Pump and method of determining a flow rate of a pump

Abstract

The invention relates to a pump (10), in particular a centrifugal pump, which comprises: a motor (11); and a controller (12) configured to control a speed of the motor (11); wherein the controller (12) comprises a trained machine learning, ML, model (13) being supplied with input data, the input data comprising at least two of: an electrical power intake of the motor (11), a speed of the motor (11), and a differential pressure value. The ML model (13) is configured to output flow value data based on the input data.

Description

[0001] PUMP AND METHOD OF DETERMINING A FLOW RATE OF A PUMP

[0002] TECHNICAL FIELD OF THE INVENTION

[0003] The invention relates to a pump, in particular a centrifugal pump, and to a method of determining a flow rate of a pump.

[0004] BACKGROUND OF THE INVENTION

[0005] A centrifugal pump is a type of pump that raises or transfers a fluid by converting rotational energy from an impeller into hydrodynamic energy of the fluid flow.

[0006] Such pumps are often equipped with flow meters which can be used for monitoring and regulating the rate of fluid flow through the pump. By providing feedback on flow rates, they enable precise control of pump operation, e.g. , to maintain a constant flow rate. However, the integration of such flow meters in the pump increases the cost and complexity of the pump.

[0007] In general, the flow depends on various performance parameters of the pump, such as pump speed or motor power. However, estimating the flow rate from these parameters would require a complex, CPU heavy model (e.g. , based on interpolation of the parameters) which might be unfeasible for a pump with a high efficiency impeller and a complex H-Q curve. If a differential pressure sensor input is further added to such a model, its complexity and the required CPU load further increases.

[0008] Thus, it is an objective to provide an improved pump and an improved method of determining a flow rate of the pump which avoid the above-mentioned disadvantages. In particular, it is an objective to determine a flow rate of a pump in an accurate and fast manner.

[0009] SUMMARY OF THE INVENTION

[0010] The object of the present invention is achieved by the solution provided in the enclosed independent claims . Advantageous implementations of the present invention are further defined in the dependent claims .

[0011] According to a first aspect, the invention relates to a pump, comprising: a motor; and a controller configured to control a speed of the motor. The controller comprises a trained machine learning (ML) model, for example a trained neural network, being supplied with input data, the input data comprising at least two of: an electrical power intake of the motor, a speed of the motor, and a differential pressure value. The ML model is configured to output flow value data based on the input data.

[0012] This achieves the advantage that a pump is provided which can accurately predict its flow rate without requiring a dedicated flow meter.

[0013] For instance, the input data comprises the differential pressure value and at least one of: the electrical power intake of the motor, and the speed of the motor.

[0014] The pump can be a centrifugal pump. The pump can comprise an impeller which is arranged in a pump housing and which is connected to the motor by a shaft. The motor can be an electric motor . For example, the input data comprises all three of: the electrical power intake, the actual speed of the motor and the differential pressure value.

[0015] The flow value data can represent a flow rate, in particular the current flow rate, of the pump.

[0016] The machine learning model can be a neural network model (or short: a neural network) . For example, the neural network model can be a convolutional neural network (CNN) model. However, other neural network models are also possible.

[0017] As an alternative to a neural network, the ML model could also be a decision tree model or a deep learning model.

[0018] In an embodiment, the controller is configured to adapt a pump setting, in particular the speed of the motor, based on the flow value data. This achieves the advantage that the pump can be precisely controlled based on the determined flow values, e.g. , to maintain a constant flow rate.

[0019] In an embodiment, the controller is configured to adjusting a maximum level of the electrical power intake of the motor based on a user input; and the ML model is supplied with the maximum level as further input data.

[0020] The pump can comprise a user interface for receiving the user input, e.g. a GUI or other input interface. The user input could also be transmitted to the pump via a wireless communication interface (e.g. , Bluetooth) .

[0021] In an embodiment, the input data further comprises a temperature, a density, and / or a viscosity of a fluid to be pumped. The fluid can be water.

[0022] For example, the input data can comprise further sensor and / or measurement data, such as system pressure of the pump.

[0023] For instance, if the pressure at some location in the pump is nearing the steam pressure of the medium (often water) , local boiling followed by implosion may occur. This phenomenon is called cavitation. Cavitation can influence the characteristics of the pump, including the flow. In order to take this into account the system pressure can be added as input data.

[0024] In an embodiment the input data further comprises a geometric parameter of the pump and / or information on its installation.

[0025] By taking into account additional input data (e.g. , information on the fluid) , the accuracy of the determined flow values can be enhanced. For instance, the ML model is trained on the relationship between the different input parameters and their influence on the flow rate.

[0026] In an embodiment, the pump further comprises a differential pressure sensor which is configured to detect the differential pressure value.

[0027] For example, the differential pressure sensor comprises two pressure sensing elements which are arranged at a pump intake and at a pump outtake, respectively.

[0028] In an embodiment, the ML model is trained on individual characteristics of the pump. This achieves the advantage that the accuracy of the generated flow values can be enhanced.

[0029] In an embodiment, for training the ML model, the pump is exposed to determined flow conditions; wherein the controller is configured to collect further input data while the pump is exposed to the determined flow conditions; and wherein the ML model is configured to use the collected further input data as training data. This achieves the advantage that the ML model can be trained efficiently on the characteristics of the individual pump .

[0030] For instance, for training the ML model, the pump can be mounted in a test rig which provides the determined flow conditions. The determined flow conditions can be "known" flow conditions.

[0031] The further input data can comprise at least two of : electrical power intake of the motor, speed of the motor, and a differential pressure value. The further input data can be collected by the controller while the pump is mounted in the test rig.

[0032] The determined flow conditions can cover the full flow range (Q- range) of the pump. In this way, the ML model can be trained for the full flow range and not only specific regions of Q and H.

[0033] In an embodiment, the ML model is configured to correlate the collected further input data with the determined flow conditions, for instance in the test rig, and to adjust its configuration based on said correlation.

[0034] In case of the ML model being a neural network, weights of the neural network can be adjusted based on said correlation. In a neural network, the weights can control the signal between two neurons (i.e. , how much influence the input has on the output) . In addition, the neural network can comprise biases which are constant and which are not influenced by a previous layer of the neural network. The biases are an additional input into a next neural network layer.

[0035] In an embodiment, the trained ML model is configured to continuously or repeatedly adjust its configuration, for instance the weights of the neural network, over the lifetime of the pump based on the input data received over time. This achieves the advantage that the ML model, for instance the neural network, can continuously adapt on the pump environment and / or use .

[0036] In an embodiment, the trained ML model is configured to further output information on a status of the pump based on the input data. For example, the information on the status can comprise information on a wear of the pump and / or on changes of the pump performance over time .

[0037] In an embodiment, the ML model is configured to output at least two flow value data estimates, wherein each flow value data estimate is determined based on a different combination of the input data; and the controller and / or the ML model are configured to detect a pump inefficiency if the at least two flow value data estimates differ from each other by more than a threshold value .

[0038] For instance, the controller and / or the ML model can track deviations of the flow value data calculated with different combinations of input data over time. In this way, an increase in pump inefficiency and / or a pump degradation over time can be tracked .

[0039] The output information on the status of the pump can comprise information on the detected pump inefficiency.

[0040] In an embodiment, the pump further comprises a communication interface which is configured to receive further input data from at least one further pump; the further input data comprising at least two of: an electrical power intake of a motor of the further pump, a speed of the motor of the further pump, and a differential pressure value related to the further pump; wherein the ML model is configured to output further flow value data relating to the at least one further pump based on the further input data.

[0041] The pump can be a primary pump and the at least one further pump can be a secondary pump. For instance, the at least one further pump does not comprise a trained ML model.

[0042] According to a second aspect, the invention relates to a method of determining a flow rate of a pump, wherein the pump comprises a motor and a controller configured to control a speed of the motor, wherein the controller comprises a trained machine learning (ML) model, for example a trained neural network. The method comprises the steps of: supplying the ML model with input data, the input data comprising at least two of: an electrical power intake of the motor, a speed of the motor, and a differential pressure value; and outputting flow value data based on the input data with the ML model.

[0043] This achieves the advantage that a method is provided to determine the flow rate of the pump in an accurate and fast manner without requiring a dedicated flow meter. For instance, the input data comprises the differential pressure value and at least one of: the electrical power intake of the motor, and the speed of the motor.

[0044] The flow value data can represent the flow rate, in particular a current flow rate, of the pump.

[0045] In an embodiment, the method comprises the further step of: adapting a pump setting, in particular the speed of the motor, based on the flow value data. This achieves the advantage that the pump can be precisely controlled based on the determined flow values, e.g. , to maintain a constant flow rate.

[0046] In an embodiment, the ML model is trained by: exposing the pump to determined flow conditions, collecting further input data while the pump is exposed to the determined flow conditions, and using the collected further input data as training data for training the ML model. This achieves the advantage that the ML model, e.g. a neural network, can be trained efficiently on the characteristics of the individual pump.

[0047] The further input data can comprise at least two of : electrical power intake of the motor, speed of the motor, and a differential pressure value.

[0048] The method may comprise the step of mounting the pump in a test rig which provides the determined flow conditions. The further input data can be collected by the controller while the pump is arranged in the test rig.

[0049] The determined flow conditions can cover the full flow range (Q- range) of the pump.

[0050] In an embodiment, the ML model is configured to correlate the collected further input data with the determined flow conditions and to adjust its configuration based on said correlation.

[0051] In an embodiment, the method comprises the further step of: continuously or repeatedly adjusting the configuration of the ML model over the lifetime of the pump based on the input data received over time.

[0052] In an embodiment, the method comprises: outputting at least two flow value data estimates with the ML model, wherein each flow value data estimate is determined based on a different combination of the input data; and detecting a pump inefficiency if the at least two flow value data estimates differ from each other by more than a threshold value.

[0053] In an embodiment, the method further comprises: receiving further input data from at least one further pump; the further input data comprising at least two of: an electrical power intake of a motor of the further pump, a speed of the motor of the further pump, and a differential pressure value related to the further pump; and the method further comprises: outputting further flow value data relating to the at least one further pump based on the further input data.

[0054] The method according to the second aspect of the invention can be carried out by the pump according to the first aspect of the invention .

[0055] BRIEF DESCRIPTION OF THE DRAWINGS Embodiments of the invention will be explained in the following together with the figures.

[0056] Fig. 1 shows a schematic diagram of a pump according to an embodiment ;

[0057] Fig. 2 shows a schematic representation of the correlation between flow rate and other pump parameters according to an embodiment;

[0058] Fig. 3 shows a flow chart of a method of determining the flow rate of a pump according to an embodiment; and

[0059] Fig. 4 shows a flow chart of a method of training an ML model of a pump according to an embodiment.

[0060] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0061] Fig. 1 shows a schematic diagram of a pump 10 according to an embodiment .

[0062] The pump 10 comprises: a motor 11; and a controller 12 configured to control a speed of the motor 11. The controller 12 comprises a trained machine learning (ML) model 13 being supplied with input data, the input data comprising at least two of: an electrical power intake of the motor, a speed of the motor, and a differential pressure value. The ML model 13 is configured to output flow value data based on the input data.

[0063] The pump 10 can be a centrifugal pump. For instance, the pump 10 can be used as a circulation pump in a water circulation system. The motor 11 can be an electric motor. The motor 11 can be configured to drive an impeller of the pump 10. The impeller can be arranged in a section of the pump housing 16 with the motor 11 being connected to the impeller via a shaft. The impeller can be a high efficiency impeller with a complex three-dimensional design. For instance, a high efficiency impeller has a non- monotonous power curve, such that that there can be two possible flow solutions to a set of power and speed input. The speed of the motor 11 can be a rotational speed of the motor (e.g. , an RPM value) . The motor can be configured to forward information on its current speed to the controller 12, e.g. , via a communication connection.

[0064] The electrical power intake of the motor 11 can refer to the electrical power which is supplied to the motor 11. The pump 10 can comprise means to detect this electrical power intake, e.g. , a current and / or voltage sensor.

[0065] The controller 12 can be a component of the pump electronics. The controller 12 can comprise at least one processor which is configured to execute the ML model 13. The controller 12 can be configured to supply the input data to the ML model 13.

[0066] For instance, the input data comprises all three of: the electrical power intake of the motor, the speed of the motor, and the differential pressure value.

[0067] The machine learning (ML) model 13 can be a neural network model. The neural network 13 can be based on a convolutional neural network (CNN) or another suitable neural network type. Alternatively, the ML model 13 could also be a decision tree model or a deep learning model.

[0068] The ML model 13 can be trained to derive the flow rate of the pump 10 from the input data. Thus, no dedicated flow meter is required which reduces the costs and complexity of the pump 10.

[0069] The resulting flow value data which are provided by the ML model 13 can represent the flow rate of the pump 10. In an example, the ML model 13 can provide the flow value data in real-time based on current input data.

[0070] Thereby, the ML model 13 can exploit the fact that the flow rate of the pump 10 depends on the power intake of the motor (power) , the speed of the motor (speed) , and the differential pressure at the pump inlet and outlet (delta p) .

[0071] Fig. 2 shows a schematic depiction of this correlation. In particular, the flow rate of the pump 10 can be expressed as a function of at least two of the three parameters: power, speed and delta p.

[0072] The performance of a pump at different flow rates can be expressed by pump characteristic curves, such as the head-flow curve (H-Q curve) or the power curve (P-Q curve) . The H-Q curve indicates the relationship between the pump head and the flow rate Q at certain speed, and the P-Q curve indicates the relationship between the power intake and flow rate Q at a certain speed. A further characteristic curve of the pump 10 is the flow over pressure curve that shows the dependency between the differential pressure and the flow rate. Besides the pump speed, the exact shape of theses curves can depend on many individual factors, such as an impeller geometry (in case of a centrifugal pump) .

[0073] For these reasons, it can be difficult to accurately predict the flow rate of pumps based on the input parameters by conventional means, e.g. , by interpolation of measured values using polynomials and pump affinity. This applies, in particular, if the pump 10 has a high efficiency impeller with complex H-Q and / or P-Q curves. For instance, the relationship between power and flow is not simply 1:1 and more than one solution for the flow Q can be possible.

[0074] The ML model 13, e.g. a neural network, can be trained to learn these complex relationships between the different parameters (power, speed and delta p) for an individual pump or for a certain pump type. For example, the ML model 13 can be trained on the individual pump (i.e. , on individual characteristics of the pump) and, thus, provide highly accurate flow rate values.

[0075] For instance, in parts of the H-Q or P-Q domain (i.e. , in different regions of the curves) , the H-Q respectively P-Q value can be dominated by only one of the parameters: power, Delta p, or motor speed. The ML model can be configured to choose the optimal parameter for a certain region of the respective curve, e.g. , depending on which region of the curve it currently operates in.

[0076] Moreover, executing the ML model 13 can require less computational resources as compared to a conventional interpolation-based approach. In particular, the ML model can be "light" both in CPU load and memory requirements, such that it can easily be hosted either on edge or in the cloud. For instance, the ML model 13 can be executed by a processor of the pump, or it can be executed by a central device which is e.g. , connected to the controller 12 of the pump 10 via a communication interface of the pump 10 (e.g. , a wireless interface such as a WiFi or a Bluetooth interface) .

[0077] The ML model 13 can be configured to output the flow value data to the controller 12. The controller 12 can then adapt one or more pump settings, in particular the speed of the motor, based on the flow value data.

[0078] In addition or alternatively, the controller 12 can output the flow value data to an interface of the pump 10. The interface can comprise a data interface (e.g. , USB) and / or a graphical user interface.

[0079] The input which is supplied to the ML model 13, and on the basis of which the ML model 13 determines the output flow value data can comprise additional data. This additional data can comprise: a temperature, a density, and / or a viscosity of a fluid to be pumped .

[0080] The input data can further comprise geometric parameters of the pump 10, e.g. a particular shape of a pump impeller or information on an installation of the pump 10.

[0081] The pump 10 can further comprise an interface 14 which is configured to receive a user input. For instance, the user input can indicate a maximum level of the electrical power intake of the pump 10.

[0082] The controller 12 can be configured to adjusting the maximum level of the electrical power intake based on the user input, and the ML model 13 can be supplied with the maximum level as further input data.

[0083] For example, the pump 10 further comprises a differential pressure sensor which is configured to detect the differential pressure value (s) . As shown in Fig. 1, the differential pressure sensor can comprise two pressure sensing elements 15a, 15b, which are arranged at a pump inlet and at a pump outlet, respectively. Thus, the differential pressure value can be representative of a pressure difference between pump inlet and outlet (e.g. , between a suction and a discharge nozzle of the pump 10) .

[0084] The differential pressure sensor can be communicatively connected to the control unit 12 for forwarding the differential pressure value (s) to the control unit 12.

[0085] Several different training strategies for the ML model 13 are possible, depending on for example whether the objective is optimal accuracy or assessment of wear.

[0086] In an example, the ML model 13 is trained by mounting the pump 10 in a test rig. The pump 10 can thereby set to a dedicated training mode for carrying out a training routine.

[0087] The test rig can provide determined flow conditions of a fluid (e.g. , water) through the pump. For example, in the test rig, known flow rates are generated through the pump. Thereby, the test rig can provide flow conditions over the entire flow range (Q range) of the pump 10.

[0088] The controller 12 can be configured to collect further input data while the pump 10 is mounted in the test rig (and the known flow rates are generated in the pump) . The ML model 13 can be configured to use the collected further input data as training data for training the ML model, e.g. the neural network. For instance, the controller 12 can collect a large number of data points (e.g. , 50.000 to 100.000 data points) to use as training data. Alternatively, a much smaller number of data points (e.g. , 1.000) could be collected. In this case, additional synthetic training data could be generated based on the collected training data and used for training.

[0089] The further input data can comprise at least two of the following parameters which are recorded while the pump 10 is mounted in the test rig: electrical power intake of the motor, speed of the motor, and differential pressure values.

[0090] During the training routine, the ML model 13 can be configured to correlate the collected further input data with the known flow conditions in the test rig and to adjust its configuration based on said correlation. In case the ML model 13 is a neural network, adjusting its configuration may correspond to adjusting its weights. In this way, the ML model can "learn" the characteristic curves of the pump (e.g. , the H-Q curve or pressure / flow curve) and, in particular, the impact of the motor speed, the electrical power intake and the differential pressure value on the flow rate Q in different regions of these curves .

[0091] For instance, the pump 10 is communicatively connected to the test rig to receive information on the provided flow rates.

[0092] In an example, the pump 10 can be equipped with a pre-trained ML model 13, which was e.g. , pre-trained on a reference pump and is further adjusted during the training routine in the test rig. Thereby, new layers can be added to the ML model 13.

[0093] In yet another example, a part of the ML model can be frozen during training, such that only some aspects (e.g. , some layers) of the neural network model are trained.

[0094] For example, to examine how temperature influences the flow rate, a trained model, which was trained without temperature variations, could be used as a starting point. The part of the model which does not receive temperature input could be frozen. Then, a parallel leg or additional layers that can be trained could be added to the model. This provides the advantage that only parts of the ML model need to be trained and old training (of other parts) could be reused as far as possible. Furthermore, the old (previously trained) parts of the ML model 13 could be corrected based on the training data (so-called "finetuning") .

[0095] In additional or alternatively, the ML model 13 can be continuously trained during normal operation of the pump 10. For instance, the ML model 13, e.g. a neural network, can be configured to further adjust its configuration (e.g. , the weights of the neural network) over its lifetime based on the input data received over time.

[0096] For instance, after changing a component of the pump (e.g. , an inlet) , information on this change can be input in the pump, and the ML model 13 can be trained on a new model or can be adjusted to take this change into account.

[0097] Besides providing flow information, the trained ML model 13 can be configured to provide additional data. For instance, the ML model 13 can be configured to further output information on a status of the pump based on the input data. The pump 10 may provide detailed reports and insights, making it easier for operators to understand and act on the information. Such an alert system may ensure that critical issues are promptly addressed, enhancing operational efficiency and reliability.

[0098] Moreover, the ML model 13 can be interrogated by a user on an interdependence of the input parameters in different regions of the H-Q curve. For example, an estimate of how much the impeller geometry deviates from a spec can be provided by the ML model 13. This information could be used for a pump calibration. Additionally or alternatively, the ML model 13 can provide information on a wear of the pump 10.

[0099] For example, the information provided by the ML model 13 can be used to improve the pump 10 design and / or to directly quantify deviations during pump production. The ML model 13 can further learn and output information on how much wear can be tolerated before it negatively affects a pump performance.

[0100] Depending on the network design, the ML model 13 can comprise multiple sub-networks, e.g. , networks which compete with a metanetwork and / or networks which only take some of the input parameters into account. In particular, these networks can exploit the fact that different input parameters are sufficient to calculate an exact flow in different work areas of the pump. For example, a first sub-network can consider the differential pressure and power intake, a second network can consider the power intake and the motor speed, and a third network can consider the differential pressure and the motor speed. According to a further example, the ML model 13 is configured to output at least two flow value data estimates, wherein each flow value data estimate is determined based on a different combination of the input data. The controller 12 and / or the ML model 13 can be configured to detect a pump inefficiency if the at least two flow value data estimates differ from each other by more than a threshold value.

[0101] Thereby, each flow value estimate can comprise flow value data output by the ML model 13 for a specific combination of the input data. For instance, the ML model 13 can determine a first flow value estimate based on the differential pressure value and the electrical power intake of the motor and a second flow value estimate based on the differential pressure value and the speed of the motor. If these two flow value estimates differ by more than the threshold value, the controller 12 can detect a pump inefficiency.

[0102] For example, the detection of pump inefficiency may be based on certain assumption, e.g. , that a differential pressure sensor is correctly calibrated and / or that the pump speed does not degrade over time. Hence, by comparing the flow values estimates, information on pump energy efficiency degradation can be derived .

[0103] In this way, the ML model 13 can detect inefficiencies in pump operation and, e.g. , quantify the degree of inefficiency. This can improve predictive maintenance capabilities, reduce downtime, and optimize pump performance.

[0104] Furthermore, pump inefficiencies can be identified and quantified by comparing the estimated flow value data with an expected flow value data for any given set of input values. Hence, an output flow data value that deviates from the expected flow data value at a given set of input values can be derived from pump inefficiencies, i.e. the pump requires more power than expected to achieve a said flow data value.

[0105] The ML model 13 can also be trained to quantify the size of the identified inefficiency based on a comparison of the expected flow data value and the output flow data value with a given set of input parameters.

[0106] In particular, the ML model 13 used by the pump 10 can identify inefficiencies with greater precision than traditional methods. By learning from extensive historical data, the models can detect subtle patterns and anomalies that might be missed by human inspection or basic threshold-based systems. Furthermore, a real-time monitoring and alert system may enable proactive maintenance strategies. By detecting inefficiencies early, the pump 10 allows operators to address potential issues before they escalate into major problems, thus reducing unplanned downtime and extending the life of the pump 10. By optimizing pump performance and reducing energy consumption, this can lead to significant cost savings. Efficient pumps use less energy, which translates to lower operating costs. Additionally, reducing the frequency and severity of pump failures minimizes repair and replacement costs.

[0107] The combination of flow estimation, inefficiency detection and quantification can provide a comprehensive diagnostic tool. This approach can ensure that all aspects of pump performance can be monitored, and any inefficiencies are not only detected but also measured in terms of their impact on overall efficiency.

[0108] According to a further example, the flow value monitoring via the ML model 13 can be scaled to monitor multiple pumps, e.g. , across various industrial applications. This makes it a versatile solution that can be implemented in a wide range of settings, from small-scale operations to large industrial plants .

[0109] The pump 10 can further comprise an optional communication interface which is configured to receive further input data from at least one further pump. The further input data can comprise at least two of: an electrical power intake of a motor of the further pump, a speed of the motor of the further pump, and a differential pressure value related to the further pump (e.g. , measured with a differential pressure sensor of the further pump) . The ML model 13 of the pump 10 can be configured to output further flow value data relating to the at least one further pump based on the further input data.

[0110] Thereby, the pump 10 can act as a primary pump and the at least one further pump can act as a secondary pump. For instance, the at least one further pump may not comprise a trained ML model.

[0111] The (primary) pump 10 and the at least one further (secondary) pump can be components of a pump system.

[0112] According to a further example, the trained ML model 13 can be configured to continuously or repeatedly adjust its configuration over the lifetime of the pump 10 based on the input data received over time. For example, the ML model 13 is continuously updated with new data, improving its accuracy and adapting to changing operational conditions. In this way, it can be ensured that the pump 10 remains effective over time and can accommodate new types of inefficiencies or changes in pump operation .

[0113] Fig. 3 shows a flow chart of a method 30 of determining the flow rate of the pump 10 according to an embodiment. For example, the method 30 can be carried out with the pump 10 as shown in Fig. 1.

[0114] The method 30 comprises the steps of: supplying 31 the ML model 13 with input data, the input data comprising at least two of: the electrical power intake of the motor, the speed of the motor 11, and the differential pressure value; and outputting 32 the flow value data based on the input data with the ML model 13.

[0115] The method 30 may comprise the additional step of: adapting 33 the pump setting, in particular the speed of the motor 11, based on the flow value data.

[0116] Fig. 4 shows a flow chart of a method 40 of training the ML model 13 of the pump 10 according to an embodiment.

[0117] This training method 40 can comprise the steps of: exposing 41 the pump 10 to determined flow conditions, collecting 42 the further input data while the pump 10 is exposed to the determined flow conditions, and using 43 the collected further input data as training data for training the ML model 13.

[0118] For instance, the pump 10 can be mounted in a test rig which provides the determined flow conditions to carry out step 41. The ML model 13 can be configured to correlate the collected further input data with the determined flow conditions (e.g. , in the test rig) , and to adjust its configuration based on said correlation .

[0119] The method 40 can comprise the further step of: continuously or repeatedly adjusting 44 the configuration of the ML model 13, e.g. a neural network) over the lifetime of the pump 10 based on the input data received over time.

Claims

Claims1. A pump (10) , comprising: a motor (11) ; and a controller (12) configured to control a speed of the motor (11) ;f wherein the controller (12) comprises a trained machine learning, ML, model (13) being supplied with input data, the input data comprising a differential pressure value and at least one of : an electrical power intake of the motor (11) , and a speed of the motor (11) ; wherein the ML model (13) is configured to output flow value data based on the input data.

2. The pump (10) of claim 1, wherein the controller (12) is configured to adapt a pump setting, in particular the speed of the motor (11) , based on the flow value data.

3. The pump (10) of claim 1 or 2, further comprising: wherein the controller (12) is configured to adjusting a maximum level of the electrical power intake of the motor (11) based on a user input; wherein the ML model (13) is supplied with the maximum level as further input data.

4. The pump (10) of any one of the preceding claims, wherein the input data further comprises a temperature, a density, and / or a viscosity of a fluid to be pumped.

5. The pump (10) of any one of the preceding claims, furthercomprising : a differential pressure sensor which is configured to detect the differential pressure value.

6. The pump (10) of any one of the preceding claims, wherein the ML model (13) is trained on individual characteristics of the pump.

7. The pump (10) of any one of the preceding claims, wherein, for training the ML model (13) , the pump (10) is exposed to determined flow conditions; wherein the controller (12) is configured to collect further input data while the pump (10) is exposed to the determined flow conditions; and wherein the ML model (13) is configured to use the collected further input data as training data.

8. The pump (10) of claim 7, wherein the ML model (13) is configured to correlate the collected further input data with the determined flow conditions and to adjust its configuration based on said correlation.

9. The pump (10) of any one of the preceding claims, wherein the trained ML model (13) is configured to continuously or repeatedly adjust its configuration over the lifetime of the pump (10) based on the input data received over time .

10. The pump (10) of any one of the preceding claims, wherein the trained ML model (13) is configured to further output information on a status of the pump based on the input data .

11. The pump (10) of any one of the preceding claims, wherein the ML model (13) is configured to output at least two flow value data estimates, wherein each flow value data estimate is determined based on a different combination of the input data; wherein the controller (12) and / or the ML model (13) is configured to detect a pump inefficiency if the at least two flow value data estimates differ from each other by more than a threshold value.

12. The pump (10) of any one of the preceding claims, further comprising : a communication interface which is configured to receive further input data from at least one further pump; the further input data comprising at least two of: an electrical power intake of a motor of the further pump, a speed of the motor of the further pump, and a differential pressure value related to the further pump; wherein the ML model (13) is configured to output further flow value data relating to the at least one further pump based on the further input data.

13. A method (30) of determining a flow rate of a pump (10) , wherein the pump (10) comprises a motor (11) and a controller (12) configured to control a speed of the motor (11) , wherein the controller (12) comprises a trained machine learning, ML, model (13) , the method (30) comprising the steps of: supplying (31) the ML model (13) with input data, the input data comprising a differential pressure value and at least oneof : an electrical power intake of the motor (11) , and a speed of the motor (11) ; outputting (32) flow value data based on the input data with the ML model (13) .

14. The method (30) of claim 13, comprising the further step of : adapting (33) a pump setting, in particular the speed of the motor (11) , based on the flow value data.

15. The method (30) of claim 13 or 14, wherein the ML model (13) is trained by: exposing (41) the pump (10) to determined flow conditions, collecting (42) further input data while the pump (10) is exposed to the determined flow conditions, and using (43) the collected further input data as training data for training the ML model (13) .

16. The method (30) of claim 15, wherein the ML model (13) is configured to correlate the collected further input data with the determined flow conditions and to adjust its configuration based on said correlation.

17. The method (30) of any one of any one of claims 13 to 16, further comprising the step of: continuously or repeatedly adjusting (44) the configuration of the ML model (13) over the lifetime of the pump (10) based on the input data received over time.

Images