Device, method and system for determining significance of pump data

Abstract

The present disclosure relates to pump industry and proposes a method for determining a significance of pump data related to a pump. The method comprises receiving the pump data from one or more sensors, and processing the pump data using a plurality of analysis modules. Each analysis module analyzes the pump data using a different analysis approach to generate a respective significance value. The method further comprises combining the respective significance values generated by the plurality of analysis modules to calculate a significance value for the pump data.

Description

[0001] GRUNDFOS HOLDING A / S P24119WO

[0002] P629O8 / WO

[0003] DEVICE, METHOD AND SYSTEM FOR DETERMINING SIGNIFICANCE OF PUMP DATA

[0004] TECHNICAL FIELD

[0005] This disclosure relates generally to the field of industrial equipment management, for example, management of a pump. For instance, this disclosure provides devices, methods and a system for determining significance of pump data.

[0006] BACKGROUND

[0007] Industrial pumps play a vital role in effectively transporting fluids across various processes and applications. In traditional pump systems, gathering and storing pump data is essential for providing, maintaining and managing pump operations.

[0008] SUMMARY

[0009] A pump nowadays is typically equipped with multiple sensors that continuously generate large volumes of measurement data that need to be processed, analyzed, and stored. Conventional approaches for handling pump data typically focus on specific aspects of data analysis, such as basic anomaly detection or threshold-based monitoring.

[0010] Conventional approaches for handling pump data present several technical challenges. First, treating all pump data in an equal manner may lead to inefficient use of computational and storage resources, as less significant data receives the same level of processing and storage as highly significant data. Second, isolated analysis approaches may miss important patterns or events. Third, the lack of a comprehensive significance assessment makes it difficult to appropriately prioritize system responses to different operational conditions and events.

[0011] In view of the above-discussed challenges, this disclosure aims to provide an improved method for handling pump data that addresses these technical challenges. An objective is to provide a solution for assessing the significance of pump data. GRUNDFOS HOLDING A / S

[0012] P24119WO

[0013] P629O8 / WO

[0014] These and other objectives are achieved by solutions of this disclosure as described in the independent claims. Advantageous implementations are further defined in the dependent claims.

[0015] A first aspect of this disclosure provides a method for determining a significance of pump data related to a centrifugal pump. The method comprises the following steps:

[0016] - receiving the pump data from one or more sensors;

[0017] - processing the pump data using a plurality of analysis modules, wherein each analysis module analyzes the pump data using a different analysis approach to generate a respective significance value; and

[0018] - combining the respective significance values generated by the plurality of analysis modules to calculate a significance value for the pump data.

[0019] The calculated significance value for the pump data may be referred to as an overall significance value. The overall significance value reflects the significance of the pump data.

[0020] The centrifugal pump maybe a smart pump, e.g., a pump having integrated monitoring and processing capabilities, sensors, software, and connectivity features, or the like, allowing it to be able to perform automated monitoring and control of the pumping operation. The smart pump’s monitoring capabilities may particularly benefit from the advantages provided by the method. The pump maybe a fluid pump, or a liquid pump, or an oil pump, or a water pump. That is, the fluid may be a liquid or water. The fluid may also be a gas.

[0021] The multiple analysis modules employ different analysis approaches such that various significance values reflecting various metrics can be obtained. For instance, the various metrics may comprise one or more of: degree of deviation from expected measurement ranges; frequency and magnitude of anomalies; changes exceeding predefined operational thresholds; and the like.

[0022] The step of combining the respective significance values comprises evaluating interdependencies between different sensor measurements, such that opposing trends GRUNDFOS HOLDING A / S P24119WO

[0023] P629O8 / WO or correlated patterns, such as one sensor measurement increasing while another decreases, are factored into the overall significance value.

[0024] For combining the multiple significance values to obtain the overall significance value, multiple metrics maybe taken into account to determine the overall importance of the pump data. For instance, the method can evaluate correlations or counteracting trends in pump data to determine the overall significance.

[0025] Optionally, various combinations of the analysis modules may be used for analyzing the pump data. The modular architecture allows for easy adaptation to various pump types, operating environments, and performance requirements. Analysis modules can be added, removed, or tailored based on the specific needs of the pump or a system thereof.

[0026] By using the overall significance value, the centrifugal pump or the system thereof can dynamically adjust pump control parameters, ensuring optimal operation under varying conditions.

[0027] Optionally, the respective significance values may be normalized to ensure they fall within a common range (e.g., o to 1, or o to 100, or the like) to avoid dominance of any single module due to scale differences.

[0028] Optionally, each significance value maybe multiplied by a weighting factor that reflects its relative importance, reliability, or relevance to the pump operation before being combined into the overall value. The weighting factors may be dynamically adjusted based on contextual factors, such as the operational mode of the centrifugal pump, environmental conditions, or specific analysis goals.

[0029] Optionally, the respective significance values may be combined using statistical techniques, such as but not limited to: weighted averages, medians, or max-min normalization to enhance robustness against outliers.

[0030] Optionally, two or more of the multiple significance values obtained by the multiple analysis modules may be multiplied to obtain the overall significance value. GRUNDFOS HOLDING A / S

[0031] P24119WO

[0032] P629O8 / WO

[0033] Optionally, two or more of the multiple significance values obtained by the multiple analysis modules maybe compared (with each other, or against a threshold), such that a subset of the two or more of the multiple significance values may be taken into account for calculating the overall significance value. For instance, if a first significance value is higher than a certain threshold, then a second significance value is not used for calculating the overall significance value.

[0034] Examples of the pump data may include but not limited to:

[0035] - vibration data: acquired using accelerometers to detect imbalances, misalignments, or bearing faults;

[0036] - temperature data: monitored to check for overheating in bearings or the motor;

[0037] - pressure data: used to measure the inlet and outlet pressures, helping to detect issues like cavitation or blockages;

[0038] - flow rate data: monitored to ensure the centrifugal pump is operating within its designed parameters;

[0039] - torque data: collected from variable frequency drives (vfds) to monitor the load on the centrifugal pump;

[0040] - acoustic data: captured using microphones to detect cavitation or other abnormal sounds;

[0041] - speed data: measured to ensure the centrifugal pump is running at the correct speed (rpm);

[0042] - current and voltage of the centrifugal pump.

[0043] In an implementation form of the first aspect, each analysis module analyzes a respective part of the pump data, wherein the respective part of the pump data is output by one of the one or more sensors.

[0044] Optionally, the pump data comprises derived data that is derived from sensor outputs. That is, the method may comprise deriving further data based on the output by the one or more sensors, and including the derived data as part of the pump data. For instance, flow may be calculated from pressure and speed measurement. GRUNDFOS HOLDING A / S P24119WO

[0045] P629O8 / WO

[0046] In a further implementation form of the first aspect, the plurality of analysis modules comprises a physics-based model that compares a respective part of the pump data against a predefined operational curve and generates an increased significance value if the respective part of the pump data deviates from the predefined operational curve.

[0047] The operational curve includes multi-dimensional representations, considering correlations between multiple variables, such as pressure, flow, and vibration. As an example, the operational curve may comprise a QH curve.

[0048] Optionally, the physics-based model incorporates tolerance zones around the operational curve, which represent acceptable deviations. Significance is only increased when data falls outside these zones.

[0049] Optionally, the method may comprise using historical sensor data to refine and validate the predefined operational curve, so as to ensure it evolves with the pump's operational history.

[0050] In a further implementation form of the first aspect, the plurality of analysis modules comprises a physics informed neural network (PINN) for analyzing at least a part of the pump data.

[0051] The PINN can integrate high-fidelity simulation outputs with real-time sensor data, combining theoretical physics principles with practical operational insights. Using the PINN incorporates a mechanism to detect and correct discrepancies between simulated and actual sensor data, ensuring the accuracy of predictions. The PINN can adapt its parameters based on contextual factors, such as changes in operational mode, environmental conditions, or pump lifecycle stage. Domain-specific knowledge, such as pump fluid dynamics, cavitation thresholds, and energy efficiency metrics, can be embedded into the neural network.

[0052] In a further implementation form of the first aspect, the plurality of analysis modules comprises an anomaly detection model for identifying irregularities in at least a part of the pump data. GRUNDFOS HOLDING A / S P24119WO

[0053] P629O8 / WO

[0054] The anomaly detection model can employ diverse methods such as but not limited to:

[0055] - statistical methods (e.g., standard deviation, cumulative sum);

[0056] - machine learning approaches (e.g., clustering, neural networks); and

[0057] - probabilistic methods (e.g., Gaussian distribution, Bayesian networks).

[0058] The anomaly detection model may be used to detect temporal anomalies, such as sudden spikes or drifts in sensor / pump data over time. Alternatively or additionally, spatial anomalies can be detected, such as data relationships between multiple sensors deviating from expected patterns.

[0059] In a further implementation form of the first aspect, the plurality of analysis modules comprises an autoencoder neural network for reconstructing a full measurement based on at least a part of the pump data, and for outputting a reconstruction error, wherein the method comprises determining a respective significance value based on the reconstruction error.

[0060] In alternative to the autoencoder neural network, unsupervised learning methods like principal component analysis (PCA) may be used.

[0061] In a further implementation form of the first aspect, the method further comprises: adjusting, by each analysis module, the respective significance value based on an associated sensor uncertainty and / or one or more production tolerances of the centrifugal pump.

[0062] Optionally, specific sensor uncertainty and / or production tolerances maybe assigned based on the type of sensor or pump component, which ensures tailored significance adjustment for each data category.

[0063] Optionally, the tolerances or uncertainties maybe fine-tuned based on environmental conditions, such as temperature, pressure, or operational load.

[0064] In a further implementation form of the first aspect, each respective significance value represents or quantifies at least one of: a degree of deviation from a preset sensor measurement range; GRUNDFOS HOLDING A / S P24119WO

[0065] P629O8 / WO a frequency of outliers or anomalies in at least a part of the pump data; a magnitude of the outliers or anomalies; a magnitude of data changes in at least a part of the pump data; sensor measurement uncertainty; a fault probability of the pump; and a level of exceedance of predefined operational thresholds in at least a part of the pump data.

[0066] In a further implementation form of the first aspect, the overall significance value is calculated based on a weighted aggregation of the respective significance values generated by the plurality of analysis modules, and represents a measure of significance of the pump data for the operation of the centrifugal pump.

[0067] In a further implementation form of the first aspect, the pump data comprises a plurality of sensor measurements from a plurality of the sensors, and the overall significance value represents a degree of correlations and / or counteracting trends between the individual sensor measurements.

[0068] In a further implementation form of the first aspect, the method comprises determining a data compression level for storing and / or transmitting the pump data based on the overall significance value.

[0069] In a further implementation form of the first aspect, the method comprises determining or adjusting one or more pump control parameters of the centrifugal pump based on the overall significance value.

[0070] In a further implementation form of the first aspect, the method comprises identifying one or more operational events related to the centrifugal pump based on the overall significance value.

[0071] A second aspect of the present disclosure provides an apparatus for determining a significance of pump data related to a centrifugal pump or a system thereof. The apparatus is configured to:

[0072] - receive the pump data from one or more sensors; GRUNDFOS HOLDING A / S P24119WO

[0073] P629O8 / WO

[0074] - process the pump data using a plurality of analysis modules, wherein each analysis module is configured to analyze the pump data using a different analysis approach to generate a respective significance value; and

[0075] - combine the respective significance values generated by the plurality of analysis modules to calculate an overall significance value for the pump data.

[0076] The apparatus of the second aspect may share the same features and advantages as the method of the first aspect.

[0077] A third aspect of the present disclosure provides a centrifugal pump or a centrifugal pump system comprising an apparatus for determining a significance of pump data related to the centrifugal pump or the centrifugal pump system according to the second aspect.

[0078] A fourth aspect of this disclosure provides a computer program product comprising instructions which, when the program is executed by a first computer, cause the first computer to perform the method of the first aspect.

[0079] All steps that are performed by the various entities described in the present application as well as the functionalities described to be performed by the various entities are intended to mean that the respective entity is adapted to or configured to perform the respective steps and functionalities. Even if, in the following description of specific embodiments, a specific functionality or step to be performed by external entities is not reflected in the description of a specific detailed element of that entity that performs that specific step or functionality, it should be clear for a skilled person that these methods and functionalities can be implemented in respective software or hardware elements or any kind of combination thereof.

[0080] BRIEF DESCRIPTION OF DRAWINGS

[0081] The above-described aspects and optional implementations will be explained in the following description of specific embodiments in relation to the enclosed drawings, in which GRUNDFOS HOLDING A / S P24119WO

[0082] P629O8 / WO

[0083] FIG. 1 shows a diagram of a method according to the present disclosure;

[0084] FIG. 2 shows an example of a diagram of a method;

[0085] FIG. 3 shows an example of a deployment of analysis modules;

[0086] FIG. 4 shows a flowchart of a method for determining a significance of pump data;

[0087] FIG. 5 shows a schematic diagram of a centrifugal pump; and

[0088] FIG. 6 shows an example of training a machine learning model for analyzing pump data.

[0089] DETAILED DESCRIPTION OF EMBODIMENTS

[0090] Illustrative embodiments may be related to a centrifugal pump and a system thereof, which are described with reference to the figures. Although this description provides a detailed example of possible embodiments and implementations, it should be noted that the details are intended to be exemplary and in no way limit the scope of the application.

[0091] In this disclosure, an embodiment or example may refer to other embodiments or examples. For example, any description including but not limited to terminology, element, process, explanation, and / or technical advantage mentioned in one embodiment or example is applicable to the other embodiments or examples. The same elements are labeled with the same reference signs and may function similarly or likewise.

[0092] FIG. 1 shows a diagram of a method according to the present disclosure.

[0093] The method may be applied to a centrifugal pump, or a control entity thereof. In this disclosure, the centrifugal pump is configured to receive pump data 101 from one or more sensors. The centrifugal pump is configured to process the pump data 101 using a plurality of analysis modules 110a, nob. Each analysis module 110a, nob analyzes GRUNDFOS HOLDING A / S

[0094] P24119WO

[0095] P629O8 / WO the pump data using a different analysis approach to generate a respective significance value 102a, 102b. These generated significance values 102a, 102b are then combined by the centrifugal pump to calculate an overall significance value for the pump data.

[0096] The present disclosure provides an approach to data significance determination specifically designed for pump applications. Unlike conventional anomaly detection methods, the solution of the present disclosure offers a comprehensive method of assessing data importance that extends beyond simple fault identification. In the context of this invention, "significance" may be referred to as a multidimensional evaluation of data that captures the broader operational context of the pump, or a pump system. The determined overall significance value according to the present disclosure can serve multiple critical functions, including but not limited to: prioritizing system reactivity, guiding failure detection strategies, and determining optimal data compression levels. The concept of the significance value may be contextual, deriving its meaning from the specific application, product, and engineering domain in which it is situated. By integrating field data with comprehensive knowledge sources, the significance determination transcends simple measurement interpretation. It represents a sophisticated synthesis of productspecific insights, pump operational understanding, physical principles, engineering expertise, and data-driven analytics.

[0097] FIG. 2 shows an example of a diagram of a method. The method in FIG. 2 may be built based on the method shown in FIG. 1. Features introduced in FIG. 2 maybe similarly applied to the method in FIG. 1.

[0098] According to the present disclosure, each analysis module is adapted to analyze a respective part of the pump data. The respective part of the pump data may be output by one of the one or more sensors of the pump, such as flow sensor 201a, vibration sensor 201b, and temperature sensor 201c. In this way, each module can analyze the data (or at least part of the data) and output their own significance value 202a, 202b, 202c. Based on the outputs of the multiple modules, an overall significance value 203 is calculated. The overall significance value 203 can be used in various application scenarios, for example but not limited to: control algorithms, failure detection and compression level selection. GRUNDFOS HOLDING A / S

[0099] P24119WO

[0100] P629O8 / WO

[0101] The plurality of analysis module comprises two or more of:

[0102] - a physics informed neural network 210a;

[0103] - an anomaly detection model 210b;

[0104] - an autoencoder neural network 210c.

[0105] Pump physics can utilize a pressure and flow model that examines whether measured data points align with established QH curves. Deviations from standard operational curves are automatically flagged as potentially significant, providing a fundamental physical assessment of pump performance. Physics informed neural networks (PINNs) are specialized neural networks that creates hybrid models by integrating high-fidelity simulation techniques with real sensor data. One way is to put trust in parts of the simulation where it is highly confident and use data to adjust part of the simulation where modelling techniques are uncertain. The present disclosure systematically captures and converts domain expertise from multiple professional sources, including skilled workers, engineering specialists, application experts, and service technicians. The specialized knowledge is transformed into operational machine learning models capable of deployment across edge and cloud computing environments, ensuring broad applicability and intelligent data interpretation.

[0106] Various types of the anomaly detection model may be used, such as unsupervised or supervised. Examples of an unsupervised anomaly detection model may be: standard deviation, cumulative sum, cluster, tree, or Gaussian based anomaly detectors. Examples of a supervised anomaly detection model may be: a neural network trained on anomalous and non-anomalous data to identify anomalies, or support vector machines. The anomaly detection model is used to scan for data irregularities and output a significance value based thereon. Other modules can be used to monitor unbalance, water hammer, overall vibration, cavitation, response etc. of the pump or the pump system. These can output a level or probability that can be used in combination to determine the significance value.

[0107] The autoencoder neural network may be pump type-specific autoencoder that is trained to reconstruct a full sensor array measurement. Based on the size of the reconstruction error, the autoencoder neural network outputs a significance value. GRUNDFOS HOLDING A / S

[0108] P24119WO

[0109] P629O8 / WO

[0110] The significance value of each module may be adjusted based on sensor uncertainty and / or production tolerances for specific product series / type (known from the manufacturing site, and / or through data from machine, and / or through end-tests).

[0111] The significance determination method incorporates advanced long-term data analysis techniques to enhance predictive capabilities. By aggregating extensive sensor data over prolonged periods, the method develops artificial intelligence models capable of identifying historical failure patterns. These artificial intelligence models can analyze the existing system failures, using this information to predict potential failure risks and adjust data significance accordingly.

[0112] The method can statistically compile control signals across extended operational periods, establishing comprehensive signal distributions. Signals deviating significantly from expected patterns are identified as potentially more significant, indicating unusual system behavior.

[0113] The method can track anomaly occurrences within specific time frames, increasing the significance value when anomalous events exceed predefined thresholds.

[0114] Sensor measurements exceeding critical operational thresholds— such as extreme temperature ranges— automatically receive elevated significance ratings. This approach recognizes that values approaching system stress limits inherently carry higher operational risk.

[0115] The modular architecture of the present disclosure enables dynamic significance block configuration, allowing analysis modules to be added or removed based on specific computational resources and analytical requirements. This flexibility permits a unified significance value that can be repurposed across multiple application domains.

[0116] The method synthesizes predictions from diverse data products, each potentially employing different analytical methodologies— ranging from purely data-driven to physics-based approaches. For instance, an impeller life prediction generating high- GRUNDFOS HOLDING A / S P24119WO

[0117] P629O8 / WO risk values would trigger specialized data processing strategies, such as targeted compression focusing on critically important parameters.

[0118] In the following, an example of data compression based on pump data is shown. In this example, the pump data used for determining data compression level comprises vibration measurements. A set of vibration measurements could be parsed to several analysis modules. Each module outputs a significance value.

[0119] An anomaly detection module is used and is adapted to analyses the deviation of the vibration measurements. By tracking the frequency and magnitude of significant variations within the dataset, the module generates a significance value that increases proportionally with the frequency of detected anomalies. For instance, the anomaly detection module analyzes whether a single vibration measurement varies too significantly from the previous ones, and how many times this happens in the dataset. If this occurs many times a high significance value would be outputted.

[0120] For specific pump types, the centrifugal pump may utilize a calibration module. The calibration module incorporates predefined vibration sensor uncertainties and manufacturing tolerances. The calibration module generates a baseline significance value that accounts for inherent measurement variability, typically represented as a negative value to calibrate the overall significance assessment.

[0121] Other modules can be used to monitor unbalance, water hammer, overall vibration, cavitation, response etc. of the signal. These would output a level or probability that can be used in combination to determine the significance value.

[0122] Control values used during the measurement time of the vibration measurement dataset could be analyzed in a separate module for outliers from the norm. The more outliers are found the larger the significance value.

[0123] The vibration measurements can be analyzed through a trained fault detector neural network to test if the trained model recognizes risk of fault. The greater the found risk of fault based on the vibration data, the greater the significance value. GRUNDFOS HOLDING A / S P24119WO

[0124] P629O8 / WO

[0125] The sum of these modules would then be compared with different defined thresholds and the lowest one that it exceeds will define the compression level. A low overall significance value triggers high-level data compression. A high significance value invokes minimal compression, potentially employing lossless compression techniques.

[0126] FIG. 3 shows an example of a deployment of analysis modules. Based on the method shown in FIGs. 1-2, the multiple analysis modules 310 are used for analyzing pump data. Examples of the analysis modules 310 includes but not limited to: pressure / pump speed data analyzer, physics informed neural network model; sensor uncertainties and tolerances module; pump failure detection Al model; autoencoder anomaly detector; Gaussian anomaly detector. Measurement pump data 301 is collected and is fed into the multiple analysis modules 310 to obtain an overall significance value 303 according to the method of FIGs. 1-2. Examples of the Measurement pump data 301 includes but not limited to: pump speed, pressure, electric current, and so on. Optionally, further pump data 3011 maybe derived based on the measurement data 301, such as flow. The measurement pump data 301 and the derived pump data 3011 may be collectively referred to as pump data. The overall significance value 303 can be used to determine a data compression level, based on which the pump data is compressed, to obtain compressed pump data 304. The compressed pump data 304 may be then sent to a server for further analysis. The overall significance value 303 may also be sent to the server, such that the server has a combination 305 of the compressed pump data 304 and the associated overall significance value 303. Based on the overall significance value 303, the server can determine a priority for analyzing the respective compressed pump data 304.

[0127] FIG. 4 shows a flowchart of a method 400 for determining a significance of pump data according to this disclosure. The method 400 comprises the following steps:

[0128] Step 401: receiving the pump data from one or more sensors;

[0129] Step 402: processing the pump data using a plurality of analysis modules, wherein each analysis module analyzes the pump data using a different analysis approach to generate a respective significance value; and

[0130] Step 403: combining the respective significance values generated by the plurality of analysis modules to calculate an overall significance value for the pump data. GRUNDFOS HOLDING A / S P24119WO

[0131] P629O8 / WO

[0132] The method 400 may share the same optional features introduced above in FIGs. 1-3, which are not repeated herein.

[0133] FIG. 5 shows a schematic diagram of a centrifugal pump 50. The centrifugal pump 50 is configured to pump a fluid, e.g., a liquid like water, oil, or a gas. The centrifugal pump 50 may be a smart pump with various sensors 51 and processing functionalities. The centrifugal pump may be installed in a household environment or in an industrial environment, or the like.

[0134] According to this disclosure, an apparatus 52 is disclosed, which utilize a plurality of analysis modules configured to analyze pump data 501, so as to calculate an overall significance value 503. of the pump data 501. The apparatus 52 is configured to perform the method introduced above in FIGs. 1-4. The apparatus 52 may be a component of the centrifugal pump 50 (as shown in FIG. 5), or may be an external component communicable with the centrifugal pump 50 (not shown). This is not limited in the present disclosure.

[0135] The apparatus 52 comprises a processor, which is configured to perform various processing steps and functionalities disclosed above in FIGs. 1-4. The processor may be a controller or of a control module for the centrifugal pump. The processing circuitry may comprise hardware and / or the processing circuitry may be controlled by software. The hardware may comprise analog circuitry or digital circuitry, or both analog and digital circuitry. The digital circuitry may comprise components such as applicationspecific integrated circuits (ASICs), field-programmable arrays (FPGAs), digital signal processors (DSPs), or multi-purpose processors. The processor may further comprise memory circuitry, which can store one or more instruction(s) that can be executed by the processor or its processing circuitry, in particular, under control of the software. For instance, the memory circuitry may comprise a non-transitory storage medium storing executable software code which, when executed by the processor or its processing circuitry, causes the various described operations disclosed above in FIGs. 1-4 to be performed.

[0136] FIG. 6 shows an example of training a machine learning model (e.g., a neural network) for analyzing pump data. The machine learning model used in the present disclosure GRUNDFOS HOLDING A / S P24119WO

[0137] P629O8 / WO maybe trained with training goals defined by a variety of sources like pump knowledge, simulations and labelled fault cases. Each of these can supply training targets for one or multiple of the analysis modules.

[0138] As shown in FIG. 6, measured and saved pump sensor data maybe collected as training data. In addition, simulated sensor data maybe added to the training data. In addition, labelled sensor data from previous fault cases maybe added to the training data. Types of the pump sensor data may include pump speed, pressure, electric current, and the like. Based on sensor measurements, other type of pump data may be derived, such as flow.

[0139] For each machine learning model, a labelled significance (a significance target) can be obtained by various sources, such as pump knowledge, data product, pump simulation software, expert knowledge, labelled previous fault cases, etc. Each machine learning model is adapted to output a significance value during the training phase.

[0140] The training error used in potential backpropagation or similar training algorithms is then calculated by comparing the error between the model-predicted significance values (one for each model) and those supplied by the varied sources. The training data excluding targets is supplied primarily by sensor measurements obtained from tests or use of pumps. The training data is labelled with a significance given by expert knowledge, data products or other sources. Some sources like the labelled fault cases can also supply the training data points themselves along with the labelled significance.

[0141] The present disclosure has been described in conjunction with various embodiments as examples as well as implementations. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed matter, from the studies of the drawings, this disclosure, and the independent claims. In the claims as well as in the description the word “comprising” does not exclude other elements or steps and the indefinite article “a” or “an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutually different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.

Claims

GRUNDFOS HOLDING A / S P24119WOP629O8 / WOClaims1. A method (400) for determining a significance of pump data (101) related to a centrifugal pump, the method comprising: receiving (401) the pump data (101) from one or more sensors; processing (402) the pump data (101) using a plurality of analysis modules (110a, nob), wherein each analysis module (110a, nob) analyzes the pump data (101) using a different analysis approach to generate a respective significance value (102a, 102b); and combining (403) the respective significance values (102a, 102b) generated by the plurality of analysis modules (110a, nob) to calculate a significance value (103) for the pump data (101).

2. The method (400) according to claim 1, wherein each analysis module (110a, nob) analyzes a respective part of the pump data (101), wherein the respective part of the pump data (101) is output by one of the one or more sensors.

3. The method (400) according to claim 1 or 2, wherein the plurality of analysis modules (110a, nob) comprises a physics-based model (210a) that compares a respective part of the pump data against a predefined operational curve and generates an increased significance value if the respective part of the pump data (101) deviates from the predefined operational curve.

4. The method (400) according to any one of claims 1 to 3, wherein the plurality of analysis modules comprises a physics informed neural network for analyzing at least a part of the pump data (101).

5. The method (400) according to any one of claims 1 to 4, wherein the plurality of analysis modules comprises an anomaly detection model (210b) for identifying irregularities in at least a part of the pump data (101).

6. The method (400) according to any one of claims 1 to 5, wherein the plurality of analysis modules comprises an autoencoder neural network (210c) for reconstructing a full measurement based on at least a part of the pump data (101), and for outputtingGRUNDFOS HOLDING A / S P24119WOP629O8 / WO a reconstruction error, wherein the method comprises determining a respective significance value based on the reconstruction error.

7. The method (400) according to any one of claims 1 to 6, comprising: adjusting, by each analysis module, the respective significance value based on an associated sensor uncertainty and / or one or more production tolerances of the centrifugal pump.

8. The method (400) according to any one of claims 1 to 7, wherein each respective significance value represents or quantifies at least one of: a degree of deviation from a preset sensor measurement range; a frequency of outliers or anomalies in at least a part of the pump data (101); a magnitude of the outliers or anomalies; a magnitude of data changes in at least a part of the pump data (101); sensor measurement uncertainty; a fault probability of the centrifugal pump; and a level of exceedance of predefined operational thresholds in at least a part of the pump data (101).

9. The method (400) according to any one of claims 1 to 8, wherein the significance value (103) for the pump data is calculated based on a weighted aggregation of the respective significance values (102a, 102b) generated by the plurality of analysis modules, and represents a measure of significance of the pump data (101) for the operation of the pump.

10. The method according to any one of claims 1 to 9, wherein the pump data (101) comprises a plurality of sensor measurements from a plurality of the sensors, and the significance value (103) for the pump data represents a degree of correlations and / or counteracting trends between the individual sensor measurements.

11. The method according to any one of claims 1 to 10, further comprising determining a data compression level for storing and / or transmitting the pump data (101) based on the significance value (103) for the pump data.GRUNDFOS HOLDING A / S P24119WOP629O8 / WO12. The method according to any one of claims 1 to 11, further comprising determining or adjusting one or more pump control parameters of the centrifugal pump based on the significance value (103).

13. The method according to any one of claims 1 to 12, further comprising identifying one or more operational events related to the pump based on the significance value (103).

14. An apparatus (52) for determining a significance of pump data (501) related to a centrifugal pump (50) or a centrifugal pump system, wherein the apparatus (52) is configured to: receive the pump data (501) from one or more sensors (51); process the pump data (501) using a plurality of analysis modules, wherein each analysis module is configured to analyze the pump data (501) using a different analysis approach to generate a respective significance value; and combine the respective significance values generated by the plurality of analysis modules to calculate a significance value (503) for the pump data (501).

15. A computer program product comprising instructions which, when the program is executed by a processor of a centrifugal pump, causes the centrifugal pump to perform the method according to one of the claims 1 to 13.

Images