Methods and systems for detecting machine defects

By defining state indicators, recording data, training machine learning algorithms, and comparing predicted values, the problem of false alarms in machine defect detection in existing technologies has been solved, achieving more accurate defect detection, especially for bearing detection in wind turbines.

CN114151289BActive Publication Date: 2026-11-10AB SKF SKF PATENT DEPARTMENT
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202111036999.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2020-09-07
Filing Date
2021-09-06
Publication Date
2026-11-10
Estimated Expiration
2041-09-06

AI Technical Summary

Technical Problem

Existing technologies, when detecting machine defects, struggle to distinguish whether changes in status indicators are due to machine defects or changes in operating conditions, leading to a high risk of false alarms and an inability to detect potential damage in a timely manner.

Method used

By defining status indicators, recording machine operating status data, training machine learning algorithms to establish the relationship between operating status and status indicators, predicting the current status indicator value, and comparing it with the actual value, we can determine whether the machine is operating normally and improve detection accuracy by using neural networks.

Benefits of technology

This technology enables more accurate detection of machine defects while taking into account the machine's current operating status, reducing false alarms and improving the sensitivity and reliability of defect detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114151289B_ABST
    Figure CN114151289B_ABST
Patent Text Reader

Abstract

A system for detecting at least one machine defect, the system comprising: a defining device for defining at least one state indicator reflecting the machine state from machine kinematic data for the defect to be monitored; a recording device for recording machine operating state data and machine state monitoring data during a predetermined period of normal machine operation and for recording current state monitoring data; a determining device for determining a state indicator value using the state monitoring data and determining a current state indicator value from the at least one state indicator and the current state monitoring data; a machine learning algorithm; a prediction device for predicting a state indicator value for the current operating state data using the machine learning algorithm; a training device for training the machine learning algorithm to establish a relationship between the operating state data and the state indicator value; and a comparison device for comparing the current state indicator value and the predicted state indicator value and for determining whether the machine is operating normally based on the comparison result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to machine diagnostics, and more specifically, to the detection of machine defects. Background Technology

[0002] Machine diagnostics is used to monitor the health of machines. The main goal of machine diagnostics is to detect defects in machines as early as possible.

[0003] Typically, condition indicators are monitored to detect machine defects; these indicators are defined by the machine's motion parameters.

[0004] In many applications where operating parameters have a significant impact on machine behavior, it is difficult to determine whether changes in condition indicators are due to machine defects or changes in operating conditions.

[0005] For example, alarm levels set on defined status indicator values ​​can be used to monitor rotating machine components (such as the main bearings of wind turbines) to detect defects.

[0006] If the status indicator value meets the alarm status set, the alarm is triggered, indicating which component displays the defect.

[0007] Ideally, condition indicators should only be sensitive to the presence of defects in the machine component to which the condition indicator pertains.

[0008] However, when a wind turbine is running, the status index values ​​can fluctuate depending on the operating conditions of the wind turbine (e.g., the absolute value and variance of the bearing rotational speed, the power output of the wind turbine, and the wind speed and direction driving the blades).

[0009] To avoid false alarms, the alarm level is set high enough to avoid discrepancies caused by operational behavior. The risk is that defects leading to wind turbine damage may not be detected.

[0010] To detect machine defects, it is necessary to consider the fluctuations in status indicator values ​​caused by operational behavior and avoid triggering false alarms. Summary of the Invention

[0011] Based on one aspect, a method for at least one machine defect is proposed.

[0012] The method includes:

[0013] a) To define at least one state index from machine kinematic data that reflects the state of the machine in response to the defects to be monitored.

[0014] b) Record the machine's operating status data and status monitoring data during the predetermined period of normal machine operation, and use the status monitoring data to determine status indicator values.

[0015] c) Train machine learning algorithms to establish relationships between operational status data and status indicator values ​​recorded during predetermined time periods.

[0016] d) Record the current status monitoring data and determine the current status indicator value from the at least one status indicator and the current status monitoring data.

[0017] e) Predict status indicator values ​​for current operational status data using machine learning algorithms.

[0018] f) Compare the current state indicator value with the predicted state indicator value, and

[0019] g) Determine whether the machine is presumed to be operating normally based on the comparison results.

[0020] Taking into account the current operating status of the machine, defect detection is made more accurate.

[0021] Preferably, the step of comparing the current state indicator value with the predicted state indicator value includes: determining the difference between the current state indicator value and the predicted state indicator value, and if the difference is within a set limit, the machine is assumed to be operating normally.

[0022] Advantageously, steps d), e), f), and g) are repeated as long as the machine is assumed to be operating normally outside the predetermined time period.

[0023] Preferably, the step of comparing the current state indicator value with the predicted state indicator value includes: determining the difference between the current state indicator value and the predicted state indicator value, wherein if the difference is outside a set limit, the machine is assumed to have a defect to be monitored.

[0024] Advantageously, machine learning algorithms include neural networks.

[0025] Preferably, the machine includes a wind turbine, operating status data, and current operating status data. The wind turbine includes bearings that support the blades of the wind turbine. The operating status data and current operating status data include the rotational speed of the bearings, the power output of the wind turbine, and / or wind speed and direction.

[0026] On the other hand, a system for detecting at least one machine defect is proposed.

[0027] The system includes:

[0028] - Define a device for defining at least one state index reflecting the state of a machine from machine kinematic data, in response to a defect to be monitored.

[0029] - A recording device for recording machine operating status data and machine status monitoring data during a predetermined period of normal machine operation, and for recording current status monitoring data.

[0030] - A determining device for determining a status indicator value using status monitoring data, and for determining a current status indicator value from the at least one status indicator and the current status monitoring data.

[0031] -Machine learning algorithms

[0032] - A predictive device used to predict status indicator values ​​for current operating status data using machine learning algorithms.

[0033] - A training device used to train machine learning algorithms to establish relationships between operational status data and status indicator values, and

[0034] - A comparison device is used to compare the current status indicator value with the predicted status indicator value, and to determine whether the machine is operating normally based on the comparison result.

[0035] Preferably, if the difference between the current state indicator value and the predicted state indicator value is within a set limit, the comparison device is configured to generate a first signal indicating that the machine is assumed to be operating normally.

[0036] Advantageously, if the difference between the current state indicator value and the predicted state indicator value is outside a set limit, the comparison device is configured to generate a second signal indicating that the machine is assumed to have a defect to be monitored.

[0037] On the other hand, a wind turbine connected to the system defined above is proposed. Attached Figure Description

[0038] Other advantages and features of the invention will become apparent from a review of the detailed description of the embodiments (in no way limiting) and the accompanying drawings, wherein:

[0039] [ Figure 1 [A schematic illustration shows an embodiment of the system for detecting defects in a wind turbine according to the present invention; and]

[0040] [ Figure 2 This illustration shows an embodiment of a method for detecting defects in a wind turbine according to the present invention. Detailed Implementation

[0041] Reference Figure 1 , Figure 1An example of a wind turbine 1, representing an embodiment of a system 2 for detecting at least one wind turbine defect.

[0042] As an alternative, system 2 can be integrated into (or incorporated into) wind turbine 1.

[0043] The wind turbine 1 includes a bearing 3 that supports the main shaft 4 of the wind turbine 1, an operating status sensor 5, and a status monitoring sensor 6.

[0044] Main shaft 4 supports blades 7.

[0045] Sensor 5 generates operating status data OCD of wind turbine 1, such as the absolute value and variance of the rotational speed of bearing 3, the power output of wind turbine 1, and the wind speed and direction of the wind driving blades 7.

[0046] Sensor 5 includes, for example, a speed sensor, a power sensor, and / or a wind sensor.

[0047] Sensor 6 generates Condition Monitoring Data (CMD) for wind turbine 1.

[0048] Sensor 6 includes, for example, an accelerometer that generates vibration data from bearing 3.

[0049] System 2 includes:

[0050] - Defining means (DM) are used to define state indicators (CI) reflecting the state of wind turbine 1 from the kinematic data of wind turbine 1, in response to defects to be monitored.

[0051] - A recording device RM is used to record the operating status data OCD and the condition monitoring data CMD of the wind turbine 1 during a predetermined period of normal operation of the wind turbine 1, and to record the current condition monitoring data CMD.

[0052] - The determination device DETM is used to determine the condition indicator value CIV using condition monitoring data CMD, and to determine the current condition indicator value CCIV from the condition indicator CI and the current condition monitoring data CCMD.

[0053] -The machine learning algorithm ALGO

[0054] - The prediction device PM is used to predict the pCIV (state indicator value) of the current operating status data COCD using the machine learning algorithm ALGO.

[0055] - Training device™ for training the machine learning algorithm ALGO to establish the relationship between runtime state data OCD and state indicator values ​​CIV (Relationship Between State and Values) REL.

[0056] - Comparison device CM is used to compare the current state index value CCIV with the predicted state index value pCIV, and to determine whether the wind turbine 1 is operating normally based on the comparison result.

[0057] Next, bearing 3 will be monitored to detect bearing defects.

[0058] The condition index CI includes, for example, the gear meshing frequency of gear 3, which is equal to the spindle speed multiplied by the number of teeth of bearing 3, where the number of teeth of bearing 3 is kinematic data of the wind turbine.

[0059] The machine learning algorithm ALGO can include, for example, neural networks.

[0060] When the wind turbine 1 is running outside the predetermined time period, the sensor 5 generates the current operating status data COCD of the wind turbine 1, and when the wind turbine 1 is running outside the predetermined time period, the sensor 5 generates the current status monitoring data CCMD.

[0061] If the current status index value CCI and the predicted status index value pCI are within the set limits, the comparison device CM generates a first signal S1 indicating that the bearing 3 is assumed to be operating normally.

[0062] When the current state index value CCI and the predicted state index value pCI are outside the set limits, the comparison device CM generates a second signal S2, which indicates that the bearing 3 is assumed to have a monitored defect.

[0063] The signals S1 and S2 generated by the comparison device CM can be processed by a processing unit (not shown), which can, for example, stop the wind turbine 1 when the bearing 3 is defective or an alarm is triggered.

[0064] The comparator CM determines the deviation value DEV.

[0065] For each current state indicator value (CCIV) and the associated predicted state indicator value (pCIV), the corresponding deviation value (DEV) is equal to:

[0066] DEV = CCIV - pCIV (1)

[0067] If the defect to be monitored occurs, the trend of the condition indicator value CCIV is increasing, while the trend of the predicted condition indicator value pCIV remains stable. The trend of the deviation value DEV is increasing.

[0068] Implement trend analysis of deviation values ​​(DEV) to determine whether the defects to be monitored have occurred.

[0069] Examples of how to determine whether a defect to be monitored has occurred have been published.

[0070] When wind turbine 1 is running outside of the predetermined time period, a set of deviation values ​​DEV are recorded during the learning period.

[0071] The setting limits, including the lower limit LL and the upper limit UL, were determined.

[0072] The mean μ and standard deviation σ of the set of deviation values ​​DEV were determined.

[0073] The upper limit UL is equal to:

[0074] UL=μ+X.σ (2)

[0075] The lower limit LL equals:

[0076] LL=μ-X.σ (3)

[0077] Where X is the detection factor.

[0078] X is, for example, included between 1 and 4, and is determined based on the detection sensitivity of the defect to be monitored.

[0079] For example, when the wind turbine 1 is running, at the end of the predetermined period and the learning period, if M out of the N values ​​of the deviation value DEV are outside the set limit, the comparison device CM generates a second signal S2.

[0080] Otherwise, the comparator CM generates the first signal S1.

[0081] For example, M and N are equal to 4 and 3 respectively.

[0082] Figure 2 This describes an implementation method for detecting defects in bearing 3.

[0083] Assume that the lower limit LL value and the upper limit UL value are determined.

[0084] In step 10, the device DM defines the status index CI.

[0085] In step 11, the recording device RM records the operating status data OCD and the condition monitoring data CMD of the wind turbine 1 during a predetermined period of normal operation of the wind turbine 1, and the determining device DETM uses the condition monitoring data CMD to determine the condition index value CVI.

[0086] In step 12, the training device TM trains the machine learning algorithm ALGO to establish the relationship REL between the running state data OCD and the state indicator values ​​CIV recorded during a predetermined time period.

[0087] In step 13, the recording device RM records the current status monitoring data CCMD, and the determining device DETM determines the current status index value CCI from the status index CI and the current status monitoring data CCMD.

[0088] In step 14, the prediction device PM uses the machine learning algorithm ALGO to predict the status index value pCI for the current operating status data COCD.

[0089] In step 15, the comparison device determines the deviation value DEV.

[0090] If the deviation value is within the set limits LL and UL (step 16), then in step 17, the comparison device CM generates the first signal S1.

[0091] In another embodiment, if M out of the N values ​​of the deviation value DEV are within the set limit (step 16), then in step 17, the comparison device CM generates a first signal S1.

[0092] Then, if wind turbine 1 is still operating outside the predetermined time period (step 18), steps 13, 14, 15 and 16 are repeated as long as it is assumed that the machine is operating normally outside the predetermined time period.

[0093] For example, repeat steps 13, 14, 15 and 16 every day, every 10 minutes.

[0094] The duration of step repetition is determined based on the sensitivity of defect detection.

[0095] If the deviation value is outside the set limits LL and UL (step 16), then in step 19, the comparator CM generates a second signal S2.

[0096] In another embodiment, if M out of the N values ​​of the deviation value DEV are outside the set limit (step 16), then in step 19, the comparison device CM generates a second signal S2.

[0097] When the wind turbine 1 is operating normally, the current state index value is predicted by the machine learning algorithm ALGO (predicted state index value pCI) trained during the learning period, and the current state index value is compared with the current state index value based on the current operating state data COCD.

[0098] The deviation between the predicted status indicator value and the current status indicator value is compared with the set limit to determine whether the deviation represents a defect to be monitored.

[0099] Taking into account the current operational status data, defect detection is made more accurate.

[0100] In the example shown, a condition index CI reflecting the condition of bearing 3 is analyzed for the defects to be monitored in wind turbine 1.

[0101] To detect more defects, more status indicators can be defined.

[0102] In another embodiment, system 2 can detect defects in another part of wind turbine 1, such as the absolute value and variance of the rotational speed of bearing 3, the power output of wind turbine 1, and the wind speed and direction of the wind driving blades 4.

[0103] In another embodiment, system 2 can detect at least one defect in another machine other than a wind turbine, such as a rotary machine or a reciprocating machine (e.g., a piston engine).

Claims

1. A method for detecting at least one machine defect, the method comprising: a) For the defect to be monitored, define at least one state index (CI) from machine kinematic data that reflects the state of the machine. b) Record the machine's operating status data (OCD) and the machine's condition monitoring data (CMD) during a predetermined period of normal machine operation, and use the condition monitoring data (CMD) to determine the condition indicator value (CIV). c) Train a machine learning algorithm (ALGO) to establish a relationship (REL) between the operational status data (OCD) and the status indicator values ​​(CIV) recorded during a predetermined time period (t1). d) Record the current status monitoring data (CCMD) and determine the current status indicator value (CCIV) from the at least one status indicator (CI) and the current status monitoring data (CCMD). e) Predict the status indicator value (pCIV) for the current running status data (COCD) using the machine learning algorithm (ALGO). f) Determine the deviation between the current state indicator value (CCIV) and the predicted state indicator value (pCIV), and g) Determine whether the machine is assumed to be operating normally by performing trend analysis of the deviation values, wherein the step of performing trend analysis of the deviation values ​​includes determining whether M out of N values ​​of the deviation values ​​are within a set limit.

2. The method according to claim 1, characterized in that, The step of determining whether the machine is assumed to be operating normally by performing trend analysis of the deviation values ​​includes: if M out of the N values ​​of the deviation values ​​are within the set limits, then the machine is assumed to be operating normally.

3. The method according to claim 1 or 2, characterized in that, Repeat steps d), e), f), and g) as long as the machine is assumed to be operating normally outside the predetermined period.

4. The method according to any one of claims 1 to 3, characterized in that, The step of determining whether the machine is assumed to be operating normally by performing trend analysis of the deviation values ​​includes: if M out of the N values ​​of the deviation values ​​are outside the set limits, then the machine is assumed to have the defect to be monitored.

5. The method according to any one of claims 1 to 4, characterized in that, The machine learning algorithm (ALGO) includes neural networks.

6. The method according to any one of claims 1 to 5, characterized in that, The machine includes a wind turbine, operating status data and current operating status data. The wind turbine includes a bearing (3) that supports the blades (4) of the wind turbine. The operating status data and the current operating status data include the rotational speed of the bearing, the power output of the wind turbine and / or the wind speed and wind direction.

7. A system (2) for detecting at least one machine defect, said system comprising: - Definition device (DM) for defining at least one condition index (CI) reflecting the state of the machine from machine kinematic data in response to defects to be monitored. - A recording device (RM) for recording the machine's operating status data (OCD) and the machine's status monitoring data (CMD) during a predetermined period of normal operation of the machine, and for recording the current status monitoring data (CCMD). - A determining device (DETM) for determining a status indicator value (CIV) using status monitoring data (CMD), and for determining a current status indicator value (CCIV) from the at least one status indicator and the current status monitoring data (CCMD). - Machine learning algorithm (ALGO) - A prediction device (PM) for predicting a status indicator value (pCIV) for the current operating status data (COCD) using the machine learning algorithm (ALGO). - A training device (TM) for training the machine learning algorithm (ALGO) to establish the relationship (REL) between the running state data (OCD) and the state indicator value (CIV), and - A comparison device (CM) is used to determine the deviation between the current status index value (CCIV) and the predicted status index value (pCIV), and to determine whether the machine is assumed to be operating normally by performing trend analysis of the deviation value, wherein the step of performing trend analysis of the deviation value includes determining whether M out of N values ​​of the deviation value are within a set limit.

8. The system according to claim 7, characterized in that, If M out of the N values ​​of the deviation are within the set limit, the comparison device (CM) is configured to generate a first signal (S1) indicating that the machine is assumed to be operating normally.

9. The system according to claim 7 or 8, characterized in that, If M out of the N values ​​of the deviation are outside the set limit, the comparison device (CM) is configured to generate a second signal (S2) indicating that the machine is assumed to have the defect to be monitored.

10. A wind turbine connected to a system (2) according to any one of claims 7 to 9.

Citation Information

Patent Citations

  • Operating state evaluation method and operating state evaluation device

    US20200263669A1