WO2022123640A1 - 時系列信号のトリガ条件決定方法、監視対象設備の異常診断方法および時系列信号のトリガ条件決定装置 - Google Patents
時系列信号のトリガ条件決定方法、監視対象設備の異常診断方法および時系列信号のトリガ条件決定装置 Download PDFInfo
- Publication number
- WO2022123640A1 WO2022123640A1 PCT/JP2020/045563 JP2020045563W WO2022123640A1 WO 2022123640 A1 WO2022123640 A1 WO 2022123640A1 JP 2020045563 W JP2020045563 W JP 2020045563W WO 2022123640 A1 WO2022123640 A1 WO 2022123640A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- monitored
- signal
- trigger condition
- time
- trigger
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 86
- 230000005856 abnormality Effects 0.000 title claims description 45
- 238000005520 cutting process Methods 0.000 claims abstract description 70
- 238000010801 machine learning Methods 0.000 claims abstract description 14
- 238000012544 monitoring process Methods 0.000 claims description 43
- 238000003745 diagnosis Methods 0.000 claims description 39
- 238000003066 decision tree Methods 0.000 claims description 18
- 238000010586 diagram Methods 0.000 description 11
- 238000004519 manufacturing process Methods 0.000 description 7
- 238000003860 storage Methods 0.000 description 7
- 230000002159 abnormal effect Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 230000000630 rising effect Effects 0.000 description 3
- 238000004513 sizing Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000005098 hot rolling Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000003825 pressing Methods 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000009628 steelmaking Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M99/00—Subject matter not provided for in other groups of this subclass
- G01M99/005—Testing of complete machines, e.g. washing-machines or mobile phones
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0221—Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
- G05B23/024—Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
Definitions
- the present invention relates to a method for determining a trigger condition for a time-series signal, a method for diagnosing an abnormality in equipment to be monitored, and a device for determining a trigger condition for a time-series signal.
- the abnormality diagnosis is specifically performed by the following procedure.
- a time-series signal at point N is cut out from the collected monitored signal.
- the cut out time-series signal is expressed at one point in the N-dimensional space, and a reference is created.
- (3) Calculate the Q statistic (distance from the principal component plane) for the signal that is the target of the abnormality diagnosis.
- the Q statistic exceeds a predetermined threshold value, it is determined to be abnormal.
- the monitored section is cut out from the monitored signal, and a model is created when the monitored equipment is operating normally.
- the monitored section is cut out from the monitored signal, and the distance from the model created from the monitored signal during normal operation is calculated.
- the condition for specifying the timing for cutting out the monitored section from the monitored signal is called a "trigger condition”
- the time-series signal that can be this trigger condition is called a "trigger candidate signal”.
- the cut out waveform is obtained when the monitored section is cut out from multiple monitored signals collected for the monitored equipment and overlapped. Must overlap to some extent.
- the waveforms of the monitored sections are superposed by the following procedure. (1) The monitored section is cut out from each of a plurality of monitored signals using the trigger condition. (2) Convert the horizontal axis of each monitored section (for example, convert the time on the horizontal axis into a crank angle or the like). (3) Convert (for example, normalize) the vertical axis of each monitored section.
- the trigger condition of (1) above is manually determined by comparing, for example, the monitored signal and the trigger candidate signal so that the cut out waveforms overlap. Therefore, the trigger condition is determined. It took time and effort to make a decision.
- the present invention has been made in view of the above, and is a method for determining a trigger condition for a time-series signal capable of automatically determining a trigger condition for cutting out a monitored section from a monitored signal, and a monitoring target equipment. It is an object of the present invention to provide an abnormality diagnosis method and a device for determining a trigger condition for a time series signal.
- the method for determining the trigger condition of the time-series signal is a time-series signal indicating the state of the monitored equipment when performing an abnormality diagnosis of the monitored equipment.
- the method for determining the trigger condition of the time-series signal which is the condition for cutting out the monitored section to be the target of the abnormality diagnosis from the monitored signal, one or more monitored signals related to the monitored equipment.
- the collection step to be performed, the cutting step of cutting out the monitored section of the monitored signal based on a predetermined standard for the signal group, and the start time of the cut out monitoring target section of the signal group are specified.
- a learning model that generates label data with the start time label turned on and other times turned off, inputs the trigger candidate signal of 1 or more at each time, and outputs the label data at each time. It is characterized by including a model generation step generated by learning and a trigger condition determination step of determining the trigger condition using the learning model for the monitored signal for performing the abnormality diagnosis.
- the method for determining the trigger condition of the time-series signal according to the present invention is the first monitored signal selected by the cutting step from a plurality of monitored signals collected in the collecting step in the above invention.
- the monitoring target section is cut out based on the equipment characteristics of the monitoring target equipment, and the monitoring target signal other than the first monitoring target signal among the plurality of monitoring target signals is the monitoring target section of the first monitoring target signal. It is characterized in that each monitored section is cut out by searching for a section having the largest correlation coefficient with the waveform included in.
- the method for determining the trigger condition of the time-series signal according to the present invention is characterized in that, in the above invention, the learning model is a decision tree.
- the trigger candidate signal is a one-pulse signal in the model generation step
- the trigger candidate signal is converted into a sawtooth wave and then the machine. It is characterized by learning.
- the method for determining the trigger condition of the time-series signal according to the present invention returns to the cutting step and previously cuts out when the discrimination accuracy cannot be obtained at the time of machine learning in the model generation step. It is characterized in that the monitored section is shifted back and forth, the monitored section of the monitored signal is newly cut out, and then the model generation step is performed again.
- the abnormality diagnosis method of the monitored equipment is the state of the monitored equipment according to the trigger condition determined by the above-mentioned method for determining the trigger condition of the time series signal. From the monitored signal, which is a time-series signal indicating the above, the signal of the monitored section to be the target of the abnormality diagnosis is cut out and accumulated, and the abnormality diagnosis of the monitored equipment is performed based on the accumulated signal. ..
- the time-series signal trigger condition determining device is a time-series signal indicating the state of the monitored equipment when performing an abnormality diagnosis of the monitored equipment.
- the collection means to be collected, the cutting means for cutting out the monitored section of the monitored signal for the signal group, and the start time of the cut out monitoring target section for the signal group are specified.
- a learning model that generates label data with the start time label turned on and other times turned off, inputs the trigger candidate signal of one or more at each time, and outputs the label data at each time. It is characterized by comprising a model generating means for generating the above by machine learning, and a trigger condition determining means for determining the trigger condition using the learning model for the monitored signal for performing the abnormality diagnosis.
- a trigger condition for cutting out a monitored section from a monitored signal by using a learning model that learns under what conditions the monitored signal and the trigger candidate signal turn on the trigger can be determined automatically.
- FIG. 1 is a block diagram showing a schematic configuration of a time-series signal trigger condition determining device according to an embodiment of the present invention.
- FIG. 2 is a flowchart showing a flow of a method for determining a trigger condition of a time-series signal according to an embodiment of the present invention.
- FIG. 3 is a diagram schematically showing the content of the cutting process of the method for determining the trigger condition of the time-series signal according to the embodiment of the present invention.
- FIG. 4 is a diagram schematically showing the content of the cutting process of the method for determining the trigger condition of the time-series signal according to the embodiment of the present invention.
- FIG. 1 is a block diagram showing a schematic configuration of a time-series signal trigger condition determining device according to an embodiment of the present invention.
- FIG. 2 is a flowchart showing a flow of a method for determining a trigger condition of a time-series signal according to an embodiment of the present invention.
- FIG. 3 is a diagram schematic
- FIG. 5 is a diagram schematically showing the contents of the model generation step of the method for determining the trigger condition of the time series signal according to the embodiment of the present invention.
- FIG. 6 is a diagram schematically showing a decision tree generated in the model generation step of the method for determining the trigger condition of the time series signal according to the embodiment of the present invention.
- FIG. 7 is a diagram schematically showing how a one-pulse signal is converted into a sawtooth wave in a model generation step of a method for determining a trigger condition of a time-series signal according to an embodiment of the present invention.
- FIG. 8 is a diagram schematically showing a sawtooth wave converted in the model generation step of the method for determining the trigger condition of the time series signal according to the embodiment of the present invention.
- FIG. 9 is a diagram showing a monitored signal and a trigger candidate signal in an embodiment of the time-series signal trigger condition determination method according to the embodiment of the present invention.
- FIG. 10 is a diagram showing a state in which the waveforms of the cut-out monitoring target sections are superimposed in the embodiment of the method for determining the trigger condition of the time-series signal according to the embodiment of the present invention.
- FIG. 11 is a diagram showing a method of determining a trigger condition using a decision tree in an embodiment of the method of determining a trigger condition of a time-series signal according to an embodiment of the present invention.
- FIG. 12 is a diagram showing a state in which the waveforms of the cut-out monitoring target sections are superimposed in the embodiment of the time-series signal trigger condition determination method according to the embodiment of the present invention.
- time-series signal trigger condition determination method the monitored equipment abnormality diagnosis method, and the time-series signal trigger condition determination device (hereinafter referred to as “learning device”) according to the embodiment of the present invention will be described with reference to the drawings. ..
- the trigger condition determination device determines the trigger condition, which is a condition for cutting out the monitored section from the monitored signal when performing an abnormality diagnosis of the monitored equipment in the production equipment such as a factory and the experimental equipment such as a research institute. It is a device.
- the monitored signal is a time-series signal indicating the state of the monitored signal, as described above.
- the monitored signal differs depending on the type of the monitored equipment. For example, when the monitored equipment is a "motor", the current or speed of the motor is used as the monitored signal.
- the trigger condition determination device constantly collects one or more monitoring target signals and corresponding trigger candidate signals.
- the trigger candidate signal is a time-series signal related to the monitored equipment, and indicates a time-series signal detected at the same time as the monitored signal.
- the trigger candidate signal is this condition, and is, for example, a signal such as On or Off. If this trigger condition is known in advance, it is easy to cut out the monitored section. However, when there are many diverse facilities and they operate in a complicated manner, it may not be possible to easily determine this trigger condition.
- the trigger candidate signal that directly indicates the trigger condition may not always be fetched into the database, and only the signal in the indirect form may exist.
- the signal group that defines the equipment operating conditions is selected as the trigger candidate signal, the operation start rule is extracted from the history of those signals by machine learning, etc., and the conditions for cutting out the monitored section are determined.
- the trigger candidate signal is preferably a signal that is not directly related to an abnormality of the target process, equipment, or the like, and a signal indicating various command values or ON / OFF of a specific event is a candidate.
- the monitored signal itself may be included in the trigger candidate signal.
- the monitored section indicates a section of the monitored signal to be cut out for performing abnormality diagnosis.
- cutting start time the time to start cutting out the monitored signal
- the width of cutting out is specified by the value of the trigger candidate signal collected at the same time as the monitored signal.
- the width of the cutout differs depending on the type of the equipment to be monitored. For example, when the equipment to be monitored is a "motor", the section in which the motor accelerates may be specified as the width of the cutout. Alternatively, if it is a product manufacturing process, it may be the width of the section from the start of manufacturing to the end of manufacturing.
- the trigger condition is a condition for cutting out a monitoring target section to be a target of abnormality diagnosis from the monitoring target signal, and specifically, the above-mentioned cutting start time and cutting width of the monitoring target signal. Shows.
- the trigger condition determination device 1 is realized by a general-purpose information processing device such as a personal computer or a workstation, and includes an input unit 10, an output unit 20, a storage unit 30, and a calculation unit 40. There is.
- the input unit 10 is an input means for the calculation unit 40, and is realized by a data acquisition device, a keyboard, a pointing device, or the like. Further, the output unit 20 is realized by a liquid crystal display or the like.
- the storage unit 30 is realized by a hard disk device or the like. In the storage unit 30, for example, data processed by the calculation unit 40 (monitoring target signal, trigger candidate signal, trigger condition, learning model, etc.) is stored.
- the arithmetic unit 40 is realized by, for example, a processor composed of a CPU (Central Processing Unit) or the like, and a memory (main storage unit) composed of a RAM (Random Access Memory), a ROM (Read Only Memory), or the like.
- the arithmetic unit 40 loads a program into the work area of the main storage unit and executes it, and controls each component unit or the like through the execution of the program to realize a function that meets a predetermined purpose.
- the arithmetic unit 40 has a collecting unit (collecting means) 41, a cutting unit (cutting means) 42, a model generating unit (model generating means) 43, and a trigger condition determining unit (trigger condition determining means) through the execution of the above-mentioned program. Functions as 44. Details of each part will be described later (see FIGS. 2 to 8).
- Trigger condition determination method The method for determining the trigger condition according to the present embodiment will be described with reference to FIGS. 2 to 8.
- the collection process, the cutting process, the model generation process, and the trigger condition determination process are performed in this order. Further, in the trigger condition determination method, as will be described later, the cutting step and the model generation step are repeated as necessary.
- the collecting unit 41 collects a signal group including a monitored signal and a trigger candidate signal (step S1). Although the case where the collection unit 41 collects a plurality of monitoring target signals will be described here, the collection unit 41 may collect only one monitoring target signal.
- the cutting unit 42 cuts out the monitored section of the monitored signal based on a predetermined standard for the signal group collected in the collecting step.
- the cutting unit 42 first roughly cuts out a signal group as shown in FIG. 3 (step S2). For example, in equipment that repeatedly operates, such as coil rolling equipment, monitoring target signals and trigger candidate signals are continuously acquired for continuously flowing coils. Therefore, in step S2, for example, in order to divide the monitored signal and the trigger candidate signal for each coil, a rough cutout of the signal group is performed.
- the rough cutting of the signal group may be performed at a preset timing according to the type of the equipment to be monitored, or as shown in the figure, a rough cutting signal is selected from a plurality of trigger candidate signals. , It may be performed at the timing when the rough cutting signal rises.
- reference numeral Sg is a signal group before rough cutting
- reference numeral Ss is a signal to be monitored before rough cutting
- reference numeral St is a trigger candidate signal before rough cutting
- reference numerals Sg1, Sg2, and Sg3 are signals after rough cutting.
- the group, reference numeral Ss1, Ss2, Ss3 indicates a monitoring target signal after rough cutting, and reference numerals St1, St2, St3 indicate a trigger candidate signal after rough cutting.
- the cutting unit 42 selects one monitoring target signal (monitoring target signal Ss1 in the figure) from the plurality of monitoring target signals roughly cut out in step S2.
- the monitoring target section Sm1 of the selected monitoring target signal Ss1 is cut out (step S3).
- the cutting condition for cutting out the monitored section Sm1 in step S3 is determined based on the equipment characteristics of the monitored equipment. For example, when the monitored equipment is a "motor” and the monitored signal Ss1 is a "motor current value", it is determined whether or not the increase in the motor current value when the motor accelerates is normal.
- the section in which the motor accelerates is set as the monitored section Sm1. That is, the time point at which the motor starts accelerating is specified as the cutout start time of the monitored section Sm1, and the section from the start of acceleration to the end of acceleration is specified as the cutout width.
- the cutting unit 42 calculates the correlation coefficient between the waveform included in the monitored section Sm1 cut out in step S3 and the waveform included in the other monitored signals Ss2 and Ss3 (step S4).
- the cutting unit 42 has the largest correlation coefficient with the waveform included in the monitoring target section Sm1 cut out in step S3 for the other monitoring target signals Ss2 and Ss3.
- the monitored sections Sm2 and Sm3 of the monitored signals Ss2 and Ss3 are cut out, respectively (step S5).
- steps S4 and S5 a waveform similar to the waveform included in the monitored section Sm1 cut out in step S3 is searched for from the waveforms included in the monitored signals at other times roughly cut out in step S2. do.
- the Euclidean distance between the data of each time-series signal may be used.
- Model generation process For each signal group (multiple monitored signals), the start time of the monitored section to be cut out is specified in advance, the label of the start time is turned on, and the label data is turned off at other times.
- a learning model is generated by machine learning, in which each value of one or more trigger candidate signals at each time is input and the label data at each time is output.
- the model generation unit 43 first monitors the monitoring target signals Ss1, p2, p3 corresponding to the start times p1, p2, p3 of the monitoring target sections Sm1, Sm2, Sm3 cut out for each signal group.
- the values of Ss2 and Ss3 and the values of the trigger candidate signals St1, St2 and St3 (hereinafter referred to as "values of the signal group") are labeled as "trigger ON", and the monitored sections Sm1, Sm2 and Sm3 are cut out.
- the value of the signal group corresponding to the time other than the start time of is given the label of "trigger OFF" (step S6).
- the label of "trigger ON” indicates that the value of the signal group to which this label is attached is the cutting start time
- the label of "trigger OFF” indicates that the value of the signal group to which this label is attached. It indicates that the value is not the cutout start time.
- the model generation unit 43 inputs the value of the signal group labeled with "trigger ON” and the value of the signal group labeled with “trigger OFF”, and inputs the label of "trigger ON” and "trigger ON”. As shown in FIG. 6, a decision tree is generated by machine learning using the label of "trigger OFF" as an output (step S7).
- learning data in which the objective variable is labeled as "trigger ON” and “trigger OFF” and the value of each trigger candidate signal corresponding to each time of “trigger ON” and “trigger OFF” is used as an explanatory variable.
- “trigger ON” may be set to “1” and “trigger OFF” may be set to "0” to be treated as a function.
- the learning model generated in step S7 is not limited to the decision tree, and may be, for example, a random forest or a neural network.
- “trigger ON” may be set to "1” and “trigger OFF” may be set to "0” to be treated as a function.
- the trigger candidate signal included in the signal group is a one-pulse signal, that is, as shown in the upper figure of FIG. 7, only one scan of the rising or falling ends of the ON-OFF signal.
- machine learning is performed after converting the trigger candidate signal into a sawtooth wave, as shown in the lower figure of the figure.
- a 1-pulse signal is a signal that is turned on only for a short time. Therefore, in the above-mentioned cutting step, when searching for a portion having a high degree of similarity in waveform, one pulse signal should be turned on at the time of "trigger ON", but the time when "trigger ON” is set. However, it may shift before and after the time when the 1-pulse signal is turned on. On the other hand, as shown in the figure, by converting the 1-pulse signal into a sawtooth wave, it is possible to solve the problem caused by the ON / OFF delay of the 1-pulse signal.
- the slope of the sawtooth wave after conversion is determined, for example, by how many seconds after the one-pulse signal is turned on, and is set so that it does not overlap with the rising edge of the next signal. Further, when converting a 1-pulse signal into a sawtooth wave, as shown in part A of FIG. 7, the cut-out of the monitored section is started before the 1-pulse signal is turned on due to the deviation of the cutting start point. It is desirable to have a margin (for example, about 5 scans) for the accident.
- the shape of the sawtooth wave is defined by the parameters tf and tb as shown in FIG. 8, and it is desirable that the relationship between the parameters tf and tb is tb ⁇ tf. Further, in the sawtooth wave, as shown in the figure, if the value z of the converted signal is within the range indicated by B, it is determined that the trigger is ON.
- the process if an error occurs during machine learning that the trigger condition cannot be generated normally, or if the discrimination accuracy cannot obtain a predetermined value, the process returns to the above-mentioned cutting process and the monitoring that was cut out last time is performed. The target section is shifted back and forth, the monitored section of the monitored signal is newly cut out, and then the model generation step is performed again. That is, after re-cutting out the monitored section of the monitored signal, the decision tree is constructed again. Then, if the trigger condition can be normally generated during machine learning, the model generation process is terminated, and if it cannot be normally generated, the process returns to the cutting process again, and the cutting process and the model generation process are repeated.
- the trigger condition determination method is a method of learning the state of the trigger candidate signal at the start time of the monitored section (monitored section Sm1) first specified in the cutting process. .. Therefore, for example, when the state of the trigger candidate signal at the start time of the monitored section designated first has no characteristic, learning cannot be performed well. Therefore, as described above, if an error occurs in the model generation process, the monitoring target section initially specified in the cutting process is shifted back and forth, and the monitoring target section is redesignated to prevent problems during learning. It can be resolved.
- the trigger condition determination unit 44 determines the trigger condition for the monitored signal for abnormal diagnosis using the decision tree (step S8). That is, the trigger condition determination unit 44 can know the timing of "trigger ON” by inputting the trigger candidate signal into the decision tree, and can extract the timing of monitoring start of the monitored signal for abnormal diagnosis. can.
- the trigger condition can be briefly described by rearranging and rewriting the conditions of the generated decision tree (see FIG. 11 described later). In this way, by using the decision tree generated in step S7, it is possible to easily grasp under what conditions the monitored signal and the trigger candidate signal are triggered. Even when a learning model other than the decision tree is used, the timing at which the trigger candidate signal at each time, which is an explanatory variable, is sequentially input and the output of "trigger ON" is obtained is set as the timing of "trigger ON”. good.
- the abnormality diagnosis method of the monitored equipment is to cut out the signal of the monitored section from the monitored signal and store it in the storage unit 30 according to the trigger condition determined by the trigger condition determination method described above, and to monitor based on the stored signal. Perform equipment abnormality diagnosis.
- the trigger condition determination device, the trigger condition determination method, and the abnormality diagnosis method of the monitored equipment according to the present embodiment as described above the trigger is turned on under what conditions the monitored signal and the trigger candidate signal are.
- the trigger condition for cutting out the monitored section from the monitored signal can be automatically determined.
- the trigger condition for cutting out the start target section of the monitored signal can be automatically determined. It is no longer necessary to manually examine and determine the trigger conditions, and it is possible to simplify the advance preparation required when diagnosing an abnormality in the monitored equipment.
- the sizing press equipment is set as the equipment to be monitored, and the present invention is applied to the actual speed of the main motor of the sizing press equipment.
- the monitored signal and the trigger candidate signal in this embodiment are shown in FIG.
- the monitored signal is the actual speed of the traction motor, and the following five trigger candidate signals are selected.
- v1 ON when the press is loaded
- v2 Speed command value of traction motor
- v3 Speed command value of width adjustment motor (lower side of drive)
- v4 ON during pressing
- v5 Sum load of press load cell
- FIG. 9 shows the section of 2500 to 2800 scan in order to see the rising edge of the signal.
- FIG. 10 shows a graph obtained by cutting out and superimposing a portion of the waveforms included in the monitored signals at other times where the correlation coefficient becomes maximum
- FIG. 11 shows a decision tree constructed in the model generation process.
- FIG. 12 shows a graph in which monitored signals (actual speed of the traction motor) are cut out and superimposed based on the trigger conditions extracted from the decision tree in the decision process.
- the monitored equipment is a steelmaking process, particularly a sizing press equipment in a hot rolling mill, has been described, but the scope of application of the present invention is not limited to this field, and petroleum-related products and chemicals. It can be applied to production equipment for all manufacturing processes such as chemicals and experimental equipment for research institutions.
- the method for determining the trigger condition for the time-series signal, the method for diagnosing the abnormality of the monitored equipment, and the device for determining the trigger condition for the time-series signal according to the present invention have been specifically described above with reference to the embodiments and examples for carrying out the invention.
- the gist of the present invention is not limited to these descriptions, and must be broadly interpreted based on the description of the scope of claims. Needless to say, various changes, modifications, etc. based on these descriptions are also included in the gist of the present invention.
- Trigger condition determination device 10
- Input unit 20
- Output unit 30
- Storage unit 40
- Calculation unit 41
- Collection unit (collection means) 42
- Cutting part (cutting means) 43
- Model generation unit (model generation means) 44
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Automation & Control Theory (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Computational Linguistics (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
Description
(1)収集した監視対象信号からN点の時系列信号を切り出す。
(2)切り出した時系列信号をN次元空間の一点に表現し、基準を作成する。
(3)異常診断の対象となる信号について、Q統計量(主成分平面との距離)を算出する。
(4)Q統計量が予め定めた閾値を超えた場合に異常と判定する。
(1)トリガ条件を用いて複数の監視対象信号からそれぞれ監視対象区間を切り出す。
(2)それぞれの監視対象区間の横軸を変換する(例えば横軸の時間をクランク角等に変換する)。
(3)それぞれの監視対象区間の縦軸を変換する(例えば正規化する)。
トリガ条件決定装置は、工場等の生産設備および研究所等の実験設備において、監視対象設備の異常診断を行う際に、監視対象信号から監視対象区間を切り出すための条件であるトリガ条件を決定する装置である。
本実施形態に係るトリガ条件決定方法について、図2~図8を参照しながら説明する。トリガ条件決定方法は、収集工程と、切り出し工程と、モデル生成工程と、トリガ条件決定工程と、をこの順で行う。また、トリガ条件決定方法では、後記するように、必要に応じて切り出し工程およびモデル生成工程を繰り返す。
収集工程では、収集部41が、監視対象信号およびトリガ候補信号からなる信号群を収集する(ステップS1)。なお、ここでは収集部41が複数の監視対象信号を収集する場合について説明するが、収集部41が収集する監視対象信号は1つでもよい。
切り出し工程では、切り出し部42が、収集工程で収集された信号群について、所定の基準に基づいて、監視対象信号の監視対象区間を切り出す。以下、切り出し工程の詳細について説明する。
モデル生成工程では、各信号群(複数の監視対象信号)について、切り出したい監視対象区間の開始時刻を予め特定し、当該開始時刻のラベルをONとし、それ以外の時刻をOFFとするラベルデータを生成し、各時刻の1以上のトリガ候補信号の各値を入力とし、各時刻のラベルデータを出力とする学習モデルを、機械学習により生成する。
トリガ条件決定工程では、トリガ条件決定部44が、異常診断を行う監視対象信号について、決定木を用いてトリガ条件を決定する(ステップS8)。すなわち、トリガ条件決定部44は、トリガ候補信号を決定木に入力することにより、「トリガON」のタイミングを知ることができ、異常診断を行う監視対象信号の監視開始のタイミングを抽出することができる。ここで、ステップS8で決定木を利用する場合は、生成された決定木の条件を整理して記述し直すことで、トリガ条件を簡潔に記述することができる(後記する図11参照)。このように、ステップS7で生成した決定木を用いることにより、監視対象信号およびトリガ候補信号がどのような条件のときにトリガONになるのかを容易に把握することができる。また、決定木以外の学習モデルを用いる場合でも、説明変数である各時刻のトリガ候補信号を順次入力して、「トリガON」の出力が得られるタイミングを、「トリガON」のタイミングとすればよい。
監視対象設備の異常診断方法は、前記したトリガ条件決定方法によって決定されたトリガ条件に従って、監視対象信号から監視対象区間の信号を切り出して記憶部30に蓄積し、蓄積した信号に基づいて監視対象設備の異常診断を行う。
v1:プレス在荷でON
v2:主電動機の速度指令値
v3:幅アジャスト電動機(ドライブ下側)の速度指令値
v4:プレス中にON
v5:プレスロードセルの和荷重
10 入力部
20 出力部
30 記憶部
40 演算部
41 収集部(収集手段)
42 切り出し部(切り出し手段)
43 モデル生成部(モデル生成手段)
44 トリガ条件決定部(トリガ条件決定手段)
Claims (7)
- 監視対象設備の異常診断を行う際に、前記監視対象設備の状態を示す時系列信号である監視対象信号から、前記異常診断の対象となる監視対象区間を切り出すための条件であるトリガ条件を決定する時系列信号のトリガ条件決定方法において、
前記監視対象設備に関する1以上の監視対象信号と、前記監視対象設備に関連し、かつ前記監視対象信号と同時刻に検出された時系列信号であって、前記トリガ条件となりうる時系列信号を示すトリガ候補信号と、からなる信号群を収集する収集工程と、
前記信号群について、所定の基準に基づいて、前記監視対象信号の監視対象区間を切り出す切り出し工程と、
前記信号群について、切り出した前記監視対象区間の開始時刻を特定し、該開始時刻のラベルをONとし、それ以外の時刻をOFFとするラベルデータを生成し、各時刻の1以上の前記トリガ候補信号を入力とし、各時刻の前記ラベルデータを出力とする学習モデルを機械学習により生成するモデル生成工程と、
前記異常診断を行う監視対象信号について、前記学習モデルを用いて前記トリガ条件を決定するトリガ条件決定工程と、
を含むことを特徴とする時系列信号のトリガ条件決定方法。 - 前記切り出し工程は、
前記収集工程で収集された複数の監視対象信号の中から選択した第一の監視対象信号について、前記監視対象設備の設備特性に基づいて監視対象区間を切り出し、
前記複数の監視対象信号のうちの前記第一の監視対象信号以外の監視対象信号について、前記第一の監視対象信号の監視対象区間に含まれる波形との相関係数が最も大きい区間を探索することにより、監視対象区間をそれぞれ切り出すことを特徴とする請求項1に記載の時系列信号のトリガ条件決定方法。 - 前記学習モデルは、決定木であることを特徴とする請求項1または請求項2に記載の時系列信号のトリガ条件決定方法。
- 前記モデル生成工程において、前記トリガ候補信号が1パルス信号である場合、前記トリガ候補信号をのこぎり波に変換した後に機械学習することを特徴とする請求項1から請求項3のいずれか一項に記載の時系列信号のトリガ条件決定方法。
- 前記モデル生成工程において機械学習の際に判別精度が所定の値を得られない場合、前記切り出し工程に戻り、前回切り出した監視対象区間を前後にシフトさせ、前記監視対象信号の監視対象区間を新たに切り出した後、前記モデル生成工程を再度行うことを特徴とする請求項1から請求項4のいずれか一項に記載の時系列信号のトリガ条件決定方法。
- 請求項1から請求項5のいずれか一項に記載の時系列信号のトリガ条件決定方法によって決定されたトリガ条件に従って、監視対象設備の状態を示す時系列信号である監視対象信号から、前記異常診断の対象となる監視対象区間の信号を切り出して蓄積し、蓄積した信号に基づいて前記監視対象設備の異常診断を行うことを特徴とする監視対象設備の異常診断方法。
- 監視対象設備の異常診断を行う際に、前記監視対象設備の状態を示す時系列信号である監視対象信号から、前記異常診断の対象となる監視対象区間を切り出すための条件であるトリガ条件を決定する時系列信号のトリガ条件決定装置において、
前記監視対象設備に関する1以上の監視対象信号と、前記監視対象設備に関連し、かつ前記監視対象信号と同時刻に検出された時系列信号であって、前記トリガ条件となりうる時系列信号を示すトリガ候補信号と、からなる信号群を収集する収集する収集手段と、
前記信号群について、所定の基準に基づいて、前記監視対象信号の監視対象区間を切り出す切り出し手段と、
前記信号群について、切り出した前記監視対象区間の開始時刻を特定し、該開始時刻のラベルをONとし、それ以外の時刻をOFFとするラベルデータを生成し、各時刻の1以上の前記トリガ候補信号を入力とし、各時刻の前記ラベルデータを出力とする学習モデルを機械学習により生成するモデル生成手段と、
前記異常診断を行う監視対象信号について、前記学習モデルを用いて前記トリガ条件を決定するトリガ条件決定手段と、
を備えることを特徴とする時系列信号のトリガ条件決定装置。
Priority Applications (5)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP20965026.6A EP4261639A4 (en) | 2020-12-08 | 2020-12-08 | METHOD FOR DETERMINING A TIME SERIES SIGNAL TRIGGER CONDITION, METHOD FOR DIAGNOSIS OF AN ANOMALY IN A MONITORED INSTALLATION, AND DEVICE FOR DETERMINING A TIME SERIES SIGNAL TRIGGER CONDITION |
KR1020237018480A KR20230098321A (ko) | 2020-12-08 | 2020-12-08 | 시계열 신호의 트리거 조건 결정 방법, 감시 대상 설비의 이상 진단 방법 및 시계열 신호의 트리거 조건 결정 장치 |
PCT/JP2020/045563 WO2022123640A1 (ja) | 2020-12-08 | 2020-12-08 | 時系列信号のトリガ条件決定方法、監視対象設備の異常診断方法および時系列信号のトリガ条件決定装置 |
CN202080107703.8A CN116601576A (zh) | 2020-12-08 | 2020-12-08 | 时间序列信号的触发条件决定方法、监视对象设备的异常诊断方法以及时间序列信号的触发条件决定装置 |
US18/039,799 US20240027304A1 (en) | 2020-12-08 | 2020-12-08 | Trigger condition determination method for time series signal, abnormality diagnosis method for equipment to be monitored, and trigger condition determination device for time series signal |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2020/045563 WO2022123640A1 (ja) | 2020-12-08 | 2020-12-08 | 時系列信号のトリガ条件決定方法、監視対象設備の異常診断方法および時系列信号のトリガ条件決定装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022123640A1 true WO2022123640A1 (ja) | 2022-06-16 |
Family
ID=81973337
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2020/045563 WO2022123640A1 (ja) | 2020-12-08 | 2020-12-08 | 時系列信号のトリガ条件決定方法、監視対象設備の異常診断方法および時系列信号のトリガ条件決定装置 |
Country Status (5)
Country | Link |
---|---|
US (1) | US20240027304A1 (ja) |
EP (1) | EP4261639A4 (ja) |
KR (1) | KR20230098321A (ja) |
CN (1) | CN116601576A (ja) |
WO (1) | WO2022123640A1 (ja) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017104305A1 (ja) | 2015-12-16 | 2017-06-22 | 三菱電機株式会社 | 故障診断装置および故障診断方法 |
WO2020136836A1 (ja) * | 2018-12-27 | 2020-07-02 | 株式会社Ihi | 異常診断装置、異常診断方法、異常診断プログラム、及び記録媒体 |
JP2020197980A (ja) * | 2019-06-04 | 2020-12-10 | Jfeスチール株式会社 | 時系列信号のトリガ条件決定方法、監視対象設備の異常診断方法および時系列信号のトリガ条件決定装置 |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100250195A1 (en) * | 2009-03-27 | 2010-09-30 | General Electric Company | Systems and Methods for Providing Dynamic Segmental Signal Analysis |
JP5431235B2 (ja) * | 2009-08-28 | 2014-03-05 | 株式会社日立製作所 | 設備状態監視方法およびその装置 |
US10565513B2 (en) * | 2016-09-19 | 2020-02-18 | Applied Materials, Inc. | Time-series fault detection, fault classification, and transition analysis using a K-nearest-neighbor and logistic regression approach |
JP6670418B2 (ja) * | 2018-02-26 | 2020-03-18 | 株式会社日立情報通信エンジニアリング | 状態予測装置および状態予測制御方法 |
JP6763910B2 (ja) * | 2018-05-30 | 2020-09-30 | 横河電機株式会社 | 異常検出装置、異常検出方法、異常検出プログラム及び記録媒体 |
CN111309539A (zh) * | 2020-03-26 | 2020-06-19 | 北京奇艺世纪科技有限公司 | 一种异常监测方法、装置和电子设备 |
-
2020
- 2020-12-08 EP EP20965026.6A patent/EP4261639A4/en active Pending
- 2020-12-08 KR KR1020237018480A patent/KR20230098321A/ko unknown
- 2020-12-08 CN CN202080107703.8A patent/CN116601576A/zh active Pending
- 2020-12-08 US US18/039,799 patent/US20240027304A1/en active Pending
- 2020-12-08 WO PCT/JP2020/045563 patent/WO2022123640A1/ja active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017104305A1 (ja) | 2015-12-16 | 2017-06-22 | 三菱電機株式会社 | 故障診断装置および故障診断方法 |
WO2020136836A1 (ja) * | 2018-12-27 | 2020-07-02 | 株式会社Ihi | 異常診断装置、異常診断方法、異常診断プログラム、及び記録媒体 |
JP2020197980A (ja) * | 2019-06-04 | 2020-12-10 | Jfeスチール株式会社 | 時系列信号のトリガ条件決定方法、監視対象設備の異常診断方法および時系列信号のトリガ条件決定装置 |
Non-Patent Citations (1)
Title |
---|
See also references of EP4261639A4 |
Also Published As
Publication number | Publication date |
---|---|
CN116601576A (zh) | 2023-08-15 |
EP4261639A1 (en) | 2023-10-18 |
US20240027304A1 (en) | 2024-01-25 |
KR20230098321A (ko) | 2023-07-03 |
EP4261639A4 (en) | 2024-01-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6386520B2 (ja) | 数値制御装置及び機械学習装置 | |
EP2363767A1 (en) | Process analysis system | |
JP5179086B2 (ja) | 工業プロセスの監視方法及び監視システム | |
US20150276557A1 (en) | State monitoring system, state monitoring method and medium | |
JP6895816B2 (ja) | 異常診断装置、異常診断方法及び異常診断プログラム | |
WO2019087508A1 (ja) | 監視対象選定装置、監視対象選定方法、およびプログラム | |
JP7003970B2 (ja) | 時系列信号のトリガ条件決定方法、監視対象設備の異常診断方法および時系列信号のトリガ条件決定装置 | |
CN112711850A (zh) | 一种基于大数据的机组在线监测方法 | |
WO2022123640A1 (ja) | 時系列信号のトリガ条件決定方法、監視対象設備の異常診断方法および時系列信号のトリガ条件決定装置 | |
WO1996005555A1 (fr) | Dispositif d'inference de causes | |
JP7251955B2 (ja) | 検出装置及び機械学習方法 | |
JP3567073B2 (ja) | 時系列データ予測方法および装置 | |
TWI765452B (zh) | 時間序列訊號之觸發條件決定方法、監視對象設備之異常診斷方法及時間序列訊號之觸發條件決定裝置 | |
WO2022123876A1 (ja) | 異常判定モデル生成装置、異常判定装置、異常判定モデル生成方法および異常判定方法 | |
WO2022123877A1 (ja) | 異常判定装置、異常判定モデル生成方法および異常判定方法 | |
CN112783935A (zh) | 分析装置 | |
CN113093584A (zh) | 焊接异常检测方法、装置、计算机设备和存储介质 | |
JPH08110262A (ja) | オンライン回転体診断装置 | |
CN114004360B (zh) | 基于模糊专家模型的智能装备诊断设备和方法 | |
EP3951530A1 (en) | Process state monitoring device and process state monitoring method | |
Khadem Hosseini et al. | A Projection-Based Support Vector Machine Algorithm for Induction Motors’ Bearing Fault Detection | |
CN117895640A (zh) | 一种适用于电网自动化维护的监控方法及系统 | |
CN114519231A (zh) | 航空发动机故障诊断方法和装置、计算机装置和存储介质 | |
JPH10225443A (ja) | 心電図データ解析装置 | |
JPH06261581A (ja) | 故障診断用推論装置 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20965026 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 20237018480 Country of ref document: KR Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18039799 Country of ref document: US |
|
WWE | Wipo information: entry into national phase |
Ref document number: 202080107703.8 Country of ref document: CN |
|
REG | Reference to national code |
Ref country code: BR Ref legal event code: B01A Ref document number: 112023010951 Country of ref document: BR |
|
ENP | Entry into the national phase |
Ref document number: 112023010951 Country of ref document: BR Kind code of ref document: A2 Effective date: 20230602 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2020965026 Country of ref document: EP Effective date: 20230710 |
|
NENP | Non-entry into the national phase |
Ref country code: JP |