Detecting acoustic anomalies in a pump system

Abstract

The present disclosure relates to pumps and pump systems, for instance, smart pumps. In particular, the disclosure is concerned with pump control and monitoring. A pump system is provided, which can monitor sound, detect acoustic anomalies, and control the pump accordingly. The pump system comprises a pump for pumping a fluid, an acoustic sensor, and a processor. The acoustic sensor is configured to capture sound of the pump environment and of the pump. Further, the acoustic sensor is configured to convert the captured sound into acoustic data. The processor is configured to compare the acoustic data with reference acoustic data for the pump, in order to detect an anomaly in the acoustic data based on the reference acoustic data.

Description

[0001] GRUNDFOS HOLDING A / S

[0002] P24118WO

[0003] P62832 / WO

[0004] DETECTING ACOUSTIC ANOMALIES IN A PUMP SYSTEM

[0005] TECHNICAL FIELD

[0006] The present disclosure relates to pumps and pump systems, for instance, to smart pumps. The disclosure is concerned with pump control and monitoring. Specifically, the disclosure proposes a pump system, which can monitor sound of the pump environment and optionally sound of the pump, detect acoustic anomalies in the sound, and optionally control the pump according to the acoustic anomalies.

[0007] BACKGROUND

[0008] Pumps and pump systems, which are integral to a wide range of industrial, agricultural, and municipal applications, are prone to failures that may stem from various root causes. These root causes may include mechanical wear and tear, cavitation, improper lubrication, misalignment, overheating, or even external environmental factors, like pressure fluctuations, corrosive elements, or temperature changes in the surroundings of the pump. The failures of the pump or pump system can result in costly downtime and maintenance issues.

[0009] Detection of root causes is thus crucial, and may allow preventing more severe breakdowns, or ensuring a more efficient and less interrupted pump operation. Unfortunately, conventional monitoring and maintenance methods often rely on reactive measures, wherein a root cause is identified only after a failure has occurred. This may result in longer downtime and worse failure prevention.

[0010] If pumps of pump systems were enabled to detect indicators of potential failures or their root causes in real-time, early warnings could be provided. In this way, the pump performance could be optimized, unexpected downtimes could be reduced, and lower repair costs would be expected. GRUNDFOS HOLDING A / S

[0011] P24118WO

[0012] P62832 / WO

[0013] SUMMARY

[0014] In view of the above, this disclosure has the objective to provide a pump system that can achieve a better performance, less downtime in case of a failure, lower maintenance effort, and a better operational security. A particular objective is to enable the pump system to detect anomalies, which indicate potential failures or their root causes, in order to avoid the failures in the first place or at least to allow addressing the failures more quickly if they occur. To this end, an objective is to integrate predictive maintenance capabilities into the pump system, and to allow the pump system to autonomously monitor its relevant conditions to detect pump related root causes or failures. Another objective is to enable the monitoring of the pump environment, in order to detect root causes or failures that are not directly pump related, but may still affect the pump performance.

[0015] These and other objectives are achieved by the solutions of this disclosure, which are described in the independent claims. Advantageous implementations are described in the dependent claims.

[0016] The solutions of this disclosure are based on the fact that many pump failures, which occur due to various root causes, can be predicted early and efficiently or can at least identified reliably based on acoustic signs, particularly, acoustic anomalies of the pump and its surroundings. The solutions may be facilitated by the fact that pump systems, especially smart pump systems, may already be equipped with at least a microphone, which is in principle able to capture sound of the pump environment and sound of the pump, respectively.

[0017] A first aspect of this disclosure provides a pump system comprising: a pump for pumping a fluid; an acoustic sensor configured to capture sound of the pump environment and optionally of the pump, and convert the captured sound into acoustic data; and a processor configured to compare the acoustic data with reference acoustic data for the pump, in order to detect an anomaly in the acoustic data based on the reference acoustic data. GRUNDFOS HOLDING A / S

[0018] P24118WO

[0019] P62832 / WO

[0020] Based on the captured sound, which is related to at least the pump environment, and may also be related to the pump itself, and based additionally on the reference acoustic data, the processor is configured to detect the acoustic anomaly, or even more than one acoustic anomaly. The acoustic anomaly may be an unexpected or not normal sound signature in the acoustic data, for instance, an unexpected sound peak or frequency of sound, e.g., sound peak or frequency not existent or not as pronounced in the normal operation of the pump. The acoustic anomaly may be indicative of a root cause for a potential failure of the pump. The failure may have already occurred or may be about to occur. The pump system may thus predict such failures early, so that they can be avoided. At least, the detection of the acoustic anomaly may allow identifying the root cause of a failure more quickly, if it has already occurred, resulting in reduced downtime of the pump. The pump system may accordingly be able to perform a smart acoustic monitoring and anomaly detection of the pump, to detect pump related root causes or failures, and thus ensure operational integrity and enhanced maintenance procedures. The pump system is particularly capable of monitoring the pump environment, in order to detect root causes or failures that are not directly pump related. As examples of the latter, such root causes or failures may include a dry run of the pump and simultaneous water leakage in the pump environment, or may include a fire alarm in the building in which the pump system is located.

[0021] The reference acoustic data may be an acoustic profile of the pump system, which is based on the sound generated under normal operation of the pump (without root cause or failure) in the pump environment. This acoustic profile may be captured at or after initialization of the pump system at its installation site. The reference acoustic data may be referred to as a baseline profile of the pump system at the installation site.

[0022] The pump of the pump system may be 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 acoustic sensor and the processing of the acoustic sensor data to detect one or more acoustic GRUNDFOS HOLDING A / S

[0023] P24118WO

[0024] P62832 / WO anomalies. The pump maybe a rotary pump or a centrifugal pump. The pump may be a fluid pump, or a liquid pump, or an oil pump, or a water pump. That is, the fluid maybe a liquid or water. The fluid may also be a gas.

[0025] The processor maybe configured to run a computer program based on the acoustic data and the reference acoustic data as inputs to the computer program, in order to detect the acoustic anomaly. The computer program maybe an algorithm, which may be based on a trained model. The processor may comprise, or may be connected, to a memory, which stores the computer program (its instruction or code) and allows the processor to invoke the computer program from the memory.

[0026] In an implementation form of the pump system, the acoustic sensor is configured to capture sound only of the pump environment, or is configured to capture sound of the pump environment and of the pump, and convert the captured sound into the acoustic data.

[0027] In an implementation form of the pump system, the processor is configured to determine the anomaly in the acoustic data based on a deviation between the acoustic data and the reference acoustic data.

[0028] For example by executing the computer program, the processor may continuously compare the incoming acoustic data provided by the acoustic sensor against the reference acoustic data, in order to detect any deviation. For example, the acoustic anomaly may be detected in dependence of the nature of the deviation, e.g., its magnitude.

[0029] In an implementation form of the pump system, the processor is configured to determine whether the anomaly relates to the pump or to the pump environment.

[0030] This distinction of the anomaly helps identifying the root cause, and thus finding and avoiding the corresponding failure, or at least addressing the corresponding pump failure purposefully. GRUNDFOS HOLDING A / S

[0031] P24118WO

[0032] P62832 / WO

[0033] In an implementation form of the pump system, the processor is configured to determine a sound source related to the anomaly.

[0034] The sound source may provide further indication regarding the root cause and / or a pump failure. The sound source may be in the pump environment, i.e., at the installation site, or may be in or at the pump.

[0035] In an implementation form of the pump system, the processor is configured to, based on the anomaly, determine a root cause and / or a respective likelihood for each of a plurality of possible root causes.

[0036] The processor can thus automatically identify or isolating the root cause. The processor can issue a corresponding notification or message to a command center. If possible, the processor can even take steps to remove the root cause. In this disclosure, sending a notification or message to a command center may mean sending the notification or message to a server or uploading it into a cloud. That is, in this disclosure, the command center may be implemented by a server or cloud. The command center may be a building management system of the building in which the pump system is located. The command center could also be implemented by an internal or external controller for the pump system.

[0037] In an implementation form of the pump system, the processor is configured to determine a location of the sound source.

[0038] This may help a pump operator or maintenance persona to identify the root cause.

[0039] In an implementation form of the pump system, the processor is configured to control the pump to perform an operation to remove an anomaly related to the pump.

[0040] The processor is configured to control the pump based on the detected anomaly. In this way, the pump system maybe autonomously able to address failures or take preemptive measures against failures. Thus, a proper operation of the pump may be ensured and / or a maintenance of the pump maybe facilitated. GRUNDFOS HOLDING A / S

[0041] P24118WO

[0042] P62832 / WO

[0043] In an implementation form of the pump system, the pump system further comprises at least one parameter sensor comprising at least one of a pressure sensor, a temperature sensor, and a flow rate sensor; wherein the processor is configured to determine the anomaly based further on sensor data provided by the at least one parameter sensor.

[0044] The parameter sensor data may provide additional information to the processor, which facilitates detecting the one or more anomalies in the acoustic data. Thus, the parameter sensor data may help to identify or predict root causes of failures of the pump, for instance, in a faster, more reliable, and more accurate manner. This is, because the underlying parameters, e.g. certain operating parameters of the pump that are monitored by the sensors, may be impacted by the root cause as well, like the acoustic profile of the pump. The pump may be a smart pump, and maybe integrated with one or more of the mentioned parameter sensors.

[0045] In an implementation form of the pump system, the anomaly indicates at least one of: an anomalous flow of the fluid; an anomalous pressure of the fluid; an anomalous fluid temperature; an anomalous pump temperature; water leakage in the pump environment; a ventilator unbalance in the pump environment; smoke, fire, or a smoke or fire alarm in the pump environment; air leakage in the pump environment.

[0046] Other root causes may be indicated by the acoustic anomaly as well, the above are only examples. The processor maybe able to identify one or more root causes based on the acoustic anomaly.

[0047] In an implementation form of the pump system, the processor is configured to execute a machine learning algorithm (MLA) to detect the anomaly in the acoustic data, wherein input data to the MLA is based on the acoustic data and the reference acoustic data.

[0048] This disclosure specifically envisages a MLA that is trained to determine potential acoustic anomalies in the pump’s surroundings and optionally within the pump, and to provide a notification, for instance, to a command center. GRUNDFOS HOLDING A / S

[0049] P24118WO

[0050] P62832 / WO

[0051] The MLA may generally be referred to as an algorithm or learning procedure. The MLA may be stored in a memory, and may be executed by the processor. The MLA may comprise a set of rules and / or may employ statistical techniques to identify one or more anomalies in the acoustic data, particularly, in comparison with the reference acoustic data. The MLA may analyze the acoustic data, and may learn from detected anomalies. For example, the MLA may learn by processing the acoustic data and using it to train a model, wherein the model may comprise a neural network or similar trainable modes. The model may represent learned patterns and / or correlations in acoustic data, and the model can be used by the MLA to make decisions or predictions autonomously based on new acoustic data input. In an example, the MLA is trained to determine the root cause related to an anomaly. Training the MLA may refer to training the model within the MLA. The MLA may accordingly encompass both the definition of the model and the process of training the model.

[0052] In an implementation form of the pump system, the processor is configured to extract one or more features of the acoustic data and input the one or more features into the MLA to compare the acoustic data and the reference acoustic data and detect the anomaly.

[0053] In this way, the anomaly detection may become more efficient, as it requires less amount of data, and is more accurate. For instance, extracting features from the acoustic data may comprise selecting or transforming the raw acoustic data into relevant, informative variables (features) that enhance the MLA’s performance. This may simplify the acoustic data, highlighting patterns and relationships, which may help the MLA to learn more efficiently and make better predictions.

[0054] In an implementation form of the pump system, the processor is integrated into the pump; or the pump system further comprises a control unit, and the processor is integrated into the control unit.

[0055] The control unit may be arranged on or in the pump, or may be connected to the pump. GRUNDFOS HOLDING A / S

[0056] P24118WO

[0057] P62832 / WO

[0058] In an implementation form of the pump system, the pump system further comprises a communication interface configured to transmit information to a server, wherein the transmitted information indicates at least one of the anomaly, the acoustic data, and a deviation between the acoustic data and the reference acoustic data.

[0059] For instance, the pump system may communicate the information to a command center, which may be implemented by a management entity like a server. The management entity maybe cloud-based, that is, it maybe in a cloud. For instance, the management entity maybe implemented on a server, or maybe the server. The management entity may also be a computer or the like, and may comprise a processor. The management entity may be configured to manage one or more pumps, specifically the pump of the pump system of the first aspect. The management entity may, to this end, interact with the pump system wirelessly.

[0060] In an implementation form of the pump system, the acoustic sensor is a microphone or a vibration sensor.

[0061] In an implementation form of the pump system, the vibration sensor is attached to a housing of the pump and configured to capture a vibration of the housing as the sound of the pump.

[0062] The sound captured by the acoustic sensor maybe in the form of vibrations, which may stem from sound, which is produced by the operation of the pump, for instance, by movement of mechanical parts, like motors, valves etc., or the pumped fluid. Sound may even be transported through the fluid to the pump housing. The vibrations sensor, being attached to the housing allows accurately and reliably measuring the vibrations, for instance, unaffected by environmental sound. The captured sound may also be environmental noise or noise-like events in the surrounding of the pump, and can be captured by the microphone. The pump may comprise both microphone and vibration sensor for picking up different kinds of sounds and vibrations. GRUNDFOS HOLDING A / S

[0063] P24118WO

[0064] P62832 / WO

[0065] In an implementation form of the pump system, the pump system comprises multiple acoustic sensors arranged in the pump environment and / or at the pump.

[0066] In an implementation form of the pump system, the processor is configured to perform beamforming with the multiple acoustic sensors to determine the sound source.

[0067] For example, the room or installation site where the pump system is installed may have multiple microphones connected to the pump. By having multiple microphones, it maybe possible to execute advanced beamforming to pinpoint the exact location of the sound source, in order to quickly identify the origin of the sound for better targeted inspection and remediation of a root cause. The beamforming with multiple microphones may be used to focus on sound from a specific direction, reducing noise from other directions for clearer audio.

[0068] In an implementation form of the pump system, the acoustic sensor is provided on a main printed circuit board (PCB) of the pump.

[0069] In an implementation form of the pump system, the processor is configured to perform noise filtering and / or noise suppression on the acoustic data and compare the resulting filtered acoustic data with the reference acoustic data, in order to detect the anomaly.

[0070] This allows capturing particular anomalies more accurately, and especially reduces the chances of false positives, for example, the detection of acoustic anomalies that do not relate to any root cause or failure.

[0071] In an implementation form of the pump system, the pump system further comprises a frequency filter arranged and configured to separate sound of the pump from sound of the pump environment.

[0072] This enables the processor to distinguish between anomalies related to the pump and anomalies related to the pump environment, respectively, and potentially their GRUNDFOS HOLDING A / S

[0073] P24118WO

[0074] P62832 / WO corresponding root causes and failures. Unwanted sound can be filtered out, if desired, in order to focus on a certain kind of anomaly.

[0075] A second aspect of this disclosure provides a method for a pump system, the method comprising: capturing sound of the pump environment and optionally of the pump, and converting the captured sound into acoustic data; and comparing the acoustic data with reference acoustic data for the pump, in order to detect a deviation between the acoustic data and the reference acoustic data.

[0076] The method of the second aspect may have implementations that correspond to the implementations of the pump system of the first aspect. The method of the second aspect and its implementations achieve the effects and advantages described above with respect to the pump system of the first aspect and its respective implementations.

[0077] A third aspect of this disclosure provides a computer program comprising instructions which, when the program is executed by a pump system, instructs the pump system to perform the method according to the second aspect or any implementation thereof.

[0078] In summary of the above aspects and implementation forms, the present disclosure provides a smart acoustic monitoring and anomaly detection system, which may be integrated within a pump system or a pump, to ensure operational integrity and enhanced maintenance procedures. The at least one acoustic sensor of the pump system is capable of detecting environmental sounds indicative of potential issues with the pump system or the pump installation site and optionally internal acoustic anomalies of the pump. The pump system, for instance, a controller thereof, may store a self-learning algorithm designed to establish an acoustic baseline profile of the pump’s normal operational and surrounding sounds, for example, during a learning phase after installation of the pump. After the learning phase, the algorithm may continuously compare incoming acoustic data against this established acoustic profile to detect deviation and corresponding anomalies. GRUNDFOS HOLDING A / S

[0079] P24118WO

[0080] P62832 / WO

[0081] The pump system can thus enable automatic acoustic anomaly detection, which may result in reduced downtime in case of a failure. By disposing the MLA at the pump, there is less data being transmitted to a command center (e.g., implemented by a server or a cloud), which leads to reduced costs, and lower bandwidth requirements, etc. By detecting anomalies not just internally but especially in the surrounding pump environment, the pump system ensures a greater level of operational security.

[0082] BRIEF DESCRIPTION OF THE DRAWINGS

[0083] The above described aspects and implementations are explained in the following description of embodiments with respect to the enclosed drawings:

[0084] FIG. 1 shows a pump system according to this disclosure with an acoustic sensor.

[0085] FIG. 2 shows an exemplary pump system according to this disclosure employing a MLA to detect acoustic anomalies.

[0086] FIG. 3 shows a method according to this disclosure for detecting acoustic anomalies.

[0087] FIG. 4 shows a method of deploying and training a MLA to monitor a pump system and detect acoustic anomalies.

[0088] FIG. 5 shows a method of analyzing acoustic data do detect anomalies and sources.

[0089] DETAILED DESCRIPTION OF EMBODIMENTS

[0090] FIG. 1 shows a pump system 10 according to this disclosure. The pump system 10 comprises a pump 11, which is configured to pump a fluid, e.g. a liquid like water or oil, or a gas. The pump maybe a centrifugal pump. The pump 11 maybe a smart GRUNDFOS HOLDING A / S P24118WO

[0091] P62832 / WO pump with various integrated sensors and processing functionalities. The pump system 10 may be installed in a house or in a factory, or the like.

[0092] The pump system 10 further comprises an acoustic sensor 12, for instance, a microphone or vibration sensor, or similar. The pump system 10 may comprise more than one acoustic sensor 12, for instance, one or more microphones and / or one or more vibration sensors. Each acoustic sensor 12 may be configured to capture sound, which is related to the pump environment and / or is related to the pump 11 itself. The one or more acoustic sensors 12 of the pump system 10 are at least used to capture sound of the pump environment, i.e., caused by something in the pump environment. Optionally, the one or more acoustic sensors 12 are also used to capture sound of the pump 11, i.e., caused by the pump 11 itself. That is, the one or more acoustic sensors 12 are either used only for capturing sound of the pump environment or both of the pump environment and the pump 11. Notably, also the sound in the pump environment can of course comprise sound that originally stemming from the pump 11 (in an embodiment, this sound may be filtered out), but on the other hand sound of the pump 11 could also include internal pump sound, which is not audible in the pump environment. Each acoustic sensor 12 is configured to convert its captured sound 15 into acoustic data 16. Notably, in case of having multiple acoustic sensors 12, at least one acoustic sensors 12 may be configured to capture sound related to the pump environment, and optionally at least one acoustic sensors 12 maybe configured to capture sound of the pump 11, and / or at least one acoustic sensor 12 maybe configured to capture both pump and environmental sound(s). For example, a vibration sensor may be preferable for some kinds of vibrational sounds typically stemming from the pump 11, while a microphone may be preferable for some kinds of sounds in the pump environment. Some acoustic sensors 12 maybe directional acoustic sensors 12, like a directional microphone, while some acoustic sensors 12 may be unidirectional.

[0093] The one or more acoustic sensors 12 maybe communicatively coupled to the pump 11, for instance, to a controller or a control module of the pump 11. The acoustic sensor(s) 12 may be of the analog or digital type, and at least one acoustic sensor 12 may be connected directly a main PCB of the pump 11, or at least one acoustic sensor 12 maybe connected via wires to the pump 11. One or more acoustic sensors GRUNDFOS HOLDING A / S P24118WO

[0094] P62832 / WO

[0095] 16 can also be connected wirelessly, if separated from the pump 11, such as via Bluetooth low energy or other wireless means. One or more acoustic sensors 12 maybe placed in a human machine interface (HMI) of the pump 11. In addition to detecting acoustic data 16, which is relevant for audible anomaly detection, one or more acoustic sensors 12, preferably one or more microphones, may be able of receiving voice commands of a user of the pump 11. Such microphone(s) may be configured to only detect sound relevant to the presence of an acoustic anomaly 18 external to the pump. At least one acoustic sensor 12 could also be an integrated stereo microphone in the pump 11, which may be capable of collecting audio directivity. This can reveal angular information of audio source related to one or more anomalies 18.

[0096] The pump system 10 further comprises a processor 13, which is configured to perform various processing steps and functionalities. The processor 13 may be a controller or of a control module for the pump 11. The processor 13, or processing circuitry thereof, is configured to perform, conduct, or initiate various operations of the pump system 10 described in this disclosure. 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 application-specific integrated circuits (ASICs), field-programmable arrays (FPGAs), digital signal processors (DSPs), or multi-purpose processors. The processor 13 may further comprise memory circuitry, which can store one or more instruction(s) that can be executed by the processor 13 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 13 or its processing circuitry, causes the various described operations of the pump system 10 to be performed.

[0097] Each acoustic sensor 12 of the pump system 10 may provide captured sound as the acoustic data 16 to the processor 13. The processor 13 is configured to compare the acoustic data 16 with reference acoustic data 17 for the pump 11. The reference acoustic data 17 maybe a baseline acoustic profile that indicates a normal (failureless) operation of the pump 11. The reference acoustic data 17 may be stored in a GRUNDFOS HOLDING A / S P24118WO

[0098] P62832 / WO memory available to the processor 13, or may be obtained from the processor 13 on-demand from a command center. Based on the comparison, the processor 13 is configured to detect an anomaly 18 in the acoustic data 16, i.e., it can detect the anomaly 18 based on the reference acoustic data 17. For example, the processor 13 may be configured to determine the anomaly in the acoustic data 16 based on a deviation between the acoustic data 16 and the reference acoustic data 17. In one example, any deviation may be related to an anomaly 18, and may be accordingly detected. In another example, certain deviations are known to the processor 13, or to the software run on the processor 13, e.g. by pre-storing a set of reference deviations which relate to anomalies 18. The processor 13 could also determine a magnitude of the deviations. For example, a certain sound (e.g., a certain frequency of sound) may be increased or decreased in volume in the acoustic data 16 compared to the reference acoustic data 17, and based thereon the processor 13 may determine an anomaly 18. For example, when the deviation of the volume is above or below a certain threshold value. The processor 13 may also evaluate the spectrum of the sound, e.g., a deviation could also be caused by frequency shifts, for instance, of sound peaks between the acoustic data 16 and the reference acoustic data 17.

[0099] Voice-isolating noise suppression may be used selectively for voice recognition, since anomaly signatures are likely to occur as part of the discarded noise content. However, other contextual noise suppression may be relevant for anomaly localization and interpretation, if used for audio feature enhancement. This could possibly be part of a MLA 23.

[0100] FIG. 2 shows an exemplary pump system 10 according to this disclosure, which builds on the pump system 10 shown in FIG. 1. Same elements in FIG. 1 and FIG. 2 are labelled with the same reference signs and maybe implemented likewise.

[0101] The pump system 10 of FIG. 2 employs a MLA 23 to detect acoustic anomalies 18. In particular, the processor 13 is configured to execute the MLA 23 to detect the anomaly 18 in the acoustic data 16. The input data to the MLA 23 is based on the acoustic data 16 and the reference acoustic data 17. GRUNDFOS HOLDING A / S

[0102] P24118WO

[0103] P62832 / WO

[0104] The MLA 23 may be stored into a controller or control modules of the pump 11, which also comprises the processor 13. The MLA 23 can be configured to establish a baseline for normal operational acoustic patterns via self-learning, and then to detect deviations from this norm to identify anomalies 18, and alert a potential issue to a command center. In particular, the pump system 10 may be configured to transmit information 24 to a server 26 of the command center, wherein the transmitted information 24 indicates at least one of the acoustic data 16, a deviation, and a detected anomaly 18. The transmission may be done via a communication network 25. Instead of the server 26, the command center maybe implemented by a cloud.

[0105] The pump system 10, for instance, the pump 11, may comprises a communication interface (not shown), which is communicatively coupled to the communication network 25 to transmit the information 24 data to the command center server 26. In non-limiting embodiments, the communication network 25 can be implemented as the internet, wide-area communication network, local area communication network, a private communication network and the like. The communication link to the communication network can be wired or wireless.

[0106] The command center server 26 can be implemented by a conventional computer server. It may be operated by the operator of the pump system 10, or by a building manager.

[0107] As shown in FIG. 2, the pump system 10 may further include at least one parameter sensor 21, potentially more than one. The one or more parameter sensors 21 of the pump system 10 comprise at least one of a pressure sensor, a temperature sensor, a flow rate sensor, and an optical sensor, or two kinds thereof, or three kinds thereof, or all kinds thereof. The processor 13 is configured to determine the anomaly 18 based further on sensor data 22 provided by the at least one parameter sensor 21. The one or more parameter sensors 21 may provide real-time data on pump operations to the processor 13.

[0108] The environment of the pump system 10 may be an installation site or room 20.

[0109] The installation room 20 is where the pump 11 is installed. How the installation GRUNDFOS HOLDING A / S

[0110] P24118WO

[0111] P62832 / WO room 20 is implemented is not limited, and may, in addition to the pump 11, be the location where other devices related to the pump system 10 are installed (for example, a water tank). Just as an example, the installation room 20 can be a dedicated area in a basement of a house, boiler room of a building (commercial, industrial, or even an apartment), an engine room of a movable object (such as a ship), or similar.

[0112] As shown in FIG. 2, the at least one acoustic sensor 12 of the pump system 10 may comprise room microphones, which maybe installed in the installation room 20. These room microphones maybe communicatively coupled, wired or wirelessly, to the pump 11, particularly the processor 13, and / or the server 26. In some implementations of the pump system 10, the room microphones are configured to detect audible anomalies 18 within the installation room 20. In additional implementations of the pump system 10, the room microphones can be used to determine the location of the audible anomaly 18. In some implementations of the pump system 10, the room microphones are each embedded within a device also stored within the room 20 (for example, additional pumps).

[0113] FIG. 3 shows a general method 30 according to this disclosure, to be performed by the pump system 10. The method 30 comprises a step 31 of capturing sound related to the pump environment and optionally capturing sound of the pump 11, and converting the captured sound 15 into acoustic data 16. This may be done by the acoustic sensor 12. The method 30 further comprises a step 32 of comparing the acoustic data 16 with reference acoustic data 17 for the pump, in order to detect an anomaly 18 in the acoustic data 16. This may be done by the processor 13.

[0114] FIG. 4 shows a method 40 of deploying and training a MLA 23 to monitor a pump system 10 and detect acoustic anomalies 18.

[0115] As has been alluded above, the MLA 23 may be configured to establish a normal operational acoustic baseline profile or pattern of the pump 11 operating in the installation room 20. Given that each pump system 10 and each installation room 20 is unique, the MLA 23 maybe configured to learn the normal operating acoustic GRUNDFOS HOLDING A / S

[0116] P24118WO

[0117] P62832 / WO conditions specific to the environment the pump 11 is installed in. The flowchart of the method 40 in FIG. 4 illustrates an example how the MLA 23 is trained and used.

[0118] At step 41, the MLA 23 is installed at the pump 11 with no prior training data. How the MLA 23 is implemented is not limited in this disclosure. The MLA 23 maybe a general-purpose anomaly detection MLA, such as an autoencoder or a one-class support vector machine designed to learn a representation of a single class (i.e. the “normal” operating state).

[0119] At step 42, the pump 11 operates under its typical conditions and the MLA 23 collects, via the one or more acoustic sensors 12, the acoustic data 16 within the pump 11 and / or near the pump 11. How long the acoustic data 16 is collected is not limited, and may be predetermined by the operator of the pump 11 (for example, one week, one month, etc.)

[0120] At step 43, the MLA 23 is configured to extract features from the collected acoustic data 16, to ensure that it is in a suitable format for training the MLA 23.

[0121] At step 44, the MLA 23 then creates a model based on the processed acoustic data 16, to learn a profile of what constitutes a normal acoustic signature for that particular pump 11 in its specific operational environment 20.

[0122] At step 45, once the MLA 23 has been trained, the MLA 23 continuously analyzes incoming acoustic data 16 of one or more acoustic sensors 12, comparing it to the baseline profile - i.e. the reference acoustic data 17 - for anomaly detection. That is, the pump system 10 performs the method 30. If the MLA 23 detects an acoustic signature that, for example, deviates above a certain threshold (e.g. determined by the operator of the server 26), the MLA 23 is configured to notify the detected anomaly 18 to, for example, the server 26. It may also indicated the server 26 to take corrective action(s) or explore the issue in more detail, such as opening a channel so that the operator can listen to the acoustic anomaly 18.

[0123] In some non-limiting examples, after the operator has determined a root cause related to the detected acoustic anomaly 18, the operator may execute a feed-back GRUNDFOS HOLDING A / S

[0124] P24118WO

[0125] P62832 / WO loop mechanism to update the MLA 23, such that in future cases with a similar anomaly 18, the MLA 23 is not only capable of detecting the anomaly 18, but also to provide a potential root cause to the anomaly 18. This continuous feedback loop mechanism is particularly useful to counter any changes in the installation room 20 after the training phase, such as removing or adding a new device within the installation room 20 (which may add sound to the pump environment).

[0126] Although the training of the MLA 23 has been explained as being made exclusively within the pump 11, it should be understood that it is not limited as such. Certain steps (such as step 44) may be done at the server 26 of the command center. Moreover, although in the present explanation, the MLA 23 is installed within the pump 11 with no prior training data, it is not always limited as such. It is contemplated that the MLA 23 may be pre-trained with certain acoustic anomaly detection functionalities, for example, which are non-dependent of the operational environment 20 of the pump 11.

[0127] FIG. 5 shows details of a method 50 of analyzing acoustic data 16 do detect anomalies and sources. This method 50 may relate to the step 32 of the general method 30, and may be performed by the processor 13, in particular, employing the MLA 23.

[0128] For example, in some implementations, in addition to detecting an acoustic anomaly 18, the MLA 23 is further configured to determine whether the acoustic anomaly 18 pertains to the pump 11 (or the pump system 10), or pertains to the installation room 20.

[0129] At step 51, the MLA 23 analyzes the acoustic data 16. How the MLA 23 analyzes the acoustic data 16 to detect an anomaly 18 is not limited, and examples have been given above. In some examples, the MLA 23 is configured to analyze the acoustic data 16 from one or more acoustic sensors 12.

[0130] At step 52, the MLA 23 may detect an anomaly 18, if the acoustic data 16 is abnormal. Examples of detecting anomalies 18 in the acoustic data 16 based on the reference acoustic data 17 have been explained previously. GRUNDFOS HOLDING A / S P24118WO

[0131] P62832 / WO

[0132] If the acoustic data 16 is abnormal, the MLA 23 may be further configured to determine the origin of the anomaly 18.

[0133] For instance, at step 53 the MLA 23 may determine that the previously detected acoustic anomaly 18 originates from the pump 11 or the pump system 10. For determining this, the MLA 23 may, for example, search for one or more correlations between the acoustic data 16 and at least one concurrently captured data event of another type than acoustic data, for instance, captured in the pump 11 or in the vicinity in the pump system 10. The correlation(s) may indicate that the detected acoustic anomaly 18 most likely originates from the pump 11 or the pump system 10. Conversely, if no such correlation exists, this may indicate that the detected acoustic anomaly 18 does not originate from the pump 11 or the pump system 10. Some non-limiting examples of potential root causes that are responsible for abnormal acoustic data 16 within the pump 11 (or pump system 10) - i.e. for an acoustic anomaly 18 - include, but are not limited to at least one of: cavitation; mechanical seal leakage; bearing failures; overheating; excessive vibration; flushing leakage; water hammer; air in the hydraulic system.

[0134] If on the other hand, the MLA 23 determines that the acoustic data 16 is not indicative of a pump issue, the MLA 23 may be configured at step 54 to determine that the anomaly 18 is related to a root cause in the installation room 20. For determining whether the anomaly 18 is related to a root cause in the installation room 20, the MLA 23 may, for example, establish a likelihood of the acoustic anomaly 18 not originating from the pump 11 or the pump system 10. The likelihood may be determined, for instance, based on the before-mentioned data event correlation analysis. The MLA 23 may then combine the likelihood with deduced spatial properties of acoustic data 16 captured by at least one acoustic sensor 12 of the pump system 10 installed in the installation room 20. Such root causes may include, but are not limited to at least one of: water leakage (from pipes, valves, etc.); non-pump related device noise; intrusion; ventilator unbalance; pressurized air leakage; smoke, fire, gas alarm. GRUNDFOS HOLDING A / S

[0135] P24118WO

[0136] P62832 / WO

[0137] At step 55, in the pump 11, the MLA 23 or the server 26 may be configured to determine a location, within the installation room 20, from which the sound causing the anomaly 18 originates from. How the location detection is made is not limited in this disclosure, and may, just as an example, be implemented by one or more of the following methods.

[0138] In one example, the at least one acoustic sensor 12 may include microphones that are mono units, and it maybe possible to execute sound-delay based triangulation and collaborative beamforming.

[0139] In another example, the at least one acoustic sensor 12 may include microphones that are stereo units, and localization could be achieved without the need for tight collaborative co-sampling of the microphones, since individual directional information would suffice as triangulation data, thereby eliminating strong correlation needs. However, detection reliability and accuracy of anomaly location within the installation room 20 may optionally be improved by using a correlation analysis between concurrent acoustic data and non-acoustic event data related to the pump 11 or the pump system 10.

[0140] In another example, an acoustic anomaly 18 with an identified acoustic signature of cavitation may be detected by the MLA 23 in the acoustic data 16 acquisitioned through at least one acoustic sensor 12 external to the pump 11. In addition, concurrent measurements of one or more pump-related parameters - e.g. inlet pressure, media temperature, and / or pump duty point - may exhibit conditions clearly prone to generating cavitation in the pump 11. In this case, a detection certainty of localizing the cavitation in the pump 11 may be high, even if other cavitation-prone components (such as a valve) may exist in the vicinity of the pump 10 and may not be directly distinguishable by means of purely acoustic localization algorithms, i.e., concluding on pump issue 53 or room issue 54.

[0141] In another example, if the MLA 23 detects an acoustic anomaly 18 without being able to determine its type, a correlation with said concurrently occurring cavitation-prone conditions and / or parameters measured in the pump 11 could GRUNDFOS HOLDING A / S P24118WO

[0142] P62832 / WO increase the certainty of concluding, for instance, that in fact the pump-cavitation is causing the acoustic anomaly 18.

[0143] In another example, if the MLA 23 detects an acoustic anomaly 18 of the type “flushing leakage”, and succeeds to acoustically localize its origin in the installation room 20 distant to the pump 11, a correlation with concurrent data of the monitored duty point of the pump 11, and / or other internal pump-related parameters, may reveal whether said leakage is or is not happening in the pump’s closed hydraulic system, or is happening in another unrelated system at the same location in the installation room 20. This information could direct warning and mitigation actions to be initialized in either case.

[0144] The above examples of correlating an acoustic anomaly 18 and a concurrent pattern in one or more monitored pump parameters could be modified by replacing the concurrent pattern data with recorded data, e.g., historic data. For example, the MLA 23, triggered by the acoustic anomaly 18, may search for relevant trending pattern developments in in-pump monitored and recorded data, wherein this recorded data was not recognized at the time of its acquisition as a pre-warning of the anomaly 18 to occur later. Also trending patterns in historic acoustic data could contribute to this analysis. Either one or both could serve as training data for refining system-specific predictive anomaly warning capabilities of the MLA 23.

[0145] For instance, revisiting the “flushing leakage” example mentioned above, possibly recurring water hammer events, which may be vaguely reflected in recorded data sets e.g. acoustically 12 or as pressure fluctuations in the pump 11, may have been deemed harmless by the MLA 23 at their acquisition time. However, these events may have cumulatively weakened the hydraulic system, and may over time have caused the leakage. In this case, training the MLA 23 on non-concurrent correlation(s) could enable the MLA 23 to assess and communicate the risk of a similar future anomaly occurring in the specific installation room 20 where it is installed. GRUNDFOS HOLDING A / S

[0146] P24118WO

[0147] P62832 / WO

[0148] In another example, it may also be contemplated that a person skilled in the art could combine the above mentioned techniques to improve the localization accuracy.

[0149] In summary, the solutions of this disclosure, which are implemented in the pump system 10, for instance, by means of the MLA 23, enable an automatic and autonomous acoustic anomaly detection at the pump system 10. This may result in reduced down time of the pump system 10 in case of a failure. In other cases, failures can even be prevented by early detection of acoustic anomalies 18.

[0150] Due to the fact that acoustic anomalies 18 cannot just be detected internally, i.e. related to the pump system 10, but also in the surrounding environment 20 of the pump 11, the pump system 10 has a greater level of operational security.

[0151] Further, since the processor 13, particularly the MLA 23, is operable at the pump system 10, there is only little data that has to be transmitted to the server 26 of the command center. This reduces the requirement for communication bandwidth and also reduces costs.

[0152] In the claims as well as in the description of this disclosure, 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 may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.

Claims

GRUNDFOS HOLDING A / SP24118WOP62832 / WOClaims1. A pump system (10) comprising: a pump (11) for pumping a fluid; an acoustic sensor (12) configured to capture sound (15) of the pump environment and optionally of the pump (11), and convert the captured sound (15) into acoustic data (16); and a processor (13) configured to compare the acoustic data (16) with reference acoustic data (17) for the pump (11), in order to detect an anomaly (18) in the acoustic data (16).

2. The pump system (10) according to claim 1, wherein the processor (13) is configured to determine the anomaly (18) in the acoustic data (16) based on a deviation between the acoustic data (16) and the reference acoustic data (17).

3. The pump system (10) according to claim 1 or 2, wherein the processor (13) is configured to determine whether the anomaly (18) relates to the pump (11) or to the pump environment.

4. The pump system (10) according to one of the claims 1 to 3, wherein the processor (13) is configured to determine a sound source related to the anomaly (18).

5. The pump system (10) according to claim 4, wherein the processor (13) is configured to determine a location of the sound source.

6. The pump system (10) according to one of the claims 1 to 5, wherein the processor (13) is configured to control the pump (11) to perform an operation to remove an anomaly (18) related to the pump (11).

7. The pump system (10) according to one of the claims 1 to 6, further comprisingGRUNDFOS HOLDING A / SP24118WOP62832 / WO at least one parameter sensor (21) comprising at least one of a pressure sensor, a temperature sensor, a flow rate sensor, and an optical sensor; wherein the processor (13) is configured to determine the anomaly (18) based further on sensor data (22) provided by the at least one parameter sensor (21).

8. The pump system (10) according to one of the claims 1 to 7, wherein the anomaly (18) indicates at least one of:- an anomalous flow of the fluid;- an anomalous pressure of the fluid;- an anomalous fluid temperature;- an anomalous pump temperature;- water leakage in the pump environment;- a ventilator unbalance in the pump environment;- smoke, fire, or a smoke or fire alarm in the pump environment;- air leakage in the pump environment.

9. The pump system (10) according to one of the claims 1 to 8, wherein the processor (13) is configured to execute a machine learning algorithm, MLA, (23) to detect the anomaly (18) in the acoustic data (16), wherein input data to the MLA (23) is based on the acoustic data (16) and the reference acoustic data(17)-10. The pump system (10) according to claim 9, wherein the processor (13) is configured to extract one or more features of the acoustic data (13) and input the one or more features into the MLA (23) to compare the acoustic data (16) and the reference acoustic data (17) and detect the anomaly(18).

11. The pump system (10) according to claim 9 or 10, wherein: the processor (13) is integrated into the pump (11); or the pump system (10) further comprises a control unit, and the processor (13) is integrated into the control unit.GRUNDFOS HOLDING A / SP24118WOP62832 / WO12. The pump system (10) according to one of the claims 1 to 11, further comprising a communication interface configured to transmit information (24) to a server (26), wherein the transmitted information (24) indicates at least one of the anomaly (18), the acoustic data (16), and a deviation between the acoustic data (16) and the reference acoustic data (17).

13. The pump system (10) according to one of the claims 1 to 12, wherein the acoustic sensor (12) is a microphone or a vibration sensor.

14. The pump system (10) according to one of the claims 1 to 13, wherein the acoustic sensor (12) is a vibration sensor attached to a housing of the pump (11) and configured to capture a vibration of the housing as the sound of the pump (11).

15. The pump system (10) according to one of the claim 1 to 14, comprising multiple acoustic sensors (12) arranged in the pump environment and / or at the pump (11); wherein the processor (13) is configured to perform beamforming with the multiple acoustic sensors (12) to determine a sound source related to the anomaly (18).

16. The pump system (10) according to one of the claims 1 to 15, wherein the acoustic sensor (12) is provided on a main printed circuit board, PCB, of the pump (11).

17. The pump system (10) according to one of the claims 1 to 16, wherein the processor (13) is configured to perform noise filtering and / or noise suppression on the acoustic data (16) and compare the resulting filtered acoustic data (16) with the reference acoustic data (17), in order to detect the anomaly (18).

18. The pump system (10) according to one of the claims 1 to 17, further comprisingGRUNDFOS HOLDING A / SP24118WOP62832 / WO a frequency filter arranged and configured to separate sound of the pump (11) from sound of the pump environment.

19. A method (30) for a pump system (10), the method (30) comprising: capturing (31) sound (15) of the pump environment and optionally of the pump (11), and converting the captured sound (15) into acoustic data (16); and comparing (32) the acoustic data (16) with reference acoustic data (17) for the pump (11), in order to detect an anomaly (18) in the acoustic data (16).

20. A computer program comprising instructions which, when the program is executed by a pump system (10), instructs the pump system (10) to perform the method (30) according to claim 19.

Images