A self-learning method and system for condition monitoring model

Abstract

The present invention relates to the field of condition monitoring technology, and in particular to a condition monitoring model self-learning method and system. The present invention relates to a condition monitoring model self-learning method, comprising: S1, a self-learning sensor selection step; S2, a self-learning permission interval setting step; S3, a drive sensor selection step; S4, a measured value vector screening step; S5, a pre-selected vector screening step; S6, a vector learning step to be learned. The present invention relates to a condition monitoring model self-learning system, comprising a measured value vector acquisition module, a measured value vector screening module and a data transmission module. The present invention is conducive to the condition monitoring model gradually covering as many normal operating modes of the monitored object as possible in the process of continuously acquiring measured value vectors, thereby reducing the false alarm rate.

Description

Technical Field

[0001] The present invention relates to the technical field of condition monitoring, and in particular to a condition monitoring model self-learning method and system. Background Art

[0002] In industrial production environments, there are a large number of devices and systems composed of multiple devices. The reliability of the equipment and systems directly determines the reliability and economy of the factory. Industrial sites usually monitor and control the equipment and systems by installing various sensors.

[0003] In order to detect signs of equipment failure early, in addition to the original fixed threshold monitoring of each sensor, the historical health operation data of the equipment and system is usually used to establish a status monitoring model. This monitoring model automatically introduces the interrelationships between the sensors it monitors, and establishes an alarm mechanism that can adapt to changes in working conditions, environment and other factors, thereby improving the accuracy of the alarm and achieving early warning of early signs of failure.

[0004] For example, the Chinese invention patent with application number CN202011388541.8 discloses an online monitoring method and device for equipment based on multidimensional parameter estimation, which includes: selecting a sensor detection signal related to equipment performance, establishing an initial state matrix according to the historical data of the sensor detection signal, and obtaining a final state matrix according to the initial state matrix; obtaining the real-time operating parameters of the equipment, performing similarity calculation based on the real-time operating parameters and the final state matrix, and obtaining the estimated parameter value of the equipment; performing residual calculation on the real-time operating parameters and the estimated parameter value, and monitoring and warning the equipment through the result of the residual calculation. This method uses the historical operating data of the sensor to create a model for real-time monitoring of the health status of key equipment or systems in the industrial field, referred to as a state monitoring model. However, in actual production, the historical data used to create the state monitoring model is often difficult to fully cover various normal operating conditions and environmental conditions. Especially in the early stage of equipment operation with a lack of historical data, the state detection model is very easy to misjudge and misalarm, and it is necessary to frequently manually update the data of the uncovered normal operating conditions and environmental conditions to the state monitoring model to avoid misalarms when the same monitoring data is obtained next time. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention proposes a self-learning method and system for a condition monitoring model, which can continuously optimize the condition monitoring model online during the actual operation of the monitored object to reduce the false alarm rate and the workload of manually updating the condition monitoring model.

[0006] The present invention is achieved through the following technical solutions:

[0007] The present invention provides a state monitoring model self-learning method, the method comprising:

[0008] S1, self-learning sensor selection step;

[0009] Selecting a self-learning sensor from among all sensors used for monitoring;

[0010] S2, self-learning license interval setting step;

[0011] Set the self-learning permission interval for each self-learning sensor;

[0012] S3, driving sensor selection step;

[0013] From the self-learning sensors, a sensor with a pre-set alarm threshold interval is selected as a driving sensor, and the alarm threshold interval of each driving sensor is read;

[0014] S4, measured value vector screening step;

[0015] Obtaining a measured value vector from the state monitoring model; screening the measured value vectors, and taking the measured value vectors whose measured values ​​of the respective learning sensors are within the corresponding self-learning permission interval and whose measured values ​​of the driving sensors are outside the corresponding alarm threshold interval as pre-selected vectors;

[0016] S5, pre-selected vector screening step;

[0017] Observe the running state of the monitored object and select pre-selected vectors, and select pre-selected vectors that have not been learned by the state monitoring model and whose corresponding running state of the monitored object is normal as the vectors to be learned;

[0018] S6, step of learning the vector to be learned;

[0019] Train the vector to be learned into the state monitoring model.

[0020] The present invention can analyze the measured value vectors obtained in real time, and train the vectors to be learned that are not covered by the state monitoring model and correspond to the normal operating state of the monitored object into the state monitoring model, so as to automatically and continuously optimize the state monitoring model online, without the need for manual addition by the model maintenance engineer, which is conducive to greatly reducing the model maintenance workload of the model maintenance engineer. In the process of continuously obtaining the measured value vectors, the state monitoring model can gradually cover as many normal operating modes of the monitored object as possible, thereby reducing the false alarm rate.

[0021] This method is also helpful for completing the model launch in the early stage of operation of the monitored object when the normal operating conditions covered by the condition monitoring model are relatively small and the condition monitoring model is not yet very mature.

[0022] Preferably, the method for selecting the self-learning sensor is: from all sensors, obtain a sensor with a set alarm threshold interval, or a sensor whose corresponding normal threshold range of the monitoring signal is known and can reflect the operating conditions of the monitored object, as a self-learning sensor.

[0023] Preferably, the basis for setting the self-learning permission interval includes: the permitted working status of the monitored object, the historical operating data of the monitored object and the sensor setting value.

[0024] Preferably, the measured value vector screening step specifically includes:

[0025] Step S41, step of obtaining a measured value vector;

[0026] Obtaining a vector of measured values ​​from a condition monitoring model;

[0027] Step S42, numerical value determination step;

[0028] In the measured value vector, the measured value measured by the self-learning sensor is compared with the corresponding self-learning permission interval, and the measured value measured by the driving sensor is compared with the corresponding alarm threshold interval;

[0029] Step S43, pre-selected vector acquisition step;

[0030] The measured value vectors in which the measured values ​​of the respective learning sensors are within the corresponding self-learning permission interval and the measured value of the driving sensor is outside the corresponding alarm threshold interval are used as pre-selected vectors.

[0031] The present invention also provides a state monitoring model self-learning system, the system comprising:

[0032] A measured value vector acquisition module is used to obtain a measured value vector composed of measured values ​​of each sensor from a state monitoring model;

[0033] The measured value vector screening module is used to screen out the vectors to be learned that have not been learned by the state monitoring model, the measured values ​​of the respective learning sensors are within the corresponding self-learning permission interval, and the measured values ​​of the driving sensors are outside the corresponding alarm threshold interval, and the vectors corresponding to the normal operating state of the monitored object are to be learned;

[0034] Data transmission module; used to transmit the vector to be learned to the condition monitoring model.

[0035] The present invention has the following beneficial effects:

[0036] 1. In the process of continuously acquiring measured value vectors, the present invention trains the to-be-learned vectors that are not covered by the state monitoring model and correspond to the normal operating state of the monitored object into the state monitoring model. The state monitoring model can gradually cover as many normal operating modes of the monitored object as possible without the need for manual addition by the model maintenance engineer, which is beneficial to greatly reduce the model maintenance workload of the model maintenance engineer.

[0037] 2. The present invention is conducive to completing the model launch at the initial operation stage of the monitored object when the normal working conditions covered by the condition monitoring model are relatively small and the condition monitoring model is not yet mature. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The present invention is a flow chart of a state monitoring model self-learning method. DETAILED DESCRIPTION

[0039] The following is a specific embodiment of the present invention, and the technical solution of the present invention is further described in conjunction with the accompanying drawings, but the present invention is not limited to this embodiment. After reading this specification, those skilled in the art can make modifications to the embodiment without creative contribution as needed, but as long as it is within the scope of the claims of the present invention, it is protected by the patent law.

[0040] The prior art uses a state monitoring model to capture and alarm early abnormal signs of the monitored object. The state monitoring model is established based on the monitoring data of all sensors used to monitor the monitored object. Among these sensors, there are sensors with alarm threshold intervals. After the state monitoring model obtains the measured value vector composed of the measured values ​​of each sensor, if the measured values ​​of the sensors with alarm threshold intervals are all within the corresponding alarm threshold intervals, it is determined that the monitored object is running normally, and the state monitoring model does not trigger an alarm; if the measured values ​​of the sensors with alarm threshold intervals are outside the corresponding alarm threshold intervals, it is determined that the monitored object is running abnormally, and the state monitoring model triggers an alarm. However, the alarm threshold interval is often small, especially in the early stage of the operation of the monitored object, when the historical data of the sensors used to establish the state monitoring model is insufficient, it is easy to have the measured data outside the alarm threshold interval, but the monitored object is running normally, so it is easy to trigger a false alarm, and then the state monitoring model needs to be manually updated frequently.

[0041] The present invention provides a state monitoring model self-learning method, which is used to enable the state monitoring model to automatically cover more normal working conditions of the monitored object to reduce the false alarm rate. Figure 1 , methods include:

[0042] S1, self-learning sensor selection step; select a self-learning sensor from all sensors used for monitoring.

[0043] The selection criteria for self-learning sensors are: among all sensors, sensors with set alarm threshold intervals, or sensors whose corresponding monitoring signal normal threshold ranges are known and can reflect the operating conditions of the monitored objects are used as self-learning sensors.

[0044] The measured data of the self-learning sensor can more comprehensively reflect the working characteristics of the monitored object.

[0045] Taking the variable frequency motor as the monitored object as an example, from all the sensors in its state monitoring model, the water pump motor current sensor A and ambient temperature sensor B with set alarm threshold intervals, and other corresponding monitoring signals with known normal threshold ranges that can reflect the operating conditions of the variable frequency motor, are selected as self-learning sensors. The bearing temperature sensor C at the transmission end of the water pump motor, the bearing temperature sensor D at the free end of the water pump motor, the stator winding temperature sensor No. 1 of the A phase of the water pump motor, the stator winding temperature sensor No. 2 of the A phase of the water pump motor, and the stator winding temperature sensor No. 1 of the B phase of the water pump motor are selected.

[0046] S2, self-learning license interval setting step;

[0047] Set a separate self-learning permission interval for each self-learning sensor.

[0048] The self-learning permission interval is determined by engineering design personnel based on information such as the permitted working status of the monitored object, the historical operating data of the monitored object, and sensor settings.

[0049] When the measured data of each self-learning sensor is within the corresponding self-learning permission interval, it is generally considered that the risk of failure of the monitored object is small; when the measured data of the self-learning sensor is outside the corresponding self-learning permission interval, it is considered that the monitored object is abnormal. The setting of the self-learning permission interval is used to prevent the data corresponding to unhealthy working conditions from being trained into the state monitoring model to ensure the quality and effectiveness of the state monitoring model.

[0050] Taking the variable frequency motor as an example, according to the permitted working status and historical operating data of the variable frequency motor, the self-learning permission range of sensor A is set to 56A to 82A; for the variable frequency motor, the ambient temperature is generally between -10°C and 42°C, therefore, the self-learning permission range of sensor B is set to -10°C to 42°C; for the bearing temperature and coil temperature, based on the historical operating data and the high alarm values ​​of 75°C and 120°C, the self-learning permission ranges of sensor C and sensor D are set to 40°C to 68°C, and the self-learning permission ranges of sensor E, sensor F, and sensor G are set to 60°C to 108°C.

[0051] S3, driving sensor selection step;

[0052] From the self-learning sensors, sensors with pre-set alarm threshold intervals are selected as driving sensors, and the alarm threshold intervals of each driving sensor are read.

[0053] Continuing with the variable frequency motor as an example, select sensor A and sensor B, which originally have alarm threshold intervals, as drive sensors. Read the alarm threshold interval 60A to 72A of sensor A originally set, and read the alarm threshold interval 5°C to 30°C of sensor B originally set.

[0054] The self-learning sensors and their corresponding ranges are shown in Table 1:

[0055] Table 1 Statistics of self-learning sensors and their corresponding intervals

[0056]

[0057] It should be emphasized that the alarm threshold interval is obtained from the established condition monitoring model.

[0058] S4, the measured value vector screening step; specifically includes:

[0059] S41, step of obtaining a measured value vector;

[0060] Obtaining a measured value vector from the state monitoring model; the measured value vector is composed of signal values ​​monitored by all sensors at all times;

[0061] S42, numerical value determination step;

[0062] In the measured value vector, the measured value measured by the self-learning sensor is compared with the corresponding self-learning permission interval, and the measured value measured by the driving sensor is compared with the corresponding alarm threshold interval;

[0063] S43, a pre-selected vector acquisition step;

[0064] The measured value vectors in which the measured values ​​of the respective learning sensors are within the corresponding self-learning permission interval and the measured value of the driving sensor is outside the corresponding alarm threshold interval are used as pre-selected vectors.

[0065] When the measured data of each self-learning sensor are within the corresponding self-learning permission interval, it is generally believed that the risk of failure of the monitored object is small. The measured value vector screening step is used to obtain the measured value vectors where the signal value monitored by the driving sensor exceeds the corresponding alarm threshold interval, and the signal values ​​monitored by each learning sensor are within the corresponding self-learning permission interval. This type generally corresponds to the normal working condition of the monitored object, but the existing state monitoring model will trigger an alarm.

[0066] Take variable frequency motor as an example:

[0067] When the obtained measured value vector one is as follows: [65.1, 26.3, 58.2, 54.3, 72.1, 72.2, 73.5...], 65.1 refers to the real-time monitoring value of sensor A is 65.1A, 26.3 refers to the real-time monitoring value of sensor B is 26.3℃, 58.2 refers to the real-time monitoring value of sensor C is 58.2℃, 54.3 refers to the real-time monitoring value of sensor D is 54.3℃, 72.1 refers to the real-time monitoring value of sensor E is 72.1℃, 72.2 refers to the real-time monitoring value of sensor F is 72.2℃, and 73.5 refers to the real-time monitoring value of sensor G is 73.5℃; since the measured values ​​of the respective learning sensors of the measured value vector one are within the corresponding self-learning permission interval, and the measured values ​​of each driving sensor are within the corresponding alarm threshold interval, the measured value vector one is not a preselected vector.

[0068] When the measured value vector 2 is obtained as follows: [65.1, 33.3, 74.2, 54.3, 72.1, 73.2, 73.5...], since the measured value of sensor C in the measured value vector 2 is outside the corresponding self-learning permission interval, the measured value vector 2 is not a pre-selected vector.

[0069] When the obtained measured value vector three is as follows: [65.1, 33.3, 63.2, 58.3, 76.1, 77.2, 76.5... ], since the measured values ​​of the respective learning sensors of the measured value vector three are within the corresponding self-learning permission interval, and the measured value of sensor B as the driving sensor is outside the corresponding alarm threshold interval, the measured value vector three is the pre-selected vector.

[0070] S5, pre-selected vector screening step;

[0071] Observe the running state of the monitored object and screen the pre-selected vectors, screen out the pre-selected vectors that have been learned by the state monitoring model or whose corresponding running state of the monitored object is abnormal, and retain the pre-selected vectors that have not been learned by the state monitoring model and whose corresponding running state of the monitored object is normal as the vectors to be learned;

[0072] This pre-selected vector screening step is beneficial to further prevent data corresponding to unhealthy conditions from being trained into the condition monitoring model, so as to ensure the quality and effectiveness of the condition monitoring model.

[0073] S6, step of learning the vector to be learned;

[0074] Train the vector to be learned into the state monitoring model.

[0075] The existing condition monitoring model triggers an alarm when the measured value of a sensor with an alarm threshold interval exceeds the alarm threshold interval. The alarm threshold interval is determined based on the historical operation data of the monitored object, which makes it difficult to fully cover all normal operating conditions of the monitored object, especially in the early stage of operation of the monitored object when the historical data required to establish the condition monitoring model is insufficient.

[0076] The self-learning permission interval of the present invention is determined by the engineering design personnel according to the permitted working range of the monitored object, the sensor setting, and the equipment operation history data and other information. The setting of the self-learning permission interval is used to cooperate with the alarm threshold interval to select the pre-selected vectors that are not covered by the state monitoring model but generally correspond to the normal operating conditions of the monitored object. Then, through the pre-selected vector screening step, the pre-selected vectors that have been learned by the state monitoring model or the corresponding operating conditions of the monitored object are screened out, and the pre-selected vectors that have not been learned by the state monitoring model and the corresponding operating conditions of the monitored object are retained as the vectors to be learned and trained into the state monitoring model, which is conducive to the state monitoring model gradually and comprehensively covering all normal operating conditions of the monitored object, so as to gradually improve the state monitoring model, and also helps to greatly reduce the model maintenance workload of the model maintenance engineer.

[0077] Furthermore, the step of learning the vector to be learned is performed before the state monitoring model determines whether the acquired measured value vector triggers an alarm, which is helpful to avoid false alarms in time.

[0078] In summary, this method enables the condition monitoring model to automatically learn the measured value vectors of the normal working conditions of the monitored object that are not covered, which is conducive to greatly reducing the model maintenance workload of the model maintenance engineer; the self-learning process is carried out before the judgment of the alarm trigger, which can avoid false alarms in time. This method is also conducive to completing the model launch in the early stage of equipment operation when the normal working conditions covered by the condition monitoring model are relatively small and the condition monitoring model is not yet mature.

[0079] The present invention also provides a state monitoring model self-learning system, the system comprising:

[0080] A measured value vector acquisition module is used to obtain a measured value vector composed of measured values ​​of each sensor from a state monitoring model;

[0081] The measured value vector screening module is used to screen out the vectors to be learned that have not been learned by the state monitoring model, the measured values ​​of the respective learning sensors are within the corresponding self-learning permission interval, and the measured values ​​of the driving sensors are outside the corresponding alarm threshold interval, and the vectors corresponding to the normal operating state of the monitored object are to be learned;

[0082] Data transmission module; used to transmit the vector to be learned to the condition monitoring model.

[0083] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. Any technician familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed by the present invention, and these modifications or substitutions should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the embodiments of the present invention may be deformed or modified in any way.

Claims

1. A self-learning method for a condition monitoring model, characterized in that the method include: S1, self-learning sensor selection step; Selecting a self-learning sensor from among all sensors used for monitoring; S2, self-learning license interval setting step; Set the self-learning permission interval for each self-learning sensor; S3, driving sensor selection step; From the self-learning sensors, a sensor with a pre-set alarm threshold interval is selected as a driving sensor, and the alarm threshold interval of each driving sensor is read; S4, measured value vector screening step; Obtaining a measured value vector from the state monitoring model; screening the measured value vectors, and taking the measured value vectors whose measured values ​​of the respective learning sensors are within the corresponding self-learning permission interval and whose measured values ​​of the driving sensors are outside the corresponding alarm threshold interval as pre-selected vectors; S5, pre-selected vector screening step; Observe the running state of the monitored object and select pre-selected vectors, and select pre-selected vectors that have not been learned by the state monitoring model and whose corresponding running state of the monitored object is normal as the vectors to be learned; S6, step of learning the vector to be learned; Train the vector to be learned into the state monitoring model.

2. A condition monitoring model self-learning method according to claim 1, characterized in that: The method for selecting the self-learning sensor is: from all sensors, a sensor with a set alarm threshold interval, or a sensor whose corresponding normal threshold range of the monitoring signal is known and can reflect the operating conditions of the monitored object is obtained as a self-learning sensor.

3. A condition monitoring model self-learning method according to claim 1, characterized in that: The basis for setting the self-learning permission interval includes: the permitted working status of the monitored object, the historical operating data of the monitored object and the sensor setting value.

4. The state monitoring model self-learning method according to claim 1, characterized in that: The measured value vector screening step specifically includes: Step S41, step of obtaining a measured value vector; Obtaining a vector of measured values ​​from a condition monitoring model; Step S42, numerical value determination step; In the measured value vector, the measured value measured by the self-learning sensor is compared with the corresponding self-learning permission interval, and the measured value measured by the driving sensor is compared with the corresponding alarm threshold interval; Step S43, pre-selected vector acquisition step; The measured value vectors in which the measured values ​​of the respective learning sensors are within the corresponding self-learning permission interval and the measured value of the driving sensor is outside the corresponding alarm threshold interval are used as pre-selected vectors.

5. A state monitoring model self-learning system, characterized in that the system include: A measured value vector acquisition module is used to obtain a measured value vector composed of measured values ​​of each sensor from a state monitoring model; Measured value vector screening module; It is used to filter out the vectors to be learned that have not been learned by the state monitoring model, the measured values ​​of the respective learning sensors are within the corresponding self-learning permission interval, and the measured values ​​of the driving sensors are outside the corresponding alarm threshold interval, and the vectors to be learned that correspond to the normal operating state of the monitored object; Data transmission module; Used to transfer the vector to be learned to the condition monitoring model.

Images