JP7625882B2 - Signal processing method, signal processing device, and computer program - Google Patents

Description

This disclosure relates to signal processing methods, signal processing devices, and computer program technologies.

Conventionally, there is known a technique for detecting change points in a signal, which is time-series data (Non-Patent Document 1). In conventional techniques, change points are detected using an algorithm called PLET (Pruned Exact Linear Time). In algorithms for detecting change points such as PELT, a point where at least one of the mean value and variance of a signal, which is time-series data, changes is assumed to be a change point. In conventional techniques, change points are detected by comparing the likelihood between the case where it is assumed that there is a change point and the case where it is assumed that there is no change point.

Optimal detection of changepoints with a linear computational cost; Killick,R., Fearnhead, P. and Eckley, I. A; October 10, 2012

Conventional technology can accurately detect change points for signals that have normality, but there may be cases where change points cannot be detected accurately for signals that do not have normality. Therefore, there has been a demand for technology that can accurately detect change points for various signals, including signals that do not have normality.

According to a first aspect of the present disclosure, a signal processing method is provided. This signal processing method includes: converting a first type signal, which is time-series data having a change point, into a second type signal that maintains information about the change point by performing signal processing on the first type signal; and detecting the change point using at least one of the first type signal and the second type signal. The conversion into the second type signal includes at least one of generating a first difference signal that is a set of differences in signal values at two first time points in the first type signal; and, when the first type signal has a discrete distribution, adding a noise signal generated by normal random numbers to the first type signal.

According to a second aspect of the present disclosure, a signal processing device is provided. The signal processing device includes a conversion unit that performs signal processing on a first type signal, which is time-series data having a change point, to convert the first type signal into a second type signal that maintains information about the change point, and a detection unit that detects the change point using at least one of the first type signal and the second type signal, and the conversion unit performs at least one of generating a first difference signal that is a set of differences in signal values at two first time points in the first type signal, and adding a noise signal generated by normal random numbers to the first type signal when the first type signal has a discrete distribution.

According to a third aspect of the present disclosure, a computer program is executed to be executed by a computer. This computer program includes a conversion function that performs signal processing on a first type signal, which is time-series data having a change point, to convert the first type signal into a second type signal that maintains information about the change point, and a detection function that detects the change point using at least one of the first type signal and the second type signal, and the conversion function includes at least one of a function of generating a first difference signal that is a set of differences in signal values at two first time points in the first type signal, and a function of adding a noise signal generated by normal random numbers to the first type signal when the first type signal has a discrete distribution.

FIG. 1 is a diagram for explaining a signal processing device according to an embodiment of the present invention. FIG. 4 is a diagram showing an example of a first type signal. 11 is a first flowchart of a detection process. 13 is a second flowchart of the detection process. FIG. 4 is a diagram showing an example of a first type signal. FIG. 4 is a diagram showing a second type signal generated by performing a first difference process. 7 is a diagram showing positions of transition points detected in the second type signal of FIG. 6; FIG. 11 is a diagram showing an example of change point information displayed on a display unit. FIG. 4 is a diagram showing a first type signal. 4 is a diagram showing a second kind of signal generated by performing a first difference processing on a first kind of signal; FIG. 11 is a diagram showing the positions of transition points detected in the second type signal of FIG. 10 . FIG. 4 is a diagram showing a first type signal. 4 is a diagram showing a second kind of signal generated by performing a first difference processing on a first kind of signal; FIG. FIG. 14 is a diagram showing positions of transition points detected in the second type signal of FIG. 13 . FIG. 13 is a diagram showing a result of executing a second difference process on a section having a slope. FIG. 4 is a diagram showing a first type signal. FIG. 4 is a diagram showing a noise signal added to a first type signal. FIG. 4 is a diagram showing a second type signal generated by adding a noise signal to a first type signal.

A. Embodiments:
FIG. 1 is a diagram for explaining a signal processing device 10 of this embodiment. FIG. 2 is a diagram showing an example of a first kind of signal FS. In FIG. 2, the horizontal X-axis represents an element corresponding to time, and the vertical Y-axis represents a signal value Va. Note that the X-axis is not limited to time as long as it is an index representing time. For example, the X-axis may be a data number that identifies each detection value when a detection value is obtained from a sensor at a predetermined time interval. In this embodiment, the number of data items in the first kind of signal FS is 200.

As shown in FIG. 1, the signal processing device 10 is configured by an electronic calculator such as a personal computer. The signal processing device 10 includes a processor 20, a storage device 30, a display unit 40, an input unit 50, and a communication unit 60.

The storage device 30 is composed of a ROM, a RAM, a HDD, and the like. Various programs for controlling the operation of the signal processing device 10 are stored in the storage device 30. The storage device 30 also stores the first kind of signal FS input to the signal processing device 10 via the input unit 50 and the communication unit 60. As shown in FIG. 2, the first kind of signal FS is time series data having a change point CP. The change point CP is a time point at which the degree of change in at least one of the mean value and the variance change occurs in the time series signal by a predetermined value or more. The predetermined value may be the same for the mean value change and the variance change, or may be different. In FIG. 2, the mean value changes by a predetermined value or more before and after the change point CP. The first kind of signal FS may or may not have normality. The first kind of signal FS is, for example, a detection signal output from a sensor that detects a physical quantity, such as an external voltage sensor or current sensor.

As shown in FIG. 1, the processor 20 functions as a conversion unit 22 and a detection unit 24 by executing various programs stored in the storage device 30. The processor 20 is composed of one or more processors. At least a part of the functions executed by the processor 20 may be realized by a hardware circuit.

The conversion unit 22 performs predetermined signal processing on the first type signal FS stored in the storage device 30, thereby converting the first type signal FS into a second type signal SS while maintaining information about the change point CP. Specifically, the conversion unit 22 converts the first type signal FS into the second type signal SS by performing at least one of signal processing including a first difference processing, a second difference processing, and an overdetection suppression processing.

In the first difference processing, the converter 22 generates a first difference signal, which is a set of differences between signal values at two first time points as two time points in the first kind signal FS, as the second kind signal SS. The two first time points are separated by Δt1 in the time series. In other words, the first difference processing is a process of differentiating the first kind signal FS arranged in a time series, and when the signal value Va at time t in the first kind signal FS is x(t), the converter 22 generates the second kind signal SS by the following formula (1). The generated second kind signal SS is stored in the storage device 30. In this embodiment, Δt1 corresponds to a time interval of 1/10 of the total number of data, which is the number of plots of the first kind signal FS.
Signal value of second type signal SS=x(t+Δt1)−x(t) (1)

In the second differential process, when a specific condition is satisfied, the converter 22 generates a second differential signal, which is a set of differences between the signal values Va at the second time points as two time points, for the first kind signal FS as the second kind signal SS. The specific condition will be described in detail later. The two second time points are separated by Δt2, which is a time interval smaller than the time interval Δt1 between the two first time points used in the first differential process. In this embodiment, Δt2 is half of Δt1. In other words, the second differential process is a process of differentiating the first kind signal FS arranged in a time series, and when the signal value Va for time t in the first kind signal FS is x(t), the converter 22 generates the second kind signal SS by the following formula (2). The generated second kind signal SS is stored in the storage device 30.
Signal value of second type signal SS=x(t+Δt2)−x(t) (2)

In the overdetection suppression process, when the signal value Va of the first type signal FS has a discrete distribution with a finite number of values, such as two or three, the conversion unit 22 generates a second type signal SS by adding a noise signal NS generated by normal random numbers to the first type signal FS. The generated second type signal SS is stored in the storage device 30. The noise signal NS may be generated by the conversion unit 22 according to the first type signal FS to which the noise signal NS is to be added, or may be stored in advance in the storage device 30.

The detection unit 24 executes a change point detection process to detect a change point CP using at least one of the first type signal FS and the second type signal SS. The detection unit 24 detects the change point CP by applying an algorithm for detecting the change point CP to at least one of the first type signal FS and the second type signal SS. Examples of algorithms for detecting the change point CP include PELT, Segment Neighborhood, and The optimal partitioning method. In this embodiment, the detection unit 24 uses PELT as the algorithm.

The display unit 40 is, for example, a liquid crystal monitor, and displays change point information IM related to the change point CP. The change point information IM is, for example, an image obtained by combining position information indicating the position of the change point CP in the first type signal FS with an image that indicates the position of the change point CP.

The input unit 50 is an input interface such as a keyboard or mouse, and accepts input information from the user. The communication unit 60 is a communication interface that can send and receive data to and from external devices via wired or wireless connections.

3 is a first flowchart of the detection process for detecting the change point CP executed by the signal processing device 10. FIG. 4 is a second flowchart of the detection process. FIG. 5 is a diagram showing an example of the first type signal FS. FIG. 6 is a diagram showing the second type signal SS generated by performing the first difference process on the first type signal FS of FIG. 5. FIG. 7 is a diagram showing the position of the change point CP detected in the second type signal SSa of FIG. 6. The first type signal FS shown in FIG. 5 has an actual change point CP at time 100 on the X-axis, and the average values AV1 and AV2 of the signal value Va change by a predetermined value or more before and after the change point CP. That is, in the first type signal FS shown in FIG. 5, the change point CP is detected at time 100 by PELT. The first type signal FS shown in FIG. 5 is also called the first type signal FSa, and the second type signal SS shown in FIG. 6 and FIG. 7 is also called the second type signal SSa. In the following description, when the first type signal FS and the second type signal SS are used to distinguish between them, different symbols are added to the end of the names. Note that the vertical axis in FIG. 6 indicates the degree of inclination, which is the value IVa after the first difference process of the first type signal FS.

As shown in FIG. 3, in step S10, the conversion unit 22 performs a first differential process, which is a signal process, on the first type signal FSa to generate a first differential signal SSa as the second type signal SSa shown in FIG. 6.

As shown in FIG. 3, in step S12, the detection unit 24 executes a change point detection process on the generated first difference signal SSa. The detection unit 24 executes a change point detection process on the first difference signal SSa shown in FIG. 6 to detect two change points CP1 and CP2. The positions of the two change points CP1 and CP2 are time 80, which is Δt1 before time 100 on the X axis, and time 100.

As shown in FIG. 3, in step S14 following step S12, the detection unit 24 performs a division process on the first differential signal SSa, which is the second type signal SSa. In the division process, as shown in FIG. 7, the detection unit 24 divides the first differential signal SSa at the positions of the change points CP1 and CP2 to divide the first differential signal SSa into a plurality of intervals It. The first differential signal SSa shown in FIG. 7 is divided into three intervals Ita1, Ita2, and Ita3.

As shown in FIG. 3, in step 16, the detection unit 24 detects whether or not there is a slope of the second type signal SSa in each of the multiple sections Ita1, Ita2, and Ita3. In this embodiment, the detection unit 24 calculates the average value AV of the second type signal SSa in each of the multiple sections Ita1, Ita2, and Ita3, and sets the calculated average value as the slope of each section Ita1, Ita2, and Ita3 of the second type signal SSa. In the second type signal SSa shown in FIG. 7, the average value AV as the slope is value AVa1 in section Ita1, value AVa2 in section Ita2, and value AVa3 in section Ita3. Values AVa1 and AV4 are "0", and value AVa2 is "2.2".

In the slope detection process of step S16, the detection unit 24 detects whether or not there is a slope in each of the sections Ita1 to Ita3 based on the average value AV. Specifically, the detection unit 24 detects that there is no slope when the average value AV is within the reference range, and detects that there is a slope when the average value AV is outside the reference range. In this embodiment, the reference range is "0". That is, the section Ita1 and the section Ita3, in which the average value is "0", are detected as having no slope, and the section Ita2, in which the average value is "2.2", is detected as having a slope. Here, as shown in FIG. 7, when the detection unit 24 satisfies a specific condition that the interval, which is the width on the X-axis of the section Ita2 in which there is a slope, is equal to or less than the interval Δt1 between the two first time points used to generate the first differential signal SSa, the detection unit 24 considers the section Ita2 that satisfies the specific condition to be an area without a slope. In this embodiment, the interval of section Ita2 is smaller than the interval Δt1, so section Ita2 is considered to be a section with no slope.

Next, as shown in FIG. 3, in step S22, the detection unit 24 detects a change point CP for a first type of section that has no slope among the multiple sections It using the first type signal FSa, which is the original signal. That is, the detection unit 24 detects a change point CP for the first type of section using PELT for the first type signal FSa. In this way, if the interval between sections Ita2 that have a slope is less than or equal to the interval between two first time points, it is highly likely that the section has no slope in the first type signal FSa. Therefore, by regarding sections that satisfy a specific condition as sections that have no slope and applying PELT to the first type signal FS, which is the original signal, the change point CP can be detected more accurately.

On the other hand, in step S18, the conversion unit 22 performs a second difference process on the second type of intervals that satisfy the specific condition of having a slope among the multiple intervals It to generate a second type signal SSa, which is a second difference signal. Then, in step S20, the detection unit 24 performs a change point detection process using the second type signal SSa, which is a second difference signal. That is, the detection unit 24 detects a change point CP for the second type of interval using PELT for the second type signal SS. Note that, when the detection unit 24 detects a change point CP of the second type signal SS, which is a difference signal, it may perform a correction process for the position of the change point CP. In the correction process, a correction value equivalent to Δt2 used in the second difference process is added to the position of the detected change point CP, and the position after the addition is regarded as the position of the change point CP. In this embodiment, the correction value is 10 points, which is the number of data present during Δt2. That is, if time 90 is detected as the change point CP in step S20, a correction value of 10 is added and time 100 is regarded as the change point CP. This correction process allows the position of the change point CP to be detected with greater accuracy. Note that in the second type signal SSa shown in Figures 5 and 6, there is no second type section, so only the process of step S22 is executed.

As shown in FIG. 4, in step S24, the detection unit 24 determines whether the detection result of the change points CP is an overdetection. For example, the detection unit 24 determines that the detection result of the change points CP detected by steps S18 to S22 shown in FIG. 3 is an overdetection when the number of change points CP detected by steps S18 to S22 is equal to or greater than a predetermined ratio to the number of data of the first type signal FSa. The predetermined ratio is set to a value that allows the first type signal FSa to be determined to have a discrete distribution, and is, for example, any value between 40% and 100%.

If it is determined in step S24 that there is no overdetection, the detection unit 24 outputs in step S30 the results of the change points CP detected in steps S18 to S22 to the display unit 40 as change point information IM.

Figure 8 is a diagram showing an example of change point information IM displayed on the display unit 40. The change point information IM has an image IM1 in which first position information IMa indicating the position of the change point CP is combined with the first type signal FSa, which is time series data, and second position information IM2 indicating the position of the change point CP. The first position information IMa includes, for example, a vertical line to indicate the position of the change point CP and a numerical value on the X-axis indicating the position of the change point CP. The second position information IM2 includes a numerical value on the X-axis indicating the position of the change point CP.

On the other hand, as shown in FIG. 4, if it is determined in step S24 that there is an overdetection, that is, if it is determined that the first type signal FS has a discrete distribution, the conversion unit 22 executes an overdetection suppression process in step S26. A specific example of the overdetection suppression process will be described in detail later. Next, in step S28, the detection unit 24 executes a change point detection process on the second type signal SS generated by the overdetection suppression process. The detection result of step S28 is displayed on the display unit 40 as change point information IM.

Next, the steps of the detection process shown in Figures 3 and 4 will be explained using first type signals FSb to FSd that are different types from the first type signal FSa shown in Figure 5.

Figure 9 is a diagram showing the first type signal FSb. Figure 10 is a diagram showing the second type signal SSb generated by performing the first difference processing on the first type signal FSb in Figure 9. Figure 11 is a diagram showing the position of the change point CP detected for the second type signal SSa in Figure 10. The first type signal FSb shown in Figure 9 has a change point CP at time 100 on the time axis, which is the X-axis, and the variance of the signal value Va changes by a predetermined value or more before and after the change point CP. In other words, the change point CP of the first type signal FSb shown in Figure 9 is detected at time 100 by PELT. In this embodiment, the number of data points of the first type signal FSb is 200.

The signal processing device 10 executes the detection process shown in Figs. 3 and 4 on the first type signal FSb, similar to the first type signal FSa shown in Fig. 5. By executing step S10, a first differential signal SSb is generated as the second type signal SSb shown in Fig. 10. The detection unit 24 executes the change point detection process of step S12 on the first differential signal SSb to detect a change point CP as shown in Fig. 11. Due to the additivity of variance, the position of the detected change point CP of the first differential signal SSb is a point in time Δt1 earlier than the position of the change point CP of the first type signal FSb, which is the original signal.

Next, the detection unit 24 divides the first differential signal SSb in FIG. 11 into a plurality of sections Itb1, Itb2 by performing a division process on the first differential signal SSb at the change point CP. In step S16, the detection unit 24 detects whether or not there is a slope in the second type signal SSb in each of the plurality of sections Itb1, Itb2. In the second type signal SSb shown in FIG. 11, the average values AVb1, AVb2 are "0" in both sections Itb1, Itb2, so it is determined that there is no slope.

Therefore, in step S22 of FIG. 3, the detection unit 24 executes a change point detection process on the first type signal FSb shown in FIG. 9. As a result, a change point CP is detected at time point 100 shown in FIG. 9. Note that, since it is determined that the detection result of the first type signal FSb is not an overdetection, step S30 is executed to output the detection result detected in step S22.

Figure 12 is a diagram showing the first type signal FSc. Figure 13 is a diagram showing the second type signal SSc generated by performing the first difference processing on the first type signal FSc of Figure 12. Figure 14 is a diagram showing the position of the change point CP detected in the second type signal SSc of Figure 13. Figure 15 is a diagram showing the second difference processing performed on the section with a slope in the first type signal FSc of Figure 12. The number of data of the first type signal FSc is 200. When the change point CP is detected by PELT in the first type signal FSc shown in Figure 12, multiple change points CP1, CP2, CP3, CP4, and CP5 are detected on the time axis, which is the X-axis. The average values AV1 to AV6 change by more than a predetermined value before and after each of the change points CP1 to CP5. On the other hand, in the first type signal FSc, the average value AVc1 of the signal value Va is "0" before the change point CP1, and the signal value Va starts to increase after the change point CP1. Therefore, the change point CP that the user actually wants to detect is only the change point CP1.

The signal processing device 10 executes the detection process shown in FIG. 3 and FIG. 4 on the first type signal FSc, similar to the first type signal FSa shown in FIG. 5. By executing step S10, a first differential signal SSc is generated as the second type signal SSc shown in FIG. 13. The detection unit 24 executes a change point detection process on the first differential signal SSc to detect a change point CP as shown in FIG. 14. Before and after the change point CP shown in FIG. 14, the average values AVc1 and AVc2 change by a predetermined value or more. The change point CP shown in FIG. 14 is also called a change point CPc. The position of the change point CPc is a time point 90, which is Δt1 before the time point 110 of the change point CP1 of the first type signal FSc, which is the original signal.

Next, in step S14, the detection unit 24 divides the first differential signal SSc in FIG. 14 into a plurality of intervals Itc1 and Itc2 by performing an interval division process that divides the first differential signal SSc at the change point CPc. In step S16, the detection unit 24 determines that there is no slope in the interval Itc1 because the average value AVc1 is "0", and determines that there is a slope in the interval Itc2 because the average value AVc2 is "2" and the interval of the interval Itc2 is greater than Δt1.

In the section from time 0 to time 90 of the first type section Itc1, which is a section with no slope, a change point CP is detected using the first type signal FSc, which is the original signal, in step S22. In other words, the detection unit 24 detects a change point in the first type section Itc1 using PELT. No change point CP is detected in the section from time 0 to time 90 of the first type signal FSc in FIG. 12.

In the section from time 90 to time 200, which is the second type section Itc2 having a slope, in step S18, the conversion unit 22 performs a second difference process on the first type signal FSc to generate the second type signal SSca, which is the second difference signal shown in FIG. 15. Then, in step S20, the detection unit 24 detects the change point CP using the second type signal SSca. The change point CP detected in FIG. 15 is also called the change point CPca. The detection unit 24 considers the time 110 to be the position of the change point CPca by adding "10" corresponding to Δt2 by correction processing to the time 100, which is the position of the change point CPca. Note that in the first type signal FSc, since it is determined that the detection result is not an overdetection, step S30 is executed to output the detection result detected in step S20 or step S22. In other words, the display unit 40, in response to a command from the processor 20, displays on the display unit 40 the change point information IM including the time point 110, which is the position of the change point CPca after the correction process.

Figure 16 is a diagram showing the first type signal FSd. Figure 17 is a diagram showing the noise signal NS added to the first type signal FSd. Figure 18 is a diagram showing the second type signal SSd generated by adding the noise signal NS to the first type signal FSd. The number of data of the first type signal FSd is 300. The first type signal FSd shown in Figure 16 has a discrete distribution. In addition, the average of the first type signal FSd changes by a predetermined value or more before and after time point 100, and the variance changes by a predetermined value or more before and after time point 200. When a change point CP is detected using PELT for the first type signal FSd, each of the discrete change points is detected as a plurality of change points CP. Note that Figure 16 shows only a portion of the detected change points CP. On the other hand, the change points CP that the user wants to detect are only the change points CPd1 and CPd2 located at time points 100 and 200 where the variance changes.

When the detection process shown in Figures 3 and 4 is performed on the first type signal FSd, it is determined in step S16 that there is "no slope" in all sections divided in step S14. Therefore, in step S22, the detection unit 24 performs a change point detection process using the first type signal FSd. In this case, a change point CP that is equal to or greater than a predetermined ratio with respect to the number of data points of the first type signal FSd is detected, and therefore a "Yes" determination is made in step S24 of Figure 4. As a result, the conversion unit 22 performs an overdetection suppression process in step S26.

In the overdetection suppression process, the conversion unit 22 generates a noise signal NS according to the first type signal FSd to which the noise signal NS is to be added. The conversion unit 22 generates a noise signal NS having a variance of, for example, 30% to 120% of the variance of the entire first type signal FSd. In this embodiment, the conversion unit 22 generates a noise signal NS having the same variance as the variance of the entire first type signal FSd. Next, the conversion unit 22 adds the generated noise signal NS to the first type signal FSd to generate the second type signal SSd shown in FIG. 18.

Next, in step S28, the detection unit 24 executes a change point detection process on the second type signal SSd generated by the overdetection suppression process. In the change point detection process of step S28, change points CPd1 and CPd2 are detected, but the other change points CP shown in FIG. 16 are not detected. The detection result of step S28 is displayed on the display unit 40 as change point information IM.

According to the above embodiment, the change point CP is detected using at least one of the first type signals FS to FSd and the second type signals SS to SSd, which are the difference signals of the first type signals FS to FSd or the signals obtained by adding a noise signal NS to the first type signals FS to FSd. This makes it possible to detect the change point CP desired by the user with higher accuracy than when the change point CP is detected using the first type signals FS to FSd, which are the original signals. For example, even if the first type signals FS to FSd are not normal signals, the change point CP can be detected with higher accuracy by converting the first type signals FS to FSd into the second type signals SS to SSd and detecting the change point CP of the second type signals SS to SSd. Specifically, as shown in FIG. 12, in the first type signal FSc in which the tendency of the change in the signal value Va, i.e., the tendency of the slope, changes at the change point CP1, the change point CP is detected using the second type signal SSca shown in FIG. 15 for the section Itc2 including the time point 100 at which the tendency of the slope changes. This allows the change point CPc at which the tendency of the slope changes to be detected with high accuracy. Also, as shown in FIG. 16, for the first type signal FSd, which is a discrete signal distribution, a noise signal NS is added to generate the second type signal SSd, and the change point CP is detected using the second type signal SSd. This allows the change point CP to be detected with high accuracy even for the first type signal FSd having a discrete distribution.

In addition, according to the above embodiment, as shown in FIG. 3, in the second type signals SS, SSa, SSb, SSc, and SSd serving as the first differential signals, the change point CP is detected in the first type section without a slope using the first type signals FS to FSd, which are the original signals. In general, detection accuracy can be improved by detecting the change point CP using the first type signals FS to FSd, which are the original signals, rather than the second type signals SS to SSd. Therefore, when the change point CP can be detected with high accuracy using the first type signals FS to FSd, that is, when detecting the point in time when the average or variance is changing, the detection accuracy of the change point CP can be improved by using the first type signals FS to FSd.

According to the above embodiment, in step S18 of FIG. 3, the conversion unit 22 generates a second difference signal that is a set of differences between two second time points that are closer to each other than the two first time points for the second type section Itc2 with a slope. The detection unit 24 then detects the change point CPca by PELT using the second difference signal as the second type signal SSca. Therefore, even if a change point CP cannot be detected using the first difference signal, it can be detected using the second difference signal that is a set of differences between two second time points that are closer to each other, and the change point CP can be detected with greater accuracy.

As described above, the signal processing device 10 of the above embodiment can detect, as change points CP, not only the points at which the mean value or variance changes for the first type signals FS to FSd, but also the points at which the slope changes and the points at which the mean value or variance changes among the signal values Va having a discrete distribution. In this way, the detection of change points CP can be performed by a single detection process shown in Figures 3 and 4, which can improve the maintainability of the signal processing device 10 in which the detection process is incorporated as a program or circuit. In other words, the maintainability of the signal processing device 10 can be improved compared to when different detection process programs or circuits are incorporated into the signal processing device 10 depending on the type of the first type signals FS to FSd.

Furthermore, by using the manufacturing data related to the manufacture of the product as first type signals FS to FSd and the detection unit 24 detecting change points CP, it is possible to estimate changes in elements that affect the quality of the product based on the change points CP. Here, the manufacturing data is data that is assigned a time point such as date and time, and is data that represents the manufacturing history such as measurement data of the product, the composition ratio of the raw materials of the product, the workers, the date and time of work, and the equipment used in manufacturing. Examples of elements that affect the product include the 4Ms (Man, Machine, Method, Material).

B. Other embodiments:
B-1. Other embodiment 1:
In the above embodiment, as shown in Fig. 3, steps S14, S16, S18, and S22 are executed, but these steps may be omitted. That is, the detection unit 24 may detect the change point CP by executing a change point detection process on the first difference signal regardless of the presence or absence of a slope. In this case, the detection unit 24 may perform a correction process for the position of the change point CP. The correction process is a process in which a correction value equivalent to Δt1 used in the first difference process is added to the position of the detected change point CP, and the position after the addition is regarded as the position of the change point CP.

C. Other Forms:
The present disclosure is not limited to the above-mentioned embodiment, and can be realized in various configurations without departing from the spirit of the present disclosure. For example, the technical features of the embodiments corresponding to the technical features in each form described below can be appropriately replaced or combined in order to solve some or all of the above-mentioned problems or to achieve some or all of the above-mentioned effects. Furthermore, if a technical feature is not described as essential in this specification, it can be appropriately deleted.

(1) According to a first embodiment of the present disclosure, a signal processing method is provided. This signal processing method includes converting a first type signal, which is time-series data having a change point, into a second type signal that maintains information about the change point by performing signal processing on the first type signal, and detecting the change point using at least one of the first type signal and the second type signal. The conversion into the second type signal includes at least one of generating a first difference signal that is a set of differences in signal values at two first time points in the first type signal, and adding a noise signal generated by normal random numbers to the first type signal when the first type signal has a discrete distribution. According to this embodiment, the second type signal is at least one of a first difference signal that is a difference signal of the first type signal, and a signal obtained by adding a noise signal to the first type signal. As a result, even if the first type signal is a signal that does not have normality, the change point can be detected with high accuracy by converting the first type signal into a second type signal and detecting the change point of the second type signal.

(2) In the above embodiment, detecting the change point may include, when the first differential signal is generated, detecting the change point of the second type signal using the first differential signal as the second type signal, dividing the second type signal into a plurality of sections by dividing the second type signal at the positions of the change points, detecting the presence or absence of a slope of the second type signal in each of the plurality of sections, and detecting the change point using the first type signal for the first type section without the slope among the plurality of sections. According to this embodiment, for the first type section without the slope of the second type signal, the change point can be detected more accurately by detecting the change point using the first type signal, which is the original signal.

(3) In the above embodiment, converting to the second type signal further includes generating, as the second type signal, a second difference signal that is a set of differences between the signal values of the first type signal at two second time points as the two time points for the second type section having the inclination among the multiple sections, and detecting the change point further includes detecting the change point using the second difference signal when the second difference signal is generated as the second type signal, and the interval between the two second time points may be smaller than the interval between the two first time points. According to this embodiment, the change point can be detected more accurately by detecting the change point using the second difference signal having a smaller difference between the intervals.

(4) In the above embodiment, detecting the presence or absence of a slope of the second type signal may include regarding the section that satisfies a specific condition that the interval between the sections with the slope is equal to or less than the interval between the two first time points as a section without the slope.

(5) According to a second embodiment of the present disclosure, a signal processing device is provided. The signal processing device includes a conversion unit that converts a first type signal, which is time-series data having a change point, into a second type signal that maintains information about the change point by performing signal processing on the first type signal, and a detection unit that detects the change point using at least one of the first type signal and the second type signal. The conversion unit executes at least one of generating a first difference signal that is a set of differences in signal values at two first time points in the first type signal, and adding a noise signal generated by normal random numbers to the first type signal when the first type signal has a discrete distribution. According to this embodiment, the second type signal is at least one of a first difference signal that is a difference signal of the first type signal, and a signal obtained by adding a noise signal to the first type signal. As a result, even if the first type signal is a signal that does not have normality, the change point can be detected with high accuracy by converting the first type signal into a second type signal and detecting the change point of the second type signal.

(6) According to a third aspect of the present disclosure, a computer program is executed to be executed by a computer. This computer program includes a conversion function for converting a first type signal, which is time-series data having a change point, into a second type signal that maintains information about the change point by performing signal processing on the first type signal, and a detection function for detecting the change point using at least one of the first type signal and the second type signal, and the conversion function includes at least one of a function for generating a first difference signal, which is a set of differences in signal values at two first time points as two time points in the first type signal, and a function for adding a noise signal generated by normal random numbers to the first type signal when the first type signal has a discrete distribution. According to this aspect, the second type signal is at least one of a first difference signal, which is a difference signal of the first type signal, and a signal obtained by adding a noise signal to the first type signal. As a result, even if the first type signal is a signal that does not have normality, the change point can be accurately detected by converting the first type signal into a second type signal and detecting the change point of the second type signal.

In addition to the above embodiments, the present disclosure can be realized in the form of a non-transitory recording medium for recording a computer program.

AV1, AVb1, AVc1, AVc2...average value, CP, CP1, CPc, CPca, CPd1...change point, FS, FSa, FSb, FSc, FSd...first type signal, IM...change point information, It, Ita1, Ita2, Ita3, Itb1, Itc1, Itc2...interval, NS...noise signal, SS, SSa, SSb, SSc, SSd...second type signal, Va...signal value, 10...signal processing device, 20...processor, 22...conversion unit, 24...detection unit, 30...storage device, 40...display unit, 50...input unit, 60...communication unit, IM...change point information, IM1...image, IMa...first position information, IM2...second position information, Ita2...interval, Ita3...interval, Itb1...interval

Claims (5)

1. A signal processing method, comprising:
performing signal processing on a first type signal, which is time-series data having a change point, to convert the first type signal into a second type signal that maintains information about the change point;
detecting the change point using at least one of the first type signal and the second type signal;
The converting into the second type signal comprises:
generating a first difference signal which is a set of differences between signal values at two first time points in the first type signal , the first time points being spaced apart by a certain time interval ;
adding a noise signal generated by Gaussian random numbers to the first kind of signal when the first kind of signal has a discrete distribution ;
Detecting the change point includes:
When the first differential signal is generated, the first differential signal is used as the second type signal, the change points of the second type signal are detected, and the second type signal is divided into a plurality of sections by dividing the second type signal at the positions of the change points.
detecting the presence or absence of a slope of the first type signal from an average value of the second type signal in each of the plurality of sections;
detecting the change point using the first type signal for a first type section having no slope among the plurality of sections;
A signal processing method comprising : 2. A signal processing method according to claim 1 , comprising:
The converting into the second type signal further includes:
generating, as the second type signal, a second difference signal which is a set of differences between the signal values of the first type signal at two second time points which are two time points separated by a constant time interval , for the second type section having the inclination among the plurality of sections;
Detecting the change point further comprises:
detecting the change point using the second differential signal when the second differential signal is generated as the second type signal;
A method of signal processing, wherein the interval between the two second time points is smaller than the interval between the two first time points. 3. A signal processing method according to claim 1 , further comprising:
Detecting the presence or absence of a slope of the first type signal from the average value of the second type signal
A signal processing method including, when a specific condition is satisfied that the interval between the sections having a slope is equal to or less than the interval between the two first points in time, regarding the sections satisfying the specific condition as sections having no slope. 1. A signal processing device, comprising:
a conversion unit that performs signal processing on a first kind of signal, which is time-series data having a change point, to convert the first kind of signal into a second kind of signal that maintains information about the change point;
a detection unit that detects the change point using at least one of the first type signal and the second type signal,
The conversion unit is
generating a first difference signal which is a set of differences between signal values at two first time points in the first type signal , the first time points being spaced apart by a certain time interval ;
and adding a noise signal generated by Gaussian random numbers to the first kind of signal when the first kind of signal has a discrete distribution .
Detecting the change point includes:
When the first differential signal is generated, the first differential signal is used as the second type signal, the change points of the second type signal are detected, and the second type signal is divided into a plurality of sections by dividing the second type signal at the positions of the change points;
detecting the presence or absence of a slope of the first type signal from an average value of the second type signal in each of the plurality of sections;
detecting the change point using the first type signal for a first type section having no slope among the plurality of sections;
A signal processing device comprising : A computer program to be executed by a computer,
a conversion function of converting a first kind of signal, which is time-series data having a change point, into a second kind of signal that maintains information about the change point by performing signal processing on the first kind of signal;
a detection function for detecting the change point using at least one of the first type signal and the second type signal,
The conversion function is a function of generating a first difference signal which is a set of differences between signal values at two first time points in the first type signal , the first time points being separated by a certain time interval ;
a function of adding a noise signal generated by a normal random number to the first kind of signal when the first kind of signal has a discrete distribution ;
The function of detecting the change point is
a function of detecting the change points of the second type signal using the first difference signal as the second type signal when the first difference signal is generated, and dividing the second type signal into a plurality of sections by dividing the second type signal at the positions of the change points;
detecting the presence or absence of a slope of the first type signal from an average value of the second type signal in each of the plurality of sections;
a function of detecting the change point using the first type signal for a first type section having no slope among the plurality of sections;
A computer program comprising :

Images