Abnormal temperature detection device, abnormal temperature detection method, and abnormal temperature detection program

Abstract

The abnormal temperature detection device includes: a temperature acquisition unit that acquires temperature data; a prior processing unit that performs prior processing to calculate a feature value of the temperature data; a synthesis unit that performs synthesis processing to synthesize the feature value and previous judgment information indicating the normality or abnormality of the temperature data obtained in the past; a learning unit that performs machine learning on the synthesis result of the synthesis unit to generate a recognition model; and a judgment unit that uses the recognition model to judge whether the temperature data for detection, which is acquired by the temperature acquisition unit, pre-processed by the prior processing unit, and synthesized by the synthesis unit, is normal or abnormal.

Description

Technical Field

[0001] The present invention relates to an abnormal temperature detection device, an abnormal temperature detection method and an abnormal temperature detection program.

[0002] This application claims priority from Japanese Patent Application No. 2020-206831 filed on December 14, 2020, the contents of which are incorporated herein by reference. Background Art

[0003] Accidents such as fires in residential buildings, fires at coal-fired power plants, and molten steel leaks at steel mills can result in loss of life and lost business opportunities. Therefore, to prevent these losses, detecting abnormal temperatures that could lead to accidents is extremely important. Furthermore, in addition to indicating a significantly elevated temperature compared to normal temperatures, abnormal temperatures can also indicate a significantly lower temperature compared to normal temperatures.

[0004] Japanese Patent Nos. 6076571 and 5808605 disclose prior art techniques for detecting abnormal conditions (including abnormal temperatures). For example, the technique disclosed in Japanese Patent No. 6076571 divides periodically recurring data into data for each cycle (periodic data), averages the multiple periodic data, and creates reference data representing normal conditions. Furthermore, abnormal conditions are detected based on the deviation between the measured periodic data and the reference data.

[0005] However, the technology disclosed in Japanese Patent No. 6076571 above determines an abnormal state when the deviation value exceeds a predetermined tolerance (threshold value). This threshold-based determination method sometimes fails to distinguish between ambient temperature changes and temperature changes caused by, for example, a fire, when the ambient temperature exhibits an upward or downward trend. In particular, when the amount of ambient temperature change and the amount of temperature change caused by, for example, a fire, are not significantly different, it is difficult to appropriately determine whether abnormal temperatures have occurred. Summary of the Invention

[0006] In order to solve the above-mentioned problem, an abnormal temperature detection device (1) involved in one embodiment of the present invention has: a temperature acquisition unit (11) that acquires temperature data (D1, D2); a prior processing unit (13) that performs prior processing to calculate a feature value of the temperature data; a synthesis unit (14) that performs synthesis processing to synthesize the feature value and previous judgment information indicating whether the temperature data is normal or abnormal, which has been obtained in the past; a learning unit (15) that performs machine learning on the synthesis result of the synthesis unit to generate a recognition model (CM); and a judgment unit (16) that uses the recognition model to judge whether the temperature data (D2) for detection, which is acquired by the temperature acquisition unit, subjected to the prior processing unit, and subjected to the synthesis processing by the synthesis unit, is normal or abnormal.

[0007] In addition, the abnormal temperature detection device involved in one embodiment of the present invention further includes a training information generating unit (12). When the machine learning is performed by the learning unit, the training information generating unit (12) generates training information (TR) with identification information indicating normal or abnormal added to the temperature data acquired by the temperature acquiring unit, the prior processing unit calculates the feature value of the temperature data contained in the training information, and the synthesis unit synthesizes the feature value, the identification information contained in the training information, and the previous judgment information.

[0008] In the abnormal temperature detection device according to one aspect of the present invention, the past determination information is at least one piece of identification information added to the past temperature data acquired by the temperature acquisition unit.

[0009] In the abnormal temperature detection device according to one aspect of the present invention, when the determination unit makes the determination, the synthesizing unit uses at least one determination result of the determination unit obtained in the past as the past determination information.

[0010] In the abnormal temperature detection device according to one aspect of the present invention, the prior processing unit calculates at least one of a standard deviation, a first-order difference, or a second-order difference of the temperature data or the averaged temperature data as the feature value.

[0011] In addition, in the abnormal temperature detection device involved in one embodiment of the present invention, the prior processing unit removes noise from the training information generated by the training information generating unit when the learning unit performs the machine learning, and calculates the feature value of the temperature data contained in the training information from which the noise has been removed.

[0012] In addition, in an abnormal temperature detection device involved in one embodiment of the present invention, the prior processing unit removes noise from the temperature data for detection acquired by the temperature acquisition unit when the determination unit determines whether the temperature data for detection is normal or abnormal, and calculates the characteristic value of the temperature data for detection from which the noise has been removed.

[0013] An abnormal temperature detection method according to one embodiment of the present invention comprises the following steps: acquiring temperature data (D1) by a temperature acquisition unit (S11); performing prior processing of calculating a characteristic value of the temperature data by a prior processing unit (S14); performing synthesis processing of synthesizing the characteristic value and previously obtained previous judgment information indicating whether the temperature data is normal or abnormal by a synthesis unit (S15); performing machine learning on the synthesis result of the first synthesis step by a learning unit to generate a recognition model (CM) (S16); acquiring temperature data for detection (D2) by the temperature acquisition unit (S21); performing prior processing on the temperature data for detection by the prior processing unit (S23); performing the synthesis processing by the synthesis unit and using the processing result of the second prior processing step (S24); and determining whether the temperature data for detection that has undergone the prior processing and the synthesis processing is normal or abnormal by a determination unit and using the recognition model (S25).

[0014] In addition, the abnormal temperature detection method involved in one embodiment of the present invention also has the following steps: when the machine learning is performed by the learning unit, a training information generating unit is used to generate training information in which identification information indicating normal or abnormal is added to the temperature data acquired by the temperature acquiring unit; the prior processing unit is used to calculate the feature value of the temperature data contained in the training information; and the synthesis unit is used to synthesize the feature value, the identification information contained in the training information, and the previous judgment information.

[0015] In the abnormal temperature detection method according to one aspect of the present invention, the past determination information is at least one piece of identification information added to the past temperature data acquired by the temperature acquisition unit.

[0016] Furthermore, the abnormal temperature detection method according to one aspect of the present invention further includes the step of using, by the synthesizing unit, at least one previously obtained determination result of the determination unit as the past determination information when the determination unit performs the determination.

[0017] In addition, the abnormal temperature detection method involved in one embodiment of the present invention further includes the following step, namely, using the prior processing unit to calculate at least one of the standard deviation, first-order difference or second-order difference of the temperature data or the averaged temperature data as the feature value.

[0018] In addition, the abnormal temperature detection method involved in one embodiment of the present invention also has the following steps: when the machine learning is performed by the learning unit, the prior processing unit is used to remove noise from the training information generated by the training information generating unit; and the prior processing unit is used to calculate the characteristic value of the temperature data contained in the training information from which the noise has been removed.

[0019] In addition, the abnormal temperature detection method involved in one embodiment of the present invention also has the following steps: when the determination unit determines whether the temperature data for detection is normal or abnormal, the prior processing unit is used to remove noise from the temperature data for detection acquired by the temperature acquisition unit; and the prior processing unit is used to calculate the characteristic value of the temperature data for detection from which the noise has been removed.

[0020] A computer-readable, non-temporary storage medium according to one embodiment of the present invention stores a program executed by a computer, wherein the program causes the computer to execute the following steps: acquiring temperature data, performing prior processing to calculate a characteristic value of the temperature data, performing synthesis processing to synthesize the characteristic value and previously obtained previous judgment information indicating whether the temperature data is normal or abnormal, performing machine learning on the synthesis result of the synthesis processing to generate a recognition model, acquiring temperature data for detection, performing the prior processing on the temperature data for detection, performing the synthesis processing using the processing result of the prior processing, and using the recognition model to determine whether the temperature data for detection that has undergone the prior processing and the synthesis processing is normal or abnormal.

[0021] In addition, one embodiment of the present invention involves a computer-readable non-temporary storage medium, wherein the program causes the computer to execute the following steps: when performing the machine learning, generating training information to which identification information indicating normality or abnormality is attached to the temperature data, calculating the feature quantity of the temperature data contained in the training information, so that the feature quantity, the identification information contained in the training information, and the previous judgment information are synthesized.

[0022] In the computer-readable non-transitory storage medium according to one aspect of the present invention, the past determination information is at least one piece of identification information added to the past temperature data acquired by the temperature acquisition unit.

[0023] In the computer-readable non-transitory storage medium according to one embodiment of the present invention, the program causes the computer to execute the step of using at least one previously obtained determination result as the past determination information when performing the determination.

[0024] In addition, in one embodiment of the present invention, the computer-readable non-transitory storage medium is provided with a program that causes the computer to execute the following steps: calculating at least one of the standard deviation, first-order difference, or second-order difference of the temperature data or the averaged temperature data as the feature value.

[0025] In addition, in one embodiment of the present invention, the computer-readable non-temporary storage medium is provided, wherein the program causes the computer to execute the following steps: removing noise from the training information when performing the machine learning, and calculating the feature value of the temperature data contained in the training information from which the noise has been removed.

[0026] An abnormal temperature detection program involved in one embodiment of the present invention causes a computer to execute the following steps: obtaining temperature data, performing prior processing to calculate a feature value of the temperature data, performing synthesis processing to synthesize the feature value and previous judgment information indicating the normality or abnormality of the temperature data obtained in the past, performing machine learning on the synthesis result of the synthesis processing to generate a recognition model, obtaining temperature data for detection, performing the prior processing on the temperature data for detection, performing the synthesis processing using the processing result of the prior processing, and using the recognition model to determine whether the temperature data for detection that has undergone the prior processing and the synthesis processing is normal or abnormal.

[0027] According to the present invention, there is an effect that abnormal temperature can be detected with high accuracy even when the ambient temperature changes.

[0028] Features and aspects of the present invention will become more apparent from the detailed description of the embodiments described below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a block diagram showing a main configuration of an abnormal temperature detection device according to an embodiment of the present invention.

[0030] Figure 2A This is a diagram for explaining feature quantities used in one embodiment of the present invention.

[0031] Figure 2B This is a diagram for explaining feature quantities used in one embodiment of the present invention.

[0032] Figure 2C This is a diagram for explaining feature quantities used in one embodiment of the present invention.

[0033] Figure 2D A diagram for explaining feature quantities used in one embodiment of the present invention.

[0034] Figure 2E A diagram for explaining feature quantities used in one embodiment of the present invention.

[0035] Figure 3 This is a flowchart showing an overview of the operation during training of the abnormal temperature detection device according to one embodiment of the present invention.

[0036] Figure 4 This is a flowchart showing an outline of the operation of the abnormal temperature detection device according to one embodiment of the present invention during detection. DETAILED DESCRIPTION

[0037] While the embodiments of the present invention are described with reference to preferred embodiments, those skilled in the art can implement various alternatives to the embodiments using the teachings of the present invention, and the present invention is not limited to the preferred embodiments described herein.

[0038] One embodiment of the present invention provides an abnormal temperature detection device, an abnormal temperature detection method, and an abnormal temperature detection program capable of detecting abnormal temperature with high accuracy even when the ambient temperature changes.

[0039] Hereinafter, an abnormal temperature detection device, an abnormal temperature detection method, and an abnormal temperature detection program according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

[0040] [summary]

[0041] The present invention is capable of detecting abnormal temperatures with high accuracy even when the ambient temperature fluctuates. For example, when the ambient temperature fluctuates with an upward or downward trend, abnormal temperatures can be detected with high accuracy even when the amount of change in the ambient temperature is not significantly different from the amount of change in the temperature caused by a fire, etc.

[0042] Setting thresholds for determining abnormal temperatures is difficult for those without a detailed understanding of the mechanisms of temperature rise. Consequently, the technology disclosed in Japanese Patent No. 6076571 requires experienced and skilled operators to modify and adjust baseline data to set thresholds in order to detect equipment anomalies in factories and other facilities. Furthermore, Japanese Patent No. 5808605 discloses a method that uses keywords to correlate maintenance history information from experienced operators with sensor data, and then uses clustering and multivariate analysis to detect signs of anomalies.

[0043] However, regarding the technology disclosed in Japanese Patent Gazette No. 6076571 and Japanese Patent Gazette No. 5808605, a skilled operator sets the threshold value using a large amount of maintenance history information. Therefore, for example, it cannot be applied to new equipment that has almost no maintenance history information. Therefore, it is conceivable that it is very difficult to set an appropriate threshold value when there is little maintenance history information. In addition, for example, when the change in ambient temperature and the change in temperature caused by a fire or the like are not significantly different, it is extremely difficult to accurately detect temperature anomalies. For example, it is conceivable that if the threshold value used to distinguish between the change in ambient temperature and the change in temperature caused by a fire or the like is set too high, the alarm will be delayed, and conversely, if the threshold value is set too low, a false alarm will be issued.

[0044] In the embodiments of the present invention, first, during the training phase, pre-processing is performed to calculate the feature values of temperature data acquired by the temperature acquisition unit. Then, synthesis processing is performed to synthesize the calculated feature values with previously acquired information indicating whether the temperature data is normal or abnormal. Furthermore, machine learning is performed on the results of the synthesis process to generate a recognition model. Next, during the detection phase, the recognition model is used to determine whether the temperature data for detection, acquired by the temperature acquisition unit and subjected to pre-processing and synthesis, is normal or abnormal. This allows for highly accurate detection of abnormal temperatures, even when the ambient temperature fluctuates.

[0045] [Details]

[0046] <Abnormal Temperature Detection Device>

[0047] Figure 1 FIG. 1 is a block diagram showing the main structure of an abnormal temperature detection device according to an embodiment of the present invention. Figure 1 As shown, the abnormal temperature detection device 1 of this embodiment has a temperature acquisition unit 11 (temperature acquisition unit), a training information generation unit 12, a prior processing unit 13 (prior processing unit), a synthesis unit 14 (synthesis unit), a learning unit 15 (learning unit), a determination unit 16 (determination unit), an output unit 17 and a storage unit 18.

[0048] This abnormal temperature detection device 1 performs machine learning using training temperature data D1 acquired by a temperature acquisition unit 11, and uses the results of the machine learning to determine whether detection temperature data D2 acquired by the temperature acquisition unit 11 is normal or abnormal. The training temperature data D1 is temperature data known to be normal or abnormal, while the detection temperature data D2 is temperature data to be determined as abnormal.

[0049] In addition, Figure 1 For ease of understanding, solid lines represent signal paths between modules during training, and dashed lines represent signal paths between modules during detection. For example, when predetermined conditions are met or when a user of abnormal temperature detection device 1 instructs the device 1 to switch between machine learning and abnormal temperature detection.

[0050] Here, abnormal temperature refers to a temperature significantly higher than normal, or a temperature significantly lower than normal. In this embodiment, the case where a temperature significantly higher than normal is detected as abnormal temperature is used as an example for explanation. Furthermore, the normal temperature waveform and the abnormal temperature waveform are not fixed waveforms and can be changed to any desired temperature waveform depending on the situation in which the abnormal temperature detection device 1 is used.

[0051] The temperature acquisition unit 11 acquires temperature data, which is time-series data representing temperature changes over time. The temperature acquisition unit 11 outputs the temperature data acquired during training (training temperature data D1) to the training information generation unit 12. Conversely, the temperature acquisition unit 11 outputs the temperature data acquired during detection (detection temperature data D2) to the prior processing unit 13.

[0052] The temperature acquisition unit 11 acquires temperature data output from a temperature sensor, such as a thermocouple, a resistance temperature detector, a thermistor, a crystal oscillator, an infrared radiation thermometer, or a fiber optic temperature sensor. The temperature acquisition unit 11 includes an analog-to-digital converter (ADC) and, when the temperature sensor outputs an analog signal, acquires digitized data as temperature data. The temperature data acquired by the temperature acquisition unit 11 can be obtained at each ADC sampling cycle, and data obtained at each ADC sampling cycle can be discarded at predetermined time intervals.

[0053] When machine learning is performed by the learning unit 15, the training information generating unit 12 generates training information TR by adding identification information indicating whether the training temperature data D1 is normal or abnormal to the training temperature data D1 acquired by the temperature acquiring unit 11. The identification information is created by, for example, a user using the abnormal temperature detection device 1.

[0054] The training information generating unit 12 generates, for example, training information TR appended with identification information indicating that the training temperature data D1 is normal, and training information TR appended with identification information indicating that the training temperature data D1 is abnormal. Furthermore, the training information generating unit 12 may only generate training information TR appended with identification information indicating that the training temperature data D1 is normal. Alternatively, the training information generating unit 12 may only generate training information TR appended with identification information indicating that the training temperature data D1 is abnormal. Furthermore, the training information TR may be generated based on a simulation calculation simulating the temperature of the object being measured by the abnormal temperature detection device 1.

[0055] The prior processing unit 13 performs necessary processing (pre-processing) in advance for machine learning or abnormal temperature detection on the training temperature data D1 included in the training information TR generated by the training information generating unit 12, or the detection temperature data D2 output from the temperature acquiring unit 11. Specifically, the prior processing unit 13 performs processing to remove noise superimposed on the training temperature data D1 or the detection temperature data D2 as necessary, and then calculates the feature value of the training temperature data D1 or the detection temperature data D2.

[0056] When the learning unit 15 performs machine learning, for example, the prior processing unit 13 calculates at least one of the standard deviation, first-order difference, or second-order difference of the training temperature data D1 (or the averaged training temperature data D1) as a feature quantity. When the determination unit 16 determines whether the detection temperature data D2 is normal or abnormal, for example, the prior processing unit 13 calculates at least one of the standard deviation, first-order difference, or second-order difference of the detection temperature data D2 (or the averaged detection temperature data D2) as a feature quantity.

[0057] Here, the standard deviation of the temperature data can be used as a feature quantity representing the magnitude of the temperature variation. Furthermore, the first-order difference of the temperature data can be used as a feature quantity representing the tendency (trend) of the temperature variation. Furthermore, the second-order difference of the temperature data can be used as a feature quantity representing the temperature variation without the trend. Furthermore, the sequence of temperature difference (first-order difference, second-order difference) can be used as a feature quantity representing the ambient temperature variation without the daily or seasonal variations.

[0058] Figures 2A to 2E This is a diagram for explaining feature quantities used in one embodiment of the present invention. Figures 2A to 2C is a graph showing an example of the change in temperature over time. Figure 2D 、 Figure 2E It means that Figures 2A to 2CGraph of characteristic quantities obtained when the temperature changes with time as an example shown in FIG. Figures 2A to 2C In the graph shown, the dotted line represents the change of ambient temperature over time, and the solid line represents the change of temperature over time with the addition of abnormal temperature caused by fire, etc. and ambient temperature. In addition, the dotted line represents the threshold temperature (threshold) set in the prior art for detecting abnormal temperature. Figures 2A to 2C In the example shown, for ease of understanding, the amount of temperature change due to a fire or the like is assumed to be the same.

[0059] Figure 2A The example shown shows that the slope of the dashed line is zero and no change in the ambient temperature occurs. Figure 2B The example shown shows that the slope of the dotted line is positive and the change in ambient temperature shows an upward trend. Figure 2B The example shown shows that the slope of the dotted line is negative and the change in ambient temperature shows a downward trend.

[0060] Here, in Figures 2A to 2C In the example shown, consider the case where abnormal temperature is detected using the threshold temperature indicated by the dot-dash line. Figure 2A In the example shown, the ambient temperature shown by the dotted line does not exceed the threshold temperature, and the temperature shown by the solid line (the temperature obtained by adding the abnormal temperature and the ambient temperature) exceeds the threshold temperature. Therefore, the abnormal temperature can be detected because the temperature shown by the solid line exceeds the threshold temperature.

[0061] In contrast, in Figure 2B In the example shown, the ambient temperature shown by the dotted line and the temperature shown by the solid line both rise and exceed the threshold temperature. Figure 2C In the example shown, only the ambient temperature indicated by the dotted line drops and exceeds the threshold temperature. Figure 2B 、 Figure 2C In the example shown, even if no abnormal temperature occurs, if the ambient temperature exceeds the threshold temperature (in the case of exceeding in the rising direction or the falling direction), it is erroneously detected that an abnormal temperature occurs. Figure 2C In the example shown, although the temperature indicated by the solid line exceeds the threshold temperature, the change in the temperature indicated by the solid line is small (approximately flat), and therefore it appears that no abnormal temperature has occurred.

[0062] Figure 2D The graph shown represents Figures 2A to 2C In the example shown, the first difference of temperature is shown by the solid line. Figure 2D The curve L1 in Figure 2A In the example shown, the first-order difference curve L2 is Figure 2B In the example shown, the first-order difference curve L3 is Figure 2C The first order difference in the example shown. Figure 2D , curves L1 to L3 all have two peaks.

[0063] but, Figure 2A In the example shown, there is no change in the ambient temperature, so the value of the portion other than the peak of the curve L1 becomes zero. Figure 2B In the example shown, an upward trend is shown, so the value of the portion other than the peak value of the curve L2 is also positive. Figure 2C In the example shown, a downward trend is observed, so the values of the portion other than the peak of curve L3 are negative. Thus, the value of the first-order difference changes according to the trend of temperature change and can be used as a characteristic quantity indicating the trend of temperature change.

[0064] Figure 2E The graph shown represents Figures 2A to 2C In the example shown, the second difference of temperature is shown by the solid line. Figure 2E , Figures 2A to 2C The second-order difference in the example is represented by the same (or substantially the same) curve. In this way, the value of the second-order difference does not change according to the tendency (trend) of the temperature change, and can therefore be used as a feature quantity representing the temperature change excluding the trend.

[0065] The synthesis unit 14 performs synthesis processing by synthesizing the feature quantity calculated by the prior processing unit 13 with the previous determination information indicating whether the temperature data obtained previously was normal or abnormal. This synthesis processing is performed because the feature quantity calculated by the prior processing unit 13 does not include the absolute value of the temperature. Therefore, abnormal temperature detection using only the feature quantity may not detect abnormal temperature.

[0066] For example, consider a situation where the temperature rises sharply and then stabilizes (no temperature change). In this case, detecting abnormal temperature using only characteristic quantities cannot distinguish between a state where the temperature remains unchanged due to abnormal temperature and a state where the temperature remains unchanged due to normal temperature. Therefore, even if abnormal temperature occurs, it may not be detected.

[0067] In this case, if abnormal temperature can be detected midway until the temperature rises and stabilizes, abnormal temperature can be detected even after the temperature rises and stabilizes. Therefore, in this embodiment, abnormal temperature is detected using the feature quantity calculated by the previous processing unit 13 and the previous determination information indicating whether the temperature data obtained in the past is normal or abnormal.

[0068] Specifically, while the learning unit 15 performs machine learning, the synthesis unit 14 performs synthesis processing to generate synthesized data X1 by synthesizing the feature values calculated by the prior processing unit 13 based on the training temperature data D1, the identification information included in the training information TR, and the previous determination information. The generated synthesized data X1 is represented, for example, by a list or vector. Furthermore, the previous determination information here can utilize at least one piece of identification information added to the previous temperature data acquired by the temperature acquisition unit 11 (e.g., identification information indicating whether the temperature data acquired just before was normal or abnormal).

[0069] Furthermore, when the determination unit 16 determines whether the detection temperature data D2 is normal or abnormal, the synthesis unit 14 performs synthesis processing to synthesize the feature quantity calculated by the prior processing unit 13 based on the detection temperature data D2 with the previous determination information to generate synthesized data X2. The synthesized data X2 generated by the synthesis unit 14 is represented by, for example, a list or a vector. Furthermore, as the previous determination information here, at least one previously obtained determination result of the determination unit 16 (for example, the determination result Y of the determination unit 16 obtained immediately before) can be used. Furthermore, it is important to note that the synthesized data X2 does not include the identification information included in the synthesized data X1.

[0070] The learning unit 15 uses the synthesized data X1 output from the synthesizing unit 14 as training data to perform machine learning. The learning unit 15 preferably uses classification (Classification) using the Random Forest technology, for example. This is because machine learning can be performed with less training data than deep learning. As a classification method, the learning unit 15 can use a binary classification method such as the Boost Decision Tree, which is a recursive partitioning method. The learning unit 15 generates a recognition model CM that determines the outputs of multiple decision trees by majority voting and provides it to the determination unit 16 (output).

[0071] Determination unit 16 uses recognition model CM output from learning unit 15 to determine whether detection temperature data D2, acquired by temperature acquisition unit 11, pre-processed by pre-processing unit 13, and synthesized by synthesis unit 14, is normal or abnormal. Determination unit 16 outputs a determination result indicating whether detection temperature data D2 is normal or abnormal to output unit 17 and storage unit 18.

[0072] The output unit 17 outputs the determination result of the determination unit 16. The output unit 17 includes a display device such as an LCD (Liquid Crystal Display) to visually display the determination result of the determination unit 16. In addition to the LCD, the output unit 17 also includes contact terminals to which alarms such as a red warning light and a buzzer can be connected, thereby notifying the determination result of the determination unit 16 using light or sound.

[0073] The storage unit 18 stores the determination results of the determination unit 16. The determination results stored in the storage unit 18 (for example, the determination result Y of the determination unit 16 obtained just before) are output to the synthesis unit 14. The storage unit 18 can be implemented by a volatile memory such as RAM (Random Access Memory), or a non-volatile memory such as flash ROM (Read Only Memory) or EEPROM (Electrically Erasable and Programmable ROM).

[0074] This abnormal temperature detection device 1 is implemented by a computer. For example, it can be implemented by a desktop, laptop, or tablet computer. When the abnormal temperature detection device 1 is implemented by a computer, each module provided in the abnormal temperature detection device 1 (temperature acquisition unit 11, training information generation unit 12, prior processing unit 13, synthesis unit 14, learning unit 15, determination unit 16, etc.) is implemented by executing a program (abnormal temperature detection program) for implementing each function by the CPU (central processing unit) provided in the computer. In other words, each module provided in the abnormal temperature detection device 1 is implemented through the collaboration of software and hardware resources.

[0075] Here, the program that implements the functions of each module provided in the abnormal temperature detection device 1 can be distributed in a state recorded on a computer-readable recording medium, or can be distributed via a network such as the Internet. In addition, the abnormal temperature detection device 1 can be implemented using hardware such as an FPGA (Field-Programmable Gate Array), an LSI (Large Scale Integration), or an ASIC (Application Specific Integrated Circuit).

[0076] In addition, the computer-readable recording medium includes, for example, a magnetic recording medium, an optical recording medium, and a semiconductor memory. Examples of magnetic recording media include floppy disks and hard disk drives. Examples of optical recording media include CD-ROMs and DVD-ROMs. Examples of semiconductor memory include flash ROMs (Read Only Memory) and EEPROMs (Electrically Erasable and Programmable ROMs).

[0077] <Abnormal Temperature Detection Method>

[0078] Next, the operation of the abnormal temperature detection device 1 will be described. The operation of the abnormal temperature detection device 1 can be broadly divided into the training phase, which performs the aforementioned machine learning to obtain the recognition model CM, and the detection phase, which detects abnormal temperatures. The following describes the training phase and the detection phase of the abnormal temperature detection device 1 in order.

[0079] Movements during training

[0080] Figure 3 This is a flowchart showing an overview of the operation during training of the abnormal temperature detection device according to one embodiment of the present invention. Figure 3 The processing of the flowchart shown is started, for example, when the user of the abnormal temperature detection device 1 operates an input unit (not shown) and issues an instruction to start machine learning.

[0081] Once the process begins, the temperature acquisition unit 11 first acquires training temperature data D1 (step S11: first temperature acquisition step). In this process, multiple training temperature data D1 are acquired once or multiple times for a predetermined period of time (e.g., approximately 3 to 5 minutes). In practical terms, for example, 100 training temperature data D1 are acquired 200 times.

[0082] Next, the training information generating unit 12 generates training information TR, which is appended with identification information indicating whether the training temperature data D1 is normal or abnormal, based on the acquired training temperature data D1 (step S12). The identification information indicating whether the acquired training temperature data D1 is normal or abnormal is input to the training information generating unit 12, for example, by a user of the abnormal temperature detection device 1 operating an input unit (not shown). Alternatively, the training information TR may be generated based on a simulation calculation simulating the temperature of the object being measured by the abnormal temperature detection device 1.

[0083] Next, the pre-processing unit 13 removes noise from the generated training information TR (step S13). For example, filtering, such as averaging, is performed on the training temperature data D1 included in the training information TR in the time axis direction to remove noise superimposed on the training temperature data D1. Furthermore, in this filtering process, an appropriate filter is selected that is suitable for the characteristics of the noise superimposed on the training temperature data D1.

[0084] Next, the prior processing unit 13 performs a process of calculating the feature quantity of the training temperature data D1 included in the training information TR (step S14: first prior processing step). Specifically, the prior processing unit 13 calculates at least one of the standard deviation, first-order difference, or second-order difference of the training temperature data D1 (or the averaged training temperature data D1) as a feature quantity. In addition, the standard deviation is a feature quantity that represents the magnitude of the change in the temperature of the training temperature data D1. The first-order difference is a feature quantity that represents the tendency (trend) of the temperature change of the training temperature data D1. The second-order difference is a feature quantity that represents the temperature change of the training temperature data D1 excluding the trend.

[0085] Next, the synthesis unit 14 performs synthesis processing to generate synthesized data X1 by synthesizing the feature value calculated based on the training temperature data D1, the identification information included in the training information TR, and the previous determination information (step S15: first synthesis step). As the previous determination information here, for example, the identification information indicating whether the temperature data is normal or abnormal, which was recently obtained by the temperature acquisition unit 11, can be used.

[0086] Once the above processing is complete, the learning unit 15 performs machine learning using the synthesized data X1 output from the synthesizing unit 14 (step S16: learning step). Here, in addition to the relationship between the feature values calculated from the training temperature data D1 and the identification information indicating whether the training temperature data D1 is normal or abnormal, the relationship with previous determination information (identification information indicating whether the temperature data is normal or abnormal, which was immediately obtained by the temperature acquiring unit 11) is also learned. After this machine learning, the learning unit 15 outputs the recognition model CM to the determination unit 16 (step S17).

[0087] also, Figure 3 The process shown in the flowchart may, for example, automatically terminate after an amount of training preset by the user of the abnormal temperature detection device 1 has been performed. Alternatively, the process may terminate after a time preset by the user of the abnormal temperature detection device 1 has elapsed, or upon a termination instruction from the user of the abnormal temperature detection device 1. Alternatively, the process may terminate if the value of the recognition model CM for predetermined time series data is less than or equal to a predetermined value.

[0088] Actions during testing

[0089] Figure 4 This is a flowchart showing an overview of the operation of the abnormal temperature detection device according to one embodiment of the present invention during detection. Figure 4 The process of the flowchart shown is started, for example, when the user of the abnormal temperature detection device 1 operates an input unit (not shown) and issues an instruction to start detection. Figure 4 The process shown in the flowchart can also be Figure 3 The process automatically starts when the process of the flowchart shown is completed.

[0090] Once the process begins, the temperature acquisition unit 11 first acquires the detection temperature data D2 (step S21: second temperature acquisition step). This process acquires only one set of training temperature data D1 corresponding to a predetermined time (e.g., approximately 3 to 5 minutes) and detection temperature data D2 of the same length (e.g., 100 detection temperature data D2). Since it is unclear whether the detection temperature data D2 is normal or abnormal, no identification information is input.

[0091] Next, the pre-processing unit 13 performs a process of removing noise from the acquired detection temperature data D2 (step S22). Figure 3 The same filter as the filter used in the process of step S13 shown above performs a process of removing noise superimposed on the temperature data D2 for detection.

[0092] Next, the prior processing unit 13 calculates the characteristic value of the detection temperature data D2 (step S23: second prior processing step). Specifically, the prior processing unit 13 calculates at least one of the standard deviation, first-order difference, or second-order difference of the detection temperature data D2 (or the averaged detection temperature data D2) as a characteristic value. The standard deviation is a characteristic value that indicates the magnitude of the temperature change in the detection temperature data D2. The first-order difference is a characteristic value that indicates the tendency (trend) of the temperature change in the detection temperature data D2. The second-order difference is a characteristic value that indicates the temperature change in the detection temperature data D2 after the trend is removed.

[0093] Next, the synthesizing unit 14 performs a synthesis process to synthesize the feature amount calculated based on the temperature data D2 for detection and the previous determination information to generate the synthesized data X2 (step S24: second synthesis step). As the previous determination information here, for example, the determination result Y stored in the storage unit 18 (the determination result of the determination unit 16 obtained just before) can be used. In addition, for example, if the first Figure 4 As in the case of the processing of the flowchart shown in FIG. 1 , when the determination result of the determination unit 16 is not stored in the storage unit 18 , for example, a normal value (eg, zero) may be used as the initial value of the determination result Y.

[0094] When the above processing is completed, the synthesized data X2 output from the synthesizing unit 14 is input to the determining unit 16. Then, the determining unit 16 uses Figure 3 The recognition model CM outputted in the process of step S17 shown is processed to determine whether the temperature data D2 for detection is normal or abnormal (step S25: determination step).

[0095] Here, as previously described, the recognition model CM learns not only the relationship between the feature values calculated from the training temperature data D1 and the identification information indicating whether the training temperature data D1 is normal or abnormal, but also the relationship with previous determination information. Furthermore, the synthesized data X2 input to the determination unit 16 is obtained by synthesizing the feature values calculated from the detection temperature data D2 with previous determination information (the determination results obtained by the determination unit 16 immediately before). Therefore, the normality or abnormality of the detection temperature data D2 is determined by taking into account the previous determination information (the determination results obtained by the determination unit 16 immediately before before).

[0096] Once this determination is made, the determination unit 16 outputs the determination result to the output unit 17 and the storage unit 18 (step S26). Consequently, information indicating whether the detection temperature data D2 is normal or abnormal is displayed on the display device in a visually recognizable manner. Alternatively, if the detection temperature data D2 is determined to be abnormal, an alarm such as a red warning light or a buzzer is used to notify the user of the abnormality through light or sound.

[0097] Furthermore, for example, when it is determined that the temperature data D2 for detection is abnormal, Figure 4 The process of the flowchart shown can end automatically. Alternatively, Figure 4 The processing of the flowchart shown can be continued in response to an instruction from the user of the abnormal temperature detection device 1 , and can be ended in response to an end instruction from the user of the abnormal temperature detection device 1 .

[0098] As described above, in this embodiment, during training, pre-processing is performed to calculate the feature values of the training temperature data D1 acquired by the temperature acquisition unit 11. Then, synthesis processing is performed to synthesize the calculated feature values with previously acquired information indicating whether the training temperature data D1 is normal or abnormal. Machine learning is then performed on the results of the synthesis processing to generate a recognition model CM. Next, during detection, the recognition model CM is used to determine whether the detection temperature data D2, acquired by the temperature acquisition unit 11 and subjected to pre-processing and synthesis processing, is normal or abnormal.

[0099] Thus, in this embodiment, in addition to the characteristic values of the training temperature data D1, machine learning is performed using past determination information (previously obtained identification information indicating whether the training temperature data D1 is normal or abnormal). Furthermore, in this embodiment, in addition to the characteristic values of the detection temperature data D2, past determination information (previously obtained determination results of the determination unit 16) is also used to determine whether the detection temperature data D2 is normal or abnormal. Therefore, in this embodiment, abnormal temperatures can be detected with high accuracy even when the ambient temperature fluctuates.

[0100] While the abnormal temperature detection device, abnormal temperature detection method, and abnormal temperature detection program according to embodiments of the present invention have been described above, the present invention is not limited to the aforementioned embodiments and can be freely modified within the scope of the present invention. For example, the abnormal temperature detection device 1 according to the aforementioned embodiment includes a learning unit 15 and a determination unit 16. This allows the abnormal temperature detection device 1 to determine whether the detection temperature data D2 is normal or abnormal, and to automatically update (auto-learn) the recognition model CM through machine learning. This automatic updating of the learning unit allows the abnormal temperature detection device 1 according to the aforementioned embodiment to be optimized for the on-site environment.

[0101] Specifically, for example, if the abnormal temperature detection device 1 operates for a predetermined period (e.g., one year) without any abnormality being determined by the determination unit 16, the temperature data obtained during that period (detection temperature data D2) can be considered normal. Therefore, the temperature data obtained during that period (detection temperature data D2) can be used as normal training temperature data D1 for automatic learning.

[0102] Furthermore, the automatic learning can be performed only during a specified period. For example, the automatic learning can be performed only on the temperature data within the last year, without using data from earlier periods. Furthermore, the automatic learning can be started or stopped based on user instructions.

[0103] In the above embodiment, the case where a temperature significantly higher than normal is detected as an abnormal temperature is used as an example. However, the present invention is not limited to the case where a temperature significantly higher than normal is detected as an abnormal temperature. It can also be applied to the case where a temperature significantly lower than normal is detected as an abnormal temperature. However, it is important to note that appropriate identification information must be added to the training temperature data D1.

[0104] The abnormal temperature detection device 1 of the above embodiment can be implemented as a standalone device or through cloud computing via a network. Here, cloud computing can be consistent with the definition of a file specified by a URL (Uniform Resource Locator) (recommended by the National Institute of Standards and Technology, USA) as follows.

[0105] http: / / nvlpubs.nist.gov / nistpubs / Legacy / SP / nistspecialpublication800-145.pdf

[0106] https: / / www.ipa.go.jp / files / 000025366.pdf

[0107] In this specification, terms indicating directions such as "front, back, up, down, right, left, vertical, horizontal, longitudinal, transverse, row, and column" refer to those directions of the device of the present invention. Therefore, the above terms in the specification of the present invention should be interpreted relatively within the device of the present invention.

[0108] The words “configured” and the like are configured to perform the function of the present invention, or are used to indicate the structure, element, or part of a device.

[0109] Furthermore, the words expressed as "means and functions" in the claims should include all structures that can be used to perform the functions included in the present invention.

[0110] The term "unit" is used to refer to a component, unit, hardware, or a part of software compiled to perform a desired function. Typical examples of hardware are devices and circuits, but are not limited to these.

[0111] While preferred embodiments of the present invention have been described above, the present invention is not limited to these embodiments. Additional, omitted, replaced, and other modifications may be made to the structure without departing from the spirit of the present invention. The present invention is not limited by the foregoing description but is defined solely by the appended claims.

Claims

1. An abnormal temperature detection device, wherein: The abnormal temperature detection device comprises: a temperature acquisition unit for acquiring temperature data; a prior processing unit that performs prior processing to calculate a feature quantity representing at least one of a standard deviation, a first-order difference, or a second-order difference of the temperature data; a synthesis unit that performs synthesis processing of synthesizing the feature amount and previously obtained past determination information indicating whether the temperature data is normal or abnormal; a learning unit that performs machine learning on the synthesis result of the synthesis unit to generate a recognition model; as well as A determination unit determines, using the recognition model, whether the detection temperature data acquired by the temperature acquisition unit, pre-processed by the pre-processing unit, and synthesized by the synthesizing unit is normal or abnormal.

2. The abnormal temperature detection device according to claim 1, wherein: The abnormal temperature detection device further includes a training information generating unit, which generates training information by adding identification information indicating whether the temperature data acquired by the temperature acquiring unit is normal or abnormal when the learning unit performs the machine learning. The prior processing unit calculates the feature value of the temperature data included in the training information, The synthesis unit synthesizes the feature amount, the identification information included in the training information, and the past determination information.

3. The abnormal temperature detection device according to claim 2, wherein: The past determination information is at least one piece of identification information added to the past temperature data acquired by the temperature acquisition unit.

4. The abnormal temperature detection device according to any one of claims 1 to 3, wherein: When the determination unit makes the determination, the synthesis unit uses at least one determination result of the determination unit obtained in the past as the past determination information.

5. The abnormal temperature detection device according to claim 2, wherein: When the learning unit performs the machine learning, the prior processing unit removes noise from the training information generated by the training information generating unit and calculates the feature value of the temperature data included in the training information from which the noise has been removed.

6. The abnormal temperature detection device according to claim 2, wherein: When the determination unit determines whether the detection temperature data is normal or abnormal, the prior processing unit removes noise from the detection temperature data acquired by the temperature acquisition unit and calculates the feature value of the detection temperature data from which the noise has been removed.

7. A method for detecting abnormal temperature, wherein: The abnormal temperature detection method comprises the following steps: Acquiring temperature data using a temperature acquisition unit; performing a prior processing of calculating a feature quantity representing at least one of a standard deviation, a first-order difference, or a second-order difference of the temperature data by a prior processing unit; performing a synthesis process of synthesizing the feature amount and previously obtained past determination information indicating whether the temperature data is normal or abnormal by a synthesis unit; Using a learning unit to perform machine learning on the synthesis result of the synthesis unit to generate a recognition model; acquiring temperature data for detection using the temperature acquisition unit; performing the prior processing on the temperature data for detection by the prior processing unit; performing the synthesis process by using the synthesis unit and the processing result of the previous processing unit; as well as The determination unit determines whether the temperature data for detection that has been subjected to the previous processing and the synthesis processing is normal or abnormal using the recognition model.

8. The abnormal temperature detection method according to claim 7, wherein: The abnormal temperature detection method further comprises the following steps: When the learning unit performs the machine learning, a training information generating unit generates training information in which identification information indicating normality or abnormality is added to the temperature data acquired by the temperature acquiring unit; calculating the feature quantity of the temperature data included in the training information using the prior processing unit; as well as The synthesizing unit synthesizes the feature amount, the identification information included in the training information, and the past determination information.

9. The abnormal temperature detection method according to claim 8, wherein: The past determination information is at least one piece of identification information added to the past temperature data acquired by the temperature acquisition unit.

10. The abnormal temperature detection method according to any one of claims 7 to 9, wherein: The abnormal temperature detection method further includes the step of using, by the synthesis unit, at least one previously obtained determination result of the determination unit as the past determination information when the determination unit performs the determination.

11. The abnormal temperature detection method according to claim 8, wherein: The abnormal temperature detection method further comprises the following steps: When the learning unit performs the machine learning, removing noise from the training information generated by the training information generating unit by the prior processing unit; as well as The feature amount of the temperature data included in the training information from which the noise has been removed is calculated by the prior processing unit.

12. The abnormal temperature detection method according to claim 8, wherein: The abnormal temperature detection method further comprises the following steps: removing noise from the detection temperature data acquired by the temperature acquisition unit by the prior processing unit when the determination unit determines whether the detection temperature data is normal or abnormal; as well as The feature amount of the temperature data for detection from which the noise has been removed is calculated by the prior processing unit.

13. A computer-readable non-transitory storage medium storing a program executed by a computer, wherein: The program causes the computer to execute the following steps: Get temperature data, performing a prior process of calculating a feature quantity representing at least one of a standard deviation, a first-order difference, or a second-order difference of the temperature data, performing a synthesis process of synthesizing the feature amount and previously obtained previous determination information indicating whether the temperature data is normal or abnormal, Performing machine learning on the synthesis result of the synthesis process to generate a recognition model, Get temperature data for detection, performing the aforementioned prior processing on the temperature data for the detection, The synthesis process is performed using the processing result of the previous process, The normality or abnormality of the temperature data for detection that has been subjected to the previous processing and the synthesis processing is determined using the recognition model.

14. The computer-readable non-transitory storage medium according to claim 13, wherein: The program causes the computer to execute the following steps: When the machine learning is performed, training information is generated in which identification information indicating normality or abnormality is added to the temperature data. Calculating the feature quantity of the temperature data included in the training information, The feature amount, the identification information included in the training information, and the past determination information are synthesized.

15. The computer-readable non-transitory storage medium according to claim 14, wherein: The past determination information is at least one piece of identification information added to the past temperature data.

16. The computer-readable non-transitory storage medium according to any one of claims 13 to 15, wherein: The program causes the computer to execute the following steps: When performing the determination, at least one determination result obtained in the past is used as the past determination information.

17. The computer-readable non-transitory storage medium according to claim 14, wherein: The program causes the computer to execute the following steps: removing noise from the training information while performing the machine learning, The feature amount of the temperature data included in the training information from which the noise has been removed is calculated.

Images