Abnormality determination device, machining system, program, and abnormality determination method

Abstract

To accurately determine whether a tool of a processing device is abnormal.SOLUTION: An abnormality determination device that determines an abnormality based on feature values extracted from detection information of physical quantities generated during execution of a series of machining operations in a machining device, includes: a calculation unit that calculates a correction parameter from a history of feature values used in multiple past abnormality determinations; a correction calculation unit that corrects the extracted feature values by adding the correction parameter; a score calculation unit that calculates a score from the feature values corrected by the correction calculation unit; and an abnormality determination unit that determines that a tool of the machining device is abnormal when the score calculated by the score calculation unit exceeds a predetermined threshold value.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to an abnormality determination device, a machining system, a program, and an abnormality determination method. [Background technology]

[0002] Conventionally, an abnormality determination device that determines abnormalities in a tool in a machine tool has been known. The conventional abnormality determination device receives vibration detection information from a vibration sensor attached to the machining device. The abnormality determination device extracts a feature value (frequency spectrum) from the detection information. The abnormality determination device also calculates a score (degree of deviation from a normal state) from the feature value. If the score exceeds a predetermined threshold, the abnormality determination device determines that the tool of the machining device is abnormal. Summary of the Invention [Problem to be solved by the invention]

[0003] However, with conventional abnormality determination devices, even when a tool of a processing device is operating normally, the detection information may fluctuate due to changes over time in the tool (for example, wear on the cutting edge). When the detection information fluctuates in this way, the feature amount fluctuates accordingly. Furthermore, when the feature amount fluctuates in accordance with the fluctuation in the detection information, the score calculated from the feature amount also fluctuates.

[0004] However, in conventional abnormality determination devices, the threshold value used for comparison with the score when determining an abnormality is a predetermined value (fixed value). Therefore, with conventional abnormality determination devices, even if the tool is operating normally, the feature amount fluctuates due to changes in the tool over time, causing the score calculated from the feature amount to exceed the predetermined threshold value, and the tool may be erroneously determined to be abnormal.

[0005] The present invention has been made in view of the above, and has an object to accurately determine whether a tool of a processing device is abnormal. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems and achieve the object, the present invention provides an abnormality determination device that determines an abnormality based on feature values ​​extracted from detection information of physical quantities generated during the execution of a series of machining operations in a machining device, and is characterized by comprising: a calculation unit that calculates a correction parameter from a history of feature values ​​used in multiple past abnormality determinations; a correction calculation unit that corrects the extracted feature values ​​by adding the correction parameter; a score calculation unit that calculates a score from the feature values ​​corrected by the correction calculation unit; and an abnormality determination unit that determines that a tool of the machining device is abnormal if the score calculated by the score calculation unit exceeds a predetermined threshold. [Effects of the Invention]

[0007] According to the present invention, it is possible to accurately determine whether a tool of a processing device is abnormal. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram showing an example of the configuration of a machining system to which a diagnostic device according to a first embodiment is applied. [Figure 2] FIG. 2 is a block diagram illustrating an example of a hardware configuration of the processing device. [Figure 3] FIG. 3 is a block diagram illustrating an example of a hardware configuration of the diagnostic device. [Figure 4] FIG. 4 is a block diagram illustrating an example of a functional configuration of the determination unit of the diagnostic device. [Figure 5] FIG. 5 is a flowchart showing an outline of the flow of the learning process. [Figure 6] FIG. 6 is a flowchart showing the flow of the normal data storage process. [Figure 7] FIG. 7 is a block diagram illustrating an example of the functional configuration of the diagnostic data correction unit. [Figure 8] FIG. 8 is a diagram illustrating an example of processing in the diagnostic data history storage unit. [Figure 9] FIG. 9 is a diagram illustrating an example of processing in the outlier removal unit and the diagnostic data prediction unit. [Figure 10] FIG. 10 is a diagram illustrating an example of processing in the diagnostic data prediction unit and the correction parameter calculation unit. [Figure 11] FIG. 11 is a flowchart showing the outline of the flow of the correction parameter calculation process. [Figure 12] FIG. 12 is a flowchart showing the flow of the diagnostic process. [Figure 13] FIG. 13 is a diagram showing a specific example of the diagnostic processing. [Figure 14] FIG. 14 is a diagram illustrating scores when / without applying the present invention. [Figure 15] FIG. 15 is a block diagram illustrating a configuration example of a machining system to which a diagnostic device according to the second embodiment is applied. [Figure 16] FIG. 16 is a flowchart showing the flow of processing in the analysis unit and the display reception unit. [Figure 17] FIG. 17 is a diagram showing an example of a display screen. DETAILED DESCRIPTION OF THE INVENTION

[0009] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of an abnormality determination device, a machining system, a program, and an abnormality determination method will be described in detail with reference to the accompanying drawings.

[0010] (First embodiment) Fig. 1 is a block diagram showing a configuration example of a machining system to which a diagnostic device according to a first embodiment is applied. As shown in Fig. 1, the machining system 1 according to the present embodiment includes a machining device 200 and a diagnostic device 100 that functions as an abnormality determination device.

[0011] The processing device 200 and the diagnostic device 100 may be connected in any manner. For example, the processing device 200 and the diagnostic device 100 may be connected by a dedicated connection line, a wired network such as a wired LAN (Local Area Network), a wireless network, or the like.

[0012] The machining device 200 includes a machine control unit 201, a tool changer 202, a display unit 203, a memory unit 204, a communication control unit 205, a numerical control unit 206, a tool information input unit 207, an alarm unit 208, an input / output unit 209, a machine tool 220, etc.

[0013] 1 and equipped with a drive unit. Z-axis stage 226 is equipped with a rotation spindle 221, which is an example of a rotation axis that constitutes processing apparatus 200. A tool holder 222 that holds a tool 223 is attached to rotation spindle 221. Machine tool 220 is equipped with an XY-axis stage 225, which is below rotation spindle 221 and is equipped with a drive unit and is movable in two axial directions in a plane perpendicular to Z-axis stage 226. XY-axis stage 225 holds workpiece 224.

[0014] The numerical control unit 206 executes machining by the machining device 200 through numerical control. For example, the numerical control unit 206 reads a machining program from the input / output unit 209, and generates and outputs numerical control data for controlling the rotation of the main spindle and the position of each axis stage. The machining program also describes the storage number of the tool changer 202, and the numerical control unit 206 performs tool change according to that description.

[0015] The numerical control unit 206 outputs the context information to the communication control unit 205. The context information is information that defines the operation of the tool 223 of the processing device 200, and a plurality of pieces of context information are defined for each type of operation of the tool 223. In this embodiment, the context information includes, for example, tool information that identifies the tool 223, rotation information of the rotating spindle 221 (for example, the spindle rotation speed that is the rotation speed of the rotating spindle 221), movement information of the Z-axis stage 226 and the XY-axis stage 225 (movement speed, information during movement), etc.

[0016] The tool information includes at least the tool type (e.g., drill, reamer, end mill) and the number of cutting edges. This tool information is input by the operator through the tool information input unit 207 in accordance with the information displayed on the display unit 203. Alternatively, the tool information can be obtained by reading a list file of the tool information from the input / output unit 209 or by inputting the information from an external computer (not shown) via the communication control unit 205. Moreover, the tool information may be stored in the storage unit 204 and can be referenced from the machining program.

[0017] The numerical control unit 206 transmits, for example, context information defining the current operation of the tool 223 to the diagnostic device 100 via the communication control unit 205. When machining the workpiece 224 in accordance with the machining program, the numerical control unit 206 controls the type of the tool 223, the positions of the Z-axis stage 226 and the XY-axis stage 225, the rotation speed of the rotary spindle 221, and the like, depending on the machining process. The numerical control unit 206 transmits, from the context information, context information corresponding to a predetermined operation to the diagnostic device 100 via the communication control unit 205. Here, the predetermined operation is a preset operation of the tool 223. In this embodiment, every time the numerical control unit 206 changes the type of operation of the tool 223, it sequentially transmits context information corresponding to the changed type of operation to the diagnostic device 100 via the communication control unit 205.

[0018] The communication control unit 205 controls communication with an external device such as the diagnostic device 100. For example, the communication control unit 205 transmits context information corresponding to the current operation of the tool 223 to the diagnostic device 100.

[0019] The physical quantity information detection unit 227 has a sensor that detects, as an analog signal, a time-varying physical quantity that occurs while the tool 223 is performing a machining operation on the workpiece 224. The physical quantity information detection unit 227 also has a function of appropriately amplifying the analog signal detected by the sensor, cutting off a desired frequency range, and then converting the analog signal into a digital signal. The physical quantity information detection unit 227 then transmits the digital signal to the diagnostic device 100 as detection information. A physical quantity is a variable used in physics. The type of sensor included in the physical quantity information detection unit 227 and the physical quantity detected may be any type. For example, the sensor included in the physical quantity information detection unit 227 may be a microphone, an acceleration sensor, an AE (acoustic emission) sensor, or the like, and each sensor outputs acoustic data, acceleration data, or data indicating AE waves as detection information. The diagnostic device 100 may have any number of physical quantity information detection units 227. For example, diagnostic device 100 may include multiple sensors that sense different physical quantities.

[0020] 1, the physical quantity information detection unit 227 has a sensor attached to a side surface of the structure that holds the rotating spindle 221 and a sensor attached to a side surface of the XY-axis stage 225. The sensor included in the physical quantity information detection unit 227 has a built-in acceleration sensor. When the machining device 200 starts machining, the physical quantity information detection unit 227 detects the acceleration of vibrations generated by the rotation of the rotating spindle 221. When the machining device 200 starts machining by bringing the tool 223 into contact with the workpiece 224, actual cutting is started, and a cutting force is generated. This acts as a vibration force, vibrating the tool 223 and the workpiece 224, and the vibration propagates between them. The physical quantity information detection unit 227 transmits the acceleration of this vibration, etc., to the diagnosis device 100 as detection information.

[0021] For example, if the cutting edge of the tool 223 breaks or chips during machining, the cutting force that was uniform for each cutting edge during normal machining becomes uneven, and the generated vibration changes. Alternatively, if chips get mixed in between the tool holder 222 and the rotating spindle 221 when replacing the tool 223 in the machining device 200, the cutting edge at the tip of the tool 223 will swing (vibrate) more significantly relative to the rotating axis. This causes the cutting amount per cutting edge of the tool 223 to become uneven, and a change in vibration occurs due to uneven cutting force, just as when the cutting edge is damaged.

[0022] The diagnostic device 100 receives the vibration detection information via the communication control unit 101. Additionally, the communication control unit 101 controls communication with the processing device 200 to receive context information from the processing device 200. The determination unit 102 determines whether the processing state of the processing device 200 is normal or not by referring to the context information and the detection information. Furthermore, if the diagnostic device 100 diagnoses that the processing state of the processing device 200 is abnormal, it transmits alert information to the processing device 200 via the communication control unit 101. Upon receiving the alert information via the communication control unit 205, the processing device 200 displays the alert information on the display unit 203 and activates the alarm unit 208. The alarm unit 208 may be a patrol lamp, a buzzer, a speaker, or the like. Furthermore, the machine control unit 201 can interrupt the operation of the processing device 200 according to the processing program and stop the processing of the processing device 200. Note that an abnormality refers to a state outside a predetermined tolerance. As an example, a tool is abnormal if, when the tool is used to machine a workpiece, the size or cutting amount of the machined workpiece is larger or smaller than a predetermined tolerance range.

[0023] Fig. 2 is a block diagram showing an example of a hardware configuration of the processing apparatus. As shown in Fig. 2, the processing apparatus 200 according to this embodiment includes a CPU (Central Processing Unit) 251, a ROM (Read Only Memory) 252, a RAM (Random Access Memory) 253 corresponding to the storage unit 204 shown in Fig. 1, a communication I / F (Interface) 254, a drive control circuit 255, a motor 256, an input / output I / F 257, an input device 258, and a display 259 corresponding to the display unit 203 shown in Fig. 1, all of which are connected via a bus 260.

[0024] The CPU 251 controls the entire processing apparatus 200. The CPU 251 controls the entire operation of the processing apparatus 200 and realizes various functions of the processing apparatus 200 by executing programs stored in the ROM 252 or the like using the RAM 253 as a work area (working region), for example.

[0025] The communication I / F 254 is an interface for communicating with external devices such as the diagnostic device 100. The drive control circuit 255 is a circuit that controls the drive of the motor 256. The rotating spindle 221, the Z-axis stage 226, and the XY-axis stage 225 each include a drive unit such as the motor 256. The sensor 270 is attached to the processing device 200 and converts a physical quantity that changes in response to the operation of the processing device 200 into an electrical signal. The signal conversion circuit 271 amplifies the electrical signal output from the sensor 270 to a desired level, removes noise components contained in the electrical signal, and converts it into a digital signal. The signal conversion circuit 271 then outputs the digital signal to the diagnostic device 100 as detection information. In other words, the sensor 270 and the signal conversion circuit 271 correspond to, for example, the physical quantity information detection unit 227 shown in FIG. 1.

[0026] The numerical control unit 206 and the communication control unit 205 shown in FIG. 1 may be realized by the CPU 251 executing a program stored in the ROM 252, i.e., by software, or may be realized by hardware such as an IC (Integrated Circuit), or may be realized by a combination of software and hardware.

[0027] Fig. 3 is a block diagram showing an example of the hardware configuration of a diagnostic device. Diagnostic device 100 according to this embodiment is, for example, a personal computer. As shown in Fig. 3, diagnostic device 100 has a configuration in which CPU 151, ROM 152, RAM 153, communication I / F 154, auxiliary storage device 155, and input / output I / F 157 are connected via bus 160.

[0028] The CPU 151 controls the entire diagnostic device 100. For example, the CPU 151 controls the operation of the entire diagnostic device 100 and realizes the diagnostic function of the processing device 200 by executing a program stored in the ROM 152 or the like using the RAM 153 as a work area.

[0029] The communication I / F 154 is an interface for communicating with external devices such as the processing device 200. The auxiliary storage device 155 stores various information such as setting information of the diagnostic device 100, context information received from the processing device 200, and detection information output from the physical quantity information detection unit 227. The auxiliary storage device 155 also stores various calculation results used to determine whether the processing state of the processing device 200 is normal or not. The auxiliary storage device 155 is made up of non-volatile storage means such as an HDD (Hard Disk Drive), an EEPROM (Electrically Erasable Programmable Read-Only Memory), or an SSD (Solid State Drive).

[0030] The CPU 151 executes a program to function as the communication control unit 101 and the determination unit 102 shown in FIG.

[0031] The input / output I / F 157 sequentially displays the detection information input from the physical quantity information detection unit 227 on the display 159, and displays the determination result by the determination unit 102. The input / output I / F 157 also accepts settings required for diagnosing the processing device 200 that are input by the user while looking at the display 159 via an input device 158 such as a keyboard or a mouse.

[0032] 4 is a block diagram showing an example of the functional configuration of a determination unit of the diagnostic device. The determination unit 102 of the diagnostic device 100 according to this embodiment includes a data acquisition unit 110, a time division unit 111, a frequency analysis unit 112, a feature extraction unit 113, a learning unit 114, a learning model storage unit 115, a normal data storage unit 116, a diagnostic data correction unit 117, a score calculation unit 118, an input unit 119, a threshold storage unit 120, an abnormality determination unit 121, and a diagnostic device display unit 130.

[0033] The data acquiring unit 110 acquires time-series data of physical properties, which are signals from the sensor 270 attached to the processing device 200. More specifically, the data acquiring unit 110 acquires context information corresponding to a predetermined operation from among a plurality of context information defining the operation of a tool attached to a rotation shaft constituting the processing device 200, rotation information of the rotation shaft, tool information of the tool, and detection information of time-varying physical quantities (time-series data of physical properties) that occur while the tool is executing a processing operation on a workpiece.

[0034] The time dividing unit 111 divides time into detected information of time-varying physical quantities (time-series data of physical properties) that occur during the execution of a series of machining operations.

[0035] The frequency analysis unit 112 performs frequency analysis on the detection information of time-varying physical quantities that occur while the tool is performing a machining operation on the workpiece. More specifically, the frequency analysis unit 112 converts time-series data into frequency domain data in units of time division by the time division unit 111.

[0036] The feature extraction unit 113 extracts features from the frequency analysis results based on context information corresponding to a predetermined operation from among multiple pieces of context information that define the operation of a tool attached to a rotary shaft that constitutes the processing device 200. More specifically, the feature extraction unit 113 extracts features in units of time division by the time division unit 111. A feature is a numerical value that quantitatively represents a feature of data. An example of a feature is a frequency spectrum. Note that the feature may be any information that indicates the feature of the detection information. For example, the feature may be energy, MFCC (Mel-Frequency Cepstrum Coefficients), etc., in addition to the frequency spectrum.

[0037] The learning unit 114 divides the detection information during normal times into time segments, and generates a learning model for each time segment from the feature amounts extracted for each time segment within that time. More specifically, the learning unit 114 generates the learning model through the learning effect of machine learning using the feature amounts extracted by the feature amount extraction unit 113.

[0038] Here, machine learning is a technology that allows a computer to acquire human-like learning capabilities, in which the computer autonomously generates algorithms necessary for judgments such as data classification from learning data that is input in advance, and applies these to new data to make predictions. The learning method for machine learning may be any of supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, and deep learning, or may be a combination of these learning methods; any learning method for machine learning is acceptable.

[0039] The learning model storage unit 115 stores the learning model generated by the learning unit 114. More specifically, the learning model storage unit 115 stores the learning model generated by the learning unit 114 together with context information of the machine tool 220.

[0040] FIG. 5 is a flowchart showing an outline of the flow of the learning process.

[0041] First, the data acquiring unit 110 receives context information from the machine tool 220 (step S1). Next, the data acquiring unit 110 receives time-series physical data from the sensor 270 (step S2).

[0042] Next, the time division unit 111 divides the time series physical data into time segments (step S3). Next, the frequency analysis unit 112 performs frequency analysis for each time segment (step S4). Next, the feature extraction unit 113 extracts features for each time segment (step S5).

[0043] Next, the learning unit 114 generates a model for each time division corresponding to the context information based on the data acquired in a normal state before diagnosis as described above (step S6). The learning model generated by the learning unit 114 can use a clustering algorithm such as a Gaussian mixture model or K-means. Learning is performed using data for multiple times of one processing cycle. Then, the learning model storage unit 115 stores the generated model for each time division (step S7).

[0044] Returning to FIG. 4, the normal data storage unit 116 divides the detection information in normal times into time divisions, calculates the representative value of the feature amount in normal times for each time division extracted within that time, and stores it.

[0045] Fig. 6 is a flowchart showing an outline of the flow of the normal data storage process. The processes from step S1 to step S5 are the same as the flow of the learning process described in Fig. 5. After the feature extraction unit 113 extracts the feature for each time division (step S5), the normal data storage unit 116 calculates a representative value of the feature for each time division corresponding to the context information (step S11) and stores the representative value of the feature for each time division (step S12).

[0046] 4, during diagnosis, the diagnostic data correction unit 117 corrects the feature amount for each time division of the data to be diagnosed (diagnostic data) using a representative value of the feature amount in a normal state. Details of the processing in the diagnostic data correction unit 117 will be described later.

[0047] The score calculation unit 118 applies the learning model for each time division to the feature amount for each time division corrected by the diagnostic data correction unit 117, and calculates a score for each time division (deviation from the normal state). Furthermore, the score calculation unit 118 calculates a representative score value corresponding to a predetermined operation of the processing device 200 from the score calculated for each time division. By using the representative score value in this way, it is possible to simplify the determination of an abnormality.

[0048] The input unit 119 receives a user input of a threshold value for determining an abnormality before a diagnosis is made.

[0049] Threshold storage unit 120 stores a predetermined threshold for determining an abnormality input by input unit 119 together with context information of machine tool 220. The predetermined threshold is a numerical value stored in the machine itself for determining an abnormality. An example of the predetermined threshold is a threshold that is received by input unit 119 as a user input and stored in threshold storage unit 120.

[0050] The abnormality determination unit 121 compares the representative score of the diagnostic data with a threshold value to determine whether the tool state or the machining state is normal or abnormal.

[0051] The diagnostic device display unit 130 displays the result of the determination by the abnormality determination unit 121 or feeds it back to the processing device 200 .

[0052] Next, the diagnostic data corrector 117 will be described in detail.

[0053] 7 is a block diagram showing an example of the functional configuration of the diagnostic data correction unit 117 according to this embodiment. The diagnostic data correction unit 117 includes a diagnostic data history storage unit 122, an outlier removal unit 123, a diagnostic data prediction unit 124, a correction parameter calculation unit 125 functioning as a calculation unit, and a correction calculation unit 126.

[0054] The diagnostic data history storage unit 122 receives a plurality of time-divided feature amounts from the feature amount extraction unit 113. The diagnostic data history storage unit 122 calculates and stores a representative value from among the plurality of time-divided feature amounts.

[0055] Here, the flow from acquiring detection information to storing a representative feature value will be described. The data acquisition unit 110 acquires detection information during diagnosis. The time division unit 111 time-divides the acquired detection information. The frequency analysis unit 112 converts the time-divided multiple pieces of detection information (time-series data of physical properties) from time-series data to frequency-domain data. The feature extraction unit 113 extracts features from the converted frequency-domain data in units of time division by the time division unit 111. The diagnostic data history storage unit 122 in the diagnostic data correction unit 117 receives the time-divided multiple feature values ​​from the feature extraction unit 113. The diagnostic data history storage unit 122 calculates a representative feature value from the time-divided multiple feature values. The representative feature value is a representative value of the feature values ​​(frequency spectrum) in one processing cycle. Specifically, the diagnostic data history storage unit 122 divides one processing cycle into time periods and calculates the average value of the frequency spectrum within each time period as the representative feature value. The diagnostic data history storage unit 122 stores the calculated feature amount representative value.

[0056] At this time, the diagnostic data history storage unit 122 stores a history of the feature quantity representative values ​​for each of the latest fixed number of time divisions of the machining operation. That is, the diagnostic data history storage unit 122 replaces the feature quantity representative value for the oldest time division with the new feature quantity representative value. The history of the feature quantity representative values ​​is the feature quantities in multiple past machining operations. As will be described later, an example of the history of the feature quantity representative values ​​is the frequency spectrum from the k-10th machining operation to the k-1st machining operation. Also, an example of the history of the feature quantity representative values ​​may be the representative value of the frequency spectrum for each time division from the k-10th machining operation to the k-1st machining operation. The history of the feature quantity representative values ​​is the respective feature quantity representative values ​​in multiple past machining operations. An example of the history of the feature quantity representative values ​​is the frequency spectrum used as the feature quantity representative value from the k-10th machining operation to the k-1st machining operation.

[0057] FIG. 8 is a diagram showing an example of processing in the diagnostic data history storage unit 122. The example shown in FIG. 8 shows up to the k-1th processed value of the feature quantity representative value for each time division. The diagnostic data history storage unit 122 stores a fixed number of histories. In the example shown in FIG. 8, the fixed number is set to "10." In the next kth processed data, the oldest k-10th processed data among the stored data is discarded and replaced with the latest kth processed data.

[0058] The outlier exclusion unit 123 determines whether or not there are statistical outliers and excludes the outliers from the history of representative values ​​of feature amounts at the time of diagnosis for each time division recorded by the diagnosis data history storage unit 122. That is, when calculating feature amounts for the current abnormality determination, the outlier exclusion unit 123 excludes outliers from feature amounts used for multiple previous abnormality determinations.

[0059] The diagnostic data prediction unit 124 predicts the representative value of the feature quantity at the time of diagnosis for the latest time division to be diagnosed, using the history of the representative value of the feature quantity at the time of diagnosis for each time division, from which outliers have been removed by the outlier removal unit 123. The prediction for which processing operation point in time is to be made may be the point in time of the latest processing count of the diagnostic data, or the next processing count. If it is the next predicted value, it is stored and used next time.

[0060] FIG. 9 illustrates an example of processing in the outlier removal unit 123 and the diagnostic data prediction unit 124. In the example illustrated in FIG. 9, the diagnostic data history storage unit 122 updates the diagnostic data 303 for the kth processed data, and a box-and-whisker plot 304 of 10 pieces of history data is shown. The box-and-whisker plot 304 is a diagram showing the variability of the data. Using the box-and-whisker plot 304, it is determined whether the data is an outlier. Upper and lower limits 305 of the whisker length are 1.5 times the interquartile range. The outlier removal unit 123 determines that the kth processed data is an outlier because it is above the upper end of the whisker. Linear approximation using the kth processed data results in a value indicated by approximate line 301. However, linear approximation excluding the kth processed data, which is an outlier, results in a value indicated by approximate line 302. The slopes of approximate line 301 and approximate line 302 differ considerably depending on whether or not there is an outlier.

[0061] The diagnostic data prediction unit 124 calculates an approximation formula using a predetermined number of feature history records recorded immediately before the operation to be diagnosed. However, if any abnormal values ​​are included in the feature history, the prediction error will be large. Therefore, the outlier removal unit 123 determines whether or not there are any statistical outliers in the feature history, and removes any outliers that are found.

[0062] The correction parameter calculation unit 125 calculates a correction parameter for correcting the feature quantity for each time division from the predicted value of the diagnostic feature quantity for each latest time division to be diagnosed and the representative value of the normal feature quantity for each time division read from the normal data storage unit 116. The correction parameter is a numerical value for correcting the feature quantity used for diagnosis.

[0063] 10 is a diagram showing an example of processing in the diagnostic data prediction unit 124 and the correction parameter calculation unit 125. In the example shown in Fig. 10, the diagnostic data prediction unit 124 calculates a predicted value 306 for the kth processed data using the diagnostic data from the (k-1)th processed data to the (k-10)th processed data, and reads normal state data 307. That is, the correction parameter calculation unit 125 calculates a feature amount for the current abnormality determination from the feature amounts used for multiple previous abnormality determinations.

[0064] Then, the correction parameter calculation unit 125 uses the calculated predicted value 306 of the kth processed item and the normal state data 307 to calculate the correction parameter according to the following formula 1. Correction parameter for the kth processing step per time division = normal data per time division - predicted value for the kth processing step per time division (Equation 1)

[0065] That is, the correction parameter calculation unit 125 calculates the correction parameter by subtracting the feature amount in the current abnormality judgment predicted for the kth time (predicted value 306 for the kth processing) from the feature amount in the normal processing (normal data 307).

[0066] The correction calculation unit 126 corrects the feature amount of the kth processing step extracted within the time division, according to the following formula 2. Corrected feature quantity of the kth machining item = feature quantity extracted for each time division of the kth machining item + correction parameter of the kth machining item for each time division (Equation 2)

[0067] That is, the correction calculation unit 126 adds a correction parameter to the feature amount extracted in the kth abnormality determination.

[0068] Here, FIG. 11 is a flowchart showing an outline of the flow of the correction parameter calculation process.

[0069] First, the diagnostic data history storage unit 122 reads history data of feature quantities for each time division (step S21). Next, the outlier removal unit 123 removes outliers from the history data for each time division (step S22). Next, the diagnostic data prediction unit 124 performs prediction using the removed history data for each time division (step S23).

[0070] The correction parameter calculation unit 125 reads normal data for each time division (step S24), and then calculates a correction parameter for each time division (step S25).

[0071] 5, the correction calculation unit 126 corrects the feature amounts for each time division to be diagnosed using the correction parameters for the same time division. More specifically, the correction calculation unit 126 reads multiple feature amounts for each time division from the feature amount extraction unit 113, and corrects the multiple feature amounts for each time division using the correction parameters for each time period calculated by the correction parameter calculation unit 125.

[0072] Next, the specific process of the diagnostic process will be described.

[0073] FIG. 12 is a flowchart showing the outline of the flow of the diagnostic processing, and FIG. 13 is a diagram showing a specific example of the diagnostic processing.

[0074] S300 shown in Fig. 13 shows an example of time-series physical data of one cycle of machining operation to be diagnosed. The example shown in Fig. 13 shows vibration data acquired by installing an acceleration pickup sensor as the sensor 270 in the machining device 200. The tool is not in contact with the workpiece, and the vibration data shows a duration of approximately one second starting immediately after an idling operation command is issued.

[0075] S301 shown in Figure 13 is a spectrogram of S300, showing the time change of the spectrum from 0 to 500 Hz. In Figure 13, the density represents the strength of the spectrum intensity. Since the idling rotation speed was set to 9000 rpm, the intensity of bin S302, which includes 150 Hz, tends to be high.

[0076] 12, first, data acquisition unit 110 receives context information from machine tool 220 (step S31). Next, data acquisition unit 110 reads a model corresponding to the context information (step S32). Next, data acquisition unit 110 receives time-series physical data from sensor 270 (step S33).

[0077] Next, the time dividing unit 111 performs time division of the time series physical data (step S34). More specifically, the data acquiring unit 110 receives the time series data S300, and the time dividing unit 111 divides it into four parts: S300a, S300b, S300c, and S300d.

[0078] Next, the frequency analysis unit 112 performs frequency analysis for each time division (step S35). More specifically, the frequency analysis unit 112 divides S300a according to windows S303a-1, S303a-2, S303a-3, S303a-4, and S303a-5 for extracting data for Fourier transform. For example, these windows each have the same time width, and their starting positions are sequentially shifted by S311. S303a-5 exceeds S300a, but if it does exceed S300a, for example, zero padding is performed or the value of S300b is compensated for. Then, the frequency analysis unit 112 performs a Fourier transform on the time series data acquired using each window.

[0079] Next, the feature extraction unit 113 extracts feature amounts for each time division (step S36). More specifically, the feature extraction unit 113 extracts feature amounts for each time division from the results of the Fourier transform. For example, the feature extraction unit 113 selects any number of frequency bins of interest to use as feature amounts. Alternatively, the feature extraction unit 113 may treat multiple bins as bands, process the values ​​of the multiple bins to use as representative values ​​for the bands, and select the frequency range of the bands and the number of bands to use as feature amounts. The representative value for the band can be selected on a case-by-case basis, such as an average or a weighted sum. Here, for simplicity, if we limit to 150 Hz bins, the features for each time division are [a1, a2, a3, a4, a5] S303a, [B1, B2, B3, B4, B5] S303b, [c1, c2, c3, c4, c5] S303c, and [d1, d2] S303d.

[0080] The normal data storage unit 116 calculates and stores representative values ​​of the feature amounts S303a, S303b, S303c, and S303d for each time division used in learning. The representative values ​​may be average values, median values, or may be left as they are.

[0081] Next, the diagnostic data corrector 117 reads the correction parameters for each time division (step S37), and then corrects the extracted feature amount for each time division (step S38).

[0082] Next, the score calculation unit 118 reads a learning model corresponding to the context information (step S39). Next, the score calculation unit 118 calculates a score for each time division from the diagnostic data and the learning model (step S40). Next, the score calculation unit 118 calculates a representative score value (step S41).

[0083] Subsequently, abnormality determination unit 121 performs abnormality determination (step S42), and diagnostic device display unit 130 outputs or transmits the determination result to processing device 200 (step S43).

[0084] Here, FIG. 14 is a diagram for explaining scores when / without applying the present invention.

[0085] As shown in FIG. 14(a), when the present invention is not applied, the score changes greatly over time, many scores exceed the threshold, and good detection is not possible.

[0086] On the other hand, as shown in FIG. 14(b), when the present invention is applied, the change in the score over time is suppressed, sudden changes can be captured, and good detection is possible.

[0087] As described above, according to this embodiment, it is possible to accurately determine whether a tool of a processing device is abnormal. More specifically, according to this embodiment, even if the score changes over time, it is possible to correctly determine an abnormality that occurs suddenly, and it is possible to achieve a highly accurate diagnosis without updating the initial learning and threshold values.

[0088] (Second embodiment) Next, a second embodiment will be described.

[0089] The second embodiment differs from the first embodiment in that it analyzes user feedback regarding the results of past anomaly determinations and proposes changes to parameters for correcting feature quantities, etc. In the following description of the second embodiment, the same parts as in the first embodiment will be omitted, and only the parts that are different from the first embodiment will be described.

[0090] FIG. 15 is a block diagram showing an example of the configuration of a machining system to which the diagnostic device according to the second embodiment is applied.

[0091] The diagnostic device 100 of this embodiment notifies the user of the judgment result of a past abnormality judgment and proposes to change the conditions for correcting the feature amount. In the machining device 200, even if the tool 223 is judged to be abnormal in the judgment result of a past abnormality judgment, there are cases where the tool 223 is found to be normal when actually checked. Therefore, in order to deal with such cases, the diagnostic device 100 of this embodiment allows the user to input feedback and proposes to change the parameters for correcting the feature amount.

[0092] As shown in FIG. 15, the diagnostic device 100 according to this embodiment includes an analysis unit 103 and a display reception unit 104 in addition to the components shown in FIG.

[0093] The analysis unit 103 analyzes the results of multiple past anomaly determinations. Furthermore, the analysis unit 103 analyzes user feedback regarding the results of past anomaly determinations using the learning effect of machine learning, and proposes changes to parameters that correct feature amounts, etc.

[0094] Here, machine learning is a technology that allows a computer to acquire human-like learning capabilities, in which the computer autonomously generates algorithms necessary for judgments such as data classification from learning data that is input in advance, and applies these to new data to make predictions. The learning method for machine learning may be any of supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, and deep learning, or may be a combination of these learning methods; any learning method for machine learning is acceptable.

[0095] The display receiving unit 104 displays the analysis results by the analysis unit 103 on the display 159 which is a display unit, and also receives changes to the setting conditions in the diagnostic device 100 based on the analysis results by the analysis unit 103 .

[0096] FIG. 16 is a flowchart showing the flow of processing in the analysis unit and the display reception unit.

[0097] 16, first, the analysis unit 103 checks the number of abnormality determinations of the tool 223 (step S51). If the analysis unit 103 determines that the number of abnormality determinations of the tool 223 is not 10 or more (No in step S52), the analysis unit 103 returns to step S51.

[0098] On the other hand, if the analysis unit 103 determines that the number of times the abnormality determination of the tool 223 has been performed is 10 or more (Yes in step S52), the display receiving unit 104 notifies the user by displaying the analysis results by the analysis unit 103 on the display 159 (step S53). For example, if the abnormality determination of the tool 223 has been performed 10 times, the abnormality determination results of the tool 223 for the past 10 times may be displayed on the screen, and an input may be received from the user as to whether the abnormality determination results of the tool 223 are correct.

[0099] Here, Fig. 17 is a diagram showing an example of the display screen. In the example of the display screen shown in Fig. 17, the abnormality determination results X of the tool 223 for the past 10 times are displayed in a list.

[0100] Additionally, in the example of the display screen shown in Fig. 17, the display reception unit 104 displays notification content Y to the user on the screen of the display 159. In the example of notification content Y to the user on the display screen shown in Fig. 17, the display reception unit 104 displays the following text: "The last 10 tool abnormality determination results will be displayed. Are the tool abnormality determination results correct (the results intended by the operator)?"

[0101] 17, the display receiving unit 104 displays on the screen of the display 159 a "Yes" button B1 and a "No" button B2 indicating a response to the notification content Y to the user.

[0102] 16, the display receiving unit 104 determines whether the user has clicked the "No" button B2 on the screen (step S54). If the display receiving unit 104 determines that the user has clicked the "Yes" button B1 on the screen (No in step S54), the tool abnormality determination in the example of the display screen shown in Fig. 17 is correct (the result intended by the user), and therefore the parameters for correcting the feature amounts are not changed and the process returns to step S51.

[0103] On the other hand, if the display receiving unit 104 determines that the user has clicked the "No" button B2 on the screen (Yes in step S54), the tool abnormality determination in the example display screen shown in FIG. 17 is not the correct result (the result intended by the user), so the display receiving unit 104 changes the parameters for correcting the feature amount (step S55) and terminates the processing.

[0104] A specific example of changing the parameters in step S55 is changing the conditions for outlier exclusion, which was explained using Fig. 9. In Fig. 9, values ​​exceeding 1.5 times the interquartile range in the box plot 304 obtained immediately before the approximation are excluded. Therefore, when the user clicks the "No" button B2 on the screen, for example, the outlier exclusion unit 123 in the determination unit 102 changes the value to be excluded as an outlier from 1.5 times to 1.8 times. By relaxing the conditions for outlier exclusion, the outlier exclusion unit 123 increases the value of the correction parameter, thereby making it possible to better prevent a tool from being erroneously determined to be abnormal.

[0105] Note that an example of changing the conditions for correcting the feature amount is not limited to this. Alternatively, when the user clicks the "No" button B2 on the screen, the display receiving unit 104 may display the conditions for excluding outliers on the screen and allow the user to change the conditions by directly inputting correction parameters.

[0106] As described above, according to this embodiment, it is possible to analyze user feedback regarding the results of past abnormality determinations and propose changes to parameters for correcting feature amounts, etc.

[0107] Each function of the above-described embodiments can be realized by one or more processing circuits. Here, the term "processing circuit" in this specification includes a processor programmed to perform each function by software, such as a processor implemented by an electronic circuit, as well as devices such as an ASIC (Application Specific Integrated Circuit), a DSP (Digital Signal Processor), an FPGA (Field Programmable Gate Array), and conventional circuit modules designed to perform each of the above-described functions.

[0108] It should be noted that diagnostic device 100 is not limited to a personal computer as long as it is a device equipped with a communication function. Diagnostic device 100 may be, for example, an image forming device, a PJ (Projector), an IWB (Interactive White Board: a white board with an electronic blackboard function that allows mutual communication), an output device such as digital signage, a HUD (Head Up Display) device, industrial machinery, an imaging device, a sound collection device, medical equipment, a network home appliance, an automobile (Connected Car), a notebook PC (Personal Computer), a mobile phone, a smartphone, a tablet terminal, a game console, a PDA (Personal Digital Assistant), a digital camera, a wearable PC, a desktop PC, or the like.

[0109] For example, aspects of the present invention are as follows. <1> An abnormality determination device that determines an abnormality based on feature quantities extracted from detection information of physical quantities generated during execution of a series of machining operations in a machining device, a calculation unit that calculates correction parameters from a history of feature amounts used in multiple past abnormality determinations; a correction calculation unit that corrects the extracted feature amount by adding the correction parameter; a score calculation unit that calculates a score from the feature amount corrected by the correction calculation unit; an abnormality determination unit that determines that a tool of the processing device is abnormal when the score calculated by the score calculation unit exceeds a predetermined threshold; An abnormality determination device comprising: <2> The score calculation unit calculates a score representative value corresponding to a predetermined operation of the processing device from the calculated score. Characterized by <1> The abnormality determination device according to claim 1. <3> an outlier exclusion unit that excludes outliers from the feature amounts used in multiple past anomaly determinations when calculating the feature amounts for the current anomaly determination; a diagnostic data prediction unit that predicts a representative value of the feature amount at the time of diagnosis for each latest time division to be diagnosed, using a history of the representative values ​​of the feature amount from which the outliers have been removed by the outlier removal unit, the correction parameter calculation unit calculates the correction parameter using the predicted value predicted by the diagnostic data prediction unit and normal state data. Characterized by <1> or <2> The abnormality determination device according to claim 1. <4> the diagnostic data prediction unit approximates a change in the feature amount by a mathematical formula, and predicts, from the approximation formula, the feature amount when the device operates normally in the next diagnosis. Characterized by <3> The abnormality determination device according to claim 1. <5> the score calculation unit corrects the feature values ​​at the time of diagnosis using the feature values ​​used in the machine learning and the predicted value predicted by the diagnostic data prediction unit, and applies the corrected feature values ​​to a learning model to calculate a score. Characterized by <3> or <4> The abnormality determination device according to claim 1. <6> a time division unit that divides the detection information of the physical quantity into time divisions, the calculation unit uses a feature amount extracted from the time-divided detection information. Characterized by <1> Or <5> 10. The abnormality determination device according to claim 9, wherein <7> an analysis unit that analyzes results of abnormality determination made by the abnormality determination unit in the past multiple times; a display accepting unit that displays the analysis result by the analyzing unit on a display unit and accepts changes to the setting conditions based on the analysis result; characterized by comprising <1> Or <6> 10. The abnormality determination device according to claim 9, wherein <8> A machining system including a machining device and an abnormality determination device that determines an abnormality based on a feature extracted from detection information of a physical quantity generated during execution of a series of machining operations in the machining device, The abnormality determination device a calculation unit that calculates correction parameters from a history of feature amounts used in multiple past abnormality determinations; a correction calculation unit that corrects the extracted feature amount by adding the correction parameter; a score calculation unit that calculates a score from the feature amount corrected by the correction calculation unit; an abnormality determination unit that determines that a tool of the processing device is abnormal when the score calculated by the score calculation unit exceeds a predetermined threshold; A processing system comprising: <9> a computer that controls an abnormality determination device that determines abnormalities based on feature quantities extracted from detection information of physical quantities generated during execution of a series of processing operations in a processing device; a calculation unit that calculates correction parameters from a history of feature amounts used in multiple past abnormality determinations; a correction calculation unit that corrects the extracted feature amount by adding the correction parameter; a score calculation unit that calculates a score from the feature amount corrected by the correction calculation unit; an abnormality determination unit that determines that a tool of the processing device is abnormal when the score calculated by the score calculation unit exceeds a predetermined threshold; A program to function as a <10> An abnormality determination method for an abnormality determination device that determines an abnormality based on feature quantities extracted from detection information of physical quantities generated during execution of a series of processing operations in a processing device, comprising: a calculation step of calculating correction parameters from a history of feature amounts used in multiple past abnormality determinations; a correction calculation step of correcting the extracted feature amount by adding the correction parameter; a score calculation step of calculating a score from the feature amount corrected in the correction calculation step; an abnormality determination step of determining that the tool of the processing device is abnormal when the score calculated in the score calculation step exceeds a predetermined threshold; An abnormality determination method comprising: [Explanation of symbols]

[0110] 1 Processing System 100 Abnormality determination device 103 Analysis Department 104 Display Reception Department 111 Time division section 117 Diagnostic data correction unit 118 Score Calculation Section 121 Abnormality determination section 123 Outlier exclusion section 124 Diagnostic Data Prediction Unit 125 Calculation Unit 126 Correction calculation unit 200 Processing equipment [Prior art documents] [Patent documents]

[0111] [Patent Document 1] Japanese Patent Publication No. 2022-072435

Claims

1. An abnormality determination device that determines an abnormality based on feature quantities extracted from detection information of physical quantities generated during execution of a series of machining operations in a machining device, a calculation unit that calculates correction parameters from a history of feature amounts used in multiple past abnormality determinations; a correction calculation unit that corrects the extracted feature amount by adding the correction parameter; a score calculation unit that calculates a score from the feature amount corrected by the correction calculation unit; an abnormality determination unit that determines that a tool of the processing device is abnormal when the score calculated by the score calculation unit exceeds a predetermined threshold; An abnormality determination device comprising:

2. The score calculation unit calculates a score representative value corresponding to a predetermined operation of the processing device from the calculated score.

2. The abnormality determination device according to claim 1.

3. an outlier exclusion unit that excludes outliers from the feature amounts used in multiple past anomaly determinations when calculating the feature amounts for the current anomaly determination; a diagnostic data prediction unit that predicts a representative value of the feature amount at the time of diagnosis for each latest time division to be diagnosed, using the history of the representative value of the feature amount from which the outliers have been removed by the outlier removal unit; Furthermore, the calculation unit calculates the correction parameter using the predicted value predicted by the diagnostic data prediction unit and normal state data.

2. The abnormality determination device according to claim 1.

4. the diagnostic data prediction unit approximates a change in the feature amount by a mathematical formula, and predicts, from the approximation formula, the feature amount when the device operates normally in the next diagnosis.

4. The abnormality determination device according to claim 3.

5. the score calculation unit corrects the feature values ​​at the time of diagnosis using the feature values ​​used in the machine learning and the predicted value predicted by the diagnostic data prediction unit, and applies the corrected feature values ​​to a learning model to calculate a score.

4. The abnormality determination device according to claim 3.

6. a time division unit that divides the detection information of the physical quantity into time divisions, the calculation unit uses a feature amount extracted from the time-divided detection information.

2. The abnormality determination device according to claim 1.

7. an analysis unit that analyzes results of abnormality determination made by the abnormality determination unit in the past multiple times; a display accepting unit that displays the analysis result by the analyzing unit on a display unit and accepts changes to the setting conditions based on the analysis result; 2. The abnormality determination device according to claim 1, further comprising:

8. A machining system including a machining device and an abnormality determination device that determines an abnormality based on a feature extracted from detection information of a physical quantity generated during execution of a series of machining operations in the machining device, The abnormality determination device a calculation unit that calculates correction parameters from a history of feature amounts used in multiple past abnormality determinations; a correction calculation unit that corrects the extracted feature amount by adding the correction parameter; a score calculation unit that calculates a score from the feature amount corrected by the correction calculation unit; an abnormality determination unit that determines that a tool of the processing device is abnormal when the score calculated by the score calculation unit exceeds a predetermined threshold; A processing system comprising:

9. a computer that controls an abnormality determination device that determines abnormalities based on feature quantities extracted from detection information of physical quantities generated during execution of a series of processing operations in a processing device; a calculation unit that calculates correction parameters from a history of feature amounts used in multiple past abnormality determinations; a correction calculation unit that corrects the extracted feature amount by adding the correction parameter; a score calculation unit that calculates a score from the feature amount corrected by the correction calculation unit; an abnormality determination unit that determines that a tool of the processing device is abnormal when the score calculated by the score calculation unit exceeds a predetermined threshold; A program to function as a

10. An abnormality determination method for an abnormality determination device that determines an abnormality based on feature quantities extracted from detection information of physical quantities generated during execution of a series of processing operations in a processing device, comprising: a calculation step of calculating correction parameters from a history of feature amounts used in multiple past abnormality determinations; a correction calculation step of correcting the extracted feature amount by adding the correction parameter; a score calculation step of calculating a score from the feature amount corrected in the correction calculation step; an abnormality determination step of determining that the tool of the processing device is abnormal when the score calculated in the score calculation step exceeds a predetermined threshold; An abnormality determination method comprising:

Images