Mechanical phenomena evaluation support device
The mechanical phenomenon evaluation support device addresses synchronization and program understanding challenges by processing asynchronous load data through frequency distributions and thresholds, achieving accurate machine learning for evaluating machining quality and tool condition.
Patent Information
- Application Number
- JP2021009226
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2021-01-24
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2041-01-24
AI Technical Summary
Existing machine learning systems for detecting abnormalities in machine tools face accuracy issues due to the inclusion of non-processing data, which is difficult to synchronize with tool position data and requires understanding of machining programs, leading to decreased performance.
A mechanical phenomenon evaluation support device that acquires and processes load time-series data without synchronization with tool position data or knowledge of machining programs, using frequency distributions, peak class extraction, and load thresholds to generate divided time-series data for machine learning, enabling accurate model generation.
Enables highly accurate machine learning for evaluating machine phenomena by generating multiple trained models from asynchronous load data, allowing for precise evaluation of machining quality, tool condition, and thermal displacement without requiring synchronized position data or program knowledge.
Smart Images

Figure 0007585807000001 
Figure 0007585807000002 
Figure 0007585807000003
Abstract
Description
[Technical field]
[0001] The present invention relates to a machine phenomenon evaluation support device. [Background technology]
[0002] Patent Document 1 describes a device that detects abnormalities in a tool of a machine tool using machine learning. Specifically, the device obtains vibration information, cutting force information, sound information, spindle load, motor current, power value, etc. as detection data related to the tool, creates a normal model learned by machine learning using data detected during machining in a normal state, and diagnoses whether the detection data is normal or abnormal using the detection data during machining after the normal model is created and the normal model. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] JP 2018-024055 A Summary of the Invention [Problem to be solved by the invention]
[0004] In order to use the above-mentioned detection data for machine learning, it is necessary to extract the detection data during processing. If the detection data during non-processing is included, it will affect the results of machine learning, causing a decrease in the accuracy of machine learning.
[0005] The detection data detected by the detector includes not only data during machining but also data during non-machining. For example, the detection data and the tool position data are associated with each other at each sampling time and acquired, and the detection data during machining is extracted from the tool position data by referring to the machining program.
[0006] However, in order to associate the detection data with the tool position data, it is necessary to acquire both data synchronously. If the two data are not synchronized, the extracted detection data may include data during non-machining. However, if the acquisition period of the tool position data is set to a short period, such as on the order of milliseconds, it is not easy to synchronize all other detection data with the tool position data.
[0007] Furthermore, when extracting detection data during machining from tool position data, the machining program is referenced, so it is necessary to know the machining program. However, when the machining program is changed, it is not easy to know which data is being machined in the tool position data.
[0008] The present invention aims to provide a machine phenomenon evaluation support device that can extract in-process data to be used for machine learning with high accuracy and utilize high-precision machine learning even when detection data is not synchronized with tool position data or when the machining program is not understood. [Means for solving the problem]
[0009] (machine Phenomenon Evaluation Support Device) machineThe phenomenon evaluation support device includes a time series data acquisition unit that acquires time series data of the load of the machine operation detected by a sensor installed on the machine, a frequency distribution generation unit that generates a frequency distribution in which a first axis represents a load class and a second axis orthogonal to the first axis represents a frequency in the load time series data, a peak class extraction unit that extracts a load class at a peak of the frequency distribution, a load threshold setting unit that sets a load threshold for dividing the load time series data based on the peak class extracted by the peak class extraction unit, a divided time series data generation unit that generates multiple divided time series data by dividing the load time series data by the load threshold, a feature determination unit that determines feature values for each of the multiple divided time series data, and a model generation unit that generates a learned model for each of the multiple divided time series data, which is a machine learning model in which the feature values are explanatory variables and the machine phenomenon is a target variable.
[0010] The load time series data is divided based on the peak class obtained from the frequency distribution to generate a plurality of divided time series data. A trained model is generated for each of the plurality of divided time series data. In other words, it is sufficient to obtain only the load time series data as the plurality of divided time series data for generating the plurality of trained models.
[0011] In other words, in order to generate a plurality of divided time series data, it is not necessary to obtain machine position data synchronized with the load time series data. Furthermore, in order to generate a plurality of divided time series data, it is not necessary to know the machine control data such as the machining program. In this way, it is possible to obtain a plurality of divided time series data at a desired timing by only obtaining the load time series data. Then, by performing machine learning using the obtained plurality of divided time series data, it is possible to generate a plurality of highly accurate trained models. [Brief description of the drawings]
[0014] [Figure 1] 1 is a diagram showing an entire system including a target machine and a machine phenomenon evaluation support device. [Diagram 2]FIG. 1 is a diagram showing an example of processing by a target machine. [Diagram 3] 1 is a flowchart showing an example of a machining method by a target machine. [Figure 4] 1 is a graph showing time series data of the load on a target machine. [Diagram 5] 1 is a functional block diagram of a first example of a machine phenomenon evaluation support device. FIG. [Figure 6] FIG. 13 is a diagram showing load time series data and its frequency distribution, as well as a plurality of divided time series data. [Figure 7] FIG. 11 is a functional block diagram of a second example of a machine phenomenon evaluation support device. [Figure 8] 1 is a graph showing time-series data of a load and a plurality of divided time-series data; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0015] (1. Overview of the Mechanical Phenomenon Evaluation Support Device) The mechanical phenomenon evaluation support device (hereinafter referred to as the "evaluation support device") is a device for evaluating various phenomena in machines, such as production machines such as machine tools, assembly equipment, and conveying equipment, vehicles such as automobiles and trains, and wind power generation equipment.
[0016] For example, in the case of a machine tool, phenomena evaluated by the evaluation support device include the machining quality of a workpiece machined by the machine tool, the presence or absence of a malfunction of the machine tool, the deterioration state of the machine tool, the amount of thermal displacement of the machine tool, etc. The machining quality of the workpiece includes the shape accuracy of the workpiece, the dimensional accuracy of the workpiece, the surface properties of the workpiece, the presence or absence of a machine-degraded layer of the workpiece, the presence or absence of scratches on the surface of the workpiece, etc.
[0017] The presence or absence of a failure of a machine tool and the deterioration state of the machine tool include the presence or absence of a failure and the deterioration state of the components that constitute the machine tool. The presence or absence of a failure and the deterioration state of the components include, for example, the presence or absence of an abnormality due to breakage or chipping of a tool (corresponding to the presence or absence of a failure), the degree of wear due to wear or the like of a tool (corresponding to the deterioration state), the deterioration state of a bearing, etc. The amount of thermal displacement of a machine tool includes the amount of deviation in the relative position between a tool and a workpiece due to thermal displacement of a component of the machine tool, the amount of thermal displacement of a component of the machine tool alone, etc.
[0018] Furthermore, in the case of an assembly device or a conveyance device, the phenomena evaluated by the evaluation support device are the assembly accuracy of a workpiece assembled by the assembly device, the conveyance state of an object conveyed by the conveyance device, etc. In the case of a vehicle or a wind power generation device, etc., the phenomena evaluated by the evaluation support device are the operating performance of each function of the automobile or the wind power generation device, etc., the presence or absence of failure and the deterioration state of parts constituting the automobile or the wind power generation device, etc.
[0019] In order to evaluate a phenomenon of a target machine, the evaluation support device acquires the load of the machine's operation detected by a sensor installed on the target machine, and evaluates the phenomenon of the target machine by machine learning based on the load of the machine's operation.
[0020] In this case, in order to evaluate the phenomenon of the target machine by machine learning, it is preferable to prepare many types of data to be used. Therefore, it is preferable to install multiple sensors on the target machine and detect the load of the machine's operation detected by the multiple sensors.
[0021] For example, in the case of a machine tool, the operation of the machine is a machining operation by the machine tool, and the load of the machine operation is a load generated on the machine tool during the machining operation. For example, the load can be a load generated on a member supporting a workpiece, a load generated on a tool that machines the workpiece, a load generated on a member that moves the workpiece or the tool, etc. Also, the load can be a force sensor that detects a force generated on the target member, a drive current of a drive device, vibration generated on the target member, etc.
[0022] In the case of a wind power generating device, the operation of the machine is, for example, the rotational operation of a wind turbine constituting the device, and the load of the operation of the machine is the load generated in the rotation of the wind turbine. For example, the load can be vibration generated in the wind turbine.
[0023] (2. Configuration of evaluation support system 100) An evaluation support system 100 including a machine tool 1 as a target machine and an evaluation support device 2 will be described with reference to Fig. 1. In this example, the target machine is exemplified as the machine tool 1. However, as described above, the target machine can be a machine other than the machine tool.
[0024] The machine tool 1 is a machine that processes a workpiece W, and performs various processes such as cutting, grinding, etc. In this example, the machine tool 1 is a machining center that performs cutting, as shown in Fig. 1.
[0025] The machine tool 1 shown in FIG. 1 is a device that processes a workpiece W by moving the tool T relative to the workpiece W. The machine tool 1 is configured to be tool replaceable, and is capable of processing according to the attached tool T. For example, the replaceable tool T may be a cutting tool such as an end mill, a milling tool, a drill, a turning tool, a threading tool, a tooth profile generating tool, or a grinding tool. Note that a tool exchange device and a tool magazine that stores the tool T are not shown in FIG. 1. In this example, the machine tool 1 is basically configured as a horizontal machining center. However, the machine tool 1 may be configured in other ways, such as a vertical machining center.
[0026] As shown in FIG. 1, the machine tool 1 has, for example, three mutually orthogonal linear axes (X-axis, Y-axis, and Z-axis) as drive axes. Here, the direction of the rotation axis of the tool T (equal to the rotation axis of the tool spindle) is defined as the Z-axis direction, and two axes perpendicular to the Z-axis direction are defined as the X-axis direction and the Y-axis direction. In FIG. 1, the horizontal direction is the X-axis direction, and the vertical direction is the Y-axis direction. Furthermore, the machine tool 1 further has, as drive axes, two rotation axes (A-axis and B-axis) for changing the relative attitude of the tool T and the workpiece W. The machine tool 1 also has a Ct axis as a rotation axis for rotating the tool T.
[0027] That is, machine tool 1 is a 5-axis machine (a 6-axis machine when the tool spindle (Ct-axis) is taken into consideration) capable of machining free-form surfaces. Here, machine tool 1 may be configured to have a Cw-axis (a rotation axis about the Z-axis in the reference state) and a B-axis, or may be configured to have an A-axis and a Cw-axis, instead of an A-axis (a rotation axis about the X-axis in the reference state) and a B-axis (a rotation axis about the Y-axis in the reference state).
[0028] In the machine tool 1, the configuration for relatively moving the tool T and the workpiece W can be appropriately selected. In this example, the machine tool 1 allows the tool T to move linearly in the Y-axis and Z-axis directions, allows the workpiece W to move linearly in the X-axis direction, and further allows the workpiece W to rotate about the A-axis and B-axis. In addition, the tool T can rotate about the Ct axis.
[0029] The machine tool 1 includes a bed 10, a workpiece holding device 20, and a tool holding device 30. The bed 10 is formed in any shape, such as a substantially rectangular shape, and is placed on the floor surface. The workpiece holding device 20 enables the workpiece W to move linearly in the X-axis direction relative to the bed 10 and to rotate about the A-axis and B-axis. The workpiece holding device 20 mainly includes an X-axis moving table 21, a B-axis rotating table 22, and a workpiece spindle device 23.
[0030] The X-axis moving table 21 is provided so as to be movable in the X-axis direction relative to the bed 10. Specifically, a pair of X-axis guide rails extending in the X-axis direction (front-rear direction in FIG. 1) are provided on the bed 10, and the X-axis moving table 21 is driven by a driving device such as a linear motor or a ball screw mechanism (not shown) to reciprocate in the X-axis direction while being guided by the pair of X-axis guide rails.
[0031] The B-axis rotating table 22 is placed on the upper surface of the X-axis moving table 21, and moves back and forth in the X-axis direction integrally with the X-axis moving table 21. The B-axis rotating table 22 is provided so as to be rotatable about the B axis relative to the X-axis moving table 21. A rotary motor (not shown) is housed in the B-axis rotating table 22, and the B-axis rotating table 22 is rotatable about the B axis by being driven by the rotary motor.
[0032] The workpiece spindle device 23 is mounted on the B-axis rotating table 22 and rotates about the B-axis integrally with the B-axis rotating table 22. The workpiece spindle device 23 supports the workpiece W so that it can rotate about the A-axis. The workpiece spindle device 23 is equipped with a rotary motor (not shown) that rotates the workpiece W. In this manner, the workpiece holding device 20 enables the workpiece W to move in the X-axis direction relative to the bed 10 and to rotate about the A-axis and B-axis.
[0033] The tool holding device 30 mainly includes a column 31, a saddle 32, and a tool spindle device 33. The column 31 is provided so as to be movable in the Z-axis direction relative to the bed 10. Specifically, the bed 10 is provided with a pair of Z-axis guide rails extending in the Z-axis direction (left-right direction in FIG. 1), and the column 31 is driven by a driving device such as a linear motor or a ball screw mechanism (not shown) to reciprocate in the Z-axis direction while being guided by the pair of Z-axis guide rails.
[0034] The saddle 32 is disposed on the side surface of the column 31 facing the workpiece W (the left side surface in FIG. 1), which is parallel to a plane perpendicular to the Z-axis direction. A pair of Y-axis guide rails extending in the Y-axis direction (the up-down direction in FIG. 1) are provided on the side surface of the column 31, and the saddle 32 reciprocates in the Y-axis direction by being driven by a driving device such as a linear motor or a ball screw mechanism (not shown).
[0035] The tool spindle device 33 is mounted on the saddle 32 and moves in the Y-axis direction together with the saddle 32. The tool spindle device 33 supports the tool T so that it can rotate about the Ct axis. The tool spindle device 33 is equipped with a rotation motor (not shown) that rotates the tool T. In this way, the tool holding device 30 holds the tool T so that it can move in the Y-axis and Z-axis directions relative to the bed 10 and can rotate about the Ct axis.
[0036] The evaluation support device 2 is connected to a sensor installed in the machine tool 1, acquires data detected by the sensor, and uses machine learning to evaluate machine phenomena, such as the machining quality of the workpiece W, the presence or absence of abnormalities (breakdowns) in the tool T, the degree of wear on the tool T, the amount of thermal displacement of the machine tool 1, etc.
[0037] (3. Examples of processing methods) The machining method for which the machine phenomenon is evaluated by the evaluation support device 2 will be described with reference to Fig. 2 and Fig. 3. The target machining method is hole-making using a drill, as shown in Fig. 2. In other words, the target machining method is a case where a drill is used as the tool T of the machine tool 1 to perform hole-making (cutting) on a workpiece W. However, the machining method for which the machine phenomenon is evaluated by the evaluation support device 2 is not limited to hole-making, and can be applied to other machining methods.
[0038] In this example, the feed rate is changed during the drilling process. In Fig. 2, in the range A at the beginning of the process, the feed rate of the tool T is the first speed Va, in the range B at the middle of the process, the feed rate of the tool T is the second speed Vb, and in the range C at the end of the process, the feed rate of the tool T is the third speed Vc. Here, the first speed Va and the third speed Vc are the same, and the second speed Vb is faster than the first speed Va and the third speed Vc. In other words, the process is performed slowly immediately after the start of the process, and then high-speed process is performed after stabilization, and the speed is slowed down before approaching the end depth of the process.
[0039] More specifically, as shown in Fig. 3, first, the tool T is moved close to the workpiece W by a fast forward operation, and the tool T is moved to a drilling position on the workpiece W (S1). Next, while moving the tool T at a first speed Va, drilling is performed on the workpiece W (S2). The range of drilling at this time is the initial processing range A in Fig. 2.
[0040] Next, immediately after the tool T is temporarily stopped, the tool T is moved at a second speed Vb faster than the first speed Va while drilling the workpiece W further (S3). The range of the drilling at this time is the range B of the middle stage of machining in FIG. 2. Next, the tool T is changed to a third speed Vc slower than the second speed Vb, and the workpiece W is drilled further (S4). The range of the drilling at this time is the range C of the final stage of machining in FIG. 2. Next, the tool T is moved out of the machined hole (S5), and the tool T is moved away from the workpiece W by a fast forward motion (S6).
[0041] (4. Load Behavior) 2 and 3, the behavior of the load acting on the machine tool 1 will be described with reference to Fig. 4. Here, the load acting on the machine tool 1 is, for example, a force detected by a force sensor installed in the tool spindle device 33. However, the load may also be the current value of a current sensor installed in a motor in the tool spindle device 33, or the magnitude of vibration detected by a vibration sensor installed in the tool spindle device 33. The load may also be the magnitude of vibration detected by a vibration sensor installed in a member supporting the workpiece W, or the force detected by a force sensor installed in the member.
[0042] Here, S1-S6 in Fig. 4 correspond to S1-S6 in the flowchart of Fig. 3. That is, in the fast-forward approach (S1), the load is very small. In the drilling at the first speed Va (S2), the load is large. In the drilling at the second speed Vb (S3), the load is even larger. In the drilling at the third speed Vc (S4), the load is reduced. And in the exit from the hole (S5) and the fast-forward retraction (S6), the load is very small.
[0043] That is, the loads in S2, S3, and S4 during machining are larger than those in S1, S5, and S6 during non-machining. Furthermore, the loads in drilling (S3) at the relatively high second speed Vb during machining are larger than those in drilling (S2, S4) at the relatively low first speed Va and third speed Vc.
[0044] (5. Configuration of the first example of the evaluation support device 2) The configuration of the evaluation support device 2 of the first example will be described with reference to Fig. 5 and Fig. 6. The evaluation support device 2 evaluates the machining quality of the workpiece W, the presence or absence of an abnormality (breakdown) of the tool T, the degree of wear of the tool T, the amount of thermal displacement of the machine tool 1, and the like, by machine learning using load data generated in the machine tool 1 detected by a sensor installed in the machine tool 1.
[0045] The evaluation support device 2 performs evaluation using only the load data detected by the sensor installed in the machine tool 1. Here, as shown in Fig. 4, the load data is a mixture of data during machining (S2-S4) and data not during machining (S1, S5-S6). Furthermore, as shown in Fig. 4, even during machining, the magnitude of the load differs if the machining conditions differ.
[0046] Therefore, the evaluation support device 2 performs a process of extracting data during processing from the load data using the load data itself. Furthermore, the evaluation support device 2 performs a process of classifying data under different processing conditions during processing using the load data itself.
[0047] The evaluation support device 2 is configured to include a processor, a storage device, an interface, etc. As shown in Fig. 5, the evaluation support device 2 includes a time-series data acquisition unit 41, a data preprocessing unit 42, a feature amount determination unit 43, a model generation unit 44, a model storage unit 45, and a judgment unit 46.
[0048] The time series data acquiring unit 41 acquires time series data D of the load of the operation of the machine tool 1 detected by a sensor installed in the machine tool 1. For example, the time series data acquiring unit 41 acquires time series data D relating to the force detected by a force sensor installed in the tool spindle unit 33 (shown in FIG. 1) as shown in FIG. 4. The load time series data D includes not only data during machining but also data not during machining. Furthermore, the load time series data D acquired by the time series data acquiring unit 41 is data asynchronous with the control data by the control program of the machine tool 1. Note that the load time series data D may be data synchronized with the control data, but does not need to be data that is actively synchronized.
[0049] The data pre-processing unit 42 generates a plurality of divided time series data Da, Db, Dc by extracting a necessary range from the load time series data D in order to use the load time series data D as explanatory variable data in machine learning. In this example, the data pre-processing unit 42 extracts divided time series data Da, Db, Dc corresponding to the processing from the load time series data D. Furthermore, the data pre-processing unit 42 extracts divided time series data Da, Dc corresponding to the drilling processing (S2, S4) at the first speed Va (similar to the third speed Vc) and divided time series data Db corresponding to the drilling processing (S3) at the second speed Vb during the processing. That is, the data pre-processing unit 42 generates a plurality of divided time series data Da, Db, Dc used for machine learning from the load time series data D.
[0050] The data preprocessing unit 42 in this example includes a frequency distribution generating unit 42a, a peak class extracting unit 42b, a load threshold setting unit 42c, and a divided time series data generating unit 42d. As shown in the right column of Fig. 6, the frequency distribution generating unit 42a generates a frequency distribution in which the first axis (vertical axis) represents the classes of loads and the second axis (horizontal axis) perpendicular to the first axis represents the frequencies in the load time series data D. The right column of Fig. 6 illustrates a histogram as an example of the frequency distribution. However, the frequency distribution may be a line graph in which the frequencies in the classes of loads are plotted as points, or may be represented by an approximation curve connecting the points of the frequencies in each class.
[0051] The peak class extraction unit 42b extracts the load class at the peak of the frequency distribution generated by the frequency distribution generation unit 42a. In this example, as shown in the right column of FIG. 6, the frequency distribution has three peak classes Pe1, Pe2, and Pe3. Peak class Pe1 mainly corresponds to the load data during non-machining (S1, S5-S6). Peak class Pe2 mainly corresponds to the load data during drilling (S2) at the first speed Va and drilling (S4) at the third speed Vc. Peak class Pe3 mainly corresponds to the load data during drilling (S3) at the second speed Vb.
[0052] 6, the load threshold setting unit 42c sets load thresholds Th1, Th2, Th3 for dividing the load time-series data D based on the peak classes Pe1, Pe2, Pe3 extracted by the peak class extraction unit 42b. The load thresholds Th1, Th2, Th3 are not set to the values of the peak classes Pe1, Pe2, Pe3 themselves, but to values slightly larger than the peak classes Pe1, Pe2, Pe3.
[0053] For example, the load thresholds Th1, Th2, and Th3 are values larger than the peak classes Pe1, Pe2, and Pe3, and are values that reach, for example, 10% or less of the frequency of the peak classes Pe1, Pe2, and Pe3. When the tail width of the distribution around the peak class Pe1 is narrow, as in the case of the peak class Pe1, the load threshold Th1 is a value very close to the peak class Pe1. Since the tail width of the peak class Pe2 is also narrow, the load threshold Th2 is a value close to the peak class Pe2. On the other hand, when the tail width of the distribution around the peak class Pe3 is wide, as in the case of the peak class Pe3, the load threshold Th3 is a value farther away than the peak class Pe3. The frequency ratio for setting the load thresholds Th1, Th2, and Th3 is not limited to 10% and can be any value.
[0054] The divided time series data generating unit 42d generates a plurality of divided time series data Da, Db, and Dc by dividing the load time series data D by the load thresholds Th1, TH2, and Th3. The divided time series data generating unit 42d first extracts intersections with the load thresholds Th1, Th2, and Th3 in the load time series data D. Then, the load time series data D is divided at the intersections.
[0055] In FIG. 6, the period of the load time series data D is divided into periods P1, P2, P3, P4, and P5. Period P1 is the period until the load threshold Th1 is reached. Period P2 is the period from when the load threshold Th1 is reached to when the load threshold Th2 is reached. Period P3 is the period from when the load threshold Th2 is reached to when the load threshold Th2 is reached again. Period P4 is the period from when the load threshold Th2 is reached again to when the load threshold Th1 is reached. Period P5 is the period after the load threshold Th1 is reached.
[0056] The periods P1 and P5 of the load time-series data D are smaller than the load threshold Th1 and are determined to be not being processed. The periods P2, P3, and P4 of the load time-series data D are determined to be being processed. Then, the data of the periods P2, P3, and P4 of the load time-series data D are generated as divided time-series data Da, Db, and Dc, respectively.
[0057] The divided time series data Da in the period P2 generally corresponds to the drilling process (S2) at the first speed Va. The divided time series data Db in the period P3 generally corresponds to the drilling process (S3) at the second speed Vb. The divided time series data Dc in the period P4 generally corresponds to the drilling process (S4) at the third speed Vc.
[0058] The feature amount determining unit 43 determines a feature amount for each of the plurality of divided time series data Da, Db, and Dc. The feature amount may be arbitrarily selected from, for example, a maximum value, an average value, a median value, a standard deviation, a variance, a first quartile, a third quartile, and the like.
[0059] The model generation unit 44 generates trained models Ma, Mb, Mc for each of the multiple divided time series data Da, Db, Dc as a learning phase of machine learning. The trained models Ma, Mb, Mc are machine learning models that use the feature amounts of each of the multiple divided time series data Da, Db, Dc as explanatory variables and the phenomenon of the machine tool 1 as a target variable.
[0060] The phenomenon of the machine tool 1, which is the objective variable, is at least one of the machining quality of the workpiece W machined by the machine tool 1, the presence or absence of a malfunction of the machine tool 1, the deterioration state of the machine tool 1, and the amount of thermal displacement of the machine tool 1. The presence or absence of a malfunction of the machine tool 1 includes, for example, the presence or absence of an abnormality in the tool T. The deterioration state of the machine tool 1 includes, for example, the degree of wear of the tool T.
[0061] For example, the model generation unit 44 generates a trained model by machine learning using each of the divided time series data Da, Db, and Dc of the load when machining a workpiece W with good machining quality as a training data set. In this case, the trained model is a model that can output whether the machining quality of the workpiece W is good or bad when the divided time series data Da, Db, and Dc of the load are used as input data. The trained model can also output the degree to which the machining quality of the workpiece W is good or bad. For example, the degree to which the machining quality is good or bad is output depending on the distance from the boundary between good and bad.
[0062] Furthermore, the model generation unit 44 generates a trained model by machine learning using each of the divided time series data Da, Db, and Dc of the load when the tool T is normal as training data. In this case, the trained model is a model that can output whether the tool T is normal or abnormal when the divided time series data Da, Db, and Dc of the load are used as input data. The trained model can also output the degree to which the tool T is normal or abnormal. For example, the degree to which the tool T is normal or abnormal is output depending on the distance from the boundary between normal and abnormal.
[0063] The model generation unit 44 separately generates a trained model Ma corresponding to the divided time series data Da of the load and a trained model Mc corresponding to the divided time series data Dc of the load. Here, since the first speed Va and the third speed Vc are the same speed, one trained model may be generated using the divided time series data Da and Dc as a training data set.
[0064] The model storage unit 45 stores multiple trained models Ma, Mb, Mc generated by the model generation unit 44. The model storage unit 45 stores the trained model Ma in association with information on the drilling process at the first speed Va (S2). The model storage unit 45 stores the trained model Mb in association with information on the drilling process at the second speed Vb (S3). The model storage unit 45 stores the trained model Mc in association with information on the drilling process at the third speed Vc (S4).
[0065] After the generation of the learned models Ma, Mb, and Mc, the judgment unit 46 evaluates the phenomenon of the machine tool 1 based on the load time series data D acquired by the time series data acquisition unit 41 and the learned models Ma, Mb, and Mc stored in the model memory unit 45, as an estimation phase of machine learning.
[0066] Particularly in this example, the determination unit 46 uses the trained model Ma to evaluate the phenomenon of the machine tool 1 using the divided time series data Da generated by the data pre-processing unit 42 as input data. Moreover, the determination unit 46 uses the trained model Mb to evaluate the phenomenon of the machine tool 1 using the divided time series data Db generated by the data pre-processing unit 42 as input data. Moreover, the determination unit 46 uses the trained model Mc to evaluate the phenomenon of the machine tool 1 using the divided time series data Dc generated by the data pre-processing unit 42 as input data.
[0067] The judgment unit 46 evaluates at least one of the machining quality of the workpiece W, the presence or absence of a malfunction of the tool T or other parts that make up the machine tool 1, the deterioration state of the tool T or other parts that make up the machine tool 1, and the amount of thermal displacement of the machine tool 1.
[0068] According to the above-described evaluation support device 2, the load time series data D is divided based on the peak classes Pe1, Pe2, and Pe3 obtained from the frequency distribution to generate a plurality of divided time series data Da, Db, and Dc. Trained models Ma, Mb, and Mc are generated for each of the plurality of divided time series data Da, Db, and Dc. In other words, it is sufficient to obtain only the load time series data D as the plurality of divided time series data Da, Db, and Dc for generating the plurality of trained models Ma, Mb, and Mc.
[0069] In other words, in order to generate a plurality of divided time series data Da, Db, Dc, it is not necessary to obtain position data of the machine tool 1 synchronized with the load time series data D. Furthermore, in order to generate a plurality of divided time series data Da, Db, Dc, it is not necessary to grasp the control data of the machine tool 1 such as a machining program. In this way, it is possible to obtain a plurality of divided time series data Da, Db, Dc at a desired timing by only obtaining the load time series data D. Then, by performing machine learning using the obtained plurality of divided time series data Da, Db, Dc, it is possible to generate a plurality of highly accurate trained models Ma, Mb, Mc. Then, by using the trained models Ma, Mb, Mc, it is possible to evaluate the phenomenon of the machine tool 1 with high accuracy. (6. Configuration of the second example of the evaluation support device 2) The configuration of the evaluation support device 2 of the second example will be described with reference to Figures 7 and 8. The evaluation support device 2 of the second example differs from the evaluation support device 2 of the first example in the data pre-processing unit 52.
[0070] Similar to the data pre-processing unit 42 of the first example, the data pre-processing unit 52 generates a plurality of divided time-series data Da, Db, and Dc by extracting a necessary range from the load time-series data D. As shown in Fig. 7, the data pre-processing unit 52 includes a slope change point extraction unit 52a and a divided time-series data generation unit 52b.
[0071] The slope change point extraction unit 52a extracts slope change points Ca, Cb, Cc, and Cd in the load time series data D. The divided time series data generation unit 52b generates a plurality of divided time series data Da, Db, and Dc by dividing the load time series data D by the slope change points Ca, Cb, Cc, and Cd. The divided time series data generation unit 52b first divides the load time series data D by the slope change points Ca, Cb, Cc, and Cd.
[0072] In FIG. 8, the period of the load time series data D is divided into periods P11, P12, P13, P14, and P15. Period P11 is the period until the slope change point Ca is reached. Period P12 is the period from when the slope change point Ca is reached to when the slope change point Cb is reached. Period P13 is the period from when the slope change point Cb is reached to when the slope change point Cc is reached. Period P14 is the period from when the slope change point Cc is reached to when the slope change point Cd is reached. Period P15 is the period after when the slope change point Cd is reached.
[0073] The periods P11 and P15 of the load time-series data D are determined to be small load data and not being processed. The periods P12, P13, and P14 of the load time-series data D are determined to be being processed. Then, the data of the periods P12, P13, and P14 of the load time-series data D are generated as divided time-series data Da, Db, and Dc, respectively.
[0074] The divided time series data Da in the period P12 generally corresponds to the drilling process (S2) at the first speed Va. The divided time series data Db in the period P13 generally corresponds to the drilling process (S3) at the second speed Vb. The divided time series data Dc in the period P14 generally corresponds to the drilling process (S4) at the third speed Vc.
[0075] In the evaluation support device 2 of the second example, the feature determination unit 43, the model generation unit 44, the model storage unit 45, and the judgment unit 46 are substantially the same as those in the first example. That is, the evaluation support device 2 of the second example generates a plurality of trained models Ma, Mb, Mc based on a plurality of divided time series data Da, Db, Dc generated by the divided time series data generation unit 52b. In addition, the evaluation support device 2 of the second example evaluates a phenomenon of the machine tool 1 using the generated trained models Ma, Mb, Mc, with the plurality of divided time series data Da, Db, Dc generated by the divided time series data generation unit 52b as input data.
[0076] In the second example of the evaluation support device 2, it is only necessary to acquire the load time series data D to acquire a plurality of divided time series data Da, Db, Dc at a desired timing. Then, by performing machine learning using the acquired plurality of divided time series data Da, Db, Dc, it is possible to generate a plurality of highly accurate trained models Ma, Mb, Mc. Then, by using the trained models Ma, Mb, Mc, it is possible to evaluate the phenomenon of the machine tool 1 with high accuracy. [Explanation of symbols]
[0077] 1: machine tool, 2: machine phenomenon evaluation support device, 20: workpiece holding device, 30: tool holding device, 41: time series data acquisition unit, 42: data pre-processing unit, 42a: frequency distribution generation unit, 42b: peak class extraction unit, 42c: load threshold setting unit, 42d: divided time series data generation unit, 43: feature amount determination unit, 44: model generation unit, 45: model storage unit, 46: judgment unit, 52: data pre-processing unit, 52a: slope change point extraction unit, 52b: divided time series data generation unit, 100: evaluation support system, Ca, Cb, Cc, Cd: slope change point, D: load time series data, Da, Db, Dc: divided time series data, Ma, Mb, Mc: trained model, Pe1, Pe2, Pe3: peak class, T: tool, Th1, TH2, Th3: load threshold, Va: 1st speed, Vb: 2nd speed, Vc: 3rd speed, W: Workpiece
Claims
1. a time series data acquisition unit that acquires time series data of a load of an operation of the machine detected by a sensor installed in the machine; a frequency distribution generating unit that generates a frequency distribution in which a first axis represents the class of the load and a second axis perpendicular to the first axis represents the frequency of the load in the time-series data; a peak class extraction unit that extracts a class of the load at a peak of the frequency distribution; a load threshold setting unit that sets a load threshold for classifying the time-series data of the load based on the peak class extracted by the peak class extraction unit; a divided time-series data generating unit that generates a plurality of divided time-series data by dividing the time-series data of the load by the load threshold; a feature determining unit that determines a feature for each of the plurality of divided time series data; a model generation unit that generates a trained model for each of the plurality of divided time-series data, the trained model being a machine learning model with the feature amount as an explanatory variable and the phenomenon of the machine as a target variable; A mechanical phenomenon evaluation support device comprising:
2. The machine operation is a machining operation by a machine tool, The machine phenomenon evaluation support device according to claim 1 , wherein the load is a load generated on the machine tool during the machining operation.
3. 3. A machine phenomenon evaluation support device as described in claim 2, wherein the phenomenon of the machine which is the objective variable is at least one of the machining quality of a workpiece machined by the machine tool, the presence or absence of a malfunction of the machine tool, the deterioration state of the machine tool, and the amount of thermal displacement of the machine tool.
4. 4. The machine phenomenon evaluation support device according to claim 2, wherein the time series data of the load acquired by the time series data acquisition unit is data asynchronous with control data generated by a control program of the machine tool.
Citation Information
Patent Citations
Abnormal working detector and method thereof
JP1984102559A
Arithmetic unit for travel required time information
JP2000285372A
Blade tool diagnostic device and diagnostic method
JP2006082154A
Abnormality sign diagnostic device and abnormality sign diagnostic method
JP2013008111A
Abnormality detection device and method for tool of machine tool
JP2018024055A
