Predictive detection and diagnostic device and method

Abstract

Identify the point at which wear or chipping occurred in the machine tool's tools. [Solution] The predictive detection and diagnostic device 1 includes a data acquisition unit 10 that acquires time-series data of the load current supplied to the motor of the machine tool 4 and at the same time acquires time-series data of the machining start trigger signal from the CNC 42; a reference data creation unit 12 that calculates reference data for kernel density estimation at each time based on a first batch data extracted from the time-series data at the time of reference data creation; and an inspection execution unit 13 that calculates a kernel density estimate at each time based on a second batch data extracted from the time-series data at the time of inspection and the reference data, and calculates a score at each time that indicates the degree of deviation of the second batch data from the first batch data based on this estimate.

Description

[Technical Field]

[0001] The present invention relates to a predictive detection and diagnostic device and method for detecting abnormalities in machine tools. [Background technology]

[0002] When cutting materials using machine tools, defective products can sometimes be produced due to tool wear or chipping. Conventionally, there is a known technique for detecting tool wear by measuring the current value of a motor connected to the spindle of a machine tool and evaluating the entire batch based on the changes in this value (see Patent Document 1). In addition, there is a known technique for detecting tool abnormalities for each type of tool by using the current value of a motor that controls tool changes, along with the current value of a motor connected to the spindle of a machine tool (see Patent Document 2).

[0003] The technology disclosed in Patent Document 1 allows for evaluation of the entire batch, but it has the problem that it cannot identify the locations where tool wear or chipping occurs in a batch consisting of multiple processing steps.

[0004] Furthermore, the technology disclosed in Patent Document 2 had the problem that a separate measuring device for the motor that controls tool changes was required in order to extract multiple machining sections within a batch. In addition, since the section is extracted each time the tool is changed, there was a problem that it was not possible to individually extract consecutive different machining operations performed with the same tool. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Patent No. 6952318 [Patent Document 2] Patent No. 6924529 [Overview of the project] [Problems that the invention aims to solve]

[0006] The present invention was made to solve the above problems and aims to provide a predictive detection and diagnostic device and method that can identify the point in time when wear or chipping occurs in the tools of a machine tool. [Means for solving the problem]

[0007] The present invention provides a predictive maintenance and diagnostic device comprising: a data acquisition unit configured to acquire time-series data of load current supplied to the motor of a machine tool, and at the same time acquire time-series data of a machining start trigger signal from a controller that controls the machine tool; a data storage unit configured to store the time-series data of the load current and the time-series data of the machining start trigger signal; a reference data creation unit configured to detect the machining section of a first batch of data extracted from the time-series data of the load current at the time of reference data creation, based on the time-series data of the first machining start trigger signal acquired simultaneously with the time-series data of the load current at the time of reference data creation, and to calculate reference data for kernel density estimation at each time based on the first batch data; and an inspection execution unit configured to calculate a kernel density estimate at each time based on the second batch data and the reference data, and to calculate a score at each time indicating the degree of deviation of the second batch data from the first batch data based on this estimate.

[0008] Furthermore, one example configuration of the predictive detection and diagnostic device of the present invention further comprises a data correction execution unit configured to correct the time positions of a plurality of first batch data and a first machining start trigger signal, and a second batch data and a second machining start trigger signal, wherein the data correction execution unit corrects the time positions of a plurality of first batch data and a plurality of time-series data of the first machining start trigger signal based on the first machining start trigger signal, corrects the time positions of a plurality of time-series data of the first machining start trigger signal based on the first machining start trigger signal, corrects the second batch data and a second machining start trigger signal based on the first and second machining start trigger signals so that their time positions match those of the first batch data and corrects the second machining start trigger signal so that their time positions match those of the first machining start trigger signal. Furthermore, one example configuration of the predictive detection and diagnostic device of the present invention is characterized by further comprising a DC component removal unit configured to remove the DC components of the first batch data and the second batch data, respectively. Furthermore, in one example configuration of the predictive detection and diagnostic device of the present invention, the DC component removal unit is characterized by calculating an offset value of the first batch data for each processing section of the first batch data and subtracting the offset value of the first batch data from the first batch data, and calculating an offset value of the second batch data for each processing section of the second batch data and subtracting the offset value of the second batch data from the second batch data. Furthermore, in one example configuration of the predictive detection and diagnostic device of the present invention, the DC component removal unit is characterized in that it removes the DC components of the first batch data and the second batch data using a digital filter. Furthermore, in one example configuration of the predictive detection and diagnostic device of the present invention, the DC component removal unit is characterized in that it converts the first batch data and each of the first batch data into frequency domain data, performs filtering to remove low-frequency components, and then converts them back into time domain data.

[0009] Furthermore, in one example configuration of the predictive maintenance detection and diagnostic device of the present invention, the reference data creation unit stores the feature quantities of a plurality of first batch data and the reference data in the data storage unit in association with the work type data input by the user, and the inspection execution unit determines the work type at the time of inspection based on the feature quantities of the first batch data and the second batch data, and calculates the kernel density estimate based on the reference data and the second batch data corresponding to the work type at the time of inspection. Furthermore, in one example configuration of the predictive detection and diagnostic device of the present invention, the feature quantities of the first batch data are the waveform and length of a representative data from among the plurality of first batch data, and the inspection execution unit performs a first variety determination by comparing the length of the representative data with the length of the second batch data, and if there are two or more variety candidates in the first variety determination, a second variety determination is performed based on the correlation coefficient or cross-correlation between the waveform of the representative data and the second batch data, and if there are two or more variety candidates in the second variety determination, the variety candidate with the largest calculation result of the correlation coefficient or cross-correlation is determined to be the variety of the workpiece at the time of inspection. Furthermore, in one example configuration of the predictive detection and diagnostic device of the present invention, the data acquisition unit acquires time-series data of the load current and time-series data of the machining start trigger signal, and acquires workpiece type data from the machine tool; the reference data creation unit stores the calculated reference data and the workpiece type data acquired by the data acquisition unit at the time of reference data creation in the data storage unit in association with each other; and the inspection execution unit acquires the reference data corresponding to the workpiece type data acquired by the data acquisition unit at the time of inspection from the data storage unit, and calculates a kernel density estimate based on this reference data and the second batch data.

[0010] Furthermore, the present invention's predictive detection and diagnostic method is characterized by including: a first step of acquiring time-series data of the load current supplied to the motor of a machine tool, and simultaneously acquiring time-series data of a machining start trigger signal from a controller that controls the machine tool; a second step of detecting the machining section of a first batch of data extracted from the time-series data of the load current at the time of creating the reference data, based on the time-series data of the first machining start trigger signal acquired simultaneously with the time-series data of the load current at the time of creating the reference data, and calculating reference data for kernel density estimation at each time based on the first batch data; a third step of calculating a kernel density estimate at each time based on the second batch data and the reference data; and a fourth step of calculating a score at each time indicating the degree of deviation of the second batch data from the first batch data based on the kernel density estimate. [Effects of the Invention]

[0011] According to the present invention, reference data for kernel density estimation is calculated time by time based on a first batch data extracted from time-series data at the time of reference data creation, a kernel density estimate is calculated time by time based on a second batch data extracted from time-series data at the time of inspection and the reference data, and a score indicating the degree of deviation of the second batch data from the first batch data is calculated time by time based on this estimate, thereby making it possible to identify the point in time when wear or chipping of the machine tool is thought to have occurred. As a result, the present invention makes it possible to improve operations, such as reviewing the processing procedure within a batch. [Brief explanation of the drawing]

[0012] [Figure 1] Figure 1 is a block diagram showing the configuration of a machine tool inspection system according to a first embodiment of the present invention. [Figure 2] Figure 2 is a flowchart illustrating the operation of the predictive detection and diagnostic device according to the first embodiment of the present invention when creating reference data. [Figure 3]FIG. 3 is a diagram for explaining a method of identifying a processing section in the first embodiment of the present invention. [Figure 4] FIG. 4 is a block diagram showing the configuration of a data correction execution unit according to the first embodiment of the present invention. [Figure 5] FIG. 5 is a flowchart for explaining jitter correction processing at the time of creating reference data by a data correction execution unit according to the first embodiment of the present invention. [Figure 6] FIG. 6 is a diagram for explaining jitter correction processing at the time of creating reference data by a data correction execution unit according to the first embodiment of the present invention. [Figure 7] FIG. 7 is a flowchart for explaining DC component removal processing at the time of creating reference data by a DC component removal unit according to the first embodiment of the present invention. [Figure 8] FIG. 8 is a flowchart for explaining the operation at the time of inspection of a prognostic detection / diagnosis device according to the first embodiment of the present invention. [Figure 9] FIG. 9 is a flowchart for explaining jitter correction processing at the time of inspection by a data correction execution unit according to the first embodiment of the present invention. [Figure 10] FIG. 10 is a flowchart for explaining DC component removal processing at the time of inspection by a DC component removal unit according to the first embodiment of the present invention. [Figure 11] FIG. 11 is a diagram showing an example of time-series data for reference data and time-series data for inspection. [Figure 12] FIG. 12 is a diagram showing an example of time-series data of scores. [Figure 13] FIG. 13 is a diagram showing the change in the average score value for each number of processing times. [Figure 14] FIG. 14 is a flowchart for explaining the operation at the time of creating reference data of a prognostic detection / diagnosis device according to the third embodiment of the present invention. [Figure 15] FIG. 15 is a diagram showing an example of the waveform of representative data of batch data. [Figure 16] FIG. 16 is a flowchart for explaining the operation at the time of inspection of a prognostic detection / diagnosis device according to the third embodiment of the present invention. [Figure 17] Figure 17 is a block diagram showing an example of the configuration of a computer that implements the predictive detection and diagnostic device according to the first to third embodiments. [Modes for carrying out the invention]

[0013] [First Embodiment] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Figure 1 is a block diagram showing the configuration of a machine tool inspection system according to a first embodiment of the present invention. The machine tool inspection system consists of a predictive detection and diagnostic device 1, a current transformer (CT) 2 which converts the load current supplied to the motor that drives the workpiece of the machine tool 4 into a current of a size that can be handled by the predictive detection and diagnostic device 1, and a display unit 3 for displaying the inspection results of the predictive detection and diagnostic device 1.

[0014] The machine tool 4 includes a cutting tool (not shown) for processing the workpiece, a motor 40 for driving the workpiece, a control unit 41 for controlling the motor 40, and a CNC (Computer Numerical Control) 42 for controlling the entire machine tool.

[0015] The predictive detection and diagnostic device 1 includes a data acquisition unit 10 that acquires time-series data of the load current waveform supplied to the motor 40 of the machine tool 4, a data storage unit 11 that stores the time-series data, a reference data creation unit 12 that calculates reference data for kernel density estimation at each time based on batch data for reference data (first batch data) extracted from the time-series data used to create the reference data, an inspection execution unit 13 that calculates an estimated kernel density at each time based on inspection batch data (second batch data) extracted from the time-series data used for inspection and the reference data, and calculates a score at each time indicating the degree of deviation of the inspection batch data from the reference data batch data based on this estimated amount, a display data generation unit 14 that generates data to be displayed on the display unit 3, and a data correction execution unit 15 that performs jitter correction processing and DC component removal processing of the batch data.

[0016] First, we will explain the operation of the predictive detection and diagnostic device 1 when creating reference data. Figure 2 is a flowchart illustrating the operation of the predictive detection and diagnostic device 1 when creating reference data. The user of the predictive maintenance and diagnostic device 1 instructs the predictive maintenance and diagnostic device 1 to create reference data when the machine tool 4 is in a normal state.

[0017] When the data acquisition unit 10 of the predictive detection and diagnostic device 1 receives instructions from the user, it acquires time-series data of the load current waveform supplied to the motor 40 of the machine tool 4 via CT2, and at the same time acquires time-series data of the machining start trigger signal from the CNC 42 (Figure 2, step S100). The machining start trigger signal is a signal synchronized with the machining section in which the workpiece is being machined. The time-series data of the load current waveform and the time-series data of the machining start trigger signal acquired together with this data are stored as a pair of data in the data storage unit 11.

[0018] The reference data creation unit 12 of the predictive detection and diagnostic device 1 extracts batch data for reference data from the time-series data of load current waveforms for reference data stored in the data storage unit 11, according to preset conditions (Figure 2, step S101). The reference data creation unit 12 extracts, for example, the time-series data of load current waveforms for one batch, from the start of machining to the end of machining, as batch data for reference data. A timing signal roughly indicating one batch can be obtained from the CNC 42. Alternatively, the reference data creation unit 12 may determine the cycle length L by determining the interval in which the current value is above a certain level. For example, the cycle length L can be determined by determining that one batch has started when the current value is 0.1A or higher, and that one batch has ended when the current value remains below 0.1A for 1 second or more. The data acquisition unit 10 and the reference data creation unit 12 repeatedly execute the processes in steps S100 and S101 until they have extracted a predetermined number of batch data.

[0019] Next, after the reference data creation unit 12 has finished extracting a predetermined number of batch data (YES in step S102 of Figure 2), it detects the processing interval within the batch data for reference data based on the time-series data of the processing start trigger signal acquired simultaneously with the batch data (step S103 of Figure 2).

[0020] For example, in the example in Figure 3, D1 represents batch data for reference data, and ST represents the machining start trigger signal. In this embodiment, the section from the falling edge of the machining start trigger signal ST to the next rising edge is detected as the machining section. In the example in Figure 3, based on the machining start trigger signal ST, 10 machining sections SP-1 to SP-10 are detected in the batch data D1.

[0021] In the calculation of reference data during the reference data creation process and the calculation of score values ​​during the inspection process, jitter in the batch data will affect the results of the reference data creation process and the inspection process. Therefore, it is desirable to appropriately correct for jitter.

[0022] The data correction execution unit 15 of the predictive detection and diagnostic device 1 corrects the jitter in both the batch data for reference data and the time-series data of the processing start trigger signal (Figure 2, step S104). Figure 4 is a block diagram showing the configuration of the data correction execution unit 15. The data correction execution unit 15 consists of a cross-correlation calculation unit 150, a jitter correction unit 151, and a DC component removal unit 152.

[0023] Figure 5 is a flowchart illustrating the jitter correction process during reference data creation by the data correction execution unit 15. The cross-correlation calculation unit 150 of the data correction execution unit 15 calculates the cross-correlation between a machining start trigger signal ST acquired simultaneously with one batch data (hereinafter referred to as representative data) from the batch data for reference data creation a predetermined number of times extracted by the reference data creation unit 12, and a machining start trigger signal ST acquired simultaneously with other batch data other than the representative data (Figure 5, step S201). In the following explanation, the representative data will be D1, the machining start trigger signal ST acquired simultaneously with the representative data D1 will be ST1, the other batch data other than the representative data will be Di, and the machining start trigger signal ST acquired simultaneously with batch data Di will be STi.

[0024] The cross-correlation calculation unit 150 calculates the cross-correlation between the machining start trigger signal ST1 and the machining start trigger signal STi for each pulse of the trigger signal. In the case of the first pulse of the machining start trigger signal ST1 and the first pulse of the machining start trigger signal STi, the cross-correlation calculation unit 150 calculates the cross-correlation from the beginning of batch data D1,Di to the beginning of the next second pulse (rising edge). In the case of the second and subsequent pulses of the machining start trigger signal ST1 and the second and subsequent pulses of the machining start trigger signal STi, the cross-correlation calculation unit 150 calculates the cross-correlation from the end of the pulse immediately preceding the target pulse (falling edge) to the beginning of the pulse immediately following the target pulse. In the case of the last pulse of the machining start trigger signal ST1 and the last pulse of the machining start trigger signal STi, the cross-correlation calculation unit 150 calculates the cross-correlation from the end of the pulse immediately preceding the last pulse to the end of batch data D1,Di.

[0025] After calculating the cross-correlation between the first pulse of the machining start trigger signal ST1 and the first pulse of the machining start trigger signal STi, the jitter correction unit 151 of the data correction execution unit 15 uses the starting point of the first pulse of the machining start trigger signal ST1 as the reference position and determines the amount of deviation Z from the reference position at the time position where the cross-correlation is maximized (Figure 5, step S202).

[0026] The jitter correction unit 151 applies jitter correction to both the machining start trigger signal STi, for which cross-correlation has been calculated, and the batch data Di acquired simultaneously with the machining start trigger signal STi (Figure 5, step S203). Specifically, the jitter correction unit 151 moves both the time position from the start of the first pulse to the end of the last pulse of the machining start trigger signal STi, for which cross-correlation has been calculated, and the time position of the batch data Di acquired simultaneously with the machining start trigger signal STi, by the amount of deviation Z calculated for the first pulse of the machining start trigger signal STi.

[0027] Next, the cross-correlation calculation unit 150 calculates the cross-correlation between the second pulse of the machining start trigger signal ST1 and the second pulse of the machining start trigger signal STi (step S201). After calculating the cross-correlation between the second pulse of the machining start trigger signal ST1 and the second pulse of the machining start trigger signal STi, the jitter correction unit 151 uses the starting point of the second pulse of the machining start trigger signal ST1 as the reference position and determines the amount of deviation Z from the reference position at the time position where the cross-correlation is maximized (step S202).

[0028] The jitter correction unit 151 moves both the time position from the start of the second pulse to the end of the last pulse of the machining start trigger signal STi, for which the cross-correlation has been calculated, and the time position from the start of the second pulse of the machining start trigger signal STi to the end of the batch data Di acquired simultaneously with the machining start trigger signal STi, by the amount of deviation Z calculated for the second pulse of the machining start trigger signal STi (step S203).

[0029] Next, the cross-correlation calculation unit 150 calculates the cross-correlation between the third pulse of the machining start trigger signal ST1 and the third pulse of the machining start trigger signal STi (step S201). After calculating the cross-correlation between the third pulse of the machining start trigger signal ST1 and the third pulse of the machining start trigger signal STi, the jitter correction unit 151 uses the starting point of the third pulse of the machining start trigger signal ST1 as the reference position and determines the amount of deviation Z from the reference position at the time position where the cross-correlation is maximized (step S202).

[0030] The jitter correction unit 151 moves both the time position from the start of the third pulse to the end of the last pulse of the machining start trigger signal STi, for which the cross-correlation has been calculated, and the time position from the start of the third pulse of the machining start trigger signal STi to the end of the batch data Di acquired simultaneously with the machining start trigger signal STi, by the amount of deviation Z calculated for the third pulse of the machining start trigger signal STi (step S203).

[0031] When the batch data Di is moved, for example in the direction of time delay, the jitter correction unit 151 deletes the trailing end of the batch data Di by the amount of the shift Z. Furthermore, since a blank space without data is created before the moved batch data Di, the jitter correction unit 151 interpolates the blank space by the amount of the shift Z using the data immediately following the blank space.

[0032] Furthermore, when the batch data Di is moved in a direction that advances time, the jitter correction unit 151 deletes the leading edge of the batch data Di by the amount of the shift Z. In addition, since a blank space without data is created after the batch data Di, the jitter correction unit 151 interpolates the blank space by the amount of the shift Z using the data immediately preceding the blank space.

[0033] Thus, the cross-correlation calculation unit 150 and the jitter correction unit 151 perform the processing in steps S201 to S203 for each pulse of the machining start trigger signal STi. When the processing in steps S201 to S203 is completed for all pulses of the machining start trigger signal STi (YES in step S200 in Figure 5), the processing in steps S200 to S203 is performed for the next batch data Di. When the processing in steps S200 to S203 is completed for all batch data Di (YES in step S204 in Figure 5), the jitter correction processing is completed. In this way, each batch data for creating reference data is corrected to be synchronized. The time-series data of the batch data D1, Di and the machining start trigger signals ST1, STi after the jitter correction processing are stored in the data storage unit 11.

[0034] Figures 6(A) to 6(F) illustrate an example of jitter correction processing. Figure 6(A) shows representative data D1, Figure 6(C) shows other batch data D2 besides representative data D1, and Figure 6(E) shows batch data D2 after jitter correction processing. However, in Figures 6(A), 6(C), and 6(E), only the machining section SP is shown for simplification, and the current waveforms of D1 and D2 are omitted. Figure 6(B) shows machining start trigger signals ST1-1 to ST1-5 acquired simultaneously with representative data D1, Figure 6(D) shows machining start trigger signals ST2-1 to ST2-5 acquired simultaneously with batch data D2, and Figure 6(F) shows machining start trigger signals ST2-1 to ST2-5 after jitter correction processing.

[0035] As described above, the cross-correlation calculation unit 150 calculates the cross-correlation between the machining start trigger signal ST1-1 and the machining start trigger signal ST2-1. The jitter correction unit 151 moves both the time position from the beginning of the machining start trigger signal ST2-1 to the end of the last machining start trigger signal ST2-5, and the time position of the batch data D2, in a direction that advances the time by the amount of deviation calculated for the machining start trigger signal ST1-1.

[0036] Next, the cross-correlation calculation unit 150 calculates the cross-correlation between the machining start trigger signal ST1-2 and the machining start trigger signal ST2-2. The jitter correction unit 151 moves both the time position from the beginning of the machining start trigger signal ST2-2 to the end of the last machining start trigger signal ST2-5, and the time position in batch data D2 from the beginning of the machining start trigger signal ST2-2 to the end of batch data D2, in a direction that advances the time by the amount of deviation calculated for the machining start trigger signal ST1-2.

[0037] Next, the cross-correlation calculation unit 150 calculates the cross-correlation between the machining start trigger signal ST1-3 and the machining start trigger signal ST2-3. The jitter correction unit 151 moves both the time position from the beginning of the machining start trigger signal ST2-3 to the end of the last machining start trigger signal ST2-5, and the time position in batch data D2 from the beginning of the machining start trigger signal ST2-3 to the end of batch data D2, in the direction of a time delay by the amount of deviation calculated for the machining start trigger signal ST1-3.

[0038] Next, the cross-correlation calculation unit 150 calculates the cross-correlation between the machining start trigger signal ST1-4 and the machining start trigger signal ST2-4. The jitter correction unit 151 moves both the time position from the beginning of the machining start trigger signal ST2-4 to the end of the last machining start trigger signal ST2-5, and the time position in batch data D2 from the beginning of the machining start trigger signal ST2-4 to the end of batch data D2, in a direction that advances the time by the amount of deviation calculated for the machining start trigger signal ST1-4.

[0039] Finally, the cross-correlation calculation unit 150 calculates the cross-correlation between the machining start trigger signal ST1-5 and the machining start trigger signal ST2-5. The jitter correction unit 151 moves both the time position from the start to the end of the machining start trigger signal ST2-5 and the time position from the start of the machining start trigger signal ST2-5 to the end of the batch data D2 in the direction of a time delay by the amount of deviation calculated for the machining start trigger signal ST1-5. In this way, jitter correction of batch data D2 and machining start trigger signals ST2-1 to ST2-5 is completed.

[0040] Next, we will explain the DC component removal process. If DC components are superimposed on batch data, it will affect the results of the reference data creation process and the inspection process. Therefore, it is desirable to appropriately remove the DC components from batch data.

[0041] The DC component removal unit 152 of the data correction execution unit 15 of the predictive detection and diagnostic device 1 removes the DC component from the batch data for reference data after jitter correction processing (Figure 2, step S105). Figure 7 is a flowchart illustrating the DC component removal process when creating reference data by the DC component removal unit 152.

[0042] The DC component removal unit 152 calculates the average or median value of the current value as an offset value for each batch of reference data for each processing section SP (Figure 7, step S301). The DC component removal unit 152 removes the DC component by subtracting the offset value calculated for this processing section SP from the batch data of the block section which is the processing start trigger signal section plus the processing section SP immediately following it (Figure 7, step S302).

[0043] The DC component removal unit 152 performs steps S301 and S302 for each processing section SP. When steps S301 and S302 have been completed for all processing sections SP (YES in step S300 in Figure 5), steps S300 to S302 are performed for the next batch of data. When steps S300 to S302 have been completed for all batch data for the jitter-corrected reference data (YES in step S303 in Figure 7), the DC component removal process is completed. The batch data after the DC component removal process is stored in the data storage unit 11.

[0044] Alternatively, the median value of the current may be used as the offset value, or the offset value may be calculated using Otsu's binarization method. Alternatively, instead of subtracting the offset value, the DC component may be removed using a digital filter with high-pass characteristics. Alternatively, the DC component of the batch data for reference data may be removed by converting the batch data for reference data into frequency domain data using frequency analysis techniques such as discrete Fourier transform or discrete cosine transform, filtering to remove low-frequency components, and then inversely transforming it back into time domain data.

[0045] Next, the reference data creation unit 12 calculates the bandwidth h and standard deviation σ necessary for calculating the kernel density estimate to be performed during inspection, based on batch data for reference data that has undergone jitter correction and DC component removal processing (Figure 2, step S106). Sample data x1, x2, ..., x n The function f calculates the kernel density estimator with the kernel function K(x) and bandwidth h as parameters. KDE (x) is given by the following equation.

[0046]

number

[0047] The kernel function K(x) is given by equation (2).

[0048]

number

[0049] The bandwidth h is given by equation (3).

[0050]

number

[0051] Equations (1) to (3) represent the function f KDEThis means that (x) is approximated by the sum of the same number of kernel functions K(x) as the sample data. In equation (3), IQR (InterQuartile Range) is the difference between the 75th percentile and the 25th percentile of the sample in the interquartile range. min(a,b) is a function that takes the smaller of a and b.

[0052] The reference data creation unit 12 calculates the time-dependent standard deviation σ and time-dependent bandwidth h of the sample data based on the sample data at the time before and after the time to be calculated. Here, time refers to the elapsed time, for example, with time 0 being the rising edge of the first processing start trigger signal ST. Sample data refers to the current value at a certain time in the batch data. The number of samples n is the value obtained by multiplying the width before and after the time to be calculated for the standard deviation σ and bandwidth h (the number of sample data points at the time before and after) by the number of batch data (the number of batches acquired by the data acquisition unit 10).

[0053] This completes the creation of the reference data. The calculation results of the reference data (standard deviation σ and bandwidth h) calculated by the reference data creation unit 12 are stored in the data storage unit 11.

[0054] Next, we will explain the operation of the predictive maintenance detection and diagnostic device 1 during inspection. Figure 8 is a flowchart illustrating the operation of the predictive maintenance detection and diagnostic device 1 during inspection. The user of the predictive maintenance and diagnostic device 1 instructs the predictive maintenance and diagnostic device 1 to perform an inspection of the machine tool 4.

[0055] When the data acquisition unit 10 of the predictive detection and diagnostic device 1 receives instructions from the user, it acquires time-series data of the load current waveform supplied to the motor 40 of the machine tool 4 via CT2, just as when creating reference data, and at the same time acquires time-series data of the machining start trigger signal from the CNC 42 (Figure 8, step S400). The time-series data of the load current waveform and the time-series data of the machining start trigger signal acquired together with this data are stored as a pair of data in the data storage unit 11.

[0056] The inspection execution unit 13 of the predictive detection and diagnostic device 1 extracts batch data for inspection from the time-series data of the load current waveform for inspection stored in the data storage unit 11, in the same way as when creating reference data, according to pre-set conditions (Figure 8, step S401).

[0057] The data correction execution unit 15 corrects the jitter in both the batch data for inspection and the time-series data of the processing start trigger signal, similar to when creating the reference data (Figure 8, step S402). Figure 9 is a flowchart illustrating the jitter correction process during inspection performed by the data correction execution unit 15.

[0058] The cross-correlation calculation unit 150 of the data correction execution unit 15 calculates the cross-correlation between the machining start trigger signal ST1 acquired simultaneously with the representative data D1 and the machining start trigger signal ST acquired simultaneously with the inspection batch data (Figure 9, step S501). In the following description, the inspection batch data will be referred to as Dtst, and the machining start trigger signal ST acquired simultaneously with the batch data Dtst will be referred to as STtst.

[0059] The cross-correlation calculation unit 150 calculates the cross-correlation between the machining start trigger signal ST1 and the machining start trigger signal STtst for each pulse of the trigger signal. In the case of the first pulse of the machining start trigger signal ST1 and the first pulse of the machining start trigger signal STtst, the cross-correlation calculation unit 150 calculates the cross-correlation from the beginning of batch data D1,Dtst to the beginning (rising edge) of the next second pulse. In the case of the second and subsequent pulses of the machining start trigger signal ST1 and the second and subsequent pulses of the machining start trigger signal STtst, the cross-correlation calculation unit 150 calculates the cross-correlation from the end (falling edge) of the pulse immediately preceding the target pulse to the beginning of the pulse immediately following the target pulse. In the case of the last pulse of the machining start trigger signal ST1 and the last pulse of the machining start trigger signal STtst, the cross-correlation calculation unit 150 calculates the cross-correlation from the end of the pulse immediately preceding the last pulse to the end of batch data D1,Dtst.

[0060] After calculating the cross-correlation between the first pulse of the machining start trigger signal ST1 and the first pulse of the machining start trigger signal STtst, the jitter correction unit 151 of the data correction execution unit 15 uses the starting point of the first pulse of the machining start trigger signal ST1 as the reference position and determines the amount of deviation Z from the reference position at the time position where the cross-correlation is maximum (Figure 9, step S502).

[0061] The jitter correction unit 151 applies jitter correction to both the machining start trigger signal STtst, for which cross-correlation has been calculated, and the batch data Dtst for inspection, which is acquired simultaneously with the machining start trigger signal STtst (Figure 9, step S503). Specifically, the jitter correction unit 151 moves both the time position from the start of the first pulse to the end of the last pulse of the machining start trigger signal STtst, for which cross-correlation has been calculated, and the time position of the batch data Dtst, by the amount of deviation Z calculated for the first pulse of the machining start trigger signal STtst.

[0062] Next, the cross-correlation calculation unit 150 calculates the cross-correlation between the second pulse of the machining start trigger signal ST1 and the second pulse of the machining start trigger signal STtst (step S501). After calculating the cross-correlation between the second pulse of the machining start trigger signal ST1 and the second pulse of the machining start trigger signal STtst, the jitter correction unit 151 uses the starting point of the second pulse of the machining start trigger signal ST1 as the reference position and determines the amount of deviation Z from the reference position at the time position where the cross-correlation is maximized (step S502).

[0063] The jitter correction unit 151 moves both the time position from the start of the second pulse to the end of the last pulse of the machining start trigger signal STtst, for which the cross-correlation has been calculated, and the time position from the start of the second pulse of the machining start trigger signal STtst to the end of the batch data Dtst for inspection, by the amount of deviation Z calculated for the second pulse of the machining start trigger signal STtst (step S503).

[0064] Next, the cross-correlation calculation unit 150 calculates the cross-correlation between the third pulse of the machining start trigger signal ST1 and the third pulse of the machining start trigger signal STtst (step S501). After calculating the cross-correlation between the third pulse of the machining start trigger signal ST1 and the third pulse of the machining start trigger signal STtst, the jitter correction unit 151 uses the starting point of the third pulse of the machining start trigger signal ST1 as the reference position and determines the amount of deviation Z from the reference position at the time position where the cross-correlation is maximized (step S502).

[0065] The jitter correction unit 151 moves both the time position from the start of the third pulse to the end of the last pulse of the machining start trigger signal STtst, for which the cross-correlation has been calculated, and the time position from the start of the third pulse of the machining start trigger signal STtst to the end of the batch data Dtst for inspection, by the amount of deviation Z calculated for the third pulse of the machining start trigger signal STtst (step S503).

[0066] Thus, the cross-correlation calculation unit 150 and the jitter correction unit 151 perform the processing in steps S501 to S503 for each pulse of the machining start trigger signal STtst. When the processing in steps S501 to S503 has been completed for all pulses of the machining start trigger signal STtst (YES in step S500 in Figure 9), the jitter correction process is considered complete. In this way, the representative data D1 and the inspection batch data Dtst are corrected to be synchronized. The time-series data of the batch data Dtst and the machining start trigger signal STtst after the jitter correction process are stored in the data storage unit 11.

[0067] Next, the DC component removal unit 152 removes the DC component from the batch data Dtst for inspection after jitter correction processing, similar to when creating the reference data (Figure 8, step S403). Figure 10 is a flowchart illustrating the DC component removal process during inspection by the DC component removal unit 152.

[0068] The DC component removal unit 152 calculates the average or median value of the current value for each machining section SP as an offset value for the batch data Dtst for inspection (Figure 10, step S601). The DC component removal unit 152 removes the DC component by subtracting the offset value calculated for this machining section SP from the batch data Dtst of the block section which is the section of the machining start trigger signal plus the machining section SP immediately following it (Figure 10, step S602).

[0069] The DC component removal process is completed when the processing in steps S601 and S602 is finished for all processing sections SP in the batch data Dtst (step S600 in Figure 10). The batch data Dtst after the DC component removal process is stored in the data storage unit 11.

[0070] Similar to the batch data used for reference data, the median of the current values ​​may be used as the offset value, or the offset value may be calculated using Otsu's binarization method. Alternatively, instead of subtracting the offset value, the DC component may be removed using a digital filter with high-pass characteristics. Alternatively, the DC component of the batch data for testing may be removed by converting the batch data for testing into frequency domain data using frequency analysis techniques such as discrete Fourier transform or discrete cosine transform, filtering to remove low-frequency components, and then inversely transforming it back into time domain data.

[0071] The inspection execution unit 13 calculates a kernel density estimate f based on the reference data (standard deviation σ and bandwidth h) and the batch data Dtst for inspection after DC component removal processing. KDE (x) is calculated for each time step (Figure 8, step S404). As with the creation of the reference data, time refers to the elapsed time, with the rising edge of the first machining start trigger signal ST being time 0.

[0072] The inspection execution unit 13 substitutes the sample data at the time to be calculated in the batch data Dtst for inspection after the DC component removal process, and the reference data (standard deviation σ and bandwidth h) corresponding to the time to be calculated, into equations (1) and (2) to obtain the kernel density estimator f at the time to be calculated. KDE We just need to calculate (x).

[0073] Next, the inspection execution unit 13 calculates the kernel density estimate f KDE Based on (x), a score SC is calculated for each time step, indicating the degree of deviation of the test sample data from the reference sample data (Figure 8, step S405). The score SC is obtained, for example, by the following formula:

[0074]

number

[0075] A higher score SC indicates a greater degree of deviation of the inspection sample data from the reference sample data. When tool wear occurs, the load current changes, so anomaly detection is possible based on the height of the score SC. The time-series data of the score SC is stored in the data storage unit 11.

[0076] The display data generation unit 14 of the predictive detection and diagnostic device 1, for example, correlates time-series data of score SC with time-series data of the load current waveform for inspection acquired by the data acquisition unit 10, graphs it, and displays it on the display unit 3 (Figure 8, step S406). The user can check the graph displayed on the display unit 3 to determine whether or not there is an abnormality in the machine tool 4.

[0077] An example of the test results according to this embodiment is shown in Figures 11(A), 11(B), 12, and 13. Figure 11(A) shows the time series data for the reference data, Figure 11(B) shows the time series data for the test, and Figure 12 shows the calculation results of the score SC. Here, although the difference between the time series data for the reference data and the time series data for the test is slight, differences occur in the flat interval from 3 seconds to 15 seconds and the flat interval from 47 seconds to 50 seconds. The calculation results of the score SC according to this embodiment are shown in Figure 12, and it can be seen that the value of the score SC is high in the interval from 3 seconds to 15 seconds and in the interval from 47 seconds to 50 seconds.

[0078] Figure 13 shows the change in the average score SC for each machining cycle. The average score SC increases mainly due to tool wear and chipping. In the case of tool wear, the average score SC increases gradually, while in the case of tool chipping, the average score SC increases sharply. As can be seen from Figure 13, it is possible to clearly distinguish between the increase in score SC due to tool wear and the increase in score SC due to tool chipping.

[0079] As described above, this embodiment makes it possible to identify the point in time when tool wear or chipping is thought to have occurred. As a result, this embodiment makes it possible to improve operations, such as reviewing the machining procedure within a batch.

[0080] [Second Example] Next, a second embodiment of the present invention will be described. In this embodiment as well, the configuration of the machine tool inspection system is the same as in the first embodiment, so it will be described using the reference numerals in Figure 1. In this embodiment, the data acquisition unit 10 of the predictive detection and diagnostic device 1 acquires time-series data of the load current waveform and time-series data of the machining start trigger signal, as well as the workpiece type data of the workpiece being processed from the CNC 42.

[0081] The reference data creation unit 12 of the predictive detection and diagnostic device 1 associates the calculated reference data (standard deviation σ and bandwidth h) with the variety data acquired by the data acquisition unit 10 during reference data creation and stores it in the data storage unit 11.

[0082] The inspection execution unit 13 of the predictive detection and diagnostic device 1 obtains reference data (standard deviation σ and bandwidth h) corresponding to the variety data acquired by the data acquisition unit 10 during inspection from the data storage unit 11, and calculates a kernel density estimate f based on this reference data and the batch data for inspection. KDE Calculate (x).

[0083] Other operations are the same as in the first embodiment. Thus, in this embodiment, the present invention can be applied to a machine tool 4 that processes multiple types of workpieces, and abnormalities in the machine tool 4 can be detected for each type of workpiece.

[0084] [Third embodiment] Next, a third embodiment of the present invention will be described. In this embodiment as well, the configuration of the machine tool inspection system is the same as in the first embodiment, so it will be described using the reference numerals in Figure 1. When processing multiple types of workpieces with a single machine tool 4, it is necessary to maintain reference data for multiple types. In the second embodiment, type data is obtained from the CNC 42, but obtaining type data often requires modification of the machine tool 4, which is time-consuming and technically demanding. Therefore, a function that identifies the type of workpiece from batch data becomes useful.

[0085] Figure 14 is a flowchart illustrating the operation of the predictive detection and diagnostic device 1 in this embodiment when creating reference data. The processing in steps S100 to S105 in Figure 14 is the same as in the first embodiment.

[0086] In this embodiment, the reference data creation unit 12 calculates the bandwidth h and standard deviation σ for each batch of reference data that has undergone jitter correction processing and DC component removal processing (Figure 14, step S106a). At this time, the reference data creation unit 12 obtains the waveform of representative data from among the multiple batch data for reference data used to calculate the bandwidth h and standard deviation σ, representative data of the processing start trigger signal acquired simultaneously with the batch data for reference data, and the cycle length L1 of the representative data of the batch data for reference data as feature quantities of the batch data for reference data.

[0087] The reference data creation unit 12 can obtain a timing signal from the CNC 42 indicating one batch, thereby determining the cycle length of the time-series data for one batch, and thus calculate the cycle length L1 of representative data from among multiple batch data. Alternatively, the reference data creation unit 12 may determine the cycle length L1 by determining the interval in which the current value is above a certain level. For example, the cycle length L1 can be determined by determining that one batch has started when the current value is 0.1A or higher, and that one batch has ended when the current value remains below 0.1A for one second or more. The reference data creation unit 12 then stores the reference data, consisting of the bandwidth h, the standard deviation σ, the waveform of the representative data of the batch data, the representative data of the machining start trigger signal, and the cycle length L1, in the data storage unit 11 in association with the workpiece type data.

[0088] The reference data creation unit 12 performs the reference data creation process described above for each type of workpiece. The workpiece type data used for reference data creation is manually entered by the user of the predictive detection and diagnostic device 1 for each type of workpiece. This enables the reference data creation unit 12 to perform the reference data creation process for each type of workpiece. Examples of representative waveform data for batch data of varieties A to E are shown in Figures 15(A) to 15(E), respectively.

[0089] Figure 16 is a flowchart illustrating the operation of the predictive detection and diagnostic device 1 in this embodiment during inspection. The processes in steps S400 and S401 in Figure 16 are the same as in the first embodiment. In this embodiment, the inspection execution unit 13 checks whether the determination by cycle length is enabled (Figure 16, step S407). Whether or not to enable the determination by cycle length is specified by the user.

[0090] If the cycle length determination is enabled, the inspection execution unit 13 determines the cycle length L2 of the batch data for inspection extracted in step S401 (Figure 16, step S408). The inspection execution unit 13 can determine the cycle length L2 of the time-series data for one batch by obtaining a timing signal indicating one batch from the CNC 42. Alternatively, the inspection execution unit 13 may determine the cycle length L2 by determining the interval in which the current value is above a certain level. For example, the cycle length L2 can be determined by determining that one batch has started when the current value is 0.1A or higher, and that one batch has ended when the current value remains below 0.1A for one second or more. If the cycle length determination is not enabled, the inspection execution unit 13 proceeds to step S416, which will be described later.

[0091] If the determination by cycle length is enabled, the inspection execution unit 13 compares the cycle length L2 obtained in step S408 with the cycle length L1 of representative data included in the standard data for each workpiece type, and determines how many workpiece types satisfy the predetermined type determination conditions for cycle length L1 (Figure 16, step S409). For example, if the cycle length L2 is within ±5% of the cycle length L1, the inspection execution unit 13 designates the workpiece type corresponding to the standard data including this cycle length L1 as the first type candidate, as it satisfies the type determination conditions.

[0092] If there is only one candidate for the first variety, the inspection execution unit 13 determines that the first candidate is the variety of the workpiece being inspected (Figure 16, step S410). If there is no candidate for the first variety, the inspection execution unit 13 determines that variety determination is not possible (Figure 16, step S411). In the examples in Figures 15(A) to 15(E), varieties A, B, D, and E are the first candidate varieties.

[0093] If there is only one candidate product type in the first inspection, the inspection execution unit 13 uses the product type of the workpiece at the time of inspection and representative data of the processing start trigger signal of the determined product type to perform batch data for inspection and jitter correction processing of the processing start trigger signal (Figure 16, step S412). The DC component removal unit 152 removes the DC component of the batch data for inspection (step S413 in FIG. 16).

[0094] Next, the inspection execution unit 13 substitutes the batch data for inspection after the DC component removal process and the reference data of the candidate varieties determined for the variety of the workpiece during inspection into equations (1) and (2), and calculates the kernel density estimator f KDE (x) at each time (step S414 in FIG. 16). Subsequently, the inspection execution unit 13 may calculate the score SC at each time based on the calculated kernel density estimator f KDE (x), for example, by equation (4) (step S415 in FIG. 16). The process of step S406 in FIG. 16 is the same as that in the first embodiment.

[0095] When it is determined in step S409 that there are two or more first candidate varieties, the inspection execution unit 13 calculates the correlation coefficient or cross-correlation between the batch data for inspection and the waveform of the representative data of the batch data included in the reference data of the first candidate varieties for each first candidate variety (step S416 in FIG. 16), and determines how many varieties of the workpiece satisfy the predetermined variety determination condition for the correlation coefficient or cross-correlation (step S417 in FIG. 16).

[0096] The inspection execution unit 13 uses the candidate varieties with a correlation coefficient or cross-correlation greater than or equal to the reference value as the second candidate varieties that satisfy the variety determination condition. In the examples of FIGS. 15(A) to 15(E), variety A is less than or equal to the reference value, and varieties B, D, and E are greater than or equal to the reference value. As a result, varieties B, D, and E become the second candidate varieties.

[0097] When there is one second candidate variety, the inspection execution unit 13 determines the second candidate variety as the variety of the workpiece during inspection (step S410). Also, when there is no second candidate variety, the inspection execution unit 13 determines that the variety cannot be determined (step S411).

[0098] If the second inspection determines that there are two or more candidate varieties, the inspection execution unit 13 determines the variety of the workpiece at the time of inspection to be the variety of the workpiece with the largest correlation coefficient or cross-correlation calculation result (Figure 16, step S418). In the above example, variety B is determined to be the variety of the workpiece at the time of inspection. The processing in steps S412 to S415 and S406 is as described above.

[0099] In this embodiment, it becomes unnecessary to acquire product type data from the machine tool 4, and only the load current needs to be acquired. Furthermore, even when cutting workpieces of the same type using the same machine tool 4, there are variations in the motor current profile due to the influence of the machining environment, random errors, and machining defects. In this embodiment, by utilizing the correlation between reference data and target data, it becomes possible to determine the product type while tolerating these variations.

[0100] The predictive detection and diagnostic device 1 described in the first to third embodiments can be realized by a computer equipped with a CPU (Central Processing Unit), a storage device, and an interface, and a program that controls these hardware resources. An example of the configuration of this computer is shown in Figure 17.

[0101] The computer comprises a CPU 200, a storage device 201, and an interface device (I / F) 202. A display unit 3 and a machine tool 4 are connected to the I / F 202. In such a computer, the program for realizing the predictive detection and diagnosis method of the present invention is stored in the storage device 201. The CPU 200 executes the processes described in the first to third embodiments according to the program stored in the storage device 201. Furthermore, at least a part of the predictive detection and diagnosis device 1 may be implemented in hardware. [Explanation of Symbols]

[0102] 1... Predictive detection and diagnostic device, 2... CT, 3... Display unit, 4... Machine tool, 10... Data acquisition unit, 11... Data storage unit, 12... Reference data creation unit, 13... Inspection execution unit, 14... Display data generation unit, 15... Data correction execution unit, 40... Motor, 41... Control unit, 42... CNC, 150... Cross-correlation calculation unit, 151... Jitter correction unit, 152... DC component removal unit.

Claims

1. A data acquisition unit is configured to acquire time-series data of the load current supplied to the motor of a machine tool, and at the same time acquire time-series data of the machining start trigger signal from the controller that controls the machine tool. A data storage unit configured to store time-series data of the load current and time-series data of the processing start trigger signal, A reference data creation unit is configured to detect a processing interval of a first batch of data extracted from the time-series data of the load current at the time of reference data creation, based on time-series data of a first processing start trigger signal acquired simultaneously with the time-series data of the load current at the time of reference data creation, and to calculate reference data for kernel density estimation at each time based on the first batch data. A predictive anomaly detection and diagnostic device comprising: an inspection execution unit configured to calculate a kernel density estimate for each time period based on the second batch data and the reference data, and to calculate a score for each time period indicating the degree of deviation of the second batch data from the first batch data based on this estimate.

2. In the predictive detection and diagnostic device according to claim 1, The system further comprises a data correction execution unit configured to correct the time positions of a plurality of the first batch data and the first machining start trigger signal and the second batch data and the second machining start trigger signal, The predictive detection and diagnostic device is characterized in that the data correction execution unit corrects the time positions of a plurality of first batch data so that they match based on the first machining start trigger signal, corrects the time positions of a plurality of time-series data of the first machining start trigger signal so that they match, corrects the second batch data so that it matches the time position of the first batch data so that it matches based on the first and second machining start trigger signals, and corrects the second machining start trigger signal so that it matches the time position of the first machining start trigger signal.

3. In the predictive detection and diagnostic device according to claim 1, The predictive detection and diagnostic device further comprises a DC component removal unit configured to remove the DC components from the first batch data and the second batch data, respectively.

4. In the predictive detection and diagnostic device according to claim 3, The DC component removal unit is characterized by calculating an offset value of the first batch data for each processing section of the first batch data and subtracting the offset value of the first batch data from the first batch data, and calculating an offset value of the second batch data for each processing section of the second batch data and subtracting the offset value of the second batch data from the second batch data.

5. In the predictive detection and diagnostic device according to claim 3, The DC component removal unit is characterized by removing the DC components of the first batch data and the second batch data using a digital filter, thereby providing a predictive detection and diagnostic device.

6. In the predictive detection and diagnostic device according to claim 3, The DC component removal unit is characterized by converting each of the first batch data into frequency domain data, performing a filter process to remove low-frequency components, and then converting them back into time domain data.

7. In the predictive detection and diagnostic device according to claim 1, The reference data creation unit stores the feature quantities of the multiple first batch data and the reference data in the data storage unit, associating them with the work type data entered by the user. The inspection execution unit determines the type of workpiece at the time of inspection based on the feature quantities of the first batch data and the second batch data, and calculates the kernel density estimate based on the reference data and the second batch data corresponding to the type of workpiece at the time of inspection, thereby providing a predictive inspection and diagnostic device.

8. In the predictive detection and diagnostic device according to claim 7, The features of the first batch data are the waveform and length of a representative data from among the multiple first batch data. The inspection execution unit performs a first type determination by comparing the length of the representative data with the length of the second batch data; if there are two or more type candidates in the first type determination, it performs a second type determination based on the correlation coefficient or cross-correlation between the waveform of the representative data and the second batch data; if there are two or more type candidates in the second type determination, it determines the type candidate with the largest result of the correlation coefficient or cross-correlation as the type of workpiece at the time of inspection.

9. In the predictive detection and diagnostic device according to claim 1, The data acquisition unit acquires time-series data of the load current and time-series data of the machining start trigger signal, and also acquires workpiece type data from the machine tool. The standard data creation unit associates the calculated standard data with the variety data acquired by the data acquisition unit during standard data creation and stores them in the data storage unit. The predictive detection and diagnostic device is characterized in that the inspection execution unit acquires the reference data corresponding to the variety data acquired by the data acquisition unit during inspection from the data storage unit, and calculates a kernel density estimate based on this reference data and the second batch data.

10. The first step involves acquiring time-series data of the load current supplied to the motor of a machine tool, and simultaneously acquiring time-series data of the machining start trigger signal from the controller that controls the machine tool. A second step involves detecting the processing interval of the first batch data extracted from the time-series data of the load current at the time of creating the reference data, based on the time-series data of the first processing start trigger signal acquired simultaneously with the time-series data of the load current at the time of creating the reference data, and calculating reference data for kernel density estimation at each time step based on the first batch data. A third step involves calculating a kernel density estimate for each time step based on the second batch data and the reference data, A predictive anomaly detection and diagnosis method characterized by comprising a fourth step of calculating, for each time step, a score indicating the degree of deviation of the second batch data from the first batch data based on the kernel density estimate.

Images