Data correction device, data correction method, and data correction program

Abstract

Perform appropriate data correction for data shifts. [Solution] The data correction device according to the present disclosure is a data correction device that performs correction on new data in which a data shift has occurred, and is characterized by comprising: a calculation unit that calculates correction parameters based on the correspondence between existing normal data and new normal data in a periodic section within the same process; and a correction unit that corrects the new data using the calculated correction parameters.

Description

Technical Field

[0001] The present invention relates to a data correction device, a data correction method, and a data correction program.

Background Art

[0002] Conventionally, a technique has been proposed in which data correction is performed using correction parameters calculated for each time based on the correspondence relationship at each time of synchronized data that has been synchronized when a data shift occurs (for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the prior art, there are cases where appropriate data correction cannot be performed for data shifts. For example, in the prior art, since correction is performed using correction parameters calculated for each time, when the times of data before and after a data shift are not synchronized, or even when time synchronization is performed but there is a deviation in the synchronization, correction parameters different from those expected may be calculated and extreme correction may be performed. Therefore, a method of data correction based on a global correspondence relationship is desired.

Means for Solving the Problems

[0005] To solve the above-mentioned problems and achieve the objective, the data correction device is a data correction device that performs correction on new data in which a data shift has occurred, and is characterized by comprising: a calculation unit that calculates correction parameters based on the correspondence between existing normal data and new normal data in a periodic section within the same process; and a correction unit that corrects the new data using the calculated correction parameters. [Effects of the Invention]

[0006] According to one aspect of the present invention, the effect is achieved that appropriate data correction can be performed in response to data shifts. [Brief explanation of the drawing]

[0007] [Figure 1] Figure 1 is a diagram illustrating an example of data shifting. [Figure 2] Figure 2 illustrates an example of a problem that exists in the conventional technology. [Figure 3] Figure 3 illustrates an example of a problem that exists in the conventional technology. [Figure 4] Figure 4 is a diagram illustrating the overall process performed by the data correction device according to this embodiment. [Figure 5] Figure 5 is a block diagram showing an example of the configuration of a data correction device according to this embodiment. [Figure 6] Figure 6 is a diagram illustrating an example of the calculation process performed by the calculation unit according to the embodiment. [Figure 7] Figure 7 is a diagram illustrating an example of the calculation process performed by the calculation unit according to this embodiment. [Figure 8] Figure 8 is a diagram illustrating an example of the correction process performed by the correction unit according to the embodiment. [Figure 9] Figure 9 is a flowchart illustrating an example of the processing flow performed by the data correction device according to the embodiment. [Figure 10] Figure 10 is a hardware configuration diagram showing an example of the configuration of a computer that implements a data correction device according to the embodiment. [Modes for carrying out the invention]

[0008] 〔background〕 In the field of process automation, early detection of anomalies is crucial for improving the safety of production equipment and the stability of product quality. One technology for detecting anomalies in production equipment from the early stages in batch processes involves generating a model based on time-series data from normal operation measured by various sensors installed on the equipment, and then independently detecting anomalies for each target by comparing this normal model with the time-series data measured for the monitored target.

[0009] In production facilities such as factories and plants, planned and large-scale inspections and repairs (hereinafter referred to as scheduled maintenance) are sometimes carried out with shutdown periods to maintain the reliability of the equipment. However, data shifts may occur in the measurement data of specific monitored targets before and after these scheduled maintenance periods. A data shift is a phenomenon in which measurement data such as current values ​​and voltage values ​​slide in the value direction due to the influence of changes in measurement characteristics or operating conditions, such as sensor calibration (adjustment of zero point and span).

[0010] Here, we will explain an example of data shift with reference to Figure 1. Figure 1 is a diagram illustrating an example of data shift. Figure 1 shows new data obtained after the data shift occurred and existing data obtained before the data shift occurred. The vertical axis represents the data value. The horizontal axis represents time. The solid line represents new data. The dotted line represents existing data. As shown in Figure 1, after the data shift occurs, the zero point of the values ​​slides in the positive direction and the span changes compared to before the shift.

[0011] The data shift shown in Figure 1 occurs because the "normal state" of the measurement data changes before and after scheduled maintenance, and does not necessarily indicate the occurrence of an anomaly. Because the normal state changes, if anomaly detection is performed using the normal model used before the scheduled maintenance and the new data measured after the maintenance, there is a risk of frequent false alarms.

[0012] Therefore, conventionally, the anomaly detection is temporarily stopped until new data after the regular maintenance is accumulated, and the operation of regenerating the normal model after the regular maintenance based on the accumulated new data and then resuming the anomaly detection is performed. However, in this operation, since the anomaly detection is stopped for a certain period, there is a risk of overlooking equipment anomalies including calibration and operation mistakes during the regular maintenance. Also, immediately after resuming the anomaly detection, the accuracy of the generated model cannot be ensured because the new data does not exist sufficiently as much as the existing data before the regular maintenance, and the anomaly detection accuracy may decrease compared to before the regular maintenance.

[0013] To avoid the above-mentioned problems, there is a technology that enhances new data by correcting a plurality of existing data to data that is assumed to be obtained under the same conditions as when measuring new data when the measurement characteristics and operating conditions are changed, and generates a new normal model from the enhanced data.

[0014] However, in the prior art, after performing time synchronization by the dynamic time stretching method between the data obtained by averaging the existing data and the new data, the differences at each time between the synchronized data are calculated as offset values respectively, and the existing data is corrected based on these. This method calculates separate correction values for each time based on a local correspondence relationship.

[0015] Here, referring to FIGS. 2 and 3, the problems existing in the prior art will be described. FIGS. 2 and 3 are diagrams for explaining an example of the problems existing in the prior art. FIG. 2 shows the new data and the existing data after time synchronization. FIGS. 2(1) and (2) show the locations where time synchronization is deviated. FIG. 3 shows the new data and the existing data after correction. FIGS. 3(1) and (2) show the locations where extreme correction is performed. The explanations of the vertical axis, horizontal axis, solid line, and dotted line in FIGS. 2 and 3 are the same as those in FIG. 1.

[0016] In the prior art, correction is performed using correction values calculated for each time. Therefore, as shown in FIGS. 2(1) and 2(2), when the time between existing data and new data is not synchronized or there is a deviation in time synchronization, inappropriate correction may be directly performed under the influence. For example, for data that alternates between two steady states, larger correction values than expected are calculated before and after the change point of the steady state with loose time synchronization shown in FIGS. 2(1) and 2(2).

[0017] As a result, as shown in FIGS. 3(1) and 3(2), extreme correction is performed. As a result, data that is not suitable to be treated as normal new data is used in the generation of the model, and there is a risk that abnormal detection will not function accurately.

[0018] Therefore, the data correction apparatus 100 corrects data based on the global correspondence between existing normal data and new normal data. Here, referring to FIG. 4, an overall view of the processing performed by the data correction apparatus 100 will be described. FIG. 4 is a diagram for explaining an overall view of the processing performed by the data correction apparatus 100 according to the embodiment.

[0019] The data correction apparatus 100 first calculates correction parameters based on the correspondence between existing normal data and new normal data in a periodic section within the same process (FIG. 4(1)). Note that the periodic section within the same process includes, for example, the target section of the batch process, and sections in which the same operation is performed in a continuous process operated in the same pattern such as cyclic operation.

[0020] For example, the data correction apparatus 100 calculates a regression coefficient obtained using a regression model created based on existing normal data and new normal data in a periodic section within the same process as a correction parameter. Then, the data correction apparatus 100 uses the calculated correction parameter to correct new data detected after a data shift has occurred (FIG. 4(2)). For example, the data correction apparatus 100 corrects new data using the regression coefficient, which is the calculated correction parameter.

[0021] In this way, the data correction device 100 corrects the new data based on correction parameters calculated from the global correspondence between existing normal data and new normal data, thereby appropriately correcting data that has experienced a data shift and accurately reducing the impact of the data shift. Furthermore, because the data correction device 100 appropriately corrects the data after a data shift has occurred, it is possible to continue using the existing normal model, thereby suppressing the oversight of anomalies due to the suspension of anomaly detection and the decrease in detection accuracy due to insufficient data.

[0022] [Configuration of the data correction device] Next, with reference to Figure 5, an example of the functional configuration of the data correction device 100 according to the embodiment will be described. Figure 5 is a block diagram showing an example of the functional configuration of the data correction device 100 according to the embodiment. The data correction device 100 shown in Figure 5 has a communication unit 110, a storage unit 120, and a control unit 130. The data correction device 100 may also have an input unit (e.g., a keyboard or mouse) for receiving various operations from the administrator of the data correction device 100, and a display unit (e.g., a liquid crystal display) for displaying various information.

[0023] (Communications Department 110) The communication unit 110 is implemented, for example, by a communication device, a communication circuit, etc. The communication unit 110 is connected to any network such as the Internet by wired or wireless connection and transmits and receives information with an external information processing device. The communication unit 110 may also be connected to an external information processing device in a communicative manner by a predetermined wireless communication function such as Wi-Fi (registered trademark). For example, the communication unit 110 transmits and receives information with an external device. In this way, the communication unit 110 is a connection means to which the data correction device 100 is connected in a communicative manner via a predetermined network.

[0024] (Storage unit 120) The storage unit 120 is implemented by, for example, semiconductor memory elements such as RAM (Random Access Memory) and flash memory, or storage devices such as hard disks and optical discs. For example, the storage unit 120 is a computer-readable recording medium that non-temporarily records data used by information processing programs. As shown in Figure 5, the storage unit 120 has an existing normal data storage unit 121, a new normal data storage unit 122, a model storage unit 123, and a correction parameter storage unit 124. The individual parts of the storage unit 120 will be described below.

[0025] (Existing normal data storage unit 121) The existing normal data storage unit 121 stores existing normal data. The existing normal data storage unit 121 stores normal data measured by various sensors installed in the equipment before a data shift occurs. Existing normal data is, for example, existing data that has been determined to be normal data by an external system or a user. The data length of each existing data item may be constant or different.

[0026] (New normal data storage unit 122) The new normal data storage unit 122 stores new normal data. For example, the new normal data storage unit 122 stores normal data measured by various sensors installed in the equipment after a data shift has occurred. New normal data is, for example, new data that has been determined to be normal by an external system or a user. The data length of the new data may be the same as or different from the existing data.

[0027] (Model memory unit 123) The model storage unit 123 stores models. The models stored by the model storage unit 123 are linear models such as linear simple regression models and linear multiple regression models, or nonlinear regression models. The models stored by the model storage unit 123 are models created using known regression methods such as least squares method and iteratively reweighted least squares (IRLS).

[0028] The linear simple regression model stored in the model memory unit 123 is, for example, a model (Equation 3) created using the mean, median, or mode (Equation 1) of existing normal data as explanatory variables and the mean, median, or mode (Equation 2) of new normal data as the dependent variable. Here, the mean, median, and mode refer to the mean, median, and mode for each time point obtained by superimposing the data measured from the start to the end point of a periodic interval so that the time or process corresponds to each other.

[0029]

number

[0030]

number

[0031]

number

[0032] In other words, the regression model stored in the model storage unit 123 is a model that shows the correspondence between new normal data and existing normal data.

[0033] In the above description, an example was given in which the model storage unit 123 stores a linear simple regression model. However, the model storage unit 123 may also store a linear multiple regression model created using additional variables such as operating conditions, temperature, and humidity of the environment where various sensors are present, as explanatory variables.

[0034] (Correction parameter storage unit 124) The correction parameter storage unit 124 stores the correction parameters. For example, the correction parameter storage unit 124 stores the regression coefficients obtained from the correspondence between new normal data and existing normal data shown by the regression model as correction parameters.

[0035] (Control unit 130) The control unit 130 may be, for example, an information processing device (control device) that controls various configurations and processes. The control unit 130 may have, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit), and may be implemented by executing a program stored inside the control unit 130 using RAM or the like as a working area. Alternatively, the control unit 130 may have, for example, an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0036] As shown in Figure 5, the control unit 130 includes an acquisition unit 131, a synchronization unit 132, a calculation unit 133, and a correction unit 134, and realizes or executes the information processing functions and operations described below. Note that the internal configuration of the control unit 130 is not limited to the configuration shown in Figure 5, and other configurations are also acceptable as long as they perform the information processing described later.

[0037] (Acquisition part 131) The acquisition unit 131 acquires measurement data. For example, the acquisition unit 131 acquires data measured by various sensors installed on the equipment. In other words, the acquisition unit 131 acquires measurement data from various sensors that is subject to data correction.

[0038] (Class of 132) The synchronization unit 132 performs temporal synchronization between existing normal data and new normal data. For example, the synchronization unit 132 performs temporal synchronization between existing normal data and new normal data using Dynamic Time Warping (DTW).

[0039] For example, first, the synchronization unit 132 uses a dynamic time stretching method to determine a warping path (WP) between existing normal data and new normal data. Then, the synchronization unit 132 stretches the time series of either the existing normal data or the new normal data so that the points connected by the determined warping path are at the same time, thereby making the data lengths of the existing normal data and the new normal data equal.

[0040] (Calculation section 133) The calculation unit 133 calculates correction parameters based on the correspondence between existing normal data and new normal data in a periodic section within the same process. Here, a "periodic section within the same process" may be, for example, a target section of a batch process that groups multiple processes together, or a section in a continuous process that operates in the same pattern, such as periodic operation, where the same operation is performed.

[0041] The calculation unit 133 calculates correction parameters using a regression model. For example, the calculation unit 133 calculates correction parameters using a regression model created based on existing normal data and new normal data in a periodic interval within the same process.

[0042] Here, an example of the calculation process of the calculation unit 133 according to the embodiment will be described with reference to Figures 6 and 7. Figures 6 and 7 are diagrams illustrating an example of the calculation process of the calculation unit 133 according to the embodiment. Figure 6 shows new normal data and existing normal data. The vertical axis represents the data value. The horizontal axis represents time. The solid line represents new normal data. The dotted line represents existing normal data.

[0043] Figure 7 is a graph plotting the relationship between the new normal data and the existing normal data's mean, as shown in Figure 6, on a graph where the vertical axis represents the new normal data values ​​and the horizontal axis represents the existing normal data's mean. The vertical axis may represent values ​​calculated based on the new normal data, such as the mean, median, or mode. Similarly, the horizontal axis may represent values ​​calculated based on the existing normal data, such as the mean, median, or mode. The dotted line indicates the point cluster obtained when the new normal data values ​​coincide with the existing normal data's mean. ◎ indicates an outlier.

[0044] In other words, Figure 7 shows the correspondence between the values ​​of new normal data and the mean values ​​of existing normal data. For example, if the line obtained from the plotted point cloud is above the dotted line, it indicates that the values ​​of the new normal data are larger than the mean values ​​of the existing normal data. On the other hand, if the line obtained from the plotted point cloud is below the dotted line, it indicates that the values ​​of the new normal data are smaller than the mean values ​​of the existing normal data.

[0045] The calculation unit 133 calculates regression coefficients a and b using a linear regression model (Equation 3) that shows the correspondence between new normal data and existing normal data, for example, as shown in Figure 7. Specifically, the calculation unit 133 calculates the slope a and intercept b of a simple linear regression model (Equation 3) that shows the correspondence between new normal data and existing normal data as regression coefficients. In the example in Figure 7, an example was shown in which the calculation unit 133 calculates correction parameters using a linear regression model, but the regression model used by the calculation unit 133 may be a nonlinear regression model, not just a linear regression model.

[0046] As an alternative, the calculation unit 133 calculates correction parameters based on the correspondence between synchronized data in a periodic interval within the same process. For example, the calculation unit 133 calculates regression coefficients using a regression model that shows the correspondence between existing normal data and new normal data that have been temporally synchronized in a periodic interval within the same process.

[0047] (Correction section 134) The correction unit 134 corrects the new data using the correction parameters calculated by the calculation unit 133. Here, an example of the correction process of the correction unit 134 according to this embodiment will be explained with reference to Figure 8. Figure 8 is a diagram illustrating an example of the correction process of the correction unit 134 according to this embodiment. In Figure 8, the corrected new data and the existing normal data are shown in a graph with the data value on the vertical axis and time on the horizontal axis. The solid line represents the corrected new data. The dotted line represents the existing normal data.

[0048] For example, the correction unit 134 corrects the new data using the regression coefficients calculated by the calculation unit 133. To give a specific example, when the regression coefficients a and b calculated by the calculation unit 133 are calculated, the correction unit 134 performs a correction on the new data z using the following equation 4 to obtain the corrected new data z' (Figure 8).

[0049]

number

[0050] The correction unit 134 uses equation 4 to perform corrections, thereby suppressing extreme corrections like those shown in Figures 3(1) and 3(2), even if there are discrepancies between existing normal data and new normal data in periodic intervals within the same process, or if there are synchronization issues. As shown in Figure 8, it can perform corrections to bring the new data obtained after the data shift closer to the data obtained before the data shift occurred.

[0051] 〔flowchart〕 From here, with reference to Figure 9, the processing flow performed by the data correction device 100 in the embodiment will be described. Figure 9 is a flowchart showing an example of the processing flow performed by the data correction device 100 in the embodiment. Each step in the flowchart may be rearranged within a range that does not contradict each other, and there may be steps that are not performed.

[0052] First, the data correction device 100 acquires existing normal data (step S101). For example, the data correction device 100 acquires existing normal data from the existing normal data storage unit 121. Next, the data correction device 100 acquires new normal data (step S102). For example, the data correction device 100 acquires new normal data from the new normal data storage unit 122.

[0053] Next, the data correction device 100 performs temporal synchronization between the existing normal data and the new normal data (step S103). For example, the data correction device 100 uses a dynamic time stretching method to make the data lengths of the existing normal data and the new normal data equal. Next, the data correction device 100 calculates correction parameters (step S104). For example, the data correction device 100 calculates regression coefficients using a regression model that shows the correspondence between the existing normal data and the new normal data in a periodic interval within the same process.

[0054] Next, the data correction device 100 acquires new data (step S105). For example, the data correction device 100 acquires data measured after the data shift occurs from various sensors installed on the equipment. Then, the data correction device 100 corrects the acquired new data (step S106). For example, the data correction device 100 corrects the new data using the calculated regression coefficients. After that, the data correction device 100 terminates its processing.

[0055] 〔effect〕 The data correction device 100 according to this embodiment is a data correction device that performs correction on new data in which a data shift has occurred, and includes a calculation unit 133 that calculates correction parameters based on the correspondence between existing normal data and new normal data in a periodic section within the same process, and a correction unit 134 that corrects the new data using the calculated correction parameters.

[0056] As a result, the data correction device 100 according to the embodiment corrects the newly acquired data after the data shift based on correction parameters calculated from the global correspondence between existing normal data and new normal data, thereby performing appropriate data correction on data affected by the data shift and accurately reducing the impact of the data shift. Furthermore, because the data correction device 100 appropriately corrects the data after a data shift occurs, it becomes possible to continue using the existing normal model, thereby suppressing the oversight of anomalies due to the cessation of anomaly detection and the decrease in detection accuracy due to insufficient data.

[0057] The data correction device 100 according to this embodiment further includes a synchronization unit 132 that performs temporal synchronization between existing normal data and new normal data, and a calculation unit 133 calculates correction parameters based on the correspondence between synchronized data in an interval.

[0058] As a result, the data correction device 100 according to the embodiment performs temporal synchronization when the data lengths of existing normal data and new normal data are different, and calculates correction parameters based on the correspondence between the synchronized data, thereby enabling appropriate data correction for data shifts. In other words, the data correction device 100 can synchronize and correct the data lengths even when the data lengths are different before and after a data shift occurs.

[0059] The calculation unit 133 of the data correction device 100 according to the embodiment calculates correction parameters using a regression model created based on existing normal data and new normal data in a periodic interval within the same process. This enables the data correction device 100 of the embodiment to perform correction using correction parameters calculated based on the global correspondence between data before and after a data shift.

[0060] The calculation unit 133 of the data correction device 100 according to this embodiment calculates correction parameters based on the correspondence between existing normal data and new normal data in the target section of the batch process. As a result, the data correction device 100 of this embodiment can perform appropriate correction to mitigate the effects of data shifts in the target section of the batch process.

[0061] [Hardware configuration] The data correction device 100 according to this embodiment is implemented by a computer 1000 having the configuration shown in Figure 10. Figure 10 is a hardware configuration diagram showing an example of a computer that implements the functions of the data correction device 100 according to this embodiment. The computer 1000 has a configuration in which a CPU 1100, RAM 1200, ROM 1300, auxiliary storage device 1400, communication interface 1500, and input / output interface 1600 are connected by a bus 1800.

[0062] The CPU 1100 operates based on programs stored in the ROM 1300 or auxiliary storage device 1400, and controls various parts. The ROM 1300 stores boot programs executed by the CPU 1100 when the computer 1000 starts up, as well as programs that depend on the computer 1000's hardware.

[0063] The auxiliary storage device 1400 stores programs executed by the CPU 1100, and data used by such programs. The communication interface 1500 receives data from other devices via a predetermined communication network NW (including wireless communication via a closed network in this embodiment) and sends it to the CPU 1100, and transmits data generated by the CPU 1100 to other devices via the predetermined communication network NW. The CPU 1100 controls output devices such as displays and printers, and input / output devices 1700 such as keyboards and mice via the input / output interface 1600. The CPU 1100 acquires data from the input / output devices 1700 via the input / output interface 1600. The CPU 1100 also outputs the generated data to the input / output devices 1700 via the input / output interface 1600.

[0064] For example, when the computer 1000 functions as one of the various devices according to this embodiment, the CPU 1100 of the computer 1000 executes a program loaded onto the RAM 1200 to realize the functions of the control unit 130.

[0065] 〔others〕 Although various embodiments have been described in detail herein with reference to the drawings, these embodiments are illustrative and are not intended to limit the present invention to these embodiments. The features described herein can be realized in various ways, including various modifications and improvements based on the knowledge of those skilled in the art.

[0066] Furthermore, the aforementioned "module (-er suffix, -or suffix)" can be reinterpreted as unit, means, circuit, etc. For example, the communication module, control module, and storage module can be reinterpreted as communication unit, control unit, and storage unit, respectively. [Explanation of Symbols]

[0067] 100 Data Correction Device 110 Communications Department 120 Storage section 121 Existing normal data storage unit 122 New normal data storage unit 123 Model Memory Unit 124 Correction parameter storage unit 130 Control Unit 131 Acquisition Department 132 Classmates 133 Calculation Section 134 Correction section

Claims

1. A data correction device that corrects new data in which a data shift has occurred, A calculation unit that calculates correction parameters based on the correspondence between existing normal data and new normal data in a periodic interval within the same process, A correction unit that corrects the new data using the calculated correction parameters. A data correction device having the following features.

2. The system further includes a synchronization unit that performs temporal synchronization between the existing normal data and the new normal data. The calculation unit described above, The correction parameter is calculated based on the correspondence between synchronized data in the aforementioned interval. The data correction device according to feature 1.

3. The calculation unit described above, The correction parameters are calculated using a regression model created based on the existing normal data and the new normal data. The data correction device according to feature 1.

4. The calculation unit described above, The correction parameter is calculated based on the correspondence between existing normal data and new normal data within the target interval of the batch process. The data correction device according to feature 1.

5. A method by which a computer performs an action. A data correction method that performs correction on new data where a data shift has occurred, A calculation step that calculates correction parameters based on the correspondence between existing normal data and new normal data in a periodic interval within the same process, A correction step in which the new data is corrected using the calculated correction parameters, A data correction method characterized by including the following.

6. A data correction program that performs corrections on new data where data shifts have occurred, A calculation step of calculating correction parameters based on the correspondence between existing normal data and new normal data in a periodic interval within the same process, A correction step in which the new data is corrected using the calculated correction parameters, A data correction program characterized by having a computer execute it.

Images