METHOD FOR DETERMINING THE DEGREE OF CONTAMINATION OF A FILTER

Description

[0001] The present invention relates to a method for determining the degree of contamination of a filter and a control device for carrying out such a method.

[0002] In conventional systems, filter contamination is predominantly determined solely by the pressure difference before and after the filter element. This is especially true in hydraulic systems where pressure monitoring is only available based on threshold values.

[0003] The respective pressure thresholds for detecting contamination are usually determined on an application-specific basis. A new, uncontaminated filter element, for example, can be identified by a pressure difference of less than 1 bar. A pressure difference of 2.5 bar or more can be considered to indicate contamination of approximately 80%. A pressure difference of 5 bar or more, for example, is considered to be 100% contamination.

[0004] Plotted over the service life, the pressure loss occurring in filters typically behaves like a kind of exponential function. This is due to increasing contamination, whereby the filter becomes less permeable and thus becomes increasingly contaminated. Thus, the contamination of the filter increases the rate of contamination, creating a positive feedback loop that causes the filter to become completely contaminated.

[0005] Monitoring the contamination of a filter by means of sensor-based pressure monitoring is a suitable approach, but under complex conditions - such as changing pressures, sudden pressure surges, significant changes in the volume flows occurring and a change in the viscous properties of the respective fluids - it sometimes does not provide sufficiently reliable and / or insufficiently accurate results.

[0006] DE102020105403A1 describes a method for determining the degree of contamination of a filter element in a component of a fluid power system, which comprises the following steps: measuring a differential pressure present in the component; measuring a temperature of a fluid flowing through the filter element; determining a tolerable differential pressure using a value matrix in which experimentally determined differential pressures and temperatures are correlated with each other at at least one predetermined tolerable filter contamination degree; comparing the measured differential pressure with the determined tolerable differential pressure, and issuing a maintenance notification when the measured differential pressure reaches or exceeds the tolerable differential pressure.The value matrix can contain multiple differential pressures, which could be recorded across multiple flow rates, temperatures, and at least one tolerable degree of contamination of the filter element. The value matrix is ​​multidimensional and can provide information about the relationship between differential pressure and degree of contamination under specific operating conditions. In particular, knowing the fluid temperature can be used to determine a tolerable differential pressure as a maximum value from the value matrix. If possible, the value matrix could also represent multiple operating scenarios of the fluid power system under consideration. The tolerable differential pressure could be different for multiple operating scenarios. The tolerable differential pressure could therefore be determined not only on the basis of the measured temperature, but also on the basis of information about the respective operating scenario.Even if this already represents an improvement in the situation with pure pressure monitoring, a more stable and situationally higher quality determination or assessment of the degree of contamination of the filter is desirable.

[0007] For example, it would be desirable to have the option of calculating the remaining service life of the filter as accurately as possible based on a determined degree of contamination.

[0008] In addition, it is desirable to determine the degree of contamination in order to avoid unplanned downtimes due to the need to change the filter at short notice.

[0009] In addition, it is desirable to determine the degree of contamination in order to avoid the risk of damage to the system due to insufficient maintenance cycles or late maintenance work or filter replacement.

[0010] Deterioration of fluid quality due to contaminated filter elements should also be avoided if possible.

[0011] In addition, detection of damage or wear in the system is desirable.

[0012] The object of the present invention is to further improve filter contamination monitoring using artificial intelligence, particularly with the aim of ensuring that this monitoring system functions reliably under the described special conditions, such as changing pressures, fluid viscosities, etc., and contributes to the safe operation of the system with the filter. In particular, the problems of known approaches mentioned above are to be at least alleviated.

[0013] This is achieved by the invention according to the features of the independent patent claims. Further advantageous embodiments are specified in the dependent claims as well as in the description and, in particular, in the description of the figures. It should be noted that the person skilled in the art can combine the individual features in a technologically expedient manner and thus arrive at further embodiments of the invention.

[0014] A method for determining the degree of contamination of a filter in a hydraulic system contributes to this, which comprises at least the following steps: a) Determining a pressure difference value that occurs across the filter. b) Determining boundary conditions that describe a relationship between the pressure difference value and the degree of contamination and that are present at least temporarily during step a). c) Assigning the (determined) pressure difference value and the (determined) boundary conditions to a pattern (or an anomaly) using a pattern database stored in a memory. d) Determining a degree of contamination associated with the pattern from the pattern database stored in the memory.

[0015] The described method goes beyond simply monitoring the current condition of the filter by simply observing the pressure difference that occurs and comparing it with threshold values.

[0016] Threshold values ​​for comparison with the pressure difference are usually based exclusively on empirical values ​​and rough curves that describe the relationship between the pressure difference and the degree of contamination. An exact prediction is not possible because the curves describing the relationship between the pressure difference and the degree of contamination are very flat, especially at the beginning of contamination, and thus even small absolute inaccuracies can lead to very significant inaccuracies. If necessary, characteristic maps can be provided that provide threshold values ​​adapted to the respective conditions.

[0017] Overall, it can be seen that the pressure difference depends on a large number of disturbances. Disturbances that should be highlighted here include, in particular, viscosity as a function of temperature, relationships between pressure and volume flow changes, and the choice of the respective fluid—for example, the selection of certain oil classes as fluids.

[0018] Until now, the available parameters were processed using analytical methods in order to determine the degree of contamination of the filter from the pressure differences that occurred.

[0019] However, the difficulty arises that the available parameters regularly represent large amounts of data, the processing of which requires long computing times. The influence of disturbances is significant, resulting in rather inaccurate statements. Due to these uncertainties, filters must be changed frequently or early (i.e. conservatively), which results in high hardware consumption, high environmental impact, and high costs. Frequently, the statement regarding the degree of filter contamination is so inaccurate that manual filter inspections are necessary. Despite all this, it is possible that filter changes are carried out too late, contaminated filters can cause contamination in the fluid, and unplanned downtimes may become necessary.

[0020] In principle, it is possible for the steps of the method to be executed (at least once) in the specified order (at least partially sequentially). It is possible for the steps to overlap at times or even be executed in parallel. In particular, the determination of different data can be overlapped in time and / or executed at different frequencies.

[0021] Step a) can involve determining a specific pressure difference value or variables derived from it. Predefined pressure difference ranges and / or pressure difference changes can be classified, and the corresponding class or class limits can be determined. The boundary conditions can be determined using sensors and / or computationally.

[0022] In step b), at least one predefined boundary condition is determined, whereby, if necessary, several or a plurality of different boundary conditions can be determined. It is possible to determine specific values ​​of a boundary condition or a variable derived from it. It is possible to classify predetermined boundary condition value ranges and / or boundary condition changes, and to determine the associated class or class limits. The boundary conditions can be determined using sensors and / or computation.

[0023] Further, according to step c), based on at least one (determined) pressure difference value and at least one (determined) boundary condition, an assignment to a (predetermined) pattern is made. This assignment accesses a pattern database stored in a memory. A pattern thus comprises information on the pressure difference value and the boundary conditions, so that it can be automatically selected depending on the specific values.

[0024] Step d) involves determining a contamination level associated with the (assigned) pattern, which is also derived from the pattern database stored in the memory. Thus, a pattern is, in particular, a multi-parameter pattern in which at least two input variables (here, the pressure difference value and the boundary condition(s)) lead to a predetermined output variable (here, the contamination level).

[0025] If no pattern can be identified in steps c) or d), it is initially considered an anomaly. If this pattern occurs multiple times, it is labeled so that similar data points can be assigned to this pattern in the future.

[0026] To construct the sample database used in step c) stored in the memory, a minimum amount of data triplets is collected, comprising pressure difference values, framework data concerning occurring framework conditions and information concerning the respective existing contamination levels.

[0027] The construction of the pattern database stored in the memory includes the assignment of data triples to clusters.

[0028] It is advantageous if the clusters are assigned along a curve that describes the expected course of the pressure difference values ​​over the lifetime of the filter.

[0029] It is particularly advantageous if an automatically learning algorithm is used to construct the pattern database stored in the memory used in step c), which processes the data triples and assigns them to clusters.

[0030] Clusters can be formed along a curve that describes the expected course of the pressure difference values ​​over the lifetime of the filter.

[0031] The formation of clusters and the construction of the pattern database preferably takes place in an initialization mode, which can also be referred to as synthesis mode or training mode. The initial formation of clusters and the construction of the pattern database can also be understood as training the algorithm for the respective application.

[0032] As a starting point for training the algorithm, measurement data or a database is imported. Preferably, these measurement data or this database produce a pressure difference curve over time for the respective filter, based on historical data or the database.

[0033] Based on the resulting curve (pressure difference over time), the data is categorized using clustering algorithms so that objects are grouped into the same cluster if they are similar according to certain metrics. For example, between three and eight, preferably five, different clusters are formed based on the life expectancy over time, each representing a time interval of the filter's life expectancy.

[0034] The approach described here uses the sample database described, which can be evaluated very efficiently based on the pressure difference values ​​in order to determine a concrete (current) degree of contamination, taking into account available information regarding the framework conditions.

[0035] With regard to the boundary conditions stored in the sample database, it should be noted that these may be current boundary conditions (those existing at the time the pressure difference value was determined). However, they may also be historical boundary conditions, such as information regarding the fluid used, e.g., a specific oil quality. Depending on the oil quality used in the period prior to the determination of the pressure difference value, the pressure difference value may, for example, indicate a different (deviating) degree of contamination.

[0036] The approach described here uses machine learning and numerical methods to preprocess large amounts of data consisting of the described data triples to build a sample database. Generating the sample database can easily involve a significant amount of computational effort because the sample database can be generated in advance (possibly with a simulation) on dedicated hardware. The total amount of data contained in the sample database can also be significantly reduced compared to the amount of raw data contained in the described data triples.When creating the pattern database, a type of information extraction is preferably carried out, in which information is extracted from the data triplets and processed in the pattern database in such a way that the question of which degree of filter contamination can be expected at which pressure drop and under which conditions can be answered efficiently and with little computational effort using the pattern database. The pattern database is not static after creation and can therefore be continuously expanded and improved. The described method can therefore also be carried out in a (small) dedicated control unit with (relatively low) computing power, where "relatively low" here means in particular that the computing power can be very small (i.e., up to 100 times smaller or even up to 1,000 times smaller) compared to the computing power of the hardware used to create the pattern database.

[0037] The described approach for constructing a pattern database and determining the degree of contamination based on this pattern database is particularly effective for eliminating disturbances. Such disturbances are typically contained in the boundary conditions determined in step b) and can be efficiently accounted for in the pattern database or during the construction of the pattern database. Even if a boundary condition is subject to significant disturbance, the pattern database provides appropriate information regarding the degree of contamination because such disturbances were efficiently accounted for during the construction of the pattern database.

[0038] It is advantageous if, based on the pattern assigned in step c) and / or the degree of contamination determined in step d), a remaining service life of the filter is determined in step e).

[0039] It is advantageous if a filter change request is issued in step e) if the degree of contamination determined in step d) exceeds a limit value.

[0040] The pattern database used in step c) can advantageously be updated in a field operation.

[0041] Preferably, the pattern database used is also updateable. For example, a control unit for implementing the described method can be updated during maintenance intervals to receive a new / updated pattern database, so that the method is then executed using the updated pattern database.

[0042] In further preferred embodiments, the described method itself is part of a continuous improvement process of the pattern database. For example, during a filter change, actual contamination data can be determined directly. If necessary, pressure drop data and boundary conditions determined during or shortly before the filter change can be entered together with the actual contamination data as a data triplet into an algorithm for improving the pattern database.

[0043] Overall, a constant improvement process can be created, which preferably forms a self-learning algorithm that continuously adapts itself or the pattern database on which the detection of the degree of contamination is based, the more data is available.

[0044] The better or more comprehensive the algorithm or pattern database, the better the prediction of the contamination level. In particular, the pattern database can detect, map, and account for anomalies and relationships between pressure drop, underlying conditions, and the contamination level that are difficult or impossible for humans to understand.

[0045] According to a further aspect, a control device is proposed which is configured to carry out the described method.

[0046] It should be noted that the particular advantages and design features described in connection with the method described above are also applicable and transferable to the control unit described below.

[0047] The invention and the technical context of the invention are explained in more detail below with reference to the figures. The figures show preferred embodiments to which the invention is not limited. It should be noted in particular that the figures, and in particular the proportions depicted in the figures, are only schematic. They show: Fig. 1: a typical hydraulic system with filters, Fig. 2: a flow diagram of the described method, Fig. 3: a diagram describing an expected pressure increase over time as a result of contamination, Fig. 4: data clusters of a sample database, Fig. 5: schematically a prediction of a remaining filter service life, determined using the described method, and Fig. 6: a cost optimization achievable using the described method.

[0048] Fig. 1shows a hydraulic system 1 with a driven hydraulic component 2, a pump 3 for generating pressure for the hydraulic valve, a supply line 6, a return line 7, and a reservoir 8 for hydraulic fluid. Filters 4 are installed in each of the supply line 6 and the return line 7. Pressure sensors 5 can be used to determine a pressure difference across the respective filters 4. The filters 4 serve to ensure oil cleanliness. These are often found in three different positions, although two different installation methods are generally common, both of which are described here.

[0049] There are filters 4 that are installed, for example, downstream of a pump 3 and upstream of the hydraulic control 9 to protect the sensitive control 9 from contamination. In such filters 4, a (defined) pressure is usually present on both sides. Therefore, two pressure sensors 5 are usually used to determine the pressure difference or temperature - a first pressure sensor 5 on one side of the filter 4 and a second pressure sensor 5 on the second side of the filter 4. Such an arrangement is shown here in the feed line 6. The pressure sensors 5 are connected to a control unit or the machine control 35 to carry out the described method.

[0050] There are also return filters 4 or bypass filters 4, which are installed, for example, in a return line 7 to clean the fluid flowing back into the reservoir 8. Since atmospheric pressure prevails on the reservoir 8 side, only one sensor 5 needs to be used to determine the differential pressure. Atmospheric pressure can simply be assumed for the side of the filter 4 facing the reservoir 8.

[0051] Based on Fig. 2 The sequence of the described procedure is now shown. The representation in Fig. 2 combines two different aspects of the described procedure. In the Fig. 2Field operation 21 is shown, in which the described method is applied to determine the degree of contamination 15 of a filter 4. Field operation 21 is the actual method performed and described here. This field operation 21 is carried out, for example, in the control system 35 of a hydraulic system. Fig. 2 Also shown is an initialization operation 22 (or synthesis operation), which serves the purpose of building or synthesizing a pattern database 14, which is required for carrying out the field operation 21 in step c). The initialization operation 22 or synthesis operation precedes the method described here and, from a higher-level perspective, can be understood as a sub-aspect of the described method. Both aspects / methods (field operation 21 and initialization operation 22) have identical steps, which are also shown here jointly for both aspects / methods.

[0052] Both field operation 21 and initialization operation 22 begin with the acquisition of pressure values ​​11 using pressure sensors, from which pressure differential values ​​12 are then generated, which describe the pressure differential across the filter 4. The respective pressure differential values ​​12 must always be viewed in the context of the respective boundary conditions 10. Boundary conditions 10 here include, for example, the applied volume flow through the filter 4, the viscosity of the respective fluid, etc. The boundary conditions 10 can be considered in various ways, some of which are listed here as examples: The boundary conditions 10 can be captured using measurements. The boundary conditions 10 can be calculated from other available parameters, or the boundary conditions 10 can be permanently stored or defined. This implementation variant is particularly possible if the boundary conditions 10 do not change.

[0053] In initialization mode 22, a clustering 13 is now carried out using the pressure difference values ​​12 and the boundary conditions 10 as well as the respective corresponding contamination data 23, and a sample database 14 is created for the first time. The sample database 14 contains data triples 33 consisting of boundary condition data relating to boundary conditions 10, pressure drop data consisting of pressure difference values ​​12, and a respective associated contamination level 15 described by contamination data 23. Based on this sample database 14, a contamination level 15 can be determined for a given pressure drop (described by a pressure difference value 12) in field operation 21 under existing / known boundary conditions 10.

[0054] Pressure difference values ​​12, framework conditions 10 and contamination data 23 each form data triples 33 in the sample database 14. The data triples 33 stored in the sample database 14 are assigned to a previously determined expected curve of the pressure difference over the entire operating time of the filter 4 with the help of clustering 13, which is then calculated using the Fig. 3 to 5 will be described in more detail below. Based on the clustering 13 or the clusters 28, the efficient assignment of determined pressure difference values ​​12 or boundary conditions 10 to a specific contamination level 15 is possible. In preferred embodiments, a (common, average) contamination level 15 is stored for all pressure difference values ​​12 or boundary conditions 10 in a cluster 28, which can then be determined in step d) of the described method in field operation 21.

[0055] In Fig. 2It is also shown that following step d), a comparison of the degree of contamination 15 with a limit value 17 can also be made and, depending on whether the degree of contamination 15 exceeds the limit value 17, a filter change request 16 is then issued.

[0056] Fig. 2also shows that field operation 21 and initialization operation 22 or synthesis operation can also merge into one another. For example, it is possible for pressure difference data 18 and framework condition data 20 from field operation 21 to also flow into an initialization operation 22 taking place parallel to field operation 21 and to be used to improve the pattern database 14. Preferably, pattern selection data 19 also flow into this initialization operation 22 or synthesis operation, indicating which patterns were assigned in step c) and which contamination levels 15 were determined thereby. By utilizing all of this data, clustering 13 and pattern database 14 can be continuously improved, and patterns unknown up to the time of measurement can be incorporated. This processing of data from field operation 21 to improve the database (the pattern database 14) takes place continuously during operation if necessary.If necessary, this improvement of the database also takes place sporadically, in which data from field operation 21 is read out at various times and assigned to a central unit (e.g., a server) to then be used to improve the pattern database 14. Contamination data 23 that were actually detected during field operation 21 when changing the filter 4 (possibly through observation by a technician changing the filter 4) can also be included here. The improved pattern database 14 can then be imported sporadically (at various times), for example, in the form of an update, into a control unit that carries out field operation 21.

[0057] Fig. 3shows a curve 26, which represents an expected course of the pressure difference value 12 during the service life 25 of the filter 4. The service life 25 of the filter 4 is shown on the time axis 24. The curve 26 represents this expected course assuming constant framework conditions 10. In Fig. 3Also recorded as an example is an actual pressure curve 27. In particular due to changing framework conditions 10, this actual pressure curve 27 fluctuates around the curve 26. The pressure difference value 12 across the filter 4 continues to rise as the filter 4 ages. This increase occurs (approximately) according to an exponential function. The degree of contamination 15 can be classified using the pressure difference value 12. The relationship between the degree of contamination 15 and the pressure difference value 12 can (but does not have to) be linear. The property of an exponential function is due in particular to the fact that with increasing degree of contamination 15, the contamination normally increases ever more rapidly, in particular because the permeability of the filter 4 decreases. The contamination of the filter 4 can therefore be understood as a positive feedback process. Fig. 3also schematically shows a limit value 17. When the contamination or the pressure difference value 12 reaches this limit value 17, a filter change request 16 should be issued. Predictions about the remaining service life of the filter can also be issued before the limit value 17 is reached. However, depending on the classification in cluster 28, a varying tolerance must be output due to measurement inaccuracies (the greater the contamination level 15, the smaller the tolerance).

[0058] Based on Fig. 4Clustering 13 is now explained. Here, the curve 26 of the pressure difference value 12 is also plotted over the service life 25 of the filter 4 on the time axis 24. Various data triplets 33 consisting of pressure difference values ​​12, boundary conditions 10, and associated contamination data 23 are divided into clusters 28, which lie along the course of the curve 26. By assigning data triplets 33 to clusters 28, an associated contamination level 15 can be efficiently determined for each cluster 28.

[0059] Fig. 5now shows how, based on curve 26 and clustering 13, a correction of the service life 25 of filter 4 during field operation 21 can be made. A first curve 26 is shown here, for example. Based on clusters 28, it is now possible to determine in which area of ​​curve 26 the pressure difference value 12 or the degree of contamination 15 currently lies. If filter 4 is not yet that contaminated, this is determined based on the (unexpectedly) low pressure difference value 12, and curve 26 can be shifted in time so that a corrected service life 34 can be determined.

[0060] Fig. 6shows how the described method can achieve an optimal value 32 of efficiency for filter changes. Shown are the downtime costs 29 that occur when the hydraulic system 1 is shut down as a result of a filter change. Also shown are storage costs 30 that arise from storing filters 4, as well as total costs 31, which are composed of downtime costs 29 and storage costs 30. The described method can achieve an optimal value 32 of minimal total costs 31. List of reference symbols

[0061] 1Hydraulic system 2Hydraulic component 3Pump 4Filter 5Pressure sensor 6Inlet line 7Return line 8Reservoir 9Control 10Framework conditions 11Pressure value 12Pressure difference value 13Clustering 14Pattern database 15Contamination level 16Filter change request 17Limit value 18Pressure difference data 19Pattern selection data 20Framework condition data 21Field operation 22Initialization operation 23Contamination data 24Time axis 25Service life 26Curve 27Actual pressure curve 28Cluster 29Downtime costs 30Storage costs 31Total costs 32Optimal value 33Data triplet 34Corrected service life 35Control unit

Claims

1. Method for determining the degree of contamination (15) of a filter (4) in a hydraulic system (1), comprising at least the following steps: a) ascertaining a pressure difference value (12) which occurs across the filter (4); b) ascertaining boundary conditions (10) which describe a relationship between the pressure difference value (12) and the degree of contamination (15) and which are present during the determined pressure difference value (12); c) assigning the pressure difference value (12) and the boundary conditions (10) to a pattern based on a pattern database (14) stored in a memory, wherein, to build the used pattern database (14) stored in the memory, a minimum amount of data triples (33), comprising - pressure difference values (12), - boundary condition data (20) relating to arising boundary conditions (10), and - information relating to the respective degree of contamination (15) present, are collected; d) ascertaining a degree of contamination (15) assigned to the pattern from the pattern database (14) stored in the memory, characterized in that the building of the pattern database (14) stored in the memory comprises the assignment of data triples (33) to clusters (28).

2. Method according to Claim 1, wherein the clusters (28) are assigned along a curve (26) which describes the expected profile of the pressure difference values (12) over a lifetime (25) of the filter (4).

3. Method according to Claim 1, wherein, to build the pattern database (14) stored in the memory and used in step c), an automatically learning algorithm which processes the data triples (33) and assigns same to clusters (28) is used.

4. Method according to Claim 3, wherein clusters (28) are formed along a curve (26) which describes the expected profile of the pressure difference values (12) over a lifetime (25) of the filter (4).

5. Method according to any one of the preceding claims, wherein, in step e), a residual lifetime of the filter (4) is determined based on the pattern assigned in step c) and / or the degree of contamination (15) ascertained in step d).

6. Method according to any one of the preceding claims, wherein, in step e), a filter change request (16) is output if the degree of contamination (15) ascertained in step d) exceeds a limit value (17).

7. Method according to any one of the preceding claims, wherein the pattern database (14) used in step c) can be updated in field operation (21).

8. Controller (35) set up to perform the method according to any one of the preceding claims.