Apparatus for predicting machining dimensions, system for predicting machining dimensions, method for predicting machining dimensions and program

Abstract

Apparatus (10) for predicting machining dimensions, comprising: a trend detection means (11) for detecting trend information for each of a plurality of workpieces (40) on which machining is carried out, wherein the trend information indicates a state trend of a machining tool (20) during a machining period from a start to an end of the machining performed by the machining tool (20); feature calculating means (14) for calculating a feature based on the trend information using the state trend in each of a plurality of sections included in the machining period; a measured value detecting means (15) for detecting a measured value of a dimension of each of the plurality of workpieces (40) after machining; and a predicting means (18) for predicting, when a new target workpiece (41) is machined, a dimension of the new target workpiece (41) after machining; characterized by a section specifying means (16) for specifying a section having a calculated feature with a highest degree of relevance as a specific section of the plurality of sections, where the relevance degree is a relevance degree for the measured value; and wherein the predicting means (18) predicts the dimension of the new target workpiece (41) after machining based on the feature calculated using the state trend in the specific portion.

Description

Technical FieldThe present disclosure relates to a machining dimension prediction apparatus, a machining dimension prediction system, a machining dimension prediction method, and a program.Prior ArtManufacturing lines at factory automation (FA) locations often include processes for machining workpieces with machining tools. The machined workpieces may be inspected to determine if the dimensions are within tolerances. However, it is difficult and time-consuming to check all the workpieces after machining. Therefore, methods for predicting machining dimensions have been developed (for example, see Patent Literature 1).Patent Literature 1 describes a method of extracting features from information on the driving state of a machining tool collected in a predetermined machining section and using the features to create a prediction model for predicting the machining dimensions. The prediction model may determine the machining quality or determine whether the machining has been normally performed without measuring the machining dimensions after the machining.Patent Literature 2 discloses a machine tool that can measure a surface state of a workpiece or determine whether a tool needs to be corrected or replaced without using a special sensor.Patent Literature 3 discloses a grinding state monitoring method, a grinding state monitoring program, and an apparatus in which prediction accuracy of a grinding quality can be improved.Patent Literature 4 discloses a machine tool that is controlled in consideration of the quality of a manufactured workpiece.Citing listPatent LiteraturePatent Literature 1: Japanese Patent No. JP 6 833 090 B2;Patent Literature 2: JP 2020-69 600 A;Patent Literature 3: JP 2021-10 980 A; andPatent Literature 4: EP 4 035 829 B1.Non-Patent LiteratureNon-Patent Literature 1: ZHANG, Jing et al., Fault Detection and Classification of Time Series Using Localized Matrix Profiles. In: 2019 IEEE International Conference on Prognostics and Health Management (ICPHM), IEEE, 2019, pp. 1-7.Overview of the InventionTechnical ProblemIn uniform cutting with a cutting tool for forming a straight groove having a uniform width, the method described in Patent Literature 1 can predict the machining dimensions with a small error from the actually measured machining dimensions. However, in curved cutting or partial cutting, a cut at measurement positions is nonuniform due to, for example, the angle at which a cutting tool comes into contact with the machining surface or the performance of curve interpolation control performed by an NC machining tool. Therefore, the features extracted from the information collected in a predetermined processing section can insufficiently reflect the dimension of a part measured in the post-processing inspection. When the prediction model is trained to predict the machining dimensions on the basis of the features described in Patent Literature 1, the prediction model may have a large error in the predicted machining dimensions. Therefore, a more accurate prediction is expected for the dimension of a workpiece machined with the machining tool.In response to the above-described problem, an object of the present disclosure is to more accurately predict the dimension of a workpiece machined by a machining tool.Solution of the ProblemTo achieve the above object, there are provided a machining dimension prediction apparatus, a machining dimension prediction method, a machining dimension prediction system and a program, each having the features of the appended independent claims. Advantageous embodiments are defined in the dependent claims, the description and the drawings. Specifically, the machining dimension prediction apparatus according to an aspect of the present disclosure includes trend acquisition means for acquiring trend information for each of a plurality of workpieces on which machining is performed indicating a state trend of a machining tool during a machining period from a start to an end of the machining performed by the machining tool, feature calculation means for calculating a feature based on the trend information using the state trend in each of a plurality of portions included in the machining period, measurement value acquisition means for acquiring a measurement value of a dimension of each of the plurality of workpieces after the machining, portion specification means for specifying, as a specific portion of the plurality of portions, a portion having a calculated feature with a highest degree of relevance to the measurement value, and prediction means for predicting a dimension of the new target workpiece after machining based on the feature calculated using the state trend in the specific section when a new target workpiece is machined.Advantageous Effects of the InventionIn the apparatus according to the above aspect of the present disclosure, the section specifying means specifies, as the specific section of the plurality of sections, the section including the calculated feature having the greatest degree of relevance to the measurement value. The prediction means predicts the dimension of the new target workpiece after machining based on the calculated feature using the state trend in the specific section. In other words, the dimension is predicted based on the feature in the portion in the machining period most relevant to the actual dimension. This enables more accurate prediction of the dimensions of the workpiece machined with the machining tool.Brief Description of the DrawingsFIG. 1 is a block diagram of a machining dimension prediction system according to Embodiment 1; FIG. 2 is a diagram of a workpiece in Embodiment 1 showing a machining example; FIG. 3 is a diagram describing the processing in Embodiment 1; FIG. 4 is a block diagram of a machining dimension prediction apparatus according to Embodiment 1, showing the hardware configuration; FIG. 5 is a block diagram of the machining dimension prediction apparatus according to Embodiment 1, which describes the prediction of a machining dimension; FIG. 6 is a functional block diagram of the machining dimension prediction device according to Embodiment 1; FIG. 7 is a graph showing an exemplary state trend of a machining tool in Embodiment 1; FIG. 8 is a graph describing the example calculation of features in Embodiment 1; FIG. 9 is a table showing an example calculation of the features in Embodiment 1; FIG. 10 is a table showing an example calculation of a degree of relevance in Embodiment 1; FIG. 11 is a graph showing an exemplary prediction model in Embodiment 1; FIG. 12 is a flowchart of a training process in Embodiment 1; FIG. 13 is a flowchart of a verification process in Embodiment 1; FIG. 14 is a flowchart of a prediction process in Embodiment 1; FIG. 15 is a graph describing the calculation of features in Embodiment 2; FIG. 16 is a table showing an example calculation of the features in Embodiment 2; FIG. 17 is a diagram describing training of a prediction model in Embodiment 3; FIG. 18 is a diagram describing a priority level in Embodiment 4; FIG. 19 is a graph describing a degree of deviation in Embodiment 5; FIG. 20 is a first graph describing the setting of sections in Embodiment 6; FIG. 21 is a second graph describing the setting of sections in Embodiment 6; and FIG. 22 is a diagram of a measurement section in a modification.DESCRIPTION OF THE EMBODIMENTSPrediction of a machining dimension according to one or more embodiments of the present disclosure will be described in detail below with reference to the drawings.Embodiment 1As shown in FIG. 1, a machining dimension prediction apparatus 10 according to the present embodiment is included in a machining dimension prediction system 100, together with a machining tool 20 for machining a workpiece 40, a sensor 21 for measuring the state of the machining tool 20, and a measurement device 30 for acquiring coordinate information about the machining tool 20 and acquiring a measurement value of the dimension of the machined workpiece 40 for quality inspection. The machining dimension prediction system 100 is part of a production system installed as a factory automation system (FA) in a factory. The machining dimension prediction system 100 can predict the machining dimensions with higher accuracy by specifying information particularly affecting the machining dimensions in the information collected by the machining tool 20. The prediction performed by the machining dimension prediction system 100 refers to estimation of the machining dimensions without actually measuring the dimension.The machining tool 20 is a numerically controlled (NC) machining tool such as a milling machine, a lathe, a drilling machine, a machining center, or a turning center. The machining tool 20 executes a predefined NC control program to machine the workpiece 40 using a tool. Examples of the tool include an end mill, a face mill, a drill, a tap, and a tip. The workpiece 40 is an object to be machined. Examples of the machining include cutting, grinding, and machining.FIGS. 2 and 3 each show an example of machining the workpiece 40. in the example of FIG. 2, the machining tool 20 includes a tool 201 that is an end mill that is brought into contact with the metal workpiece 40 as indicated by the arrow to cut the workpiece 40 and form a helical blade. More specifically, the machining tool 20 includes the tool 201 moved in the direction of the thick arrow in FIG. 3 to form an inward surface 401 indicated by the thick line and then moved in the direction of the dashed arrow to form an outward surface 402 indicated by the dashed line. The workpiece 40 thus processed is used as a component of a scroll compressor. The vane dimensions of the scroll compressor must be very accurate. Therefore, machining of the blade requires accurate dimensional prediction.The machining tool 20 is connected to the machining dimension prediction device 10 via a communication line or an analog signal line, and transmits, to the machining dimension prediction device 10, state trend information indicating the state trend of the machining tool 20 that is periodically detected with a built-in sensor. The state trend of the machining tool 20 reflects a change in the wear state of the tool, the state of the workpiece 40, or the contact state between the tool and the workpiece 40 caused by the machining. The state of the machining tool 20 is indicated by, for example, the value of the torque or current flowing through a spindle motor for rotating the tool, the value of the torque or current flowing through a motor of a feed screw for a table to which the workpiece 40 is attached, the coordinate values of the spindle or table, or the temperature or the vibration amount of the machining tool 20. The state trend information may be transmitted during the machining in real time, or the results of the measurement during the machining may be transmitted together after the machining.The sensor 21 is a state trend information collector for periodically acquiring information on the state of the machining tool 20. In some embodiments, the sensor 21 may be incorporated into the machining tool 20 as a functional component as described above. The sensor 21 is connected to the machining dimension prediction device 10 via a communication line or an analog signal line, and transmits the collected result to the machining dimension prediction device 10. The result detected by the sensor 21 is assigned to a single identification number, such as a production serial number corresponding to the single workpiece 40. The machining dimension prediction device 10 extracts, from the result collected by the sensor 21, information related to the machining performed by the machining tool 20 based on the individual identification number and the machining progress (state trend) notified of by the machining tool 20.The machining dimension prediction system 100 can eliminate either the transmission of the result collected by the machining tool 20 or the transmission of the result collected by the sensor 21. The machining dimension prediction system 100 can eliminate the sensor 21 when the transmission of the measurement result from the sensor 21 is eliminated.The measuring device 30 is connected to the machining dimension predicting device 10 via a communication line. The measurement device 30 measures the dimension of the machined workpiece 40 and transmits the measurement result associated with the single identification number to the machining dimension prediction device 10. The measurement performed by the measurement device 30 corresponds to the inspection process performed on the workpiece 40 after the machining process. The dimension to be measured by the measurement device 30 is typically the dimension of a portion of the workpiece 40 that has been machined by the machining tool 20. In the example of FIG. 3, a length D 1 and a length D 2 are measurement targets. The length D1 extends from the reference surface to a predetermined position on the inward surface 401 of the blade. The length D 2 corresponds to the thickness of a predetermined part of the blade. The measurement targets are not limited to the example of FIG. 3. One measurement target or three or more measurement targets may be measured.The machining dimension prediction device 10 is, for example, an industrial personal computer (IPC), a programmable logic controller (PLC), or another FA device. As shown in FIG. 4, the machining dimension prediction apparatus 10 includes, as hardware components, a processor 51, a main memory 52, an auxiliary memory 53, an input device 54, an output device 55, and a communicator 56.The processor 51 includes, as an integrated circuit, a central processing unit (CPU), a micro processing unit (MPU), a graphics processing unit (GPU) for calculation, or a floating point unit (FPU). The processor 51 executes a program P 1 stored in the auxiliary memory 53 to implement various functions of the machining dimension prediction apparatus 10 and perform processes described later.The main memory 52 includes a random access memory (RAM). The program P1 is loaded from the auxiliary memory 53 into the main memory 52. The main memory 52 is used as a work area for the processor 51.The auxiliary memory 53 includes a nonvolatile memory such as an electrically erasable programmable read only memory (EEPROM). The auxiliary storage 53 stores, in addition to the storage of the program P 1, various items of data used for the processing performed by the processor 51. Auxiliary memory 53 provides data to processor 51 that may be used by processor 51, as directed by processor 51, and stores data provided by processor 51.The input device 54 includes, for example, an input key, a button, a switch, a keyboard, a pointing device, or a digital input contact (DI) (photocoupler input). The input device 54 acquires information input from the user to the machining dimension prediction device 10 or other information provided from an external device, and provides the acquired information to the processor 51.The output device 55 includes, for example, a light emitting diode (LED), a liquid crystal display (LCD), a digital output contact (DO) (photocoupler output), or a speaker. The output device 55 displays various information to the user or outputs such information to an external device upon instruction of the processor 51.The communicator 56 includes a network interface circuit for communicating with an external device and an analog signal circuit. The communicator 56 receives a signal from an external device and outputs information indicated by the signal to the processor 51. The communicator 56 transmits a signal indicating the information output from the processor 51 to an external device or outputs an analog signal.The machining dimension prediction apparatus 10 includes the above-mentioned hardware components that cooperate with each other to train, for each of a plurality of workpieces 40, the relationship between the actual dimension value acquired by the measurement device 30 and the state trend of the machining tool 20 during machining acquired by the machining tool 20 and the sensor 21. When a new target workpiece 41 is machined, as shown in FIG. 5, the machining dimension prediction device 10 predicts the machining dimensions of the new target workpiece 41 on the basis of the state trend of the machining tool 20 notified from the machining tool 20 and the sensor 21 and on the basis of the relationship resulting from the training.More specifically, as shown in FIG. 6, the machining dimension prediction apparatus 10 includes, as functional components, a trend detector 11 for detecting trend information indicating the state trend of the machining tool 20 during the machining period from the start to the end of the machining, a type number detector 12 for detecting the type and the manufacturing serial number of each of the workpieces 40 and 41, a portion definer 13 for defining a plurality of portions based on the trend information detected by the trend detector 11, a feature calculator 14 for calculating features using the state trend in each portion defined by the portion definer 13, a measurement value detector 15 for detecting a measurement value of the dimension of each workpiece 40 after the machining, a portion specifier 16 for specifying at least one portion including the calculated feature, The measuring value is relevant to a large degree, a trainer 17 for training a prediction model to predict the dimension based on the feature in the specified section, a predictor 18 for predicting dimensions by inputting trend information on the new target workpiece 41 into the prediction model, and a notifier 19 for notifying a result of determination as to whether the workpiece is acceptable based on the prediction result.The trend detector 11 is mainly implemented by the communicator 56. The trend detector 11 acquires, for each of the workpieces 40 and 41 on which the same machining is performed, trend information from either the machining tool 20 or the sensor 21 or from both the machining tool 20 and the sensor 21, for example, the trend detector 11 acquires, as the state trend of the machining tool 20, trend information indicating the time-series waveform of the machining torque indicated by the line L 1 in FIG. 7. The machining torque corresponds to the current value of the spindle motor of the machining tool 20. the trend detector 11 in the machining dimension prediction apparatus 10 corresponds to an example of trend detection means for detecting trend information indicating the state trend of the machining tool 20 during the machining period from the start to the end of machining performed by the machining tool 20 for each workpiece 40 on which machining is performed.The type number detector 12 is mainly implemented by the communicator 56. The type number detector 12 acquires type information and the manufacturing serial number of each of the workpieces 40 and 41 from the machining tool 20 and the measurement device 30. the type number detector 12 outputs the acquired type information and the acquired manufacturing serial number to the section definer 13 and the measurement value detector 15. The workpieces 40 and 41 on which the same machining is performed have the same type. The type information is used to distinguish the workpieces 40 and 41 on which the same machining is performed from workpieces on which other machinings are performed. The production serial number is used to identify the individual workpieces 40 and 41.The section definer 13 divides the machining period in the acquired trend information into a plurality of sections under a trend information definition condition preliminarily related to the type information, and outputs the sections to the feature calculator 14. The section definer 13 also divides the trend information about the new target workpiece 41 into a plurality of sections under the trend information defining condition preliminarily related to the type information, and outputs the sections to the feature calculator 14.The machining period may not correspond to the entire machining process for each of the workpieces 40 and 41. for example, when the machining process for each of the workpieces 40 and 41 includes a process of roughing, a process of forming the inward surface 401 and the outward surface 402 using the tool 201, and a process of forming tapped holes using another tool, the above machining period may correspond to the process of forming the inward surface 401 and the outward surface 402. The machining period may include the time at which the parts of the workpieces 40 and 41 to be measured by the measuring device 30 are machined.For example, as shown in the example of FIG. 7, the section definer 13 defines a plurality of sections A 1 to A 18 included in the machining period. The sections A 1 to A 18 include the section A 1 of a user preset period from the start of the machining and the subsequent sections A 2 to A 18 each defined by repeatedly displacing the section A 1 by a predetermined displacement width. Although the displacement width is equal to the length of each portion in the example of FIG. 7, the displacement width may be shorter or longer than the length of each portion. The length of the last portion A 18 may be changed to be included in the machining period. The sections are usually defined for a plurality of workpieces 40 of the same type.The feature calculator 14 is mainly implemented by the processor 51. The feature calculator 14 calculates a plurality of feature types using the state trend of the machining tool 20 in each section. For example, as shown in FIG. 8, the feature calculator 14 calculates, as the features of the waveform in the section A 4, a feature F 1 that is the maximum value, a feature F 2 that is the minimum value, a feature F 3 that is the average value, and a feature F 4 that is the median value.FIG. 9 is an exemplary table including a plurality of feature types calculated for each section included in the machining period for the plurality of workpieces 40. In FIG. 9, a first workpiece 40 is identified by an identifier 40-01 and a second workpiece 40 is identified by an identifier 40-02. Returning to FIG. 6, the feature calculator 14 outputs the features calculated for the plurality of workpieces 40 to the section specifier 16. The feature calculator 14 also outputs the features calculated for the new target workpiece 41 to the predictor 18. The feature calculator 14 in the machining dimension prediction device 10 corresponds to an example of a feature calculation device for calculating the features using the state trend in each section included in the machining period based on the trend information acquired for each workpiece.The measurement value detector 15 is mainly implemented by the communicator 56. The measurement value detector 15 acquires, from the measurement device 30, the measurement value of the dimension of each workpiece 40 after machining. The measurement value acquisition unit 15 also acquires the type information and the production serial number from the type number acquisition unit 12. The measurement value detector 15 outputs the measurement value detected for each workpiece 40 to the section specifier 16. The measurement value acquisition unit 15 in the machining dimension prediction device 10 corresponds to an example of measurement value acquisition means for acquiring the measurement value of the dimension of each workpiece 40 after machining.The section specifier 16 is mainly implemented by the processor 51. The section specifier 16 specifies, in descending order of the degree of relevance to the measurement value, a plurality of combinations of the feature types and the sections including the calculated features. The degree of relevance is an index value indicating to what extent the calculated feature is relevant to the measurement value of the machining dimensions for each workpiece 40. The degree of relevance is, for example, equal to the correlation coefficient. FIG. 10 is an example table including correlation coefficients as degrees of relevance between the measurement value and multiple feature types calculated using the state trend in each section. In the example of FIG. 10, two combinations, specifically, the combination of the portion A 14 and the feature F 4 and the combination of the portion A 5 and the feature F 3 are indicated in descending order of the degree of relevance and are boldly printed. Each correlation coefficient is a real number in the range of -1 to 1. a larger correlation coefficient, or more specifically, a correlation coefficient having a larger absolute value, is more relevant to the measurement value.The number of combinations specified by the section specifier 16 is preset by the user. For example, the inward surface 401 in FIG. 3 is formed in a first sub-period in FIG. 7, and the outward surface 402 in FIG. 3 is formed in a second sub-period in FIG. 7. When the length D 2 in FIG. 3 is the measurement target to be measured by the measurement device 30, the portions having particularly high relevance to the measurement value are portions in the first sub-period in which the measurement target portion is processed and portions in the second sub-period in which the measurement target portion is processed. Thus, when the length D 2 is the measurement target, two sections can be specified. In this case, the user may set the number of combinations of sections and features to be specified to two or more. When the length D 2 alone is the measurement target, the user can set the number of combinations to be specified to two.When a combination of a portion and a feature is specified, combinations of the same portion with other features can be excluded from the combinations to be specified. More specifically, when the combination of the portion A 14 and the feature F 4 is specified as shown in FIG. 10, a combination of another portion and an arbitrary feature may be specified instead of the combination of the same portion A 14 and the feature F 3.When the length D 1 in FIG. 3 is the measurement target, a single portion in which the measurement target portion corresponding to the length D 1 is processed has a particularly large degree of relevance to the measurement value. Thus, when only the length D 1 is the measurement target, the user can set the number of combinations to be specified to one. In other words, the section specifier 16 may specify at least one combination having the greatest degree of relevance.In the above-described example, the number of combinations specified by the section specifier 16 is equal to the number of sections including the times of machining of the part to be measured by the measurement device 30. However, these numbers may differ from each other. For example, when the lengths D 1 and D 2 in FIG. 3 are both the measurement targets, the section specifier 16 may specify less than three or more than three combinations.The number of combinations specified by the section specifier 16 need not be preset. For example, the section specifier 16 may specify all combinations whose calculated degrees of relevance are greater than a predetermined threshold. Any combination having a feature with the second greatest degree of relevance or less for a specific portion may be excluded from all combinations whose calculated degrees of relevance are greater than the predetermined threshold.The section specifier 16 outputs the specified combinations, the features in the specified sections, and the measurement value to the trainer 17. The section specifier 16 in the machining dimension prediction device 10 corresponds to an example of a section specification means for specifying the section including the calculated feature having the greatest degree of relevance to the measurement value as a specific section of the plurality of sections.The trainer 17 is mainly implemented by the processor 51. The trainer 17 trains the prediction model to predict machining dimensions based on the features in the section specified by the section specifier 16 using the relation between the features in the section and the measurement value of the machining dimensions for each workpiece 40. For two combinations specified by the section specifier 16, as shown in FIG. 10, the trainer 17 trains the prediction model expressed by the following regression equation (1).In the above-described equation, B is the predicted value for the machining dimensions, C1 is a first coefficient, F3A5 is the feature F3 in the section A5, C2 is a second coefficient, F4A14 is the feature F4 in the section A14, and C3 is a constant term. Thus, the trainer 17 trains the prediction model by regression analysis. The trainer 17 outputs the trained prediction model to the predictor 18. The prediction model corresponds to an example of a second prediction model for predicting the dimension of the workpiece 40 based on the feature corresponding to the specific portion. The trainer 17 in the machining dimension prediction apparatus 10 corresponds to an example of training means for training the second prediction model.The predictor 18 acquires, from the feature calculator 14, the features calculated using the trend information on the new target workpiece 41 and inputs, from the acquired features, the features in the section specified by the section specifier 16 into the trained prediction model to estimate the dimension of the new target workpiece 41. For example, the predictor 18 acquires, as the predicted value, the value of the machining dimensions corresponding to the feature of the new target workpiece 41 in the regression line L 10 in FIG. 11. The predictor 18 substitutes the value of F3A5and the value of F4A14of the new target workpiece 41 into the regression equation (1) described above to acquire the predicted value B of the machining dimensions. The predictor 18 then outputs the prediction result on the machining dimensions to the notifier 19. The predictor 18 in the machining dimension prediction apparatus 10 corresponds to an example of prediction means for predicting the dimension of the new target workpiece 41 after machining based on the feature calculated using the state trend in the specific section. More specifically, the predictor 18 corresponds to an example of prediction means for predicting dimensions of the new target workpiece 41 after machining on the basis of the feature of the type specified by the section specification means and calculated using the state trend in the specific section specified by the section specification means together with the feature type.The notifier 19 is mainly implemented by the output device 55. The notifier 19 determines whether the new target workpiece 41 is acceptable by determining whether the value of the machining dimensions predicted by the predictor 18 is outside a predetermined tolerance. When it is determined that the predicted value is out of tolerance, the notifier 19 notifies the user that the machining is defective. The notifier 19 may output the notification by displaying information on a screen on the machining dimension prediction device 10, lighting an LED, or generating a summary sound. The notifier 19 may notify the machining tool 20 that the machining is defective to stop the machining with an alarm display for the machining tool 20. The notifier 19 may compensate the tool length (coordinates) so that the predicted dimensional value is within the tolerance.The tool length compensation refers to the compensation of, for example, the coordinates of the tool and the coordinates of the table based on a change in the tool length caused by wear, in order to improve the contact between the tool and the workpiece 41 in the subsequent machining, in order to determine the workpiece 41 as acceptable. The notifier 19 may include a tool length compensator 191 for performing tool length compensation. The tool length compensator 191 corresponds to an example of a compensating means that compensates the tool length set for the machining tool for machining the workpiece when the dimension predicted by the predicting means is out of tolerance. The tool length compensator 191 may be disposed outside the notifier 19, or the tool length compensator 191 replaces the notifier 19 in the machining dimension prediction apparatus 10.The notifier 19 may also notify the result of the prediction performed by the predictor 18. More specifically, the notifier 19 may display information on the predicted value on the screen. The notifier 19 in the machining dimension prediction device 10 corresponds to an example of a notifier that indicates a notification that the dimension predicted by the notifier is outside the predetermined tolerance.The notifier 19 may also store the measurement values of the machining dimensions of the workpieces 40 and the predicted values of the machining dimensions of the new target workpieces 41. Based on this history of the machining dimensions, the notifier 19 can notify the prediction result of the number of machinings likely to be performed before the machining dimensions fall outside the tolerance due to the cutting tool length changed by wear. As the tool wears and deteriorates due to the many machinings, the machining dimensions may generally have a larger deviation from the design value. This enables a sufficiently accurate prediction of the number of machining operations to be performed presumably before the machining dimensions fall out of tolerance. For example, the notifier 19 may perform regression analysis to predict the expected machining dimensions based on the history of the machining dimensions or to predict the expected error magnitude based on the error history in the machining dimensions. The notification may include the number of processings to be performed or the time until the number of processings to be performed reaches zero. The notifier 19 may provide information on the number of machinings to be performed.FIGS. 12, 13 and 14 are flowcharts showing a training process, a verification process, and a prediction process, respectively, performed by the machining dimension prediction apparatus 10.The training process in FIG. 12 is the training of the prediction model that starts when a specific application installed in the machining dimension prediction apparatus 10 is started by the user. In the training process, the trend acquirer 11 collects trend information for each workpiece 40 on which the same type of machining is performed (step S 11). The section definer 13 divides the state transition indicated by the trend information into a plurality of sections according to a definition rule predetermined for the machining type (step S 12). The feature calculator 14 then calculates the features using the state trend in each section included in the machining period (step S 13).The measurement value acquisition unit 15 then acquires the measurement values of the machining dimensions of the plurality of workpieces 40 for which trend information is acquired in step S 11, and assigns the measurement values to the trend information on the workpieces 40 acquired in step S 11 using the manufacturing series numbers (step S 14). The section specifier 16 specifies a plurality of combinations of the sections and the feature types calculated in step S 13 in descending order of the degree of relevance for each measurement value acquired in step S 14 (step S 15). The trainer 17 then trains the prediction model to predict machining dimensions based on the features in the sections specified in step S 15 (step S 16). The training process thus ends.The verification process shown in FIG. 13 is the verification of the prediction accuracy with the prediction model trained in the training process. In the verification process, the trend detector 11 acquires trend information for a plurality of workpieces 40 different from the workpieces 40 used for data acquisition in step S 11 in the training process but having undergone the same type of processing as those workpieces 40 (step S 21).The feature calculator 14 then calculates the features using the state trend in each section included in the machining period and set for the type, based on the trend information acquired in step S 21 (step S 22). The predictor 18 predicts the machining dimensions based on the characteristics calculated in step S 22 using the prediction model trained in the training process (step S 23).The measurement value detector 15 then detects the measurement values of the machining dimensions of the plurality of workpieces 40 for which trend information has been detected in step S 21 (step S 24). The predictor 18 compares the predicted values of the machining dimensions acquired in step S 23 with the measurement values acquired in step S 24 (step S 25). For example, the predictor 18 records the deviation between the predicted values and the measurement values.The predictor 18 then determines whether the prediction accuracy acquired in step S 25 is within the predetermined tolerance (step S 26). In step S 26, the determination may be made as to whether the adjusted determination coefficients acquired based on the dimension values predicted for the multiple machining operations and the measurement values of the workpieces 40 are within the tolerance. When it is determined that the prediction accuracy is out of tolerance (No in step S 26), the notifier 19 notifies the user of an error indicating that the resultant prediction model has insufficient prediction accuracy (step S 27). This causes the user to give an instruction to perform the training process again, such as with new training target data. The prediction model can thus be trained again in order to achieve sufficient prediction accuracy. When it is determined that the prediction accuracy is within the tolerance (Yes in step S 26), or after step S 27 is performed, the verification process ends.The prediction process shown in FIG. 14 is prediction of the machining dimensions of the new target workpiece 41 using the prediction model. In the prediction process, the trend detector 11 acquires trend information for the new target workpiece 41 (step S 31). The section definer 13 divides the state transition indicated by the trend information acquired in step S 31 into a plurality of sections according to a definition rule predetermined for the type (step S 32). The feature calculator 14 calculates the features from the sections resulting from the division in step S 32 into the sections set in step S 15 in the training process (step S 33).The predictor 18 then predicts the machining dimensions of the new target workpiece 41 using the prediction model whose sufficient prediction accuracy has been verified in the verification process (step S 34). The notifier 19 acquires and stores the predicted dimension of the new target workpiece 41 (step S 35). The notifier 19 may also record the measurement values of the workpieces 40 used for the training process and the measurement values of the workpieces 40 used for the verification process.The notifier 19 then determines whether the dimension predicted in step S 34 is out of tolerance (step S 36). When it is determined that the dimension is out of tolerance (Yes in step S 36), the notifier 19 notifies the user of the result of the determination (step S 37). To perform the tool length compensation, the tool length (coordinates) is transmitted to the machining tool 20 so that the workpiece 41 has a dimension within the tolerance in the subsequent machining (step S 38). When it is determined that the dimension is within the tolerance (No in step S 36), the notifier 19 notifies the number of machinings to be performed presumably, based on the recorded dimension history, before the dimension falls outside the tolerance (step S 39). Thus, the prediction process ends.As described above, the section specifier 16 of multiple sections specifies the section including the calculated feature having the greatest degree of relevance. The predictor 18 predicts the dimension of the new target workpiece 41 after machining on the basis of the feature calculated using the trend of the state trend in the section specified by the section specifier 16. In other words, the machining dimensions are predicted based on the feature in the portion in the machining period that is most relevant to the actual dimension. This enables more accurate prediction of the dimension of the new target workpiece 41 machined by the machining tool 20.In order to increase prediction accuracy which is low, known methods train the prediction model by manually changing the feature calculation section or manually changing the feature types. In contrast, the machining dimension prediction apparatus 10 according to the present embodiment collects and registers a sufficient amount of time-series data in the machining period and the dimensions measured after machining for workpieces of the same type on which machining is performed under the same machining conditions. This enables accurate training of the prediction model and determination as to whether the new target workpiece 41 is acceptable based on the value predicted with the prediction model.The section specifier 16 specifies two or more sections in descending order of the degree of relevance. This enables accurate prediction of the machining dimensions of the new target workpiece 41 when two or more portions of the workpiece 40 machined by the machining tool 20 are to be measured by the measurement device 30. The use of more parameters can typically increase prediction accuracy.The section specifier 16 specifies the type of the feature having a high degree of relevance to the measurement value, in addition to specifying the section having a high degree of relevance to the measurement value. The predictor 18 measures the machining dimensions of the new target workpiece 41 based on the feature of the specified type. This enables highly accurate prediction of the machining dimensions based on the feature of the appropriate type.The section specifier 16 specifies the section having the largest correlation coefficient with the measurement value as the section having the largest degree of relevance. The correlation coefficients can be calculated with a relatively low computational effort. Thus, the structure can reduce the amount of calculation for specifying the section. The structure can reduce the calculation amount particularly when more sections and more feature types are used and when the section specifier 16 searches a larger area to specify the combination of the section and the feature type.The feature calculator 14 calculates the feature for each displacement of a portion having a predetermined length within the machining period. This structure can reduce the calculation amount for the calculation of the features.The machining dimension prediction apparatus 10 includes the trainer 17 for training the prediction model. This enables accurate estimation of the machining dimensions using the prediction model acquired with a large amount of data.The machining dimension prediction apparatus 10 includes the notifier 19 that provides a notification indicating that the predicted machining dimensions are out of tolerance. With this structure, it can be determined whether the new target workpiece 41 is acceptable without checking the workpiece 41 using the measurement device 30. Insufficient machining by the wear tool may be further compensated by tool length compensation (coordinate compensation) to enable acceptable machining over a prolonged period of time.The notifier 19 provides information on how often the machining is expected to be performed before the machining dimensions fall outside the tolerance based on the history of the predicted dimension. This informs the user of the time at which the tool is to be replaced and a new tool is to be prepared.In the above-described example, the features are calculated for the sections defined by shifting a section having a fixed length. However, the example is not limited thereto. For example, the feature calculator 14 may calculate the features for each section resulting from the division of the machining period. The sections A 1 to A 18 in the example of FIG. 7 may result from the uniform division of the machining period into 18 sections. The machining period may be divided into a preset number of sections or sections having a length close to a predetermined length. With this structure, the calculation amount for the calculation of the features can also be reduced.Embodiment 2Embodiment 2 will be described focusing on the differences from Embodiment 1. The components that are the same or similar to those of Embodiment 1 are denoted by the same reference numerals and will not be described or will be described only briefly. The present embodiment is different from Embodiment 1 in that each feature calculation section is a combination of a plurality of sub-sections included in the machining period.In the present embodiment, the portions A 1 to A 18 shown in FIG. 7 in Embodiment 1 are referred to as sub-portions. Two of these sub-sections are combined into one section, and the feature calculator 14 calculates the features using the waveform in this section. More specifically, as shown in FIG. 15, the sub-sections corresponding to the sections A5 and A14 in Embodiment 1 are combined into one section, and the features F1 to F4 are calculated using the combined waveform in this section. As shown in FIG. 16, the feature calculator 14 calculates the features for the sections that are all combinations of sub-sections.As described above, in the present embodiment, the section corresponds to a period resulting from the combination of a first subsection included in the machining period and a second subsection starting at a time point later than an end time point of the first subsection. When two portions are to be measured by the measuring device 30 after being processed by the processing tool 20, such as the length D 2 in FIG. 3, the feature may be more relevant to the measurement value when the feature is calculated using the waveform resulting from the combination of the waveforms in the two subsections including the time at which the processing is performed on these portions than when the feature is calculated using the waveforms in the individual subsections. The structure can thus provide a more suitable feature for predicting the machining dimensions.Embodiment 3Embodiment 3 will be described centering on the differences from Embodiment 1. The components that are the same or similar to those of Embodiment 1 are denoted by the same reference numerals and will not be described or will be described only briefly. The machining dimensions are influenced by ambient temperature. Therefore, for example, a prediction model suitable for summer may be used in summer and a prediction model suitable for winter may be used in winter, instead of using a common prediction model for summer and winter. In the present embodiment, the prediction model is trained with data acquired when a specific condition is satisfied, and the prediction model is used to predict the machining dimensions of the new target workpiece 41 when the condition is satisfied.As shown in FIG. 17, the machining dimension prediction apparatus 10 collects a large amount of data. Each individual item of data corresponds to the trend information and the measurement value of the workpiece 40 of the same type. The trainer 17 extracts a data set satisfying a predetermined condition from the large data amount and performs cross-validation of the extracted data. More specifically, the trainer 17 divides the extracted information into verification target information for the verification process and training target information for the training process, and generates a plurality of prediction models by changing the target dataset. The trainer 17 quantizes and evaluates the prediction accuracy based on, for example, the adjusted determination coefficients, and determines the prediction model having the highest prediction accuracy. For example, the trainer 17 extracts the data set satisfying a condition from the summer period data and determines the prediction model for the summer having the highest prediction accuracy. Similarly, the trainer 17 extracts the data set satisfying a condition from the data of the winter period and determines the prediction model for the winter with the highest prediction accuracy. Under the surrounding conditions of winter, for example, the prediction model suitable for winter with high prediction accuracy may be selected and used for the processing.The machining dimension prediction apparatus 10 extracts data and performs cross validation also for another condition to acquire a prediction model suitable for the other condition. For example, the machining dimension prediction apparatus 10 acquires a prediction model suitable for the time of day with a larger amplitude of the ground vibrations and a prediction model suitable for the night time with a smaller amplitude of the ground vibrations. When the condition under which the prediction accuracy is verified is satisfied for the new target workpiece 41 as shown in FIG. 17, the predictor 18 uses the prediction model suitable for the condition to estimate the machining dimensions of the new target workpiece 41.As described above, the machining dimension prediction apparatus 10 acquires a prediction model suitable for the condition. This enables accurate prediction of the machining dimensions of the new target workpiece 41 when the condition is satisfied.The trainer 17 in the present embodiment corresponds to an example of a training means for repeatedly extracting training target information for training and verification target information other than training information from trend information that satisfies a predetermined condition in the trend information acquired for each workpiece while changing the training target information in the trend information, and repeatedly verifying prediction accuracy with the second prediction model trained with the training target information on the basis of the verification target information and the measurement values of workpieces corresponding to the verification target information in the trend information. The predictor 18 corresponds to an example of prediction means for, when the above-mentioned condition is satisfied, predicting the dimension of the new target workpiece using the second prediction model trained with the training target information.Embodiment 4Embodiment 4 will be described centering on the differences from Embodiment 1. The components that are the same or similar to those in Embodiment 1 are denoted by the same reference numerals and will not be described or will be described only briefly. The present embodiment is different from Embodiment 1 in that an index value different from the correlation coefficient with the measurement value is used as the degree of relevance.FIG. 18 is a schematic diagram showing determination of the degree of relevance in the present embodiment. As shown in FIG. 18, the trend detector 11 acquires the trend information, the feature calculator 14 sets the sections and calculates the features of all types, and the measurement value detector 15 acquires the measurement values similarly to steps S 11 to S 14 in FIG. 12.The section specifier 16 then trains a measurement value prediction model to predict the measurement value using the sections and the feature types as explanatory variables and using the measurement value as a target variable. For example, the section specifier 16 uses a gradient boosting tree to train the measurement prediction model. The measurement prediction model may be trained to determine, for each section and for each feature type, the priority level for the measurement or the degree of contribution to the measurement. In other words, the priority level is the degree to which each section and feature type is relevant to the measurement value. The section specifier 16 specifies the sections and the feature types in descending order of the priority level. The section specifier 16 may specify at least the section having the highest priority level and the feature type having the highest priority level.As described above, in this embodiment, the priority level in the measurement value prediction model is used as the degree of relevance to specify the feature types. The prediction model thus enables a very accurate prediction of the machining dimensions. The section specifier 16 corresponds to an example of the section specifying means for training a first prediction model to predict the measurement value based on the plurality of feature types, the plurality of feature types being used as explanatory variables and the measurement value being used as a target variable, and the feature type having the highest priority level in the trained first prediction model being specified as the feature type having the highest degree of relevance.In the above-described example, both the sections and the feature types are used as explanatory variables. However, the example is not limited thereto. For example, the features of all types calculated in the entire machining period may be used as explanatory variables, and the measurement value may be used as a target variable. First, the feature types having higher priority levels may be specified, and then the portions having higher correlation coefficients between the features of the specified types and the measurement value may be specified.Embodiment 5Embodiment 5 will be described focusing on the differences from Embodiment 1. The components that are the same or similar to those in Embodiment 1 are denoted by the same reference numerals and will not be described or will be described only briefly. The present embodiment is different from Embodiment 1 in that an index value different from the correlation coefficient with the measurement value is used as the degree of relevance.FIG. 19 shows a portion where a workpiece 40 determined to be unacceptable has a state trend greatly deviating from the state trends of the workpieces 40 determined to be acceptable. This portion has high relevance to the measurement value used for determining whether the workpiece is acceptable. Therefore, in the present embodiment, the section specifier 16 specifies, as sections having a large degree of relevance, a plurality of sections in descending order of the degree of deviation. The deviation degree is the degree in which the state trend of the machining tool 20 for the erroneously machined workpiece 40 deviates from the state trend of the machining tool 20 for the normally machined workpiece 40. The degree of deviation can be specified, for example, by the feature of the local matrix profile (LMP) described in NonPatentliteratur 1. A larger LMP feature indicates a larger deviation in the waveform at the time.In the above-described embodiment, the degree of deviation is used as the degree of relevance. The portion having a greater deviation between the unacceptable workpiece and the acceptable workpiece is selected. Thus, the section is specified so that the prediction model can be accurate. Although the section is specified based on the degree of deviation in the present embodiment, the feature type may be specified based on another index value after the section is specified.The degree of relevance may be other than the correlation coefficient, the priority level in Embodiment 4, or the degree of deviation in Embodiment 5.Embodiment 6Embodiment 6 will be described focusing on the differences from Embodiment 1. The components that are the same or similar to those of Embodiment 1 are denoted by the same reference numerals and will not be described or will be described only briefly. The present embodiment is different from Embodiment 1 in that the variable width portions are set.As shown in FIG. 20, the feature calculator 14 in the present embodiment calculates the features in the portions corresponding to the line segments included in a polygonal line L 3 approximating the waveform representing the state trend of the machining tool 20. Examples of the polygonal approximation include the Ramer-Douglas-Peucker algorithm.Variable length portions corresponding to the waveform may be set. The portion having a large degree of relevance to the measurement value can be specified.The method of setting the variable width portions is not limited to the polygonal approximation. For example, as shown in FIG. 21, the section definer 13 may use a predetermined state transition model for the state trend to predict the transition between model states corresponding to the waveform, and may set sections corresponding to the respective model states. The state transition model may be, for example, the one described in International Publication No. WO 2020 / 234961 use hidden Markov chains.Although one or more embodiments of the present disclosure have been described above, the present disclosure is not limited to the above-described embodiments.In the above-described embodiments, the machining dimension prediction apparatus 10 acquires, for example, the trend information and the measurement value by communication. However, the machining dimension prediction device 10 may acquire the trend information and the measurement value by loading data from the auxiliary storage 53 or from a location specified by the user on an external server.In the above-described embodiments, the state of the machining tool 20 is shown by a series of scalar values. However, the state may also be shown by a series of vector values.The feature may be other than the features described in the above-described embodiments. The feature can be the difference value between the immediately preceding measured value and the current measured value or the position deviation. The feature calculator 14 may calculate, as features, the integral value corresponding to the area between the waveform and the horizontal axis in the section in FIG. 8 and the variance value based on the feature F 3. In some embodiments, the feature calculator 14 may calculate an arithmetic feature other than the difference value, the integral value, or the variance value.In the above-described embodiments, the section specifier 16 specifies the feature type having the greatest degree of relevance. However, the section specifier 16 may use any other method to specify one of a plurality of feature types. For example, multiple prediction models may be trained using one or more combinations selected by the user from the combinations of features and sections whose degree of relevance (or correlation coefficient) is greater than or equal to 0.4. Of the trained prediction models, the model having the smallest deviation in the verification process may be used for the prediction process to select the feature type.In the above-described embodiments, one or two portions of the workpiece 40 after the machining by the machining tool 20 are measured by the measurement device 30. However, the example is not limited thereto. When the measurement target is an angle as shown in v 22, the dimension can be measured with four measurement sections E 1, E 2, E 3, and E 4, for example. The measurement portions E 1 and E 2 define a reference direction D 11. The measurement portions E 3 and E 4 define a direction D 12 defining an angle θ with the direction D 11. In this example, four sections may be specified.The functions of the machining dimension prediction apparatus 10 may be implemented by dedicated hardware or a general computer system.For example, the program P 1 executable by the processor 51 may be stored in a non-transitory computer readable recording medium for distribution. The program P 1 is installed in a computer to provide a device that performs the above-described processing. Examples of such non-transitory recording media include a flexible disk, a compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), and a magneto-optical disk (MO).The program P 1 may be stored in a disk device belonging to a server in a communication network such as the Internet, and may be superimposed on a carrier wave, for example, to be downloaded to a computer.The above-described processing may also be performed by the program P 1 being activated and executed while being transmitted through a communication network.The above-described processing may be performed by completely or partially executing the program P 1 on a server while a computer sends and receives information on the processing via a communication network.In the system having the above-described functions, which may be implemented in part by the operating system (OS) or by cooperation between the OS and applications, parts executable by applications other than the OS may be stored in a non-transitory recording medium that may be distributed or downloaded to a computer.The means for implementing the functions of the machining dimension prediction apparatus 10 are not limited to software. The functions may be partially or fully implemented by dedicated hardware including circuits.The foregoing describes some embodiments for explanatory purposes. Although specific embodiments have been presented in the foregoing discussion, it will be understood by those skilled in the art that changes in form and detail may be made without departing from the broader spirit and scope of the invention. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense. This detailed description is therefore not to be taken in a limiting sense, and the scope of the invention is defined only by the appended claims, along with the full scope of equivalents to which such claims are entitled.Industrial applicabilityThe present disclosure may be used to predict dimensions of workpieces being machined with machining tools.List of reference characters100 Machining dimension prediction system, 10 Machining dimension prediction device, 11 Trend detector, 12 Type number detector, 13 Section definer, 14 Feature calculator, 15 Measurement value detector, 16 Section specifier, 17 Trainer, 18 Predictor, 19 Notifier, 191 Tool length compensator, 20 Machining tool, 201 Tool, 21 Sensor, 30 Measurement device, 40, 41 Workpiece, 401 Inward surface, 402 Outward surface, 51 Processor, 52 Main memory, 53 Auxiliary memory, 54 Input device, 55 Output device, 56 Communicator, 57 Internal bus, A1 to A18 Section, L1, L3 Line, L10 Regression line, P1 Program

Claims

An apparatus (10) for predicting machining dimensions, comprising: trend acquisition means (11) for acquiring trend information for each of a plurality of workpieces (40) on which machining is performed, the trend information indicating a state trend of a machining tool (20) during a machining period from a start to an end of the machining performed by the machining tool (20); feature calculation means (14) for calculating a feature based on the trend information using the state trend in each of a plurality of portions included in the machining period; measurement value acquisition means (15) for acquiring a measurement value of a dimension of each of the plurality of workpieces (40) after the machining; and prediction means (18) for predicting, when a new target workpiece (41) is machined, a dimension of the new target workpiece (41) after the machining; characterized a section specifying means (16) for specifying a section having a calculated feature having a highest degree of relevance as a specific section of the plurality of sections, wherein the degree of relevance is a degree of relevance to the measurement value; and wherein the predicting means (18) predicts the dimension of the new target workpiece (41) after the machining on the basis of the feature calculated using the state trend in the specific section.The machining dimension prediction apparatus (10) according to claim 1, wherein each of the plurality of sections corresponds to a period resulting from a combination of a first subsection included in the machining period and a second subsection starting at a time point later than an end time point of the first subsection.The machining dimension prediction apparatus (10) according to claim 1 or 2, wherein the section specification means (16) specifies two or more of the plurality of sections in descending order of the degree of relevance as specific sections.The machining dimension prediction apparatus (10) according to claim 1 or 2, wherein the feature calculation means (14) calculates, based on the trend information, a plurality of feature types using the state trend in each of the plurality of sections included in the machining period, wherein the section specification means (16) specifies a feature type of the plurality of feature types having the greatest degree of relevance and specifies, as the specific section, a section including the calculated feature having the greatest degree of relevance, and wherein, when the new target workpiece (41) is machined, the prediction means predicts the dimension of the new target workpiece (41) after the machining based on the feature of the type specified by the section specification means (16) and using the state trend in the specific section, which has been specified by the section specifying means (16), is calculated together with the feature type.The machining dimension prediction apparatus (10) according to claim 4, wherein the section specification means (16) specifies two or more combinations of the specific sections and the feature types in descending order of the degree of relevance.The machining dimension prediction apparatus (10) according to claim 4 or 5, wherein the section specification means (16) trains a first prediction model to predict the measurement value based on the plurality of feature types using the plurality of feature types as explanatory variables and using the measurement value as an objective variable, and specifies, as the feature type having the highest degree of relevance, a feature type having the highest priority level in the trained first prediction model.The machining dimension prediction apparatus (10) according to any one of claims 1 to 6, wherein the section specification means (16) specifies, as the specific section, a section having a largest correlation coefficient with the measurement value.The machining dimension prediction apparatus (10) according to any one of claims 1 to 6, wherein the section specification means (16) specifies, as the specific section, a section having a largest degree of deviation, the degree of deviation being a degree in which the state trend for a defectively machined workpiece (40) deviates from the state trend for a normally machined workpiece (40).The machining dimension prediction apparatus (10) according to any one of claims 1 to 8, wherein the feature calculation means (14) calculates the feature for each displacement of a portion having a predetermined length within the machining period.The machining dimension prediction apparatus (10) according to any one of claims 1 to 8, wherein the feature calculation means (14) calculates the feature in each of the plurality of sections resulting from division of the machining period.The machining dimension prediction apparatus (10) according to any one of claims 1 to 8, wherein the feature calculation means (14) calculates the feature in each of the plurality of sections corresponding to line segments included in a polygonal line approximating a waveform representing the state trend.The machining dimension prediction apparatus (10) according to any one of claims 1 to 8, wherein the feature calculation means (14) calculates the feature using the state trend in each of the plurality of sections corresponding to the respective model states in the transition between model states predicted using a predetermined state transition model for the state trend.The machining dimension prediction apparatus (10) according to any one of claims 1 to 12, further comprising: training means (17) for training a second prediction model to predict a dimension of a workpiece (40) based on the feature corresponding to the specific portion, wherein the prediction means (18) predicts the dimension of the new target workpiece (41) after machining using the second prediction model trained by the training means (17).The machining dimension prediction apparatus (10) according to claim 13, wherein the training means (17) repeatedly extracts, while changing the training target information in the trend information, trend information that satisfies a predetermined condition in the trend information acquired for each of the plurality of workpieces (40), training target information that is information for the training, and verification target information that is different from the information for the training, wherein the training means (17) repeatedly verifies, based on the verification target information and the measurement value of a workpiece (40) that corresponds to the verification target information in the trend information, prediction accuracy with the second prediction model that has been trained with the training target information, and wherein the prediction means (18), when the predetermined condition is satisfied, predicting the dimension of the new target workpiece ( 41) using the second prediction model trained with the training target information.The machining dimension prediction apparatus (10) according to any one of claims 1 to 14, further comprising: notification means (19) for providing notification indicating that the dimension is outside the predetermined tolerance when the dimension predicted by the prediction means (18) is outside a predetermined tolerance.The machining dimension prediction apparatus (10) according to claim 15, wherein the notification means (19) provides information on a number of machining operations to be expected to be performed before the machining-resultant dimension falls outside the predetermined tolerance based on the history of the machining dimension predicted by the prediction means (18).The machining dimension prediction apparatus (10) according to claim 15 or 16, further comprising: compensating means (191) for compensating a tool length set for the machining tool (20) for machining a workpiece (40) when the dimension predicted by the predicting means (18) is outside the predetermined tolerance.A machining dimension prediction system (100) comprising: the machining dimension prediction device (10) according to any one of claims 1 to 17; and a measurement device (30) for measuring a dimension of a workpiece (40).A method for predicting machining dimensions, comprising: calculating a feature using a state trend of a machining tool (20) in each of a plurality of portions included in a machining period from a start to an end of machining for each of a plurality of workpieces (40) on which the machining is performed by the machining tool (20); and predicting a dimension of the new target workpiece (41) when a new target workpiece (41) is machined characterized specifying, from the plurality of sections, a section including a calculated feature having a greatest degree of relevance, the degree of relevance being a degree of relevance to a measurement value of a dimension of each of the plurality of workpieces (40); and predicting the dimension of the new target workpiece (41) based on the feature calculated using the state trend in the specified section.A program (P1) that causes a computer to perform operations comprising: calculating a feature using a state trend of a machining tool (20) in each of a plurality of sections included in a machining period from a start to an end of machining for each of a plurality of workpieces (40) on which the machining is performed by the machining tool (20); predicting a dimension of the new target workpiece (41) when a new target workpiece (41) is machined; characterized specifying, from the plurality of sections, a section including a calculated feature having a greatest degree of relevance, the degree of relevance being a degree of relevance to a measurement value of a dimension of each of the plurality of workpieces (40); and predicting the dimension of the new target workpiece (41) based on the feature calculated using the state trend in the specified section.

Images