Machine learning device, processing state prediction device, and control device

Abstract

The present invention relates to making a learning completion model and using the learning completion model to predict the processing state with high accuracy. The learning completion model can output the processing state of the workpiece after polishing under specified processing conditions with high accuracy without actually performing processing operation or simulation. The machine learning device has: an input data acquisition unit, which acquires processing information as input data in the polishing process of surface treatment of the processing surface of any workpiece with any tool, and the processing information includes the workpiece information before the polishing process and the processing condition information; a label acquisition unit, which acquires label data representing the processing state information, and the processing state information includes the processing state of the workpiece after the polishing process and the surface roughness of the workpiece when the processing state is normal; a learning unit, which uses the acquired input data and label data to perform supervised learning and generate a learning completion model, which inputs the processing information of the polishing process to be performed thereafter and outputs the processing state information in the polishing process.

Description

Technical Field

[0001] The present invention relates to a machine learning device, a processing state prediction device and a control device. Background Art

[0002] Polishing is a finishing method for controlling the surface roughness of workpieces such as machine parts after cutting to a certain value.

[0003] like Figure 8 (equivalent to Patent Document 1 described later) Figure 1 ), polishing is a plastic working method in which a roller (or spherical) tool T is pressed against the processing surface of a metal workpiece W and rolled simultaneously, thereby finishing the surface to be smooth. Since the vicinity of the processed part is hardened and compressive residual stress is applied, it is expected that the wear resistance and fatigue strength of the processed part will be improved. Therefore, it is mainly used in the finishing of automobile parts, etc.

[0004] In addition, Figure 8 In region A, the tool T is pressed against the surface of the workpiece W and gradually pressurized. In region B, the contact pressure exceeds the yield point of the workpiece W, causing plastic deformation. In region C, the tool T is separated from the machined surface, and the metal elasticity of the workpiece W recovers a small distance Dr. As a result, the surface irregularities of the portion where the tool T passes are flattened and the roughness is reduced.

[0005] Here, the distance Dr is also referred to as the elastic recovery amount. In addition, the distance Dv is the distance that the tool T is pressed from the surface of the workpiece W, also referred to as the rolling amount (or polishing amount). In addition, the distance Ds from the surface before polishing to the surface after polishing and elastic recovery of the distance Dr is also referred to as the dimensional change amount.

[0006] In this regard, the following technology is known: a response surface is prepared to show the relationship between the rolling amount and the surface roughness before processing for the surface roughness after processing, and the roller speed, feed speed, rolling amount, etc. of the polishing tool are determined based on the prepared response surface.

[0007] Prior art literature

[0008] Patent Literature

[0009] Patent Document 1: Japanese Patent No. 5527543 Summary of the invention

[0010] Problems to be solved by the invention

[0011] In the polishing process, the processing conditions such as the rolling amount, the relative rotation speed (roller speed) between the tool and the workpiece, and the relative feed speed between the tool and the workpiece must be appropriately adjusted according to the size and surface roughness of the material required after processing.

[0012] Especially in the case of hollow workpieces, such as Fig. 9 As shown, if the wall thickness of the workpiece is thin, it may cause fluctuations and deformation due to different rolling amounts, and it is difficult to adjust the rolling amount corresponding to the workpiece material (hardness) and wall thickness. Therefore, the processing conditions are often determined based on the experience of skilled workers.

[0013] In addition, plastic processing is accompanied by elastic recovery (the action of returning to the original shape), and the amount of elastic recovery varies depending on the workpiece material, hardness, and the polishing processing conditions. Therefore, when the required size and surface roughness cannot be achieved, it is necessary to readjust the processing conditions and reprocess. However, readjusting the processing conditions and reprocessing take time.

[0014] Therefore, it is desired to create a learning completion model and use the learning completion model to predict the processing state of the workpiece after processing with high accuracy, wherein the learning completion model can output the processing state of the workpiece after processing when polishing is performed under specified processing conditions with high accuracy without actually performing processing operations or simulations.

[0015] Means for solving problems

[0016] (1) One mode of the machine learning device disclosed in the present invention comprises: an input data acquisition unit, which acquires processing information as input data in a polishing process in which an arbitrary tool is pressed against a processing surface of an arbitrary workpiece for surface treatment, and the processing information at least includes information related to the workpiece before the polishing process and information related to processing conditions of the polishing process; a label acquisition unit, which acquires label data representing processing status information, and the processing status information includes the processing status of the workpiece after the polishing process and the surface roughness of the workpiece when the processing status is normal; and a learning unit, which uses the input data acquired by the input data acquisition unit and the label data acquired by the label acquisition unit to perform supervised learning, and generates a learning completion model, and the learning completion model is used to input processing information of a polishing process to be performed thereafter and output processing status information in the polishing process to be performed thereafter.

[0017] (2) One method of the processing state prediction device disclosed in the present invention comprises: a learning completion model, which is generated by the machine learning device of (1), and the learning completion model inputs processing information of a polishing process to be performed later and outputs the processing state information in the polishing process to be performed later; an input unit, which inputs processing information before the polishing process, and the processing information includes information related to the processing conditions of the polishing process to be performed later and information related to the workpiece as the processing object; and a prediction unit, which inputs the processing information input by the input unit into the learning completion model, thereby predicting the processing state information in the polishing process to be performed later output by the learning completion model.

[0018] (3) One aspect of the control device disclosed herein includes the machining state prediction device of (2).

[0019] Effects of the Invention

[0020] According to one embodiment, a learning model can be created that can accurately output the processing state of a workpiece after polishing under specified processing conditions without actually performing a processing operation or simulation. Furthermore, by using the learning model, the processing state of the workpiece after processing can be predicted with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a functional block diagram showing an example of a functional configuration of a prediction system according to one embodiment.

[0022] Figure 2 This is a diagram showing an example of information related to a workpiece, information related to processing conditions of a buffing process, and label data.

[0023] Figure 3 Yes means Figure 2 An example of the surface roughness before and after processing of each workpiece is shown in the figure.

[0024] Figure 4 It means providing to Figure 1 A diagram of an example of a learning completion model of a processing state prediction device.

[0025] Figure 5 This is a flowchart for explaining the prediction processing of the machining state prediction device in the operation phase.

[0026] Figure 6 This is a diagram showing an example of the configuration of a prediction system.

[0027] Figure 7 This is a diagram showing an example of the configuration of a prediction system.

[0028] Figure 8This is a schematic diagram for explaining the principle of polishing.

[0029] Fig. 9 This diagram shows an example of deformation caused by excessive rolling when the workpiece is hollow. DETAILED DESCRIPTION

[0030] Hereinafter, one embodiment of the present disclosure will be described using the drawings.

[0031] <One Implementation Method>

[0032] Figure 1 1 is a functional block diagram showing an example of a functional configuration of a prediction system according to one embodiment. Figure 1 As shown, the prediction system 1 includes a machine tool 10 , a machining state prediction device 20 , and a machine learning device 30 .

[0033] The machine tool 10, the processing state prediction device 20, and the machine learning device 30 can be directly connected to each other via a connection interface not shown in the figure. In addition, the machine tool 10, the processing state prediction device 20, and the machine learning device 30 can also be connected to each other via a network not shown in the figure, such as a LAN (Local Area Network), the Internet, etc. In this case, the machine tool 10, the processing state prediction device 20, and the machine learning device 30 have a communication unit not shown in the figure for communicating with each other through such a connection. In addition, as described later, the processing state prediction device 20 may include the machine learning device 30. In addition, the machine tool 10 may include the processing state prediction device 20 and the machine learning device 30.

[0034] The machine tool 10 is a well-known machine tool for those skilled in the art, and is assembled to have a control device 101. The machine tool 10 operates according to the operation command of the control device 101. As described later, the machine tool 10 can output processing information to the processing state prediction device 20 via a communication unit (not shown) of the machine tool 10 before the polishing process, and the processing information includes information related to the processing conditions of the polishing process to be performed thereafter and information related to the workpiece to be processed.

[0035] In addition, if Figure 8 As shown, the information related to the processing conditions of the polishing process may include the relative rotation speed of the tool T and the workpiece W, the relative feed speed of the tool T and the workpiece W, and the rolling amount Dv. In addition, the information related to the workpiece W to be processed may include the material, hardness, thickness, and surface roughness before processing of the workpiece W. Among them, the thickness (wall thickness) of the workpiece W may also be included in the information related to the workpiece W to be processed only when the workpiece W is a hollow workpiece.

[0036] The control device 101 is a numerical control device known to those skilled in the art, and generates an action command according to a program for polishing, and sends the generated action command to the machine tool 10. Thus, the control device 101 can cause the machine tool 10 to perform polishing. In addition, the control device 101 can output processing information to the processing state prediction device 20 via a communication unit of the machine tool 10 (not shown), instead of the machine tool 10, wherein the processing information includes information related to the processing conditions of the polishing process to be performed thereafter and information related to the workpiece to be processed.

[0037] In addition, the control device 101 may be a device independent of the machine tool 10 .

[0038] In the operation stage, the processing state prediction device 20 obtains processing information including information related to the processing conditions of the polishing process to be performed later and information related to the workpiece to be processed before the polishing process. In addition, the processing state prediction device 20 inputs the obtained processing information of the polishing process to be performed later into the learning completion model provided by the machine learning device 30 described later. Thus, the processing state prediction device 20 can predict the processing state information in the polishing process to be performed later.

[0039] The processing state information includes whether the processing state indicating no undulation or deformation in the processing target workpiece after the subsequent polishing process is normal or abnormal, and the surface roughness after processing when the processing state is normal.

[0040] Therefore, before explaining the machining state prediction device 20 , machine learning for generating a learned model will be explained.

[0041] <Machine Learning Device 30>

[0042] The machine learning device 30 acquires, for example, processing information including information on the workpiece before polishing and information on processing conditions of the polishing in polishing in which a tool is pressed against a processing surface of a workpiece to perform surface treatment.

[0043] In addition, the machine learning device 30 acquires data representing machining state information including the machining state of the workpiece after the polishing process and the surface roughness of the workpiece when the machining state is normal in the acquired input data as a label (correct solution).

[0044] The machine learning device 30 performs supervised learning using training data of a combination of acquired input data and labels, and constructs a learning completion model described later.

[0045] Thus, the machine learning device 30 can provide the constructed learning model to the processing state prediction device 20.

[0046] The machine learning device 30 will be described in detail.

[0047] like Figure 1 As shown, the machine learning device 30 includes an input data acquisition unit 301 , a label acquisition unit 302 , a learning unit 303 , and a storage unit 304 .

[0048] During the learning phase, the input data acquisition unit 301 acquires processing information as input data from the machine tool 10 via a communication unit (not shown), wherein the processing information includes information related to the workpiece before polishing in a polishing process in which an arbitrary tool is pressed onto a processing surface of an arbitrary workpiece for surface treatment, and information related to processing conditions of the polishing process.

[0049] Figure 2 This is a diagram showing an example of information related to a workpiece, information related to processing conditions of a buffing process, and label data.

[0050] like Figure 2 As shown, the workpiece-related information included in the processing information includes the "material", "Brinell hardness", "thickness" and "surface roughness before processing" of the workpiece.

[0051] In the case of carbon steel, the "material" included in the workpiece-related information includes "C15C", "S50C", "S55C", "S60C", etc. in addition to "S45C". In the case of cast iron, the "material" included in the workpiece-related information includes "FC100", "FC150", "FC200", "FC250", "FC300", "FC350", etc. In the case of aluminum alloy, the "material" included in the workpiece information includes "A4032", "A5052", "A5083", "A6061", "A7075", etc. in addition to "A5056" and "AC3A". In the case of magnesium alloy, the material included in the workpiece information includes "AZ31", "AZ91", etc.

[0052] As the “hardness” included in the information on the workpiece, “Brinell hardness” is set. Alternatively, “Vickers hardness” or the like may be set as the “hardness” included in the information on the workpiece.

[0053] As described above, "thickness" included in the workpiece-related information is set only when the workpiece is a hollow workpiece. Therefore, since the workpieces whose "material" is "S45C (carbon steel)", "SCM440 (chrome-molybdenum steel)", "SUS303 (chrome stainless steel)" and "A5056 (aluminum alloy)" are not hollow workpieces, "thickness" stores "-" indicating a blank column.

[0054] The "surface roughness before processing" included in the workpiece-related information is measured in advance before processing using a surface roughness measuring device based on a stylus, laser beam, etc., and is set using the value of arithmetic mean roughness (Ra). Among them, the "surface roughness before processing" can also be set using the value of maximum height (Ry) or ten-point average roughness (Rz).

[0055] The arithmetic mean roughness (Ra), the maximum height (Ry), and the ten-point average roughness (Rz) can be calculated using a known method (for example, refer to JIS B 0601: 1994, JIS B 0031: 1994, etc.), and a detailed description thereof will be omitted.

[0056] Then, if Figure 2 As shown, the information related to the processing conditions of the polishing process includes the "relative rotation speed (circumferential speed)", "relative feed speed" and "rolling amount" in the polishing process performed on each workpiece represented by the above-mentioned workpiece-related information. In addition, the "relative rotation speed (circumferential speed)" is, for example, Figure 8 As shown, it represents the relative rotation speed of the tool T and the workpiece W. In addition, the "relative feed speed" represents the relative feed speed of the tool T and the workpiece W. In addition, the "crushing amount" represents the distance Dv.

[0057] Then, the input data acquisition unit 301 stores the acquired input data in the storage unit 304 .

[0058] The label acquisition unit 302 acquires processing state information as label data (correct solution data), wherein the processing state information includes the processing state of the workpiece after polishing of each input data and the surface roughness of the workpiece when the processing state is normal. The label acquisition unit 302 stores the acquired label data in the storage unit 304.

[0059] Specifically, for example Figure 2 As shown, the label acquisition unit 302 acquires the "processing state" in which the absence of undulation and deformation is set as "normal" and the presence of undulation and deformation is set as "abnormal" in the workpiece after polishing as label data. In addition, the label acquisition unit 302 acquires the "surface roughness after processing" measured in the workpiece with a normal "processing state" as label data, for example, using a surface roughness measuring device based on a stylus, a laser beam, etc., as in the case of the "surface roughness before processing".

[0060] also, Figure 2The "processing state" of the label data is, for example, the result of visually judging the state of the workpiece by the operator after processing, and can be input by the operator. In addition, the "surface roughness after processing" of the label data is, for example, the surface roughness of the workpiece measured by the operator using a surface roughness measuring machine after processing, and can be input by the operator. Alternatively, the "processing state" and "surface roughness after processing" of the label data can be, for example, the state of the workpiece judged by image processing of the workpiece image taken after processing, and the measured surface roughness.

[0061] In addition, the label is not limited to "processing status" normal or abnormal, and may be represented by two values ​​such as "1" and "0".

[0062] In addition, Figure 2 The workpiece whose "Material" is "AC3A (aluminum alloy)" is a hollow workpiece, such as Fig. 9 As shown, undulations, deformation, etc. are generated due to the polishing process, so the "processing state" is determined to be "abnormal" and "-" indicating a blank column is stored in the "surface roughness after processing".

[0063] Figure 3 Yes means Figure 2 An example of the surface roughness before and after processing of each workpiece is shown in the figure.

[0064] Here, the "surface roughness after processing" is set by the value of arithmetic mean roughness (Ra), but it can also be set by the value of maximum height (Ry) or ten-point average roughness (Rz).

[0065] The learning unit 303 receives the above-mentioned input data and the label combination as training data. The learning unit 303 performs supervised learning using the received training data, thereby inputting processing information including information indicating processing conditions for the subsequent polishing process and information related to the workpiece as the processing object, and constructing a learning completion model 250 for outputting processing state information indicating the processing state of the processing object workpiece after the subsequent polishing process.

[0066] Then, the learning unit 303 provides the constructed learning completed model 250 to the machining state prediction device 20 .

[0067] In addition, it is preferable to prepare a plurality of training data for supervised learning. For example, the training data may be acquired from machine tools 10 actually operated in various locations such as a customer's factory.

[0068] Figure 4 It means providing to Figure 1 FIG. 2 is a diagram showing an example of a learning completion model 250 of the machining state prediction device 20. Figure 4As shown, the learning completion model 250 example is a multi-layer neural network that uses information related to the processing object workpiece and information related to the processing conditions of the subsequent polishing process as an input layer, and uses processing state information representing the processing state of the processing object workpiece after the subsequent polishing process as an output layer.

[0069] Here, the information related to the workpiece to be processed includes the material, Brinell hardness, thickness, and surface roughness before processing. In addition, the information related to the processing conditions of the subsequent polishing process includes the relative rotation speed between the tool and the workpiece, the relative feed speed between the tool and the workpiece, and the rolling amount. In addition, the processing state information includes "normal or abnormal processing state" of the workpiece to be processed, and "surface roughness after processing".

[0070] In addition, after constructing the learned model 250 , when new training data is obtained, the learning unit 303 may update the previously constructed learned model 250 by further performing supervised learning on the learned model 250 .

[0071] Thus, training data can be automatically obtained from the normal machining operation of the machine tool 10 , and thus the prediction accuracy of the machining state of the workpiece can be improved on a daily basis.

[0072] The above-mentioned supervised learning can be performed through online learning, batch learning, or small batch learning.

[0073] Online learning refers to a learning method in which supervised learning is performed immediately whenever training data is generated while machining is performed on the machine tool 10. In addition, batch learning refers to a learning method in which multiple training data corresponding to the repetitions are collected while machining is performed on the machine tool 10, and supervised learning is performed using all the collected training data. Furthermore, mini-batch learning refers to a learning method between online learning and batch learning in which supervised learning is performed whenever a certain amount of training data is accumulated.

[0074] The storage unit 304 is a RAM (Random Access Memory) or the like, and stores input data acquired by the input data acquisition unit 301 , label data acquired by the label acquisition unit 302 , and the learned model 250 constructed by the learning unit 303 .

[0075] In the above, the machine learning for generating the learned model 250 included in the machining state prediction device 20 has been described.

[0076] Next, the machining state prediction device 20 in the operational stage will be described.

[0077] <Processing state prediction device 20 in the operation stage>

[0078] like Figure 1 As shown, the machining state prediction device 20 in the operation phase is configured to include an input unit 201 , a prediction unit 202 , a determination unit 203 , a notification unit 204 , and a storage unit 205 .

[0079] In addition, the processing state prediction device 20 is used to achieve Figure 1 The processing state prediction device 20 has a CPU (Central Processing Unit) and other arithmetic processing devices (not shown) for controlling the operation of the functional blocks. In addition, the processing state prediction device 20 has a ROM (Read Only Memory) for storing various control programs, an HDD (Hard Disk Drive) and other auxiliary storage devices (not shown), and a RAM (not shown) for storing data temporarily required when the arithmetic processing device executes the program.

[0080] Furthermore, in the machining state prediction device 20, the operation processing device reads the OS and application software from the auxiliary storage device, expands the read OS and application software in the main storage device, and performs operation processing based on these OS and application software. The machining state prediction device 20 controls each hardware according to the operation result. In this way, the machining state prediction device 20 realizes the machining state prediction based on the processing result. Figure 1 That is, the processing state prediction device 20 can be implemented by the cooperation of hardware and software.

[0081] Before the buffing process, the input unit 201 inputs processing information including information on processing conditions of the buffing process to be performed later and information on the workpiece to be processed from the machine tool 10 , and outputs the input processing information to the prediction unit 202 .

[0082] In addition, the information related to the processing conditions of the polishing process performed thereafter may include the relative rotation speed between the tool and the workpiece, the relative feed speed between the tool and the workpiece, and the rolling amount. In addition, the information related to the workpiece to be processed may include the material, hardness, thickness (in the case of a hollow workpiece), and the surface roughness before processing.

[0083] The prediction unit 202 inputs information related to the processing conditions of the polishing process to be performed later and information related to the workpiece to be processed, which are included in the processing information input by the input unit 201. Figure 3 The learning model 250 is thus completed. Thus, the prediction unit 202 can predict the "normal or abnormal processing state" of the workpiece in the subsequent polishing process and the "surface roughness after processing" when the "processing state is normal" based on the output of the learning model 250.

[0084] The decision unit 203 determines whether the “machining state” predicted by the prediction unit 202 is normal or abnormal.

[0085] When the "machining state is normal" predicted by the prediction unit 202, the decision unit 203 compares the "surface roughness after machining" with the preset required accuracy α to determine whether the "surface roughness after machining" is within the required accuracy α. When the "surface roughness after machining" is within the required accuracy α, the decision unit 203 determines to cause the machine tool 10 to perform a polishing process on the workpiece based on the input machining information.

[0086] On the other hand, when the “machining state is not normal” (ie, the machining state is abnormal) or when the “surface roughness after machining” is not within the required accuracy α, the decision unit 203 decides to change the machining information of the subsequent polishing process.

[0087] Thus, the machining state prediction device 20 can prompt the operator of the machine tool 10 to re-evaluate the machining conditions of the polishing machining to be performed later so that the “machining state is normal” and the “surface roughness after machining” is within the required accuracy α.

[0088] The required accuracy α can be appropriately set according to the cycle time, machining accuracy, and the like required of the machine tool 10 .

[0089] When the notification unit 204 receives a decision from the decision unit 203 to cause the machine tool 10 to perform a polishing process based on the input processing information, it can output an execution instruction for the polishing process based on the input processing information to an output device such as a liquid crystal display included in the machine tool 10 and / or the control device 101 (not shown).

[0090] On the other hand, upon receiving a decision to change the machining information from the decision unit 203 , the notification unit 204 may output an instruction to change machining conditions of the subsequent polishing process to an output device (not shown) of the machine tool 10 and / or the control device 101 .

[0091] Furthermore, the notification unit 204 may make notifications by sound via a speaker (not shown).

[0092] The storage unit 205 is a ROM, a HDD, or the like, and can store the learned model 250 and the required accuracy α together with various control programs.

[0093] <Prediction Processing of the Machining State Prediction Device 20 in the Operation Phase>

[0094] Next, the operation related to the prediction process of the machining state prediction device 20 according to the present embodiment will be described.

[0095] Figure 5This is a flowchart for explaining the prediction processing of the machining state prediction device 20 in the operation phase. The flow shown here is repeatedly executed every time machining information is input.

[0096] In step S11 , the input unit 201 acquires, from the machine tool 10 , processing information including information on processing conditions of the subsequent buffing process and information on the workpiece to be processed, before the buffing process.

[0097] In step S12 , the prediction unit 202 inputs the processing information input in step S11 into the learned model 250 , thereby predicting “whether the processing state is normal or abnormal” and “the surface roughness after processing” of the processing target workpiece.

[0098] In step S13, the decision unit 203 determines whether the processing state predicted in step S12 is normal or abnormal. If the processing state is normal, the process proceeds to step S14. On the other hand, if the processing state is abnormal, the process proceeds to step S16.

[0099] In step S14, when the "processing state is normal" determined in step S13, the decision unit 203 compares the "processed surface roughness" with the preset required accuracy α to determine whether the "processed surface roughness" is within the required accuracy α. When the "processed surface roughness" is within the required accuracy α, the process proceeds to step S15. On the other hand, when the "processed surface roughness" is not within the required accuracy α, the process proceeds to step S16.

[0100] In step S15 , the notification unit 204 outputs an instruction to execute the polishing process based on the input processing information to an output device (not shown) of the machine tool 10 and / or the control device 101 .

[0101] In step S16, the notification unit 204 outputs a change instruction of the processing information of the polishing process to be performed thereafter to the output device (not shown) of the machine tool 10 and / or the control device 101. In this case, the operator of the machine tool 10 re-evaluates the processing conditions of the polishing process to be performed thereafter so that the "processing state is normal" and the "surface roughness after processing" is within the required accuracy α, and inputs the re-evaluated processing information to the processing state prediction device 20 (i.e., returns to the process of step S11).

[0102] Through the above, the processing state prediction device 20 of one embodiment inputs the processing information including the information related to the processing conditions of the polishing process to be performed later and the information related to the workpiece as the processing object into the learning completion model 250 before the polishing process, thereby predicting whether the processing state of the workpiece as the processing object of the polishing process to be performed later is normal or abnormal, and the surface roughness after processing.

[0103] Thus, the machining state prediction device 20 can predict the machining state of the workpiece after the machining when the buffing machining is performed under the designated machining conditions, without actually performing machining operation or simulation.

[0104] Specifically, after constructing the learning model, without performing polishing operations or measuring the workpiece after processing, it is possible to easily predict whether processing can be performed normally without causing fluctuations or deformation under the processing conditions to be implemented later, and further, what degree of surface roughness can be obtained.

[0105] This can eliminate the labor of trial and error such as readjustment of the processing conditions of the polishing process and reprocessing.

[0106] Although one embodiment has been described above, the machining state prediction device 20 and the machine learning device 30 are not limited to the above-described embodiment, and include modifications and improvements within a range that can achieve the purpose.

[0107] <Modification 1>

[0108] In the above-mentioned embodiment, the machine learning device 30 is illustrated as a device different from the machine tool 10 , the control device 101 , and the processing state prediction device 20 , but the machine tool 10 , the control device 101 , or the processing state prediction device 20 may have some or all of the functions of the machine learning device 30 .

[0109] <Modification 2>

[0110] In addition, for example, in the above-mentioned embodiment, the machining state prediction device 20 is illustrated as a device different from the machine tool 10 and the control device 101 , but the machine tool 10 or the control device 101 may have a part or all of the functions of the machining state prediction device 20 .

[0111] Alternatively, for example, the server may include a part or all of the input unit 201, prediction unit 202, determination unit 203, notification unit 204, and storage unit 205 of the machining state prediction device 20. In addition, the various functions of the machining state prediction device 20 may be implemented on the cloud using a virtual server function or the like.

[0112] Furthermore, the machining state prediction device 20 may be a distributed processing system in which the functions of the machining state prediction device 20 are appropriately distributed to a plurality of servers.

[0113] <Variation 3>

[0114] In addition, for example, in the above-mentioned embodiment, the processing state prediction device 20 uses the processing information of the polishing process to be performed later provided by the machine learning device 30 and outputs the learning completion model 250 based on the processing state information of the workpiece to be processed, and predicts the "processing state is normal or abnormal" and "the surface roughness after processing" of the workpiece to be processed in the polishing process to be performed later based on the input processing information, but it is not limited to this. Figure 6 As shown, the server 50 can store the learning model 250 generated by the machine learning device 30, and share the learning model 250 with m machining state prediction devices 20A(1) to 20A(m) connected to the network 60 (m is an integer greater than or equal to 2). Thus, even if a new machine tool and machining state prediction device are configured, the learning model 250 can be applied.

[0115] Furthermore, each of the machining state prediction devices 20A( 1 )- 20A(m) is connected to each of the machine tools 10A( 1 )- 10A(m).

[0116] In addition, each of the machine tools 10A(1)-10A(m) is connected to Figure 1 Each of the machining state prediction devices 20A(1)-20A(m) corresponds to the machine tool 10. Figure 1 The processing state prediction device 20 corresponds to.

[0117] Or, if Figure 7 As shown, the server 50 can operate as a processing state prediction device 20, for example, and predict the processing state information of "processing state is normal or abnormal" and "surface roughness after processing" of the workpiece based on the processing information of the polishing process to be performed later, for each of the machine tools 10A(1)-10A(m) connected to the network 60. Thus, even if a new machine tool is configured, the learning completion model 250 can be applied.

[0118] <Variation 4>

[0119] In addition, for example, in the above-mentioned embodiment, the machine learning device 30 performs supervised learning, but is not limited to this, and a learning model may be constructed by other learning methods (for example, reinforcement learning with + reward / - reward, etc.).

[0120] In addition, each function included in the machining state prediction device 20 and the machine learning device 30 in one embodiment can be realized by hardware, software, or a combination thereof. Here, realization by software means realization by reading a program into a computer and executing it.

[0121] The various structural parts included in the processing state prediction device 20 and the machine learning device 30 can be implemented by hardware including electronic circuits, software or a combination thereof. When implemented by software, the program constituting the software is installed on a computer. In addition, these programs can be recorded on a removable medium to be released to users, or they can be released by downloading them to the user's computer via a network. In addition, when composed of hardware, for example, integrated circuits (ICs) such as ASIC (Application Specific Integrated Circuit), gate arrays, FPGA (Field Programmable Gate Array), and CPLD (Complex Programmable Logic Device) can constitute part or all of the functions of the various structural parts included in the above-mentioned device.

[0122] The program can be stored using various types of non-transitory computer readable media (Non-transitorycomputer readable medium) and provided to the computer. Non-transitory computer readable media include various types of tangible storage media (Tangible storage medium). Examples of non-transitory computer readable media include magnetic recording media (e.g., floppy disks, magnetic tapes, hard disk drives), optical magnetic recording media (e.g., optical magnetic disks), CD-ROM (Read Only Memory), CD-R, CD-R / W, semiconductor memory (e.g., mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM). In addition, the program can also be provided to the computer through various types of transient computer readable media (Transitory computer readable medium). Examples of transient computer readable media include electrical signals, optical signals, and electromagnetic waves. Transitory computer readable media can provide the program to the computer via wired communication paths such as wires and optical fibers or wireless communication paths.

[0123] Furthermore, the steps describing the program recorded in the recording medium naturally include processing that is performed in time series in the order in which they are performed, and also include processing that is not necessarily performed in time series but is executed in parallel or individually.

[0124] In other words, the machine learning device, processing state prediction device, and control device of the present disclosure can take various embodiments having the following structures.

[0125] (1) The machine learning device 30 disclosed in the present invention comprises: an input data acquisition unit 301, which acquires processing information as input data in a polishing process in which an arbitrary tool is pressed onto a processing surface of an arbitrary workpiece for surface treatment, and the processing information at least includes information related to the workpiece before the polishing process and information related to the processing conditions of the polishing process; a label acquisition unit 302, which acquires label data representing processing status information, and the processing status information includes the processing status of the workpiece after the polishing process and the surface roughness of the workpiece when the processing status is normal; and a learning unit 303, which uses the input data acquired by the input data acquisition unit 301 and the label data acquired by the label acquisition unit 302 to perform supervised learning and generate a learning completion model 250, which is used to input processing information of the polishing process to be performed later and output processing status information in the polishing process to be performed later.

[0126] According to the machine learning device 30, it is possible to generate a learned model 250 that can output the machining state of a workpiece after polishing under specified machining conditions with high accuracy without actually performing machining operation or simulation.

[0127] (2) In the machine learning device 30 described in (1), the information related to the workpiece may include at least one of the material, hardness, thickness and surface roughness of the workpiece before processing, and the information related to the processing conditions of the polishing process may include at least one of the relative rotation speed between the tool and the workpiece, the relative feed speed between the tool and the workpiece and the grinding amount.

[0128] Thus, the machine learning device 30 can generate the learning completed model 250 that outputs the machining state information corresponding to the information on the workpiece before the polishing process and the information on the machining conditions of the polishing process.

[0129] (3) In the machine learning device 30 described in (1) or (2), the processing state information may include at least one of a normal or abnormal processing state indicating no undulation and deformation in the workpiece after polishing, and a surface roughness after processing when the processing state is normal.

[0130] Thus, the machine learning device 30 can generate the learned model 250 that outputs the machining state information of the machined workpiece corresponding to the polishing process to be performed later with high accuracy.

[0131] (4) The processing state prediction device 20 disclosed in the present invention comprises: a learning completion model 250, which is generated by the machine learning device 30 described in any one of (1) to (3), and the learning completion model 250 inputs processing information of the polishing process to be performed later and outputs processing state information in the polishing process to be performed later; an input unit 201, which inputs processing information before the polishing process, and the processing information includes information related to the processing conditions of the polishing process to be performed later and information related to the workpiece as the processing object; and a prediction unit 202, which inputs the processing information input by the input unit 201 into the learning completion model 250, thereby predicting the processing state information in the polishing process to be performed later output by the learning completion model 250.

[0132] According to the machining state prediction device 20, it is possible to predict the machining state of a workpiece after machining when buffing is performed under designated machining conditions without actually performing machining operation or simulation.

[0133] (5) In the processing state prediction device 20 described in (4), the information related to the workpiece may include at least one of the material, hardness, thickness and surface roughness of the workpiece before processing, and the information related to the processing conditions of the polishing process may include at least one of the relative rotation speed between the tool and the workpiece, the relative feed speed between the tool and the workpiece and the grinding amount.

[0134] Thus, the machining state prediction device 20 can predict machining state information corresponding to the information on the workpiece before the buffing and the information on the machining conditions of the buffing.

[0135] (6) In the processing state prediction device 20 described in (4) or (5), the processing state information may also include: in the workpiece as the processing object after the subsequent polishing process, whether the processing state without undulation and deformation is normal or abnormal, and at least one of the surface roughness after processing when the processing state is normal.

[0136] Thus, the machining state prediction device 20 can predict the machining state information of the machined workpiece corresponding to the polishing process to be performed later with high accuracy.

[0137] (7) In the processing state prediction device 20 described in (6), the processing state prediction device 20 may also include: a determination unit 203, which compares the surface roughness after processing predicted by the prediction unit 202 with a preset required accuracy α, and determines whether the surface roughness after processing is within the required accuracy α.

[0138] Thus, the machining state prediction device 20 can prompt the operator of the machine tool 10 to re-evaluate the machining conditions of the polishing process to be performed later so that the machining state is normal and the surface roughness after machining is within the required accuracy α.

[0139] (8) In the machining state prediction device 20 described in any one of (4) to (7), the learned model 250 may be provided in a server 50 that is connected to the machining state prediction device 20 via the network 60 so as to be accessible.

[0140] Thus, even if a new machine tool 10 , control device 101 , and machining state prediction device 20 are installed, the machining state prediction device 20 can apply the learned model 250 .

[0141] (9) In the machining state prediction device 20 described in any one of (4) to (8), the machining state prediction device 20 may include the machine learning device 30 described in any one of (1) to (3).

[0142] Thus, the machining state prediction device 20 can obtain the same effect as any of the above-mentioned (1) to (8).

[0143] (10) The control device 101 of the present disclosure includes the machining state prediction device 20 according to any one of (4) to (9).

[0144] According to the control device 101, the same effect as any one of the above-mentioned (4) to (9) can be obtained.

[0145] Explanation of symbols

[0146] 10 Machine Tools

[0147] 101 Control Device

[0148] 20 Processing state prediction device

[0149] 201 Input

[0150] 202 Forecasting Department

[0151] 203 Decision Department

[0152] 204 Notification Department

[0153] 205 Storage

[0154] 250 Learning Complete Model

[0155] 30 Machine Learning Devices

[0156] 301 Input data acquisition unit

[0157] 302 Label acquisition unit

[0158] 303 Learning Department

[0159] 304 Storage

[0160] 50 servers

[0161] 60 Network.

Claims

1. A machine learning device, characterized in that: have: An input data acquisition unit that acquires processing information as input data in a polishing process in which a tool of any roller or spherical shape is pressed against a processing surface of any workpiece while pressing the workpiece without cutting to finish the surface to be smooth, the processing information at least including information related to the workpiece before the polishing process and information related to processing conditions of the polishing process; a label acquisition unit that acquires label data indicating processing status information, the processing status information including the processing status of the workpiece after the polishing process and the surface roughness of the workpiece when the processing status is normal; and a learning unit that performs supervised learning using the input data acquired by the input data acquisition unit and the label data acquired by the label acquisition unit, and generates a learning completion model that inputs processing information of a polishing process to be performed later and outputs processing state information in the polishing process to be performed later, The workpiece-related information includes at least one of the material, hardness, thickness and surface roughness of the workpiece before processing. The information related to the processing conditions of the polishing process includes at least one of a relative rotation speed between the tool and the workpiece, a relative feed speed between the tool and the workpiece, and a rolling amount.

2. The machine learning device according to claim 1, characterized in that The processing state information includes at least one of whether the processing state indicating no undulation and deformation in the workpiece after the polishing process is normal or abnormal, and the surface roughness after processing when the processing state is normal.

3. A processing state prediction device, characterized in that: have: A learning completion model generated by the machine learning device according to claim 1 or 2, wherein the learning completion model inputs processing information of a polishing process to be performed thereafter and outputs the processing state information in the polishing process to be performed thereafter; an input unit for inputting processing information before the polishing process, the processing information including information related to processing conditions of the polishing process to be performed thereafter and information related to a workpiece to be processed; and a prediction unit that inputs the processing information input by the input unit into the learning completion model, thereby predicting the processing state information in the polishing process to be performed thereafter output by the learning completion model, The workpiece-related information includes at least one of the material, hardness, thickness and surface roughness of the workpiece before processing. The information related to the processing conditions of the polishing process includes at least one of a relative rotation speed between the tool and the workpiece, a relative feed speed between the tool and the workpiece, and a rolling amount.

4. The processing state prediction device according to claim 3, characterized in that: The processing state information includes at least one of whether the processing state is normal or abnormal, indicating no undulation or deformation, in the workpiece to be processed after the subsequent polishing process, and the surface roughness after processing when the processing state is normal.

5. The processing state prediction device according to claim 4, characterized in that: The machining state prediction device includes a determination unit that compares the post-machining surface roughness predicted by the prediction unit with a preset required accuracy and determines whether the post-machining surface roughness is within the required accuracy.

6. The processing state prediction device according to claim 3, characterized in that: The learned model is provided in a server connected to the machining state prediction device in an accessible manner via a network.

7. The processing state prediction device according to claim 3, characterized in that: The processing state prediction device has the machine learning device described in claim 1 or 2.

8. A control device, characterized in that: have: The machining state prediction device according to any one of claims 3 to 7.

Images