Equipment diagnostic system

Abstract

To solve the problem that an analysis process cannot be divided easily according to torque, speed change, etc.SOLUTION: A machine diagnostic system includes: a threshold calculation unit that sets multiple thresholds for reference data, using part of time-series data measured during operation of a machine as the reference data; a data division unit that divides the time-series data that meet conditions determined by the multiple thresholds as statuses for respective ranges; a feature amount calculation unit that calculates a feature amount (basic statistics) from the time-series data for each status; and an abnormality degree calculation unit that analyzes and calculates an abnormality degree, which is the degree of abnormality in inspection data, for each status using the other part of the time-series data measured during the operation of the machine as the inspection data and using the feature amount for the inspection data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an equipment diagnostic system that uses sensors to collect and analyze data from equipment to diagnose the state of a machine, and in particular to an equipment diagnostic system that divides the process to be analyzed when performing the analysis. [Background technology]

[0002] An industrial equipment system is made up of several machines that are used to manufacture products. If one of the machines breaks down, it can have a major impact on operations due to an unplanned shutdown, resulting in huge losses.

[0003] For this reason, each machine is equipped with mechanical sensors and temperature sensors that measure mechanical physical quantities such as machine vibration and pressure, and machine data is collected and analyzed to monitor numerical values ​​related to the machine's condition. Furthermore, if the machine is driven by a motor, the machine's condition can also be detected by collecting and analyzing electrical data such as motor current using electrical sensors that measure electrical physical quantities such as current and voltage. Furthermore, moving objects such as automobiles, trains, and elevators also measure electrical and mechanical physical quantities, temperature, and even the position of the moving object.

[0004] In this way, conventionally, the state of the mechanical system is monitored by diagnosing the machine using sensors, and if the machine is likely to break down, repairs are made in advance to prevent unplanned shutdowns.

[0005] For example, Patent Document 1 proposes to grasp time-series data as feature data, determine a threshold for the feature data, and divide the raw data starting from peaks that exceed the threshold, thereby dividing the process into multiple peaks. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Publication No. 2020-135808 Summary of the Invention [Problem to be solved by the invention]

[0007] For example, in machines that perform tasks such as creating, removing, and receiving products, the operating patterns of the machine, such as its torque consumption, power, and speed, may change over time. When analyzing the data obtained to monitor the condition of such machines, dividing the process into processes in which the operating patterns are as uniform as possible makes it easier to identify the parameters you want to examine through analysis. However, when dividing the process manually, the time and effort required by the operator is a problem.

[0008] Furthermore, in the case of Patent Document 1, although it is possible to divide a periodic operation pattern, it does not deal with cases where torque, speed, etc. change depending on the operation pattern. Therefore, when a change in torque, speed, etc. occurs, it is time-consuming to divide the process separately.

[0009] As described above, the conventional technology has a problem in that it is not possible to easily divide the analysis process according to torque, speed changes, etc. [Means for solving the problem]

[0010] From the above, in the present invention, "a part of the time series data measured during the operation of the equipment" of As the reference data, The aforementioned Time series data measured during equipment operation Different from the reference data part of Inspection Data Let's say"An equipment diagnostic system comprising: a data selection unit; a threshold calculation unit that sets a plurality of thresholds for the reference data; a data division unit that divides the reference data that meets the conditions determined by the plurality of thresholds in the threshold calculation unit into status ranges; a feature calculation unit that calculates feature amounts that are basic statistics from the reference data for each status divided by the data division unit; and an abnormality degree calculation unit that uses the feature amounts for the inspection data to analyze and calculate an abnormality degree that is the degree of abnormality in the inspection data for each status." [Effects of the Invention]

[0011] According to the present invention, when analyzing time-series data in an equipment diagnostic system, it is possible to reduce the time and effort required to divide the data into processes according to changes in torque, power, speed, etc. This has the effect of reducing work time and personnel costs. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a configuration diagram of a device diagnosis system according to a first embodiment. [Figure 2] FIG. 1 is a conceptual diagram of a first embodiment. [Figure 3] FIG. 10 is a configuration diagram relating to data division and selection according to the second embodiment. [Figure 4] FIG. 10 is a diagram relating to data selection in the second embodiment. [Figure 5] This relates to the time series data and its division in the third embodiment. [Figure 6] FIG. 10 is a configuration diagram of a device diagnostic system according to a fourth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. [Example]

[0014] As a first embodiment, a device diagnostic system for industrial devices will be described with reference to Fig. 1 and Fig. 2. Fig. 1 shows an example of the configuration of a device diagnostic system according to a first embodiment of the present invention. Fig. 2 shows a conceptual diagram of the concept of the present invention.

[0015] Time-series data D measured in a device to be diagnosed is input to a data selection unit 1 of a device diagnosis system 10 in Fig. 1. In the data selection unit 1, a part of the input time-series data D is set as reference data, and the other part is set as test data.

[0016] Figure 2 shows the time on the horizontal axis and the magnitude of the physical quantity to be measured on the vertical axis. Time series data D of reference data D1 This shows an example in which the value repeatedly changes in the order of high, low, and intermediate values. For the time series data D, the threshold calculation unit 2 in FIG. 1 sets the thresholds L as L1, L2, L3, and L4 (L1>L2>L3>L4). The data division unit 3 divides the values ​​of the physical quantities within the range of the two sets of thresholds determined by these. Data to be analyzed sent to the feature calculation unit 4 In the example of Figure 2, four sets of thresholds are set, and the time series data between the thresholds L1 and L2 is divided into the first reference data D1A (divided data of status StA), and the time series data between the thresholds L3 and L4 is divided into the second reference data D1B (divided data of status StA). D1B (Divided data of status StB) It can be said that the divided data of status StA is the highest value extracted from the time-series data D as the reference data, and the divided data of status StB is the intermediate value extracted from the time-series data D as the reference data.

[0017] In this way, a portion of the time series data D is used as reference data D1, and the threshold calculation unit 2 determines the threshold. At this time, the time series data D is set to fall between two thresholds (for example, L1 and L2). This allows the time series data D to be divided so that it matches within a certain range. The data division unit 3 also divides the data using the intersection of the thresholds (for example, L1 and L2) and the time series data D. The data division unit 3 also selects the data to be analyzed and sends the data to the feature calculation unit 4.

[0018] 2, for time-series data D in which a certain physical quantity transitions over time, the threshold calculation unit 2 sets thresholds L (L1, L2, L3, L4) parallel to the time axis as shown by the dashed lines to determine the range that contains the certain time-series data.The time-series data is then divided by drawing perpendicular intersections of the thresholds and the time-series data as shown by the dashed-dotted lines.

[0019] 1, the reference data D1 that has been divided and separated into statuses is subjected to feature calculation by the feature calculation unit 4, where feature values ​​are determined for each status (for example, the statuses for the high value and the intermediate value in the example of FIG. 2). The feature values ​​calculated by the feature calculation unit 4 are basic statistical quantities such as the average value, maximum value, minimum value, variance, and standard deviation.

[0020] On the other hand, a part of the time-series data D is set as test data D2, and the abnormality degree calculation unit 5 calculates the abnormality degree of the test data D2 using this test data D2 and the above-mentioned feature amount. Note that the test data D2 may be from any point in time of the time-series change in FIG. 2. When calculating the abnormality degree, the abnormality degree is calculated based on the statistical distance (general Mahalanobis distance) between the above-mentioned reference data D1 and the above-mentioned test data D2. Specifically, the feature amount calculation unit 4 calculates the feature amount from the reference data D1, and the difference (statistical distance) between the feature amount and the test data D2 is set as the abnormality degree.

[0021] In this embodiment, for machines performing operations such as excavating ground where the load fluctuates during operation, draining water using a pump, or performing operations with non-periodic load fluctuations, the time series data of the machine during the operation can be easily divided into similar states (processes).Furthermore, by analyzing the divided time series data of the same level, the state transition of the machine can be observed.

[0022] In the configuration of Figure 1, the reference data and the inspection data are different physical quantities, and a feature amount is detected from time-series data when the first physical quantity is divided into statuses, while time-series data for the second physical quantity during the same status period is used as inspection data and compared in the anomaly degree calculation unit, thereby performing a cross-correlation diagnosis. [Example]

[0023] As a second embodiment, a device diagnostic system for industrial devices will be described with reference to Fig. 3, which explains the differences from the first embodiment. Fig. 3 shows the selection of analysis data in the data dividing unit 3. In the data dividing unit 3, the time data dividing unit 11 divides the data from the intersection of the threshold value and the reference data.

[0024] The example in Figure 3 shows that using three thresholds L1, L2, and L3 (L1>L2>L3), data D11 between times t1 and t2, divided under the condition of being less than or equal to L1 and greater than or equal to L2, data D13 between times t4 and t5, data D12 between times t2 and t3, divided under the condition of being greater than or equal to L1, and data D14 between times t3 and t4, divided under the condition of being less than or equal to L2 and greater than or equal to L3, are extracted as reference data.

[0025] The data D11, D12, D13, and D14 divided by the time data division unit 11 in Fig. 3 enter the use data selection unit 13 and are selected into use data and unused data in accordance with the conditions from the predetermined condition calculation unit 12. The specific methods for data division by the time data division unit 11, the predetermined conditions obtained by the predetermined condition calculation unit 12, and the selection of data to be analyzed by the use data selection unit 13 will be described with reference to Fig. 4.

[0026] 3, of the divided data D11, D12, D13, and D14, the used data selection unit 13 adopts D12 and D13 as used data and determines D11 and D14 as unused data. In this case, the condition from the predetermined condition calculation unit 12 is, for example, that the time length of the collected data is equal to or longer than a certain time, taking into consideration that data collected for a short time is data in a transient state and therefore is unsuitable for obtaining feature quantities.

[0027] 4 shows a method for dividing and selecting time series data in this embodiment. In FIG. 4, in the time data dividing unit 11 of FIG. 3, three bold dotted thresholds L (L1, L2, L3 (L1>L2>L3)) are set for the time series data D shown in bold lines, and vertical lines (dashed lines) are drawn from the intersections of the time series data D and the thresholds L (L1, L2, L3), thereby dividing the time series data D into processes for each of four statuses St (St1, St2, St3, St4). At this time, if the reference data D pulsates, it is advisable to reduce the pulsating component using a moving average or a filter.

[0028] In this example, the status is divided into status St1 below threshold L1, status St2 between thresholds L1 and L2, status St3 between thresholds L2 and L3, and status St4 below threshold L1. The data obtained at each status St is labeled D11, D12, D13, and D14, respectively, and the period during which this data was collected is listed at the bottom of Figure 4.

[0029] Next, if we want to calculate the abnormality levels of statuses St1, St3, and St4, we set the following predetermined conditions in the predetermined condition calculation unit 12 in Figure 3, and in the usage data selection unit 13, data that does not comply with these conditions will be excluded from the analysis calculation.

[0030] For example, the feature quantity has a value according to frequency, and the frequency resolution for identifying that frequency is Δf. Δf is, for example, equal to or less than 1 / 10 of that frequency. On the other hand, if the time width of the divided process data is Δt, the frequency resolution is 1 / Δt. Therefore, by setting the required frequency resolution Δf≦1 / Δt, only data with a large Δt is analyzed and calculated.

[0031] Furthermore, if the average rate of change of the divided process data is equal to or less than the rate of change of the speed or torque of the equipment during operation, transient phenomena due to acceleration of the machine can be removed and excluded from the analysis.

[0032] This allows transient phenomena such as the rapid rise and fall of time series data and the peaks on the right side, as shown in Figure 4, to be excluded from the analysis. [Example]

[0033] As a third embodiment, a device diagnostic system for industrial devices will be described with reference to Fig. 5, showing differences from the first and second embodiments. Fig. 5 shows the simplest form of change in time-series data, where the data consists of only one high part and one low part.

[0034] When analyzing such data, two thresholds L1 and L2 are set so that the areas close to the high and low parts are treated as statuses, as shown in Figure 5. This allows statuses St1, St2, and St3 to be defined, and analysis is performed using St1 and St3, with status St2 data not being analyzed. For this reason, at least two thresholds are set, and the data is divided into three statuses. In this embodiment, when the time series data fluctuates, it is possible to analyze only the individual status data, excluding transient states. [Example]

[0035] As a fourth embodiment, a device diagnostic system for industrial devices will be described with reference to Fig. 6, which explains the differences from the first to third embodiments. Fig. 6 is a diagram showing a device diagnostic system including this embodiment, particularly a means for outputting analysis results.

[0036] FIG. 6 shows an apparatus equipped with an equipment diagnostic unit 20 that diagnoses the state of an apparatus as shown in the first embodiment and calculates an abnormality degree. The equipment diagnostic unit 20 includes an abnormality degree calculation unit 21 that divides the time-series data D and calculates an abnormality degree for each status St, as in the first embodiment. The calculated abnormality degree for each status is displayed on a display unit 22. In this embodiment, the calculated state of the apparatus (abnormality degree) is displayed, allowing the operator to understand the state of the apparatus. When displaying the abnormality degree, a display provided near the equipment diagnostic unit 20 may be used as the display unit 22, or the result of the abnormality degree calculation may be sent to a remote location via a wired or wireless connection. [Explanation of symbols]

[0037] 1...data selection unit, 2...threshold calculation unit, 3...data division unit, 4...feature calculation unit, 5...abnormality degree calculation unit, 11...time data calculation unit, 12...predetermined condition calculation unit, 13...usage data calculation unit, 20...equipment diagnosis unit, 21...abnormality degree calculation unit, 22...display unit.

Claims

1. a data selection unit that selects a part of the time series data measured during operation of the equipment as reference data and selects a part of the time series data measured during operation of the equipment that is different from the reference data as inspection data; a threshold calculation unit that sets a plurality of thresholds for the reference data; a data division unit that divides the reference data that meets a condition determined by a plurality of thresholds in the threshold calculation unit into statuses of each range; a feature amount calculation unit that calculates feature amounts, which are basic statistics, from the reference data in each status divided by the data division unit; an abnormality degree calculation unit that uses the feature amount of the inspection data to analyze and calculate an abnormality degree, which is the degree of abnormality of the inspection data, for each status.

2. The device diagnosis system according to claim 1, An equipment diagnostic system, wherein the reference data and the inspection data are time-series data for the same physical quantity measured during operation of the equipment, but are data from different times.

3. The device diagnosis system according to claim 1, The device diagnosis system is characterized in that the data dividing unit has a usage data selecting unit that sets predetermined conditions for selecting data to be used for calculating the feature amount and data not to be used, and excludes from the usage data, statuses that do not satisfy the predetermined conditions set among the divided multiple statuses.

4. The device diagnosis system according to claim 1, An equipment diagnosis system, characterized in that, when the abnormality degree calculation unit calculates the degree of abnormality, a difference between the inspection data and a feature amount of the reference data is calculated as the degree of abnormality.

Images