state as the effect, and the inference unit includes an evaluation unit that evaluates the next physical state inferred by the inference unit to calculate the evaluation value, and the evaluation value calculated by the evaluation unit is obtained as the processing result.

8. A manufacturing system as described in claim 2 or 3, further comprising a loop section which executes a series of repeated processes by the setting section, the inference section, and the determination section to determine a set of two or more actual action parameters indicating the two or more actions to be applied to the work in sequence, wherein the inference section infers a physical state of the work affected by the action as at least a part of the processing result, the setting section sets the candidate action parameter in each repetition, the inference section inputs the candidate action parameter and the previously inferred physical state of the work to the inference device in each repetition to infer the processing result for the candidate action parameter, the determination section determines the actual action parameter based on the processing result in each repetition, and the execution section applies the two or more actions to the work in sequence using the set of two or more actual action parameters obtained by the repetition.

9. The manufacturing system described in claim 8, further comprising a measurement unit that measures the physical state of the work to which the two or more actions have been applied, wherein the setting unit sets new candidate action parameters, the inference unit identifies the measured physical state as the current physical state, and inputs the new candidate action parameters and the identified current physical state to the inference device to infer a new processing result of the work, the determination unit determines new actual action parameters based on the new processing result, and the execution unit applies a new action to the work using the new actual action parameters.

10. The manufacturing system according to claim 9, further comprising a robot that transports the workpiece between the execution unit and the measurement unit.

11. A manufacturing system comprising: an inference unit that inputs a current physical state of a workpiece into an inference device that accepts input of the physical state of the workpiece and infers an action parameter indicating an action to be applied to the workpiece, and infers the action parameter corresponding to the current physical state as an actual action parameter, wherein the action parameter indicates at least one of the position of the action, the magnitude of the action, and the direction of the action; and an execution unit that applies the action to the workpiece based on the actual action parameter inferred from the current physical state.

12. A manufacturing method executed by a manufacturing system having at least one processor, comprising: a step of setting candidate action parameters indicating an action on a workpiece, the candidate action parameters indicating at least one of the position of the action, the magnitude of the action, and the direction of the action; a step of inputting the candidate action parameters to an inference device that accepts input of the candidate action parameters and infers the effect of the action on the workpiece, thereby inferring a processing result of the workpiece using the candidate action parameters; a step of determining actual action parameters based on the processing result; and a step of applying the action to the workpiece using the actual action parameters.

13. A manufacturing program that causes a computer to execute the following steps: setting candidate action parameters indicating an action on a workpiece, the candidate action parameters indicating at least one of the position of the action, the magnitude of the action, and the direction of the action; inputting the candidate action parameters into an inference device that accepts input of the candidate action parameters and infers the effect of the action on the workpiece, and inferring a processing result of the workpiece using the candidate action parameters; determining actual action parameters based on the processing result; and applying the action to the workpiece using the actual action parameters.

Images