Abnormality detection method

Abstract

To provide an abnormality detection method that can continue abnormality detection without interfering with principal component analysis even when signal data is deficient.SOLUTION: When signal data required for principal component analysis is deficient, the deficient signal data is estimated and complemented by using normal signal data adjacent to or related to the deficient signal data. When principal component analysis is performed and a principal component vector is calculated, the estimated and complemented signal data is used with intentional variation. When Q statistic is calculated from the principal component vector, the estimated and complemented signal data is used without intentional variation.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an anomaly detection method, and more particularly to an anomaly detection method for detecting an anomaly in equipment from the results of principal component analysis of multiple signal data of the same type in the equipment. [Background technology]

[0002] An example of such an anomaly detection method is described in the following Patent Document 1. This anomaly detection method uses output signals from a plurality of temperature sensors provided in the mold of the continuous casting equipment, and predicts a breakout of the continuous casting equipment from the results of principal component analysis of the signal data (temperature data). In this anomaly detection method, a principal component vector is calculated from a plurality of detected temperature data of the mold by principal component analysis, and when the Q statistics using this principal component vector exceeds a predetermined value, it is considered to be a sign of a breakout, and an anomaly of the equipment is detected. Since a certain amount of data (signal data) is required for the principal component analysis, a method of estimating and supplementing the signal data by interpolation or extrapolation is also described in the patent document 1 when the signal data is insufficient due to, for example, a failure (anomaly) of a temperature sensor or a shortage of the number of temperature sensors due to equipment replacement. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6950860 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the estimated and supplemented signal data may not be sufficiently accurate. In such a case, the calculation accuracy of the principal component vector by the principal component analysis may decrease, and the Q statistic may increase, resulting in a false diagnosis of an equipment abnormality, i.e., a breakout. If a breakout is falsely diagnosed, the operation of the continuous casting equipment must be stopped, and the production efficiency of cast steel decreases. For example, if the lack of signal data is due to a failure of the temperature sensor, the temperature sensor or the equipment (mold) including the temperature sensor may be replaced during the equipment downtime without stopping the operation of the continuous casting equipment, but there is a limit to improving the estimation accuracy of the supplemented signal data. In other words, an anomaly detection method that can continue anomaly detection without interfering with the principal component analysis even when the signal data is insufficient is desired.

[0005] The present invention has been made in consideration of the above-mentioned problems, and its object is to provide an anomaly detection method that can continue anomaly detection without interfering with principal component analysis even when signal data is insufficient. [Means for solving the problem]

[0006] In order to achieve the above-mentioned object, an anomaly detection method according to one embodiment of the present invention is an anomaly detection method for detecting an anomaly in equipment from the results of principal component analysis of multiple signal data of the same type in the equipment, wherein, when signal data required for the principal component analysis is insufficient, the missing signal data is estimated and complemented using normal signal data that is adjacent to or related to the missing signal data, and when performing the principal component analysis to calculate a principal component vector, the estimated and complemented signal data is used with intentional variation added, and when calculating a Q statistic, the estimated and complemented signal data is used without intentional variation.

[0007] Furthermore, a further aspect of the present invention is characterized in that the shortage of signal data is caused by any one of the following: a sensor failure, a shortage of the number of sensors due to equipment replacement, or a shortage of the number of signal data outputs from the sensor due to dimensional or positional fluctuations of the product. In a further aspect of the present invention, the signal data is an output signal of a temperature sensor inside a mold of continuous casting equipment, and the abnormality is a prediction of a breakout in the continuous casting equipment. Effect of the Invention

[0008] According to the anomaly detection method of the present invention, even if there is a shortage of signal data, this does not impede principal component analysis, so that anomaly detection can be continued without reducing the operation of the equipment. [Brief description of the drawings]

[0009] [Figure 1] 1 is a schematic configuration diagram showing an embodiment of a continuous casting facility to which an abnormality detection method of the present invention is applied; [Diagram 2] FIG. 2 is an explanatory view of a mold of the continuous casting machine of FIG. [Diagram 3] 3 is a graph showing an example of temperature data detected by the temperature sensor in FIG. 2. [Figure 4] 4 is a graph showing another example of the temperature data detected by the temperature sensor of FIG. 2. [Diagram 5] 5 is a graph of the Q statistic calculated using the temperature data of FIGS. 3 and 4. FIG. [Figure 6] 11 is a graph showing an example of estimated and interpolated temperature data used to calculate a principal component vector. [Figure 7] 7 is a graph of the Q statistic calculated using the temperature data of FIG. 6. [Figure 8] 3 is a graph showing an example of temperature data detected by the temperature sensor of FIG. 2. [Figure 9] 9 is a graph of the Q statistic calculated using the normal temperature data of FIG. 8. [Figure 10] 9 is a graph of the Q statistic calculated by intentionally shifting one of the temperature data in FIG. 8. [Figure 11] 2 is a graph showing a Q statistic calculated by intentionally shifting and intentionally introducing variation into one of the temperature data in FIG. 1. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, an embodiment of the anomaly detection method of the present invention will be described in detail with reference to the drawings. The embodiments shown below are examples of devices and methods for embodying the technical idea of ​​the present invention, and the technical idea of ​​the present invention does not specify the materials, shapes, structures, arrangements, etc. of the components to the embodiments described below. In addition, the drawings are schematic. Therefore, it should be noted that the relationship and ratio between thickness and planar dimensions are different from the actual ones, and the drawings include parts where the relationship and ratio of dimensions differ from each other.

[0011] Fig. 1 is a schematic diagram showing an embodiment of a continuous casting facility to which the anomaly detection method is applied, and is the same as that described in the above-mentioned Patent Document 1. This anomaly detection method is for detecting the occurrence of a breakout in a continuous casting facility. In a continuous casting machine 1 of this embodiment, molten steel 2 injected into a tundish 3 is poured into a mold 5 through a submerged nozzle 4, and a semi-solidified slab 6 pulled out of the mold 5 is supported and transported by a plurality of slab support rolls 7. A plurality (a large number) of temperature sensors 8 are embedded in the mold 5, and a judgment unit 20 detects a sign of a breakout based on the temperature of the mold 5 detected by these temperature sensors 8.

[0012] FIG. 2 shows an example of the arrangement of temperature sensors 8 embedded in the mold 5. The slab 6 is drawn in the pouring direction A, and a plurality (a large number) of temperature sensors 8 are disposed at equal intervals in the mold 5 at fixed positions in the pouring direction A, i.e., at height positions. These temperature sensors 8 are embedded at a predetermined depth from the outer wall surface of the mold 5, and are actually disposed in a plurality of rows, for example, three rows, in the height direction. For a precursor phenomenon of a breakout, please refer to the above-mentioned Patent Document 1. In short, while the temperature of the molten steel 2 or the slab 6 detected at the same height position is almost constant, an imbalance occurs in the temperatures detected at the same height position as a precursor of a breakout. Therefore, a principal component analysis is performed on the temperatures (temperature data) detected at the same height position, and the deviation from the principal component vector is calculated as a Q statistic. When this Q statistic becomes equal to or exceeds a predetermined threshold, it is detected as a precursor of a breakout. In addition, since the continuous casting machine 1 also produces cast pieces 6 with different specifications such as plate width, in order to continuously detect signs of breakout for such cast pieces 6 with different specifications, a principal component analysis is performed by interpolating the temperature data detected by the temperature sensor 8 so that the number of data in the principal component analysis is constant (the number of calculation cells is constant), and the Q statistics are calculated from the principal component vector.

[0013] When calculating the Q statistics from the results of the principal component analysis, for example, the temperature sensor 8 installed in the mold 5 may break down and abnormal temperature data may be detected. If such abnormal temperature data is involved in the principal component analysis, the principal component vector cannot be accurately obtained, and the Q statistics may become unnecessarily large, which may result in erroneous detection of an abnormality, i.e., a sign of a breakout. In addition, the temperature data may be insufficient due to a failure of the temperature sensor 8. As described above, the same number of temperature data is required for the principal component analysis, so in the above-mentioned Patent Document 1, the insufficient temperature data is estimated and supplemented from the temperature data of an adjacent temperature sensor 8 or a temperature sensor 8 that is considered to be highly related in advance, and the principal component analysis is continued. If a sign of a breakout is truly detected, it is necessary to reduce the operation of the continuous casting equipment, but if, for example, the temperature sensor 8 is only broken, there is a demand to estimate and supplement the temperature data of the broken temperature sensor 8 and continue the detection of a sign of a breakout. The broken temperature sensor 8 may be replaced, for example, when the equipment is down.

[0014] However, the estimated and complemented temperature data is not necessarily accurate enough to calculate the principal component vector with high accuracy by principal component analysis using the estimated and complemented temperature data, and as a result, the Q statistics may become large and a breakout sign may be erroneously detected. For example, FIG. 3 shows the signal data, i.e., temperature data, of a faulty temperature sensor 8, and FIG. 4 shows the temperature data of a non-faulty temperature sensor 8 (hereinafter, the adjacent temperature sensor 8) next to the faulty temperature sensor 8. As is clear from the figure, the temperature data of FIG. 3 shows a significant variation with respect to the temperature data of FIG. 4. Normally, the temperature data of the faulty temperature sensor 8 shows values ​​roughly similar to those of the temperature data of FIG. 4. In this example, the temperature data of the faulty temperature sensor 8 shown in FIG. 3 showed a significant variation from, for example, time t0, but the Q statistics were calculated from the principal component analysis results using the temperature data until time t1, and after time t1, the Q statistics were calculated from the principal component analysis results (influence coefficients) before time t1. The calculation results of the Q statistics for this time period are shown in FIG. 5. As is clear from the figure, the Q statistics are calculated to be large not only when the temperature data from the failed temperature sensor 8, which includes significant fluctuations, is used, but also when the temperature data (constant value) from the adjacent temperature sensor 8 is used. Normally, the Q statistics at every time point exceeds the breakout prediction threshold.

[0015] In contrast, FIG. 6 shows temperature data in which intentional variation has been added to the temperature data of the adjacent temperature sensor 8 shown in FIG. 4. The temperature data with this variation is obtained by adding the same amplitude to the temperature data shown in FIG. 4 at approximately the same short period. In this example, assuming that the failure of the temperature sensor 8 that failed at time t0 was discovered, the temperature data of FIG. 6 was used from time t0 to perform principal component analysis to obtain a principal component vector, and the deviation value for this principal component vector, i.e., the Q statistic, was calculated. The failure of the temperature sensor 8 may be, for example, a temperature range that would not be possible under normal conditions, or a temperature that fluctuates at a physically impossible speed (timing, temperature change rate). In addition, the temperature data from which variation has been removed (= the temperature data of FIG. 4) is used to calculate the Q statistic. The calculation result of the Q statistic for this time period is shown in FIG. 7. As is clear from the figure, the Q statistic is within a small value over the entire time period, and therefore the breakout prediction threshold is not exceeded, and the breakout prediction detection can be continued. This is because intentional variability in the temperature data results in a smaller scale for the variable resulting from standardization compared to other variables, which results in a relatively smaller influence coefficient for the corresponding variable in the principal component vector when principal component analysis is performed, and this allows intentionally lowering the sensitivity to anomaly detection. In this example, since there is only one missing temperature data, it is sufficient to provide variability using the above method, but if there is multiple signal data to be estimated and supplemented, it is desirable to, for example, change the period or amplitude in the above method for each data item so that there is no correlation between the data items.

[0016] Next, a different example will be described in which the estimated and complemented temperature data is intentionally varied, a principal component analysis is performed, and the Q statistics are calculated from the principal component vector. FIG. 8 shows an example of temperature data actually detected by the temperature sensor 8. That is, the temperature data shown in this example is normal temperature data detected by the temperature sensor 8 that is not broken. A larger amount of temperature data (50 points as an example) is used to calculate the Q statistics by the principal component analysis, and FIG. 9 shows the result of performing the principal component analysis using these normal temperature data and calculating the Q statistics from the principal component vector. Because the scale of the vertical axis is small, the Q statistics looks large, but the actual value is small. The following simulation is intended to observe the influence of the temperature data that has been shifted from the normal value when one of the temperature data in FIG. 8 (actually one of many temperature data) is shifted from the normal value (by performing a low-precision estimated and complemented) and a principal component analysis is performed, and then the temperature data in FIG. 8, i.e., the normal value, is returned, and the Q statistics is calculated from the previous principal component vector.

[0017] First, one of the thermometers breaks down, and the adjacent temperature data is multiplied by 1.2 to perform principal component analysis under the assumption that the estimation accuracy is low. After that, the thermometer is restored to a normal state at time t2, and the Q statistics are calculated from the principal component vector before that. The calculation results are shown in Figure 10. As is clear from Figure 10, the adjacent temperature data of the broken thermometer is multiplied by 1.2 to perform principal component analysis, and after that temperature data is restored to the normal state, the Q statistics calculated from the principal component vector before that deviates from the Q statistics based on the normal temperature data in Figure 9. In contrast, the adjacent temperature data of the broken thermometer (the same data as above) is multiplied by 1.2, and then the data is intentionally varied (±10°C) to perform principal component analysis. After that, the thermometer is restored to a normal state at time t2, and the Q statistics are calculated from the principal component vector before that. The calculation results are shown in Figure 11. As is clear from the figure, compared to the case where the adjacent temperature data is simply multiplied by 1.2, when the adjacent temperature data is intentionally given a variation of 1.2 times, the calculated Q statistic is almost the same as the Q statistic obtained using normal temperature data. In this case, if the absolute value of the variation, set to ±10°C, is increased, the deviation from the Q statistic obtained using normal temperature data can be further reduced, while if the magnitude of the variation is reduced, the deviation from the normal Q statistic can be increased.

[0018] The above has been described as a method for estimating and supplementing temperature data assuming that the temperature sensor 8 has failed, but the method can be widely applied to cases where there is a shortage of temperature data required for principal component analysis and the shortage needs to be supplemented by estimating the shortage. For example, the method can be used in cases where the number of temperature sensors has changed (decreased) due to equipment replacement, etc. Alternatively, the method can be applied to a shortage of the number of signal data outputs from a temperature sensor due to dimensional or positional fluctuations of the product, i.e., the cast piece.

[0019] Although the abnormality detection method according to the embodiment has been described above, the present invention is not limited to the configuration described in the above embodiment, and various modifications are possible within the scope of the gist of the present invention. For example, in the above embodiment, only the abnormality detection method for detecting a sign of a breakout in a continuous casting facility as an abnormality has been described in detail, but the abnormality detection method of the present invention can be applied to any abnormality detection method that performs principal component analysis using multiple (a large number) of signal data of the same type in the facility and calculates Q statistics from the principal component vector.

[0020] In this embodiment, when signal data required for principal component analysis is insufficient, normal signal data adjacent to or related to the insufficient signal data is used to estimate and complement the insufficient signal data, and when performing principal component analysis to calculate principal component vectors, the estimated and complemented signal data is used with intentional variations, and when calculating Q statistics from the principal component vectors, the estimated and complemented signal data is used without intentional variations. This makes it possible to continue detecting anomalies in equipment, for example, even if a sensor that outputs signal data breaks down and the accuracy of the estimated and complemented output data of the broken sensor is not sufficient.

[0021] In addition, it can also deal with a shortage of signal data due to sensor failure, a shortage of sensors due to equipment replacement, or a shortage of signal data outputs from sensors due to dimensional or positional fluctuations of the product, thereby expanding the applicability to cases of shortage of signal data and improving the continuity of equipment abnormality detection. [Explanation of symbols]

[0022] 1. Continuous casting machine 2. Molten Steel 3 Tundish 4 Submerged Nozzle 5. Mold 6. Casting 7. Cast piece support roll 8 Temperature Sensor 20 Judgment section

Claims

1. 1. An anomaly detection method for detecting an anomaly in a facility from a result of principal component analysis of a plurality of signal data of the same type in the facility, comprising: When signal data required for the principal component analysis is insufficient, the missing signal data is estimated and supplemented using normal signal data adjacent to or related to the missing signal data, and when performing the principal component analysis to calculate a principal component vector, the estimated and supplemented signal data is used with intentional variation added, and when calculating a Q statistic, the estimated and supplemented signal data is used without intentional variation.

2. 2. The abnormality detection method according to claim 1, wherein the lack of signal data is caused by any one of the following: a sensor failure, a shortage of the number of sensors due to equipment replacement, or a shortage of the number of signal data outputs from the sensor due to dimensional or positional fluctuations of the product.

3. 3. The abnormality detection method according to claim 1, wherein the signal data is an output signal from a temperature sensor inside a mold of a continuous casting facility, and the abnormality is a prediction of a breakout in the continuous casting facility.

Images