DE102022211058B4 - Computer-implemented monitoring method, hydraulic system and computer-readable storage medium - Google Patents

Abstract

The present disclosure relates to a computer-implemented monitoring method for detecting leaks in a hydraulic system (1) with the following steps. Sensors of the hydraulic system (1) record measured values of the hydraulic system (1), in particular pressure in lines and/or from actuators of the hydraulic system (1). The recorded measured values are entered into a cluster analysis model (7). The recorded measured values are divided into different predefined clusters by the cluster analysis model (7).

Description

Technical background

The present disclosure relates to a computer-implemented monitoring method for detecting leaks in a hydraulic system according to the preamble of claim 1, a hydraulic system and a computer-readable storage medium.

In known hydraulic systems, an oil level in a tank of the hydraulic system is only monitored by predefined warning and alarm limits. This means that an alarm is triggered when the oil level in the tank falls below a certain limit. However, the oil level in the oil tank is subject to temperature-related fluctuations. Furthermore, the level of the oil tank can also fluctuate due to the operation of the hydraulic system, for example due to cylinder stroke. This means that the limit for triggering the alarm must be relatively low in order to avoid false alarms. This means that if there is a leak in the hydraulic system, a certain amount of oil can escape before the leak is detected. The leaking oil must be replaced and is also an environmental risk. Oil leaks from the hydraulic system should therefore always be avoided or reduced.

Alternatively, the oil level is checked manually through a viewing window in a side wall of the oil tank. It is particularly difficult to detect an oil loss manually in large hydraulic systems with correspondingly large oil tanks. For example, it is very difficult to detect a leakage of 20l in a hydraulic system with an oil tank volume of 600l. Due to the temperature-related and/or operational fluctuations in the oil level, it is therefore difficult to detect an oil loss by checking the oil level in the oil tank (manually).

In summary, in known hydraulic systems, oil level monitoring can only be carried out manually or oil level monitoring only triggers an alarm when the level falls below a limit value, which can cause a considerable amount of oil or working fluid to escape from the hydraulic system before the leak is discovered.

The EN 10 2020 124 238 A1 discloses a monitoring method according to the preamble of claim 1.

Summary of the present disclosure

It is therefore the object of the present disclosure to overcome or at least reduce the disadvantages of the prior art and preferably to provide a method with which leaks in a hydraulic system can be detected effectively and inexpensively. Particularly preferably, the method should not require any additional sensors other than the sensors of the hydraulic system.

This object is achieved by a computer-implemented monitoring method according to the features of claim 1. The object of the present disclosure is further achieved by a hydraulic system according to the features of claim 8 and a computer-readable storage medium according to the features of claim 9.

The present disclosure relates to a computer-implemented monitoring method for detecting leaks in a hydraulic system. The monitoring method has the following steps. Sensors of the hydraulic system record measured values of the hydraulic system, in particular oil level, oil temperature and/or (oil) pressure of the hydraulic system. The recorded measured values are entered as input values into a trained cluster analysis model. The trained cluster analysis model assigns the recorded measured values to different predefined clusters. Preferably, a control unit of the hydraulic system can issue a warning to a user via an output unit based on the result of the cluster analysis of the cluster analysis model.

The measured values recorded can preferably be an oil level in an oil tank of the hydraulic system and/or an oil temperature in the oil tank. Furthermore, pressure values in lines and/or actuators, such as pumps, of the hydraulic system can be recorded. Operating states of valves and/or the actuators of the hydraulic system can also be recorded as measured values.

The measured values recorded can also be all measured values from all sensors in the hydraulic system. Since the measured values of these sensors are recorded when the hydraulic system is in operation, recording the measured values does not represent any additional effort compared to operation without the monitoring method. Preferably, only measured values from sensors that are used in conventional and known hydraulic systems are used for the monitoring method.

A leak in the hydraulic system can occur, for example, if a line in the hydraulic system bursts or leaks.

In an exemplary embodiment of the present disclosure, the oil level and oil temperature in the oil tank of the hydraulic system and pressure values on various lines of the hydraulic system are detected by sensors. The detected measurements are input into the trained cluster analysis model, which classifies the detected measurements into the different clusters.

The monitoring method according to the present disclosure can detect leakage in the hydraulic system. In particular, a slow or creeping leak can also be detected before a limit value for the oil level in the oil tank is reached. This prevents oil from leaking undetected. Furthermore, the monitoring method can compensate for temperature-related fluctuations and/or operational fluctuations and detect whether a low oil level is caused by a leak or the regular operation of a hydraulic system. False alarms caused by operational fluctuations can be avoided in this way.

By detecting the leak early, a defective component can be identified and replaced in good time without having to shut down the entire hydraulic system. The unsupervised training process allows the recorded measured values to be processed in real time. Furthermore, the training data set can only contain the raw data recorded by the sensors and does not have to have an associated class. Recorded measured values can therefore be added to the training data set in real time, increasing the accuracy of the cluster analysis model.

In other words, the present disclosure is based on the task of monitoring the creeping leakage in a hydraulic system in such a way that permanent monitoring and continuous analysis is possible depending on the operating conditions and the monitoring of the hydraulic medium (quality, quantity). With the help of machine learning, patterns and regularities are recognized on the basis of existing data sets (ideal/creeping/extremely decreasing leakage in different operating states) and algorithms, and the system is permanently monitored and analyzed.

The object of the present disclosure is further achieved by a hydraulic system. The hydraulic system has a number of sensors that record measured values of the hydraulic system and a monitoring device for detecting leaks in the hydraulic system using a cluster analysis model. The cluster analysis model is adapted and designed to classify the measured values recorded by the sensors into different clusters.

The sensors of the hydraulic system record a number of measurement data / data during operation of the hydraulic system. The monitoring device for detecting leaks has the cluster analysis model, which is configured to classify the measured values into different clusters based on the recorded measured values. This makes it possible to distinguish between different operating states of the hydraulic system. If the cluster analysis model determines that there is a leak in the hydraulic system, the user of the hydraulic system can preferably be warned or notified by an output unit.

The monitoring device may include a CPU, a storage unit and a RAM.

The output unit can be a display that shows the user the operating status of the hydraulic system. The output unit can provide the user with the results of the cluster analysis or warnings.

The monitoring device can monitor the hydraulic system for possible leaks in real time. Recorded measured values can also be added to the cluster analysis model's data set in real time, thereby improving the accuracy of the cluster analysis model. This means that the cluster analysis model can be continuously trained due to the scalability of the cluster analysis model, so that the cluster analysis model is further optimized as the data input increases.

Advantageous further developments of the present disclosure are the subject of the appended subclaims.

According to the present disclosure, the cluster analysis model has three clusters. A first cluster is preferably an operation of the hydraulic system with no leakage, a second cluster is preferably an operation of the hydraulic system with a gradual leakage, and a third cluster is preferably an operation of the hydraulic system with a rapidly/extremely decreasing leakage.

The cluster analysis model can therefore divide the measured values of the hydraulic system into three clusters. This allows the cluster analysis model can determine whether there is a creeping leak, a rapid leak or no leak based on the measured values recorded. The hydraulic system can be operating with a creeping leak, for example, if a line or a seal is leaking and working fluid is escaping. The hydraulic system can be operating with a rapid leak, for example, if a line has burst.

It is of course conceivable that the cluster analysis model divides the recorded measured values into more than three clusters. This allows the individual operating states to be differentiated even more precisely.

According to a further optional feature of the present disclosure, a warning is issued to the user by an output unit if the cluster analysis model assigns the acquired measured values to the cluster of the operation of the hydraulic system with the creeping leak or to the cluster of the operation of the hydraulic system with the rapid/extremely decreasing leak. In both cases, the leak can lead to the critical oil level being undershot. Preferably, the user can only be given a warning if a creeping leak is detected. If the rapid leak is detected, an alarm can be triggered and/or the operation of the hydraulic system can be interrupted.

Preferably, the monitoring method comprises a step in which the cluster analysis model is trained with a training data set having historical measured values of the hydraulic system as input values.

The cluster analysis model can be trained, for example, using a training data set. The training data set preferably contains historical measurement data. Historical measurement data in this context can mean that the measurement values were recorded during the previous operation of the hydraulic system. The cluster analysis model can be trained using the training data set using an unsupervised training procedure. In particular, the cluster analysis model is trained before the measurement values recorded by the sensors are entered into the cluster analysis model.

The training data set can be prepared, for example, by selecting certain relevant sections from the time-dependent curves of the historical measured values. The measured values can also be pre-treated, for example by only considering a first derivative and/or a second derivative of the measured values.

During the process of training the cluster analysis model with the training data set, the cluster analysis model can calculate centers of the various clusters in the historical measured values starting from (random) starting points. The centers can iteratively approach a computational optimum. If measured values recorded (in real time) are then entered into the trained cluster analysis model, the measured values recorded can be assigned to the next center and thus to a cluster. Since the training of the cluster analysis model is preferably unsupervised, the measured values recorded can be added to the cluster analysis model, thus increasing the accuracy of the cluster analysis model as the amount of data increases.

Preferably, the cluster analysis model is based on a k means algorithm. The advantage of using the k means algorithm is that the number of clusters is known. This can increase the accuracy of training the cluster analysis model.

Preferably, the cluster analysis model is based on a DBSCAN algorithm.

According to a further optional feature of the present disclosure, the monitoring method comprises the following steps, which are preferably carried out after the step of classifying the acquired measurement values into the different clusters. A prediction model of the cluster analysis model predicts future measurement values based on the acquired measurement values. Based on the predicted future measurement values, a time period until a critical oil level is reached in an oil tank of the hydraulic system is calculated. The output unit outputs the calculated time period as a warning or information to the user.

This means that if the monitoring procedure has determined that a leak, especially a creeping leak, is present, the forecast model can predict future measured values based on the measured values recorded.

This makes it possible to calculate how long it will take until the oil level in the oil tank falls below a critical limit. This time period can be issued to the user as a warning via the output unit. This means that the leak can be dealt with promptly before the oil level falls below the critical limit. For example, the user can be given information on how many Hours or days it takes until the critical oil level is reached.

The forecast model can therefore predict or extrapolate a course of the recorded measured values. The forecast model can be, for example, an artificial neural network. The artificial neural network can be specialized in processing time-dependent measured values, for example a convolutional neural network (CNN) or a long short time memory (LSTM) network. The forecast model can also be based on linear and/or polynomial extrapolation.

Preferably, the forecast model is trained with a training data set that has at least one time period of the historical measured values as input values and another time period of the historical measured values as output values. The forecast model can thus output a forecast measured value for input measured values.

The calculation of the time until the critical oil level is reached can be carried out by a control unit of the hydraulic system or by the cluster analysis model.

By predicting when the oil level will fall below the critical level, an early warning can be given to the user. This allows oil to be topped up before the critical level is reached and a possible cause of the leak or leak can be found and corrected.

According to a further optional feature of the present disclosure, the cluster analysis model can be trained with a training data set that includes historical measurements in which defects occurred in the operation of the hydraulic system. If the recorded measurements are classified by the model as similar to the historical measurements, a warning can be issued to the user. The user can be prompted to replace or maintain a component. This can prevent a failure of the component and thus a downtime of the hydraulic system.

Preferably, the monitoring device has a forecast model for forecasting the future course of the measured values based on the recorded measured values. The forecast model can output the forecast course of the measured values. Based on the forecast course, a warning can be issued that warns of reaching the critical oil level.

Preferably, the monitoring device comprises the output unit that issues a warning to a user when a leakage is detected by the cluster analysis model.

The present disclosure further relates to a computer-readable storage medium. The storage medium comprises instructions which, when executed by a computer, cause the computer to carry out the monitoring method according to one of the above aspects.

Short description of the characters

• 1 shows a schematic representation of a hydraulic system according to the present disclosure;
• 2 shows a flowchart of a monitoring method according to the present disclosure;
• 3 shows a flowchart of a monitoring method according to another embodiment of the present disclosure;
• 4 shows a time course of an oil level in an oil tank of the hydraulic system according to the present disclosure without leakage;
• 5 shows an example representation of different clusters of a cluster analysis model; and
• 6 shows a time course of an oil level in an oil tank of the hydraulic system according to the present disclosure with leakage.

Detailed description of the figures

1 shows a schematic representation of a hydraulic system 1 according to the present disclosure. The hydraulic system 1 has a number of lines and actuators (not shown) and a number of sensors 2. The sensors 2 record measured values of the hydraulic system 1, for example pressure values of the lines and/or actuators and/or states of actuators. At least one of the lines is fluidically connected to an oil tank 3 of the hydraulic system 1. The oil or working fluid of the hydraulic system 1 is stored in the oil tank 3. The hydraulic system 1 further comprises an (optional) control unit 4, an output unit 5 for outputting a warning or information to a user and a monitoring device 6 for detecting leaks in the hydraulic system 1. The monitoring device 6 comprises a cluster analysis model 7 and optionally a forecasting device 8. The cluster analysis model 7 divides the measured values recorded by the sensors 2 into different clusters. The prediction device 8 can predict a future course of the measured values recorded by the sensors 2 and thus predict when the oil level in the oil tank 3 falls below a critical limit. The monitoring device 6 can thus detect leaks in the hydraulic system 1 and warn the user before a large amount of working fluid escapes from the hydraulic system 1. The escaping working fluid could, for example, cause environmental damage.

2 shows a flow chart of a monitoring method according to the present disclosure. In step S1, the sensors 2 of the hydraulic system 1 record the measured values of the hydraulic system 1. In step S2, the recorded measured values are entered into the trained cluster analysis model 7. The cluster analysis model 7 is trained with historical measured values of the hydraulic system 1. In step S3, the cluster analysis model 7 assigns the recorded measured values to the different clusters. The different clusters are a cluster for operation of the hydraulic system 1 without leakage, a cluster for operation of the hydraulic system 1 with gradual leakage and cluster for operation of the hydraulic system 1 with rapid leakage. Optionally, a warning can be issued to the user of the hydraulic system 1 via the output unit 5 if the monitoring device 6 determines that a leak is present.

Of course, the cluster analysis model 7 can be trained with any other system or model.

3 shows a flow chart of a monitoring method according to a further embodiment. Steps S101 to S103 correspond to steps S1 to S3 and are therefore not explained in detail. Step S104 is carried out if the cluster analysis model 7 in step S103 has classified the recorded measured values as operation with a leak in the hydraulic system 1. In step S104, the forecast model 8 of the cluster analysis model 7 predicts a course of future measured values based on the recorded measured values. In step S105, a time period until a critical oil level is reached in the oil tank 3 of the hydraulic system 1 is calculated from the forecast future measured values. In step S106, a warning or the result of the calculation from step S105 is output to the user by the output unit 5. If the cluster analysis model 7 in step S103 comes to the conclusion that there is no leak, then the course of the measured values is not predicted. The program is terminated or restarted in step S107.

The calculation in step S105 can be performed both by the control unit 4 and by the monitoring device 6.

4 shows a temporal progression of the oil level in the oil tank 3 during operation of the hydraulic system 1 without leakage. The actual oil level fluctuates over time due to temperature changes and/or changes in the operation of the hydraulic system 1. For example, working fluid from the oil tank 3 can be provided in the lines and thereby reduce the oil level in the oil tank 3. On average, however, the oil level of the hydraulic system 1 is constant. Thus, the oil level in the hydraulic system 1 is 4 shown oil level curve around the oil level when operating the hydraulic system 1 without leakage.

5 shows an example of the three clusters with the respective center of the cluster. Preferably, the recorded measured values are divided into the three clusters. A first cluster C1 is the operation of the hydraulic system 1 without leakage. A second cluster C2 is an operation of the hydraulic system 1 with gradual leakage. A third cluster C3 is an operation of the hydraulic system 1 with extreme leakage.

The cluster analysis model 7 searches for the respective centers (marked with X) of the various clusters in a data set and assigns the measured values recorded by the sensors 2 to individual centers and thus to individual clusters.

It is of course conceivable that the recorded measured values are divided into more than three clusters.

6 shows a temporal progression of the oil level in the oil tank 3. Here, too, there are temporal fluctuations in the level due to different distribution of working fluid in the hydraulic system 1. However, the trend/average of the level of the oil tank 3 is pointing downwards. Therefore, this progression is a hydraulic system 1 with a leak. The horizontal line can illustrate a target level of the oil tank 3.

List of reference symbols

1

hydraulic system

2

Sensors

3

Oil tank

4

Control unit

5

Output unit

6

Monitoring device

7

Cluster analysis model

8th

Forecast model

C1

first cluster

C2

second cluster

C3

third cluster

Claims (9)

Computer-implemented monitoring method for detecting leaks in a hydraulic system (1), comprising the steps of: - recording measured values of the hydraulic system (1) by sensors (2), in particular oil temperature, oil level and/or pressure values of the hydraulic system (1); - entering the recorded measured values into a trained cluster analysis model (7); - dividing the recorded measured values into different predefined clusters by the trained cluster analysis model (7), characterized in that the trained cluster analysis model (7) has at least three clusters, wherein a first cluster (C1) is preferably a regular operation of the hydraulic system (1), a second cluster (C2) is preferably an operation of the hydraulic system (1) with a gradual leak, and a third cluster (C3) is preferably an operation of the hydraulic system (1) with a rapid leak. Monitoring procedures according to Claim 1 , characterized in that a warning is issued to a user by an output unit (5) when the trained cluster analysis model (7) assigns the recorded measured values to the cluster of operation of the hydraulic system (1) with the creeping leakage or to the cluster of operation of the hydraulic system (1) with the rapid leakage. Monitoring procedure according to one of the Claims 1 or 2 , characterized by a step in which the cluster analysis model (7) is trained with a training data set which has historical measured values of the hydraulic system (1) as input values, in particular before the step of entering the recorded measured values. Monitoring procedure according to one of the Claims 1 until 3 , characterized in that the cluster analysis model (7) is based on a k-means algorithm or DBSCAN algorithm. Monitoring procedure according to one of the Claims 2 until 4 , characterized by the steps: - predicting a future course of measured values based on the recorded measured values by a trained forecast model (8) of the cluster analysis model (7); - calculating a time period until a critical oil level is reached in an oil tank (3) of the hydraulic system (1) from the predicted future measured values; - issuing a warning to the user by the output unit (5). Monitoring procedures according to Claim 5 , characterized in that the forecast model (8) is trained with a training data set which has at least one temporal section of the historical measured values as input values and another temporal section of the historical measured values as output values, in particular before the forecasting step. Monitoring procedure according to one of the Claims 1 until 6 , characterized by the step of training the cluster analysis model (7) with a training data set which has historical measurement data of an operation of the hydraulic system (1) with a defect, so that the cluster analysis model (7) recognizes an operation with the defect. Hydraulic system (1) with a number of sensors (2) which record measured values of the hydraulic system (1), and a monitoring device (6) for detecting leaks in the hydraulic system (1) with a cluster analysis model (7) with at least three clusters, wherein a first cluster (C1) is preferably a regular operation of the hydraulic system (1), a second cluster (C2) is preferably an operation of the hydraulic system (1) with a creeping leak and a third cluster (C3) is preferably an operation of the hydraulic system (1) with a rapid leak, wherein the model (7) is adapted and designed to classify the measured values recorded by the sensors (2) into the various clusters. Computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the monitoring method according to one of the Claims 1 until 8th to execute.

Images