A method, arrangement, and system for analyzing the state of at least one entity

Abstract

A computer-implemented method for analyzing the state of at least one entity, the method comprising obtaining signals (002) indicative of vibrations occurring in the entity, utilizing at least one artificial intelligence model to analyze (004) the obtained signals, determining, based on the analysis, if the obtained signals are indicative of regular behavior (006) of the entity, and determining an anomaly (008) in the state of the entity based on predetermined criteria if at least one obtained signal or sequence of signals is not indicative of regular behavior of the entity. A related arrangement and system.

Description

[0001] A METHOD, ARRANGEMENT, AND SYSTEM FOR ANALYZING THE STATE OF AT LEAST ONE ENTITY

[0002] TECHNICAL FIELD OF THE INVENTION

[0003] The invention relates to processing of data in general. More specifically, the invention relates to a method, arrangement, and system for analyzing the state of at least one entity based on obtained signals indicative of vibrations occurring in the entity.

[0004] BACKGROUND OF THE INVENTION

[0005] There are several known sensors, systems, and methods for identifying, monitoring, and analyzing vibration signals emitted from various mechanical systems.

[0006] Existing sensors are applied in vibrating arrangements containing moving masses, such as paper making machines, oil and gas processing and chemical plants, pipelines, turbines, power and pumping stations, various vehicles, railways, metros, ships, airplanes, space vehicles, cranes, drones, rockets and underwater vehicles and structures, engines, motors, pumps, drives, gears, conveyers, piping systems with flowing liquids and gases, and various military and security applications.

[0007] Vibrating elements may also be various mechanical components and elements, such bearings, gears, chain drives, circuit breakers, clutches, fans, valves, items with unbalanced and reciprocating masses.

[0008] Known sensors are also used for monitoring stationary structures where vibrations are caused by events in the environment such as seismic phenomena, movements in earth’s crust, storms, waves, landslides, snow and ice loads, floods, atmosphere, or cosmic phenomena. Such systems are used for instance for safety and security fencing and safety arrangements, monitoring of buildings, foundations, roads, bridges, dams, waterways, under water sound monitoring, channels and locks, landslides, earthworks, and earthquakes.

[0009] There are several prior art vibration sensors which use piezoelectric coaxial type cables and accelerometers. For instance, known solutions provide coaxial cables where narrow dielectric piezo polymer ribbons are wrapped between a center core and an outer braid. Such sensors are used for a variety of applications including measuring fluctuating accelerations and amplitudes in vibrating environments. Maintenance professionals use vibration sensors for monitoring, avoidance of breakdowns and planning maintenance of the machinery and systems, to reduce costs and increase the performance.

[0010] However, the vibration sensors and related systems and methods of the prior art have several problems, limitations, and disadvantages. The sensors have limited performance in terms of sensitivity, frequency and amplitude ranges. Prior art systems and vibration sensors, measuring acceleration and amplitude, are typically designed and calibrated for measuring vibrations of a specific and designated component or element, such as for instance a bearing, electric motor, or an unbalanced element.

[0011] Yet, large, complex, and expensive arrangements, such as for instance a paper making machines, chemical plants, oil refineries and offshore oil I gas production platforms can have even thousands of vibrating elements, resulting in an exceedingly complex six-dimensional vibration environment. State of the art vibration sensors cannot sufficiently monitor such arrangements.

[0012] Prior art sensors must be placed very close to the vibration source because the amplitudes are attenuated. Yet, the signal may be contaminated by other vibration sources as the distance between the sensor and the source to be monitored increases. In some cases, it is difficult or even impossible to place known sensors in locations that are sufficiently close for effective monitoring of vibrations.

[0013] A further serious disadvantage of prior art sensors is that they are easily damaged and disturbed by mechanical, electromagnetic, and various environmental events, which reduce their applicability and sensitivity. Also, sensors require an outside power source which typically generates a 50 Hz harmful interference for such sensors.

[0014] Prior art systems for monitoring large and complex arrangements or structures also require a wide array of vibration sensors distributed around a large area. For instance, a paper machine and power station require a large number of expensive sensors with associated cabling and data processing centers. All this results in expensive and complex systems.

[0015] Furthermore, complex systems with a large amount of sensor elements are heavy and considerably large in size. This limits their use in weight and volume sensitive applications such as ships and boats, aviation, space and underwater vehicles, as well as in offshore and space structures.

[0016] Existing vibration sensors and associated measuring and monitoring systems and methods are thus often very expensive to design and build, while incorporation of such systems into existing structures is often also difficult and expensive. Maintenance of the systems is also laborious and these systems are often sensitive to damage.

[0017] It would be advantageous to provide a highly sensitive vibration sensor giving information regarding a plurality of types of vibrations that could be used for a variety of purposes.

[0018] Relating to monitoring entities such as elements, structures, or arrangements in general, it is known to provide edge computing units that may provide processing of data close to the data source, reducing volumes of data that should be transferred and the associated traffic with transferring data to remote locations. This may provide lower latency and reduce transmission costs associated with data transfer.

[0019] In connection with processing signals relating to vibration sensors, prior art solutions have several problems, limitations, and disadvantages when processing weak signals from vibration sensors. In prior art systems, the signal from a vibration sensor is typically connected to a computing unit with a cable connector or other element which introduces the need for a signal to pass through a discontinuity or interphase between electrical components. This causes an additional contaminating disturbance to the weak signal obtained from the vibration sensor.

[0020] Furthermore, other disturbances to the signal obtained from the vibration sensor, which induce contamination of the signal, may be caused by electromagnetic disturbances around or within a computing unit. Such disturbance can be caused for instance by any item using electric AC- power, such as for instance an electric motor, transformer or power supply, typically at 50Hz of 60 Hz frequency and their harmonics. Further disturbances may be caused by contamination by a grounding voltage of various electronic components that are provided in a computing unit.

[0021] Another problem with state-of-the-art systems relates to the dynamics of analog signals from vibration sensors. Especially when receiving and processing analog vibration signals from a large and complex vibrating arrangement, such as for instance oil I gas production or chemical plants, signals can periodically contain wide and unexpected swings in amplitude and frequency. These can occur due to environmental phenomena as well as due to accidents or other issues in the e.g. plant’s systems. Such large, unexpected analog signals are usually filtered out. Present systems are not designed to monitor such occurrences, although detection of these and their analysis may be advantageous.

[0022] Known computing systems and methods in connection with vibration sensors are typically designed for a specific use case and usually contain specialized hardware and software adaptations for the known use case and environment. Artificial intelligence (Al) systems are replacing more traditional expert systems in fault detection and monitoring. As large amounts of data can be collected cheaper and faster than ever before, building machine learning apparatuses and methods to monitor and analyze systems may become an attractive option in many situations. The state-of- the-art machine learning systems achieve good results, and when correctly applied, their use leads to increased efficiency due to their robustness and flexibility.

[0023] Building advanced machine learning models to monitor vibrating systems may also be an attractive option for many applications.

[0024] There exist several apparatuses, methods, arrangements and measuring units including Al systems for analyzing vibration signals within and around vibrating entities.

[0025] While state-of-the-art systems and methods can provide information about the entities they are monitoring, they suffer from have several problems, limitations and disadvantages. Firstly, in order to operate reliably, they require data of all eventual failure modes of the various components and arrangements they are designed to monitor and analyze. This means that failure and fault conditions of important components and arrangements should be induced repeatedly and in various ways in order to acquire enough data regarding all eventual failure modes. Without such data the system is unable to detect any anomalies and problems that were not included in such training data. This is obviously a serious and costly problem even for a moderately complex arrangement, consisting of several components and various elements.

[0026] Secondly, the reliance on specific, labeled data leads to situations where the monitoring and analysis system is built around a specific and unique dataset. This means that while a model may work well for monitoring and analyzing that specific entity, it cannot be applied to a different entity and environment. Therefore, any new model must be built from the ground up, based on data of the new entity.

[0027] SUMMARY OF THE INVENTION

[0028] An object of the invention is to alleviate at least some of the problems relating to the known prior art. In one aspect of the invention a computer- implemented method for analyzing the state of at least one entity is provided, the method comprising

[0029] - obtaining signals indicative of vibrations occurring in the entity,

[0030] - utilizing at least one artificial intelligence model to analyze the obtained signals,

[0031] - determining, based on the analysis, if the obtained signals are indicative of regular behavior of the entity, and

[0032] - determining an anomaly in the state of the entity based on predetermined criteria if at least one obtained signal or sequence of signals is not indicative of regular behavior of the entity.

[0033] With the present invention, one or more vibration analysis artificial intelligence models or machine learning algorithms may be used in detecting anomalies and monitoring change as compared with established vibration signature patterns of an entity when it is operating faultlessly or in its regular or desired state. The entity may be a single entity or arrangement comprising a plurality of components and may be related to e.g. machinery or other structures such as buildings etc. as is known in connection with detection of vibrations.

[0034] Anomalies may be caused by excessive wear of breakage of any component or composition inside the monitored system, or some outside event, such as environmental phenomena, earthquake, collision with some outside object, etc. Furthermore, anomalies in the vibration spectrum monitored can indicate other issues such as lubrication breakdown and bearing defects sufficiently far in advance to permit timely repair, prolonging the lifetime of the asset.

[0035] By using a sufficient sampling rate, such as 1-192000 / s, and monitoring continuously, a very large amount of data (signals) may be obtained. This data may be stored into a data storage unit and then analyzed for creating patterns for various vibration phenomena in the monitored entity.

[0036] The signals indicative of vibrations occurring in the entity may be obtained and analyzed as essentially unfiltered signals.

[0037] The one or more artificial intelligence models may comprise at least one large language model.

[0038] The one or more artificial intelligence models may be models employing multimodality and may comprise e.g. a multimodal large language model. A multimodal language model may utilize e.g. text, image, audio, video, and code as inputs and may also provide e.g. text, image, speech, video, and actions as outputs.

[0039] The one or more artificial intelligence models may be trained with previously obtained signals indicative of vibrations occurring in the entity or the method may comprise training of the one or more artificial intelligence models with obtained training signals indicative of vibrations occurring in the entity to classify signals and / or sequences of signals as being signals that are indicative of regular behavior of the entity and belonging to a first class of signals or sequences of signals.

[0040] Utilized training signals may comprise unlabeled data. A unique feature of the present invention may be the use of machine learning algorithms for fault / anomaly detection and / or prediction in connection with vibration analysis that does not need labeled data to function. Therefore, it can be applied “universally” across a wide range of environments and applications with minimal modifications.

[0041] The one or more artificial intelligence models may comprise at least one machine learning algorithm, preferably wherein the machine learning algorithm is trainable by self-supervised learning.

[0042] The machine learning algorithm may comprise at least one autoencoder module, optionally a vector-quantized variational auto-encoder module.

[0043] The machine learning algorithm may comprise at least one transformer model, optionally a performer model.

[0044] Several different kinds of neural networks may thus in some embodiments be used in combination and employed for use in a new kind of system and method that may receive and processes vibration data and utilize this data to detect even previously undetected anomalies in the vibrating behavior of an entity.

[0045] A unique self-learning feedback algorithm may be triggered when new vibrational signals are detected, e.g. when an obtained signal or sequence of signals is detected that is determined as not belonging to a first class of signals or determining that an obtained signal or sequence of signals differs from predicted signal or sequence of signals according to selected criteria, such as by differing by over a threshold amount. The new signals may then be monitored and analyzed to create new patterns, i.e. classify the new signals as belonging to the first class of signals or sequences of signals, optionally by a conversion unit which may be an edge type computing unit. A newly identified and analyzed anomaly pattern may be stored in a local data storage. Information on this new anomaly pattern can then be used for maintenance and repair of the monitored entity.

[0046] The method may comprise providing a notification if an anomaly in the state of the entity is determined.

[0047] The signals may be obtained from at least one piezoelectric sensor, preferably wherein the signals comprise information relating to a magnitude of vibrations in six vibrational modes. The signals may indicate no vibration in some mode if such is not detected.

[0048] The analysis may comprise determining at least one predicted future signal or predicted future sequence of signals indicative of vibrations occurring in the entity based on the obtained signals, obtaining further signals indicative of vibrations occurring in the entity and determining at least one observed signal or observed sequence of signals, and comparing at least one predicted future signal or predicted future sequence of signals to at least one observed signal or observed sequence of signals.

[0049] The method may comprise determining that at least one observed signal or observed sequence of signals is not indicative of regular behavior of the entity if at least one predicted future signal or predicted future sequence of signals differs from at least one corresponding observed signal or observed sequence of signals by over a threshold amount.

[0050] During a training phase, the one or more artificial intelligence models may learn the vibrational patterns (sequences of signals indicative of vibrations) of the entity which correspond to a regular state of the entity. The one or more artificial intelligence models may be models that have been developed for language models and may be configured to originally be utilized in determining / predicting a subsequent word based on a sequence of words obtained previously. In the present invention, such models may be used to construct a “language” corresponding to the vibrations of the entity and instead of determining or predicting a subsequent word, a subsequent signal indicative of vibrations occurring in the entity may be determined. Multimodal language models may be employed. If an observed subsequent signal differs from the determined one by over a threshold amount, the one or more artificial intelligence models may determine that the entity is not exhibiting a vibrational state indicative of regular behavior of the entity, and an anomaly may be determined.

[0051] An efficiency of the analysis may be evaluated based on determining at least one difference between at least one predicted future signal or predicted future sequence of signals and at least one corresponding signal or sequence of signals known to be regular behavior of the entity, preferably wherein a plurality of predicted future signals or predicted future sequences of signals may be determined comprising an order of certainty, and an average difference between a selected number of predicted future signals and observed signals or a difference between a selected number of predicted future sequences of signals and observed sequences of signals may be determined.

[0052] The invention also relates to a computer-implemented method for training at least one artificial intelligence model, the method comprising - obtaining signals indicative of vibrations occurring in an entity,

[0053] - training of the one or more artificial intelligence models with the obtained training signals indicative of vibrations occurring in the entity to classify signals and / or sequences of signals as being signals that are indicative of regular behavior, wherein the at least one artificial intelligence model comprises at least one transformer model, optionally a performer model.

[0054] In a further aspect of the invention, an arrangement for analyzing the state of an entity may be provided, the arrangement comprising at least one conversion unit configured to receive a plurality of first signals indicative of vibrations occurring in the entity, wherein the plurality of first electrical signals comprise at least a plurality of analog electrical signals which are obtained via at least one piezoelectric sensor, wherein the conversion unit is configured to deliver at least one second signal as an output, wherein said second signal is a digital signal indicative of the first signals, and at least one remote processing unit, wherein the remote processing unit is configured to receive the second signals, wherein the at least one of the conversion unit or remote processing unit is further configured to

[0055] - utilize at least one artificial intelligence model to analyze the obtained first and / or second signals,

[0056] - determine, based on the analysis, if the signals are indicative of regular behavior of the entity, and

[0057] - determine an anomaly in the state of the entity based on predetermined criteria if at least one obtained signal or sequence of signals is not indicative of regular behavior of the entity.

[0058] The arrangement may be configured to utilize a plurality of first and / or second signals as training data for the at least one artificial intelligence model to determine a first class of signals and / or signal sequences which are indicative of regular behavior of the entity. The present invention may enable processing and compression of data indicative of vibrations occurring in an entity and wireless transmission of the processed and compressed data for remote processing.

[0059] In alternative embodiments, data indicative of vibrations occurring in an entity, preferably processed and compressed data, may be transmitted for remote processing in wired manner.

[0060] In a further aspect of the invention, a system for analyzing the state of an entity may be provided, the system comprising at least one arrangement as specified herein and at least one piezoelectric sensor, preferably an elongated cable-like sensor, configured to provide signals indicative of a magnitude of vibrations in six vibrational modes. The piezolelectric sensor utilized may be a piezolelectric sensor as specified further herein. Signals indicative of vibrations occurring in an entity may still of course be obtained by using some other type of sensor.

[0061] An entity may be an object such as a structure or an environment, for instance. An entity may for example be an engine, a bridge, or a road. Vibrations are e.g. caused in various engines, such as diesel engines, where moving elements such as piston mechanisms, gears, rotating arrangements, valve systems, turbochargers, moving gases and liquids cause vibrations.

[0062] “Regular behavior” of the entity may for example refer to behavior that is expected for the entity, behavior that corresponds to a natural state of the entity without influence from the environment, or behavior that corresponds to behavior of the entity in a known state of the entity, such as functioning in a selected state. Regular behavior of an entity may in some cases e.g. refer to essentially no vibrations or vibrations only e.g. below a certain threshold occurring in the entity, such as when the entity is a construction that should not be moving (or only moves relatively little) in its regular state.

[0063] In one embodiment of a system, an elongated, cable-like piezoelectric sensor may be utilized, the piezoelectric sensor comprising at least a conducting core portion having a longitudinal axis that is substantially larger than a cross-sectional axis of the core portion, said core portion exhibiting a preferably essentially round cross-sectional axis, the piezoelectric sensor additionally comprising at least two helical piezoelectric portions comprising at least a first piezoelectric helical portion and a second piezoelectric helical portion, wherein the first piezoelectric helical portion is helically wound about the core portion to exhibit a helical axis essentially codirectional with the longitudinal axis of the core portion, wherein the first piezoelectric helical portion essentially corresponds to a first helix comprising a first helix angle and first handedness, and wherein the second piezoelectric helical portion is helically wound about the core portion and the first piezoelectric helical portion to exhibit a helical axis essentially codirectional with the longitudinal axis of the core portion, wherein the second piezoelectric helical portion essentially corresponds to a second helix comprising a second helix angle and second handedness, the piezoelectric sensor additionally comprising at least a first conducting outer portion essentially surrounding the core portion and the first piezoelectric helical portion and the second piezoelectric helical portion.

[0064] The first handedness is opposite to the second handedness.

[0065] The first helix angle and the second helix angle are advantageously essentially equivalent, such that an angle of about 90 degrees is formed between the first helical portion and the second helical portion at a location where the first helical portion and second helical portion cross each other. The first helix angle and second helix angle are advantageously about 45 degrees.

[0066] With signals being obtained via the piezoelectric sensor described herein, a sensitive and robust piezoelectric sensor with a wide measurement frequency and amplitude range may be provided, which may effectively and economically measure vibrations in a total of six possible vibrational dimensions, providing information regarding a total vibrational environment.

[0067] Mechanical vibrations may occur in a six-dimensional vibrational environment. These different vibration dimensions or modes comprise three motion dimensions along x, y and z axes and three rotational vibration dimensions around these axes. Also, vibration characteristics in dimensions, frequencies and amplitudes can change over time.

[0068] The six-axial piezoelectric sensor may be highly sensitive, the high sensitivity leading to higher accuracy.

[0069] Versatile collection of data relating to vibrations may be enabled, and the comprehensive obtained may enable determining discrete locations from which the vibrational data is acquired.

[0070] Due to the piezoelectric material being arranged with helical piezoelectric portions as disclosed herein, the structure of the sensor allows detection of vibrations in all six dimensions as a vector sum. The signal-generating preferably polarized piezo crystals of the helical piezoelectric portions may be arranged so that vibrations in six dimensions may be effectively detected.

[0071] The piezoelectric sensor of the invention may be provided in a cable-like extended form, whereby measurements or sensing may be performed in a large area, along the whole length of the sensor. A maximum length of the piezoelectric sensor may be about 100 m.

[0072] As the sensor may be very robust, it will not break upon bending, and e.g. bending or stretching of the sensor may also be detected. The piezoelectric helical portions being wound about the core portion such that they correspond to helices with different handedness may enhance the robustness of the sensor while increasing sensitivity.

[0073] In various applications, one or more sensors can be used. Sensors may extend over a selected area (wide or large area). An entity that is monitored may be large compared to the size of the sensor, due to the sensitivity of the sensor. Vibrational signals may be picked up from sources even e.g. 1.5 km away from the sensor (depending, of course, on the vibration source and the environment where the sensor is located / fixed). With the described sensor, a large entity or large area associated with an entity may be monitored with less sensors, making the measurement arrangement simpler, more robust, easier to install, and more economical.

[0074] Only one piezoelectric sensor may be required in some cases. With the multilayer structure of the sensor of the invention, the resistance may be reduced and sensitivity may be increased, whereby an elongated form of the sensor, where the length of the cable-type piezoelectric sensor may be increased.

[0075] The elongated form cable-type piezoelectric sensor can be used in long cable form or a large area form e.g., a plate, band, or disc sensor. Here, the piezoelectric material may be provided to form such e.g. plate shape. A limited number of individual sensors may replace several prior art sensors, associated cables, fittings, computers, etc. This aspect may be especially important for large entities or arrangements comprising several entities to be monitored occupying large areas, such as paper machines, power stations, ship machinery, pipeline pumping stations, or various plants processing oil, gas, or chemicals.

[0076] The piezoelectric sensor may enable integration of one or more further sensor types, such as thermal or optical sensors, for instance

[0077] A further aspect is that the piezoelectric sensor does not have to be placed close to the vibration source in order to detect vibrations. The sensor may be placed at a distance, while still effectively measuring vibrations. Thus, the sensor may be used in a challenging environment and without e.g. having to disassemble any entity that is to be monitored.

[0078] The sensor can therefore be fixed onto an entity such as a structure to be monitored or to e.g. another structure that is external to the structure to be monitored. The sensor may be fixed to a structure by any fastening means, such as adhesive or magnetic means.

[0079] In some environments, the sensor(s) may be embedded into a material of an entity to be monitored. A sensor may be embedded underground or into / under a road, for example. In another example, a sensor may be embedded into material forming a bridge. Here, monitoring of such environments may be effectively realized to an extent that has not even been possible before. Less robust sensors could not be embedded into such materials, e.g. concrete. The sensor may thus be configured to be at least partially embedded into a material, said material being part of or coupled to a structure the vibrations of which are to be detected with said sensor.

[0080] The described sensor may accurately measure weak signals within a frequency range of 0.2 - 96 kHz, which may be a wider frequency range than in the prior art. Prior art sensors are tuned for narrower frequency ranges, such as 10 - 200 Hz or 5 - 2000 Hz.

[0081] In some embodiments, the effects of impedance fluctuations and outside electromagnetic disturbances may be significantly reduced or essentially eliminated, which allows use of the sensor in demanding environments. Due to high sensitivity and protection from outside disturbances, the piezoelectric sensor may successfully be used to detect vibrations from a complex entity without having the sensors placed in close vicinity of the source. For example, vibrations from a steel frame of complex machinery at distance of several meters may be measured.

[0082] Tests have been carried out for instance by measuring vibrations from the cylinder block of main diesel engines onboard a cruise ship. During these tests, vibration characteristics of all 16 cylinders of Caterpillar MaK 16 M 46 DF engines was monitored in six dimensions by one invented sensor, which was fixed to the engine block, even from a distance of 5 meters.

[0083] The piezoelectric sensor does not need an outside power source because piezocrystals of the sensor sufficiently convert the mechanical vibrations into electricity. This may eliminate disturbances caused by outside electricity supply systems and increases sensitivity especially in low frequency signals.

[0084] The piezoelectric sensor may be light in weight and small, which allows its use in weight and volume sensitive applications such as for instance in ships and boats, aviation, and underwater vehicles, as well as in offshore and space technology applications.

[0085] The simple and robust design of the described sensor may make it easy to install, operate and repair, ensuring low lifetime costs.

[0086] A piezoelectric sensor may additionally comprise at least a third piezoelectric helical portion disposed between the core portion and the first piezoelectric helical portion, wherein the third piezoelectric helical portion is helically wound about the core portion to exhibit a helical axis essentially codirectional with the longitudinal axis of the core portion, wherein the third piezoelectric helical portion essentially corresponds to a third helix comprising a third helix angle and first handedness. The third helix angle of the third helical portion is advantageously smaller than the first helix angle and second helix angle. The third helix angle may be for example 1-25 degrees, such as 3-15 degrees or 5-10 degrees.

[0087] Addition of piezoelectric crystals to the sensor may enhance the sensitivity of the sensor. This is advantageously done by adding the number of piezoelectric helical portions used. Especially sensitivity to longitudinal vibrations may be enhanced. A handedness of an additional piezoelectric helical portion should be opposite to the handedness of a previous piezoelectric helical portion.

[0088] In case of three piezoribbons or piezoelectric helical portions, the inner ribbon (third helical portion) may be helically wound around the inner conducting portion at a very small longitudinal angle, and the two outer ribbons (first and second helical portions) are wound in opposite directions at a higher angle. This arrangement effectively and economically generates signals of the total vibration environment in six vibration dimensions because their signal generating polarized piezo crystals are arranged in the most advantageous way.

[0089] Sensitivity of the piezoelectric sensor may be enhanced using helical portions or piezoribbons as separate layers on top of each other not only extending the piezoelement material thickness.

[0090] The helical piezoelectric portions may comprise polarized piezoelectric film.

[0091] The core portion of the piezoelectric sensor may comprise at least 25 conductor elements. In embodiments with more than 25 strands / conductors in the core portion, the sensor may be more sensitive than prior art sensors due to a larger surface area of strands.

[0092] The conductor elements of the core portion may be made of silver or may comprise silver, preferably as a silver coating. Silver may provide a material with enhanced conductivity. Silver provides a better conducting material than copper, which may be used in the prior art sensors. On the other hand, alternative aluminum strands / conductors in a core portion may be much lighter than copper. Thus, one alternative may be to use e.g. aluminum material that is coated with silver in conductor elements of the core portion. A resistance of the core portion may be about 30 Q I km.

[0093] The first conducting outer portion may comprise copper, for instance a copper braid.

[0094] In one embodiment, the conducting outer portion may comprise a resistance of over 35 Q I km, such as between 35 and 60 Q / km , e.g. 47 Q / km. This may increase the measured vibration signal as compared to prior art.

[0095] The piezoelectric sensor may additionally comprise at least an outermost protective portion, wherein the outermost protective portion constitutes an outer layer of the piezoelectric sensor. The outermost protective portion may optionally comprise a metal or polymer material, such as such as aluminum pipe or PTFE fluoropolymer. The outermost protective layer may be made according to the requirements of the installation site, such as sufficient resistance against abrasion, mechanical impacts, excessive temperatures, and various electromagnetic disturbances.

[0096] In some embodiments, a used piezoelectric sensor may additionally comprise at least one further conducting outer portion essentially surrounding the first conducting outer portion. The further conducting outer portion may optionally comprise aluminum. The further conducting outer portion may provide shielding from radiofrequency signals that may cause interference. A thickness of the further conducting outer portion may be e.g. 20-80pm, such as 50 pm.

[0097] The piezoelectric sensor may additionally comprise at least one armoring conductive portion essentially surrounding at least the first conducting outer portion. The piezoelectric sensor may comprise at least one additional protective portion disposed between the armoring conductive portion and the first conductive outer portion or further conductive outer portion, if present, wherein the armoring conductive portion may comprise steel.

[0098] The piezoelectric sensor may be configured / optimized for a specific use case or vibrational environment. For example, sensitivity of the sensor may be prioritized or robustness of the sensor may be prioritized.

[0099] A conversion unit may comprise at least one amplifier unit, at least one analog-to-digital converter unit, and at least one data processing unit, wherein the conversion unit is configured to receive a plurality of first electrical signals as an input, wherein the plurality of first electrical signals comprise at least a plurality of analog electrical signals which are obtained via one piezoelectric sensor, and the conversion unit is configured to deliver at least one second signal as an output, wherein said second signal is a digital signal indicative of the first signals.

[0100] The conversion unit may perform real-time processing of (big) data close to the source of data generated by piezo-electric sensors. This may improve considerably response times and save bandwidth when sending data to more distant users or processing units for further processing of data.

[0101] The conversion unit may enable monitoring and processing weak and dynamic analog vibration signals from mechanical systems, preferably in all six dimensions, generated by piezoelectric sensors.

[0102] The conversion unit may be provided as a hermetically sealed unit, preferably inside one hermetically sealed box or housing. The conversion unit may be protected against mechanical impacts and external noise, vibration, and electromagnetic disturbances, which may solve several prior art problems. The conversion unit 200 may e.g. comprise a metallic housing to shield the conversion unit 200 from electromagnetic disturbances.

[0103] The conversion unit may be placed close to the vibration sources or entities. The conversion unit may be configured to be coupled to one or more piezoelectric sensors. The coupling to the one or more piezoelectric sensors may be an essentially direct coupling, such that the piezoelectric sensor is essentially directly connected to the conversion unit or there may only be a relatively short transfer cable between the conversion unit and the at least one piezoelectric sensor. This may reduce contaminating disturbances from discontinuity between the sensor and the conversion unit.

[0104] When the piezoelectric sensors do not require an external power source, disturbances / contamination from such power sources to the signal may also be avoided.

[0105] The conversion unit may operate within a wide frequency and amplitude range, receiving even weak signals from a six-dimensional vibration environment, over time, for achieving a set of measurement data from one or more piezo-electric sensors.

[0106] The piezo-electric sensors are of an extended form, whereby the measurement is performed along the total area of the sensor for example within 1 Hz - 96 kHz frequency range and at 0 - 192000 samples per second.

[0107] In some embodiments, the conversion unit may be provided on one grounded substrate, e.g. electric component plate. This may resolve the grounding problems in prior art arrangements. The conversion unit may use an advantageous method for converting analog signals into digital signals. Instead of measuring, as in prior art methods, voltage fluctuations across resistance between center and outer electric conduits of a piezoelectric sensor, in embodiments of the present invention, at least charge of the sensor may be measured as an active capacitor. The piezolelectric sensor may be considered as an active capacitor that is charged by the piezoelectric helical portions when mechanical forces affect its piezo crystals. In this situation the sensor converts mechanical vibration energy in six vibrating modes into electric charges using a vast number of polarized piezo crystals. The charge measurement may provide some advantages over voltage measurement. For example, if there is a transfer cable between the conversion unit and sensor, displacement of the sensor could cause a change in measured voltage, while charge measurement is not affected.

[0108] Yet, measurements could include voltage measurement of the sensor in addition to charge measurement. The voltage measurement could provide information that is unavailable through charge measurement, such as enabling localization of an impact along the length of the piezoelectric sensor.

[0109] With the conversion unit, the output may be provided to a remote processing unit and an analog data stream from one or more sensors may be monitored. The vibrations may advantageously be monitored in all six vibration dimensions.

[0110] The amplifier unit may be reprogrammed / adjusted online via a data link to a remote processor to process desired frequency and amplitude ranges, as well as sample rate.

[0111] The conversion unit may be configured to convert the at least one first signal into the at least one second signal as an unfiltered signal. The unfiltered signal may be transmitted to the remote processing unit, optionally after initial data processing occurring at the conversion unit.

[0112] With the present conversion unit, the vibration signals may be more effectively processed, and signals that are unexpectedly high or low may not have to be filtered out but may be analyzed. With the large amount of vibrational data obtained with a piezoelectric sensor according to the invention, unfiltered vibrational data may be processed, leading to more information being obtainable from the vibrational data. A more comprehensive account of the vibrational environment may be obtained.

[0113] The functionality of the conversion unit may open new possibilities when analyzing new and / or unexpected vibration phenomena and locating and / or solving various problems. This may allow enhancing of weak signals and filtering out contaminated unwanted signals.

[0114] The present conversion unit may also provide savings in required electronic components in the conversion unit by reducing the number of SMDs (Surface Mounted Devices).

[0115] The exemplary embodiments presented in this text are not to be interpreted to pose limitations to the applicability of the appended claims. The verb "to comprise" is used in this text as an open limitation that does not exclude the existence of also unrecited features. The features recited in depending claims are mutually freely combinable unless otherwise explicitly stated.

[0116] The novel features which are considered as characteristic of the invention are set forth in particular in the appended claims. The invention itself, however, both as to its construction and its method of operation, together with additional objects and advantages thereof, will be best understood from the following description of specific example embodiments when read in connection with the accompanying drawings.

[0117] BRIEF DESCRIPTION OF THE DRAWINGS

[0118] Next the invention will be described in greater detail with reference to exemplary embodiments in accordance with the accompanying drawings, in which:

[0119] Figure 1 illustrates a flow chart of a method,

[0120] Figure 2 illustrates a flow chart of a further method,

[0121] Figure 3 shows an illustration of data flow in one embodiment of the method,

[0122] Figure 4 illustrates a piezoelectric sensor,

[0123] Figure 5 shows at 5A piezoelectric sensor and at 5B a cross-sectional view of the sensor of Fig. 5A, Figure 6 shows a piezoelectric sensor,

[0124] Figure 7 shows a piezoelectric sensor,

[0125] Figure 8 illustrates an arrangement according to one embodiment of the invention,

[0126] Figure 9 illustrates an arrangement according to one embodiment of the invention,

[0127] Figure 10 illustrates an arrangement according to one embodiment of the invention,

[0128] Figure 11 shows an example of a system according to one embodiment of the invention,

[0129] Figure 11 shows an example of a system according to one embodiment of the invention, and

[0130] Figure 12 shows an example of a system according to one embodiment of the invention.

[0131] DETAILED DESCRIPTION

[0132] Figure 1 shows a flow chart of a method for analyzing the state of at least one entity according to one embodiment of the invention. The method comprises obtaining 002 signals indicative of vibrations occurring in the entity. The method can involve one or a plurality of entities the vibrational state of which are to be analyzed. Analyzing the state of the entity may refer to monitoring the state of the entity. The signals are preferably obtained via at least one piezoelectric sensor.

[0133] The signals may be obtained through a piezoelectric sensor that is coupled to the entity or other vibration source, the vibrations of which are indicative of a state of the entity. The signals may be obtained essentially continuously, such that the state of the entity may be monitored.

[0134] The method then comprises utilizing 004 at least one artificial intelligence model to analyze the obtained signals and determining 006, based on the analysis, if the obtained signals are indicative of regular behavior of the entity. The analyzing 004 may be carried out using one more artificial intelligence or machine learning models or algorithms.

[0135] Advantageously, the artificial intelligence models comprise at least one machine learning algorithm, wherein the machine learning algorithm is trainable by self-supervised learning.

[0136] The one or more artificial intelligence models may be trained with previously obtained signals indicative of vibrations occurring in the entity or the method comprise training or pretraining of the one or more artificial intelligence models with obtained training signals indicative of vibrations occurring in the entity to classify signals and / or sequences of signals as being signals that are indicative of regular behavior of the entity and belonging to a first class of signals or sequences of signals. The training may optionally continue throughout the method or training may be triggered in certain cases, such as when irregular behavior is determined.

[0137] The artificial intelligence model(s) may comprise various types of e.g. neural networks or other types of Al models. Yet, in an advantageous embodiment, the at least one artificial intelligence model may comprise at least one transformer model, which is preferably a performer model.

[0138] The analysis 004 may comprise determining at least one predicted future signal or predicted future sequence of signals indicative of vibrations occurring in the entity based on the obtained signals, obtaining further signals indicative of vibrations occurring in the entity and determining at least one observed signal or observed sequence of signals, and comparing at least one predicted future signal or predicted future sequence of signals to at least one observed signal or observed sequence of signals.

[0139] It may then be determined 006 that at least one observed signal or observed sequence of signals is not indicative of regular behavior of the entity if at least one predicted future signal or predicted future sequence of signals differs from at least one corresponding observed signal or observed sequence of signals by over a threshold amount.

[0140] The method then comprises determining 008 an anomaly in the state of the entity based on predetermined criteria if at least one obtained signal or sequence of signals is not indicative of regular behavior of the entity. An anomaly may be determined 008 by tracking, over a selected amount of time, signals or sequences of signals that are determined 006 to not be indicative of regular behavior of the entity. If the signals or sequences of signals deviate from expected signals or sequences of signal by over a threshold amount during a selected time interval, an anomaly may be determined. Any physical object in motion may be characterized by its own vibration characteristics, such as amplitude, intensity and frequency and their changes and various patterns. For example, a motor running in its regular state will have a specific vibrational pattern. The vibrational pattern while the motor is running may also be different from the vibrational state when the motor is started, while both of these patterns are regular behavior of the motor. Use of vibrational signals obtained at different times relating to different (regular) states of an entity may aid in gaining insights into the characteristics and health of the vibrating entity.

[0141] Figure 2 shows one further method according to an embodiment of the invention. The method may comprise a pretraining phase and a use phase. The training phase may comprise obtaining 010 training signals, which are signals indicative of vibrations occurring in the entity. The training signals may be unlabeled data. The at least one artificial intelligence model may employ at least one machine learning paradigm, which may advantageously comprise self-supervised learning.

[0142] At 012, the at least one artificial intelligence model may predict at least one next or future signal in a sequence of signals based on a sequence of obtained signals. At 014, the at least one artificial intelligence model may compare at least one subsequently obtained training signal to the predicted next signal(s). The predicted signal(s) may be scored according to its accuracy as compared to the subsequently obtained training signal.

[0143] Based on the comparison, the at least one artificial intelligence model may be updated 016 upon need, e.g. if the predicted signal differs from the obtained signal by over a threshold amount. The at least one artificial intelligence model is preferably updated iteratively.

[0144] The one or more artificial intelligence models may thus be pre-trained with data collected from the entities to be monitored and analyzed, and can learn to detect the vibrational patterns of the entity in “normal”, or “expected” operation, discarding noise and randomness and detecting the most information-dense parts of the measurement data (i.e. signals).

[0145] In some embodiments, during training of the one or more artificial intelligence models, secondary vibrations may be induced in the entity by using one or more auxiliary vibrators to produce the secondary vibrations. The auxiliary vibrator may be configured to provide vibrations at a plurality of vibrational frequencies. The secondary vibrations may be induced at a plurality of locations at the entity, and obtained signals indicative of vibrations in the entity may be utilized to obtain information relating to a location on the entity from which the signals originate. The signals indicative of vibrations in the entity that are obtained by utilizing secondary vibrations may be used to train the one or more artificial intelligence models with respect to how vibrations propagate in an entity.

[0146] The one or more artificial intelligence models may comprise at least one transformer model, preferably a performer model, whereby an Al-model originally developed for Natural Language Processing (NLP) has been modified and given the problem of constructing a unique “language” for a monitored entity and learning its syntax and grammar rules. The monitored vibrations, as obtained signals indicative of vibrations, may be considered as a piece of text that the Al model is receiving, and the Al model may predict the likely signals that should be next in a sequence of signals when the entity is in its regular state. The performer model is described in Choromanski, Krzysztof, et al. "Rethinking attention with performers." arXiv preprint arXiv:2009.14794 (2020), the contents of which are incorporated herein by reference.

[0147] Transformers scale quadratically with the number of tokens in the input sequence, being based on a self-attention mechanism. Performers, however, estimate regular softmax full-rank attention with only linear space and time complexity and are based on the Fast Attention Via positive Orthogonal Random features (FAVOR+) mechanism, which approximates the self-attention mechanism using structured random projections.

[0148] The Al model may be configured to create prediction targets based on the time-series input data (signals) that it is obtained. This may eliminate the need for a separate test dataset.

[0149] The method may be applied in connection with any type of entity without having to tailor a model or related system for a specific use case scenario.

[0150] After a pre-training phase, the method may then involve the use phase, which may essentially correspond to the method described in connection with Fig. 1 . Upon determining 008 an anomaly in the state of an entity, for instance by determining that an obtained signal or sequence of signals differs from predicted signal or sequence of signals according to selected criteria, the training phase may be initiated for at least a selected amount of time. This may enable the model to be updated to learn characteristics of a new state of the entity which is still to be regarded as being indicative of regular behavior of the entity.

[0151] These new states could e.g. be related to changes either in the load or rotations-per-minute in the entity or related arrangement. Models may then be created for each distinct phase relating to an entity or its use, such as start-up and shut-down cycles and commonly encountered load I rpm combinations. If the load and rpm values can be delivered in real-time during the measurement, the at least one Al model can be modified to take these into account directly, eliminating the need for separate models.

[0152] The detection and learning of new states that are to be regarded as regular behavior of the entity by triggering a training phase during the use phase may avoid falsely flagging an anomaly where the entity is exhibiting regular behavior, but has not been exhibited in the pretraining phase.

[0153] An efficiency of the analysis in use phase or an efficiency of the at least one model or the efficiency of the pretraining may be evaluated based on determining at least one difference between at least one predicted future signal or predicted future sequence of signals and at least one corresponding signal or sequence of signals known to be regular behavior of the entity, e.g. through obtained training signals. A plurality of predicted future / next signals or predicted future / next sequences of signals may be determined comprising an order of certainty, and an average difference between a selected number of predicted future signals and observed (training) signals or a difference between a selected number of predicted future sequences of signals and observed (training) sequences of signals may be determined.

[0154] The evaluation of efficiency of the at least one Al model may be a modified version of top-k metric that observes the n top suggestions / predictions the model gives and measures goodness / efficiency based on the average distance between the model’s output and these suggestions. This may differ from the original metric by measuring differences in arbitrary dimensions and looking at a user-defined number of top suggestions, instead of only the best suggestion.

[0155] For training and calibration purposes, the at least one Al model may determine a distribution of its mean-squared-errors (MSEs) relating to predicted future signals in order to enable efficient evaluation between “normal” and “anomalous” datasets. This may directly inform the decisionmaking regarding how well the model is functioning, and how to define anomalous behavior during use phase.

[0156] Figure 3 shows a schematic illustration of data flow in one embodiment of the method. Signals may be obtained from at least one sensor 018. The signals may then be transmitted to a Digital Signal Processing (DSP) module 020. At the DSP 020, outliers and corrupted measurements may be removed before computing Fourier transforms to present the signals in a time-frequency format for the Al models.

[0157] The signals may be obtained at a vector-quantized variational auto-encoder (VQ-VAE) module 022. VQ-VAE models are described in Aaron van den Oord, Oriol Vinyals, and Koray Kavukcuoglu. “Neural discrete representation learning” arXiv preprint arXiv: 1711 .00937 (2017), the contents of which are incorporated herein by reference.

[0158] The VQ-VAE module 022 may comprise an encoder 024, discretization bottleneck 026, and encoder 028. The discretization bottleneck 026 may take as input an output from the encoder 024 and provide discrete latent variables by a nearest neighbor look-up using a shared embedding space. The discrete latent variables may then be passed to the decoder 028. As the VQ-VAE model may make effective use of the latent space, features that span many dimensions in data space can be efficiently modelled.

[0159] The performance of the VQ-VAE module 022 may depend on whether the module is used in training phase or use phase. In training (or pretraining) phase, a full VQ-VAE model may be used to ensure the encoder 024 learns to accurately compress the incoming data into latent space. Here, the training data (training signals) are compressed, then decoded from the latent representation and finally reconstructed to be as close to the original data as possible. The output from the VQ-VAE module 022 may then be passed to a performer module 030. In use phase, the decoding step may be omitted and the output from the encoded latent or bottleneck layer may be directly fed to the performer module 030, as this contains the representation the prediction of which is of interest via the performer module 030. The performer module 030 may then provide an output that may be utilized in determining a future predicted signal.

[0160] The VQ-VAE module is utilized to receive data from the DSP module and compress the data into high-quality, low-dimensional representation of the vast amount of data to be processed. This is advantageous, as using uncompressed data may be too complex and resource-intensive for efficient and economical processing in the steps performed by e.g. a performer module.

[0161] In advantageous embodiments of the invention, the encoding space may be mapped into vocabulary tokens in such a way that indices near each other in the encoding space are also close to each other in the vocabulary. This may ensure the difference and thus the error between similar tokens stays small.

[0162] To seamlessly connect the VQ-VAE module 022 and the performer module 030 (or models associated with said modules), the data being transmitted between the modules may be formatted in a selected manner. The formatting may comprise formatting the shape of any related data point and / or ensuring that integrity of the data remains intact.

[0163] The VQ-VAE module 022 may be configured to handle input data that is split in arbitrary ways or not split at all. “Splitting” the data may refer to choosing lower and / or upper bounds for a frequency spectrum of vibrational signals that are of interest in the method, whereby the full spectrum does not need to be used if it is not necessary.

[0164] Figures 4-7 show embodiments of a piezoelectric sensor that may be used to obtain the signals indicative of vibrations occurring in an entity.

[0165] Figure 4 shows at least a portion of a piezoelectric sensor 100 that may be utilized in obtaining signals indicative of vibrations in an entity. It should be noted that the figures are not drawn to scale and are schematic illustrations. The sensor 100 may comprise at least a conducting core portion 102. The core portion 102 (and the sensor as a whole) may have a longitudinal axis that is substantially larger than a cross-sectional axis of the core portion, i.e. the sensor 100 may exhibit a cable-like shape. The core portion 102 may exhibit a preferably essentially round cross-sectional axis.

[0166] The core portion may comprise a plurality, preferably over 25, elongated conductor elements that may comprise silver, preferably at least a silver coating. The core portion may comprise a radius of e.g. 10 - 2000 micrometers.

[0167] The piezoelectric sensor 100 additionally comprises at least two helical piezoelectric portions. The helical piezoelectric portions comprise at least a first piezoelectric helical portion 104 and a second piezoelectric helical portion 106. The first piezoelectric helical portion 104 is helically wound about the core portion 102 to exhibit a helical axis essentially codirectional with the longitudinal axis of the core portion, wherein the first piezoelectric helical portion essentially corresponds to a first helix comprising a first helix angle and first handedness.

[0168] The at least second piezoelectric helical portion 106 is helically wound about the core portion 102 and the first piezoelectric helical portion 104 to exhibit a helical axis essentially codirectional with the longitudinal axis of the core portion, wherein the second piezoelectric helical portion essentially corresponds to a second helix comprising a second helix angle and second handedness. The second helix angle is preferably essentially equivalent to the first helix angle and is about 45 degrees.

[0169] The piezoelectric helical portions 104, 106 may comprise a polarized piezoelectric film material comprising piezoelectric crystals. The material of the different piezoelectric helical portions may be the same or may be different. The piezoelectric helical portions may comprise for instance polyvinylidene fluoride (PVDF).

[0170] The piezoelectrical helical portions 104, 106 may additionally comprise a silver coating.

[0171] The piezoelectric sensor 100 additionally comprises at least a first conducting outer portion 108 essentially surrounding the core portion 102, the first piezoelectric helical portion 104, and the second piezoelectric helical portion 106. The first conducting outer portion 108 may comprise a copper braid. The piezoelectric sensor 100 may additionally comprise an outermost protective portion 110, wherein the outermost protective portion 100 constitutes an outer layer of the piezoelectric sensor 100. The outermost protective portion 110 may comprise a metal or polymer material and may be tailored to a specific use case or environment.

[0172] The stiffness of outermost protective portion may be increased to increase the sensitivity of the sensor 100.

[0173] A sensor 100 may additionally comprise further portions that may be e.g. used for different measurement purposes.

[0174] A diameter of the sensor 100 may be a few millimeters, e.g. between 1 and 5 mm, such as about 2.7 mm.

[0175] The sensor 100 may be attached / fixed to the vibration source, i.e. the entity to be monitored, on a vibrating surface using various methods. For temporary and testing purposes a preferred method may be the use of magnetic fixing elements. For permanent installations a preferred method may be mechanical fixing or use of adhesive.

[0176] To enhance the sensitivity of the sensor 100, the sensor 100 can be integrated with graphene material components. The sensor 100 may further be provided with a heating element to allow low temperature use, and / or the sensor 100 may be provided with optical cables for communication and signal analysis of the environment.

[0177] The sensor 100 is advantageously placed against the vibrating surface so that the sensor 100 may detect vibrations in all six dimensions. The sensor may be fixed to the surface of an entity the vibrations of which are to be detected. Alternatively, the sensor 100 may be encompassed in the entity. For example, the entity may have the sensor 100 integrated within the entity during production.

[0178] Figure 5 shows at 5A piezoelectric sensor 100 according to one embodiment of the invention and at 5B a cross-sectional view of the sensor of Fig. 5A. The sensor 100 comprises a core portion 102 and first and second piezoelectric helical portions 104 and 106 as in the embodiment of Fig. 4. The sensor of Fig. 5 additionally comprises a third piezoelectric helical portion 112 disposed between the core portion 102 and the first piezoelectric helical portion 104. The third piezoelectric helical portion 112 may be helically wound about the core portion 102 to exhibit a helical axis essentially codirectional with the longitudinal axis of the core portion, where the third piezoelectric helical portion 112 essentially corresponds to a third helix comprising a third helix angle and first. The handedness of the third piezoelectric helical portion is opposite to the handedness of a previous piezoelectric helical portion, i.e. the second piezoelectric helical portion 106. The third helix angle may essentially correspond to or approach zero helix angle or at least be smaller than the first and second helix angle. The third piezoelectric helical portion 112 may be provided essentially as a layer portion disposed about the core portion 102. The third piezoelectric helical portion 112 may be tightly wound about the core portion 102. The sensor 100 may then further comprise a first conducting outer portion 108 and outermost protective portion 110.

[0179] Figure 6 shows a piezoelectric sensor according to a further embodiment of the invention. The sensor of Fig. 6 is otherwise similar to that of the sensor 100 of Fig. 5, but an additional protective portion 116 is provided as disposed the first conductive outer portion 108. An armoring conductive portion 114 is provided as disposed over the additional protective portion 116. An outermost protective portion 110 is provided over the armoring conductive portion 114.

[0180] Figure 6 shows a piezoelectric sensor according to one more embodiment of the invention. The sensor 100 comprises a core portion 102 and first, second, and third piezoelectric helical portions 104, 106, 112. A first conducting outer portion 108 is disposed over the third piezoelectric helical portion 112. A further conducting outer portion 118 is provided as disposed over the first conducting outer portion 108. An outermost protective portion 110 is provided over the further conducting outer portion 118.

[0181] Figures 8-10 show embodiments of an arrangement 300 according to embodiments of the invention. An arrangement 300 may comprise at least one conversion unit 200 configured to receive a plurality of first signals indicative of vibrations occurring in an entity, wherein the plurality of first electrical signals comprise at least a plurality of analog electrical signals which are obtained via at least one piezoelectric sensor 100. The conversion unit 200 is configured to deliver at least one second signal as an output, wherein said second signal is a digital signal indicative of the first signals.

[0182] The arrangement 300 may additionally comprise at least one remote processing unit 002, wherein the remote processing unit is configured to receive the second signals. The remote processing unit 002 may refer to a processor that is accessible via internet or other communication means. The remote processing unit 002 may refer to one processor of at least one virtual processor comprised in a plurality of locations which may be configured to execute at least a portion of the procedures presented herein.

[0183] At least one of the conversion unit 200 or the remote processing unit 002 is further configured to

[0184] - utilize at least one artificial intelligence model to analyze the obtained first and / or second signals,

[0185] - determine, based on the analysis, if the signals are indicative of regular behavior of the entity, and

[0186] - determine an anomaly in the state of the entity based on predetermined criteria if at least one obtained signal or sequence of signals is not indicative of regular behavior of the entity.

[0187] The previously described functionalities, models, and modules may thus be implemented in at least one of the conversion unit or remote processing unit.

[0188] A remote processing unit 002 may serve a plurality of conversion units 200, which may each receive first signals from one or more respective piezoelectric sensors.

[0189] A plurality of conversion units 200 that are capable of communicating with each other may also form a network, e.g. a plurality of conversion units 200 residing in a building. The network of conversion units 200 may form an internal network that is separate from e.g. a main wireless network. The internal network may be configured to remain operational even if the main network is not available. The internal network may be used for data processing and communication and may be utilized in providing notifications regardless of a state of the main network.

[0190] Figure 8 schematically illustrates a conversion unit 200 and remote processing unit 002. The conversion unit 200 may comprise at least one amplifier unit 202, at least one analog-to-digital converter (ADC) unit 204, and at least one data processing unit 206. The conversion unit may be configured to receive a plurality of first electrical signals as an input, wherein the plurality of first electrical signals comprise at least a plurality of analog electrical signals which are obtained via one piezoelectric sensor, and the conversion unit 200 may be configured to deliver at least one second signal as an output, wherein said second signal is a digital signal indicative of the first signals.

[0191] All components of the conversion unit 200, or at least the amplifier unit 202, analog-to-digital converter (ADC) unit 204, and data processing unit 206 may be provided on one common grounded substrate such as circuit board.

[0192] In further embodiments, for example at least the amplifier unit 202 and ADC unit 204 may be provided separately (at least on a separate substrate) from the processing unit 206.

[0193] The conversion unit 200 (or at least its components that are advantageously provided on a common substrate) may be provided as a single entity inside a hermetically sealed housing.

[0194] The conversion unit 200 may be an edge type computing unit that is placed close to the source of the first signal, i.e. piezoelectric sensor. The first signals are received as analog signals at the amplifier unit 202. The amplifier unit 202 may be a charge amplifier comprising an operational amplifier and a negative feedback capacitor. The amplifier unit 202 may convert the charge output from the piezoelectric sensor (i.e. the first signals) into a voltage signal that is indicative of the charge of the sensors and thus the vibration that has occurred, as the charge of the piezoelectric sensor is directly proportional to outside force applied on the sensor.

[0195] The signal from the amplifier unit 202 is then delivered to the ADC unit 204. The ADC unit 204 converts the analog voltage signal to a digital voltage signal. The ADC unit 204 may be a relatively simple ADC unit, for instance such that is regularly used in connection with HIFI audio systems.

[0196] The digital voltage signal is then delivered to the processing unit 206. The processing unit 206 may for instance perform minimal processing or preprocessing of the digital voltage signal, such as filtering of the signal. A data link to the remote processing unit 002 is provided for delivering the second signal from the conversion unit 200 as a digital output signal. The data link may be enabled through an applicable communication medium such as the internet. For communication, the arrangement 300 may comprise one or more communication interfaces, which may herein refer to e.g. wireless and / or wired transmitters, receivers, or transceivers operable in a target communication infrastructure such as wired or wireless network. The communication interface may comply with a selected WLAN (wireless local area network), LAN such as Ethernet, and / or cellular standard.

[0197] The conversion unit 200 may be configured to receive a plurality of first electrical signals as an input, wherein the plurality of first electrical signals comprise at least a plurality of analog electrical signals which are obtained via one piezoelectric sensor, whereby the conversion unit may be configured to deliver at least one second signal as an output, wherein the second signal is a digital signal indicative of the first signals. The second signal is delivered to the remote processing unit 002 that is external to the conversion unit 200.

[0198] Figures 9-10 show arrangements 300 with further embodiments of conversion units 200. In the conversion unit 200 of Fig. 9, the ADC unit 204 is associated with a field-programmable gate array (FPGA) unit 208 and the processing unit 206 is associated with an extended memory unit 210.

[0199] In the conversion unit of Fig. 10, the conversion unit 200 comprises a plurality of ADC subunits 212 which comprise amplifier units 202 and ADC units 204, where the ADC units are associated with FPGA units 208. Each ADC subunit may receive input signals from different sources. A computing unit 214 may comprise at least the processing unit 206, optionally an FPGA unit 208 and extended memory unit 210. As may be appreciated by the skilled person, the conversion unit 200 may comprise one or more functionally connected units for data processing, such as one or more FPGAs, microprocessors, microcontroller, and / or signal processors, for example.

[0200] Figure 11 schematically illustrates one embodiment of a system 400. The system 400 comprises at least one conversion unit 200, at least one remote processing unit 002, and at least one piezoelectric sensor 100. The construction and functioning of the conversion unit 200, remote processing unit 002, and piezoelectric sensor 100 may be as described herein elsewhere.

[0201] A system 400 may comprise at least one remote processing unit 002 that serves a plurality of conversion units 200, which may each be coupled to one or more piezoelectric sensors 100.

[0202] Figure 12 shows a further schematic illustration of a system 400 and associated functionalities and components. Item 004 refers to functionality and components that may be associated with one or more remote processors 002 which generally reside at a remote location as compared to the entity to be monitored, while item 006 refers to functionality and components that may be associated with a conversion unit 200 and / or which may reside close to the location of the entity to be monitored.

[0203] At the portion of the system 400 locating close to the entity, the system may comprise one or more piezoelectric sensors 100. Components and functionality generally associated with the conversion unit 200 may comprise a process controller 008 such as computing unit 214, a data analysis module 010 (implemented via one or more processing units of the conversion unit 200), orchestrator I text used interface 012 for communication, and one or more data storage units 014.

[0204] At the portion of the system 400 generally associated with the at least one remote processing unit 002, the system may comprise an application programming interface 016, graphical user interface or text user interface 018, one or more artificial intelligence modules 020, and one or more data storage units 022.

[0205] The invention has been explained above with reference to the aforementioned embodiments, and several advantages of the invention have been demonstrated. It is clear that the invention is not only restricted to these embodiments but comprises all possible embodiments within the spirit and scope of the inventive thought and the following patent claims.

[0206] The features recited in dependent claims are mutually freely combinable unless otherwise explicitly stated.

Claims

CLAIMS1 . A computer-implemented method for analyzing the state of at least one entity, the method comprising- obtaining signals indicative of vibrations occurring in the entity,- utilizing at least one artificial intelligence model to analyze the obtained signals,- determining, based on the analysis, if the obtained signals are indicative of regular behavior of the entity, and- determining an anomaly in the state of the entity based on predetermined criteria if at least one obtained signal or sequence of signals is not indicative of regular behavior of the entity.

2. The method of claim 1 , wherein the one or more artificial intelligence models comprise at least one machine learning algorithm, preferably wherein the machine learning algorithm is trainable by self-supervised learning.

3. The method of any claim 2, wherein the one or more artificial intelligence models are trained with previously obtained signals indicative of vibrations occurring in the entity or wherein the method comprises training of the one or more artificial intelligence models with obtained training signals indicative of vibrations occurring in the entity to classify signals and / or sequences of signals as being signals that are indicative of regular behavior of the entity and belonging to a first class of signals or sequences of signals.

4. The method of claim 3, wherein the training signals comprise unlabeled data.

5. The method of any previous claim, wherein the at least one artificial intelligence model comprises at least one autoencoder model, optionally a vector-quantized variational auto-encoder model.

6. The method of any previous claim, wherein the at least one artificial intelligence model comprises at least one transformer model, optionally a performer model.

7. The method of any previous claim, wherein the analysis comprises determining at least one predicted future signal or predicted futuresequence of signals indicative of vibrations occurring in the entity based on the obtained signals, obtaining further signals indicative of vibrations occurring in the entity and determining at least one observed signal or observed sequence of signals, and comparing at least one predicted future signal or predicted future sequence of signals to at least one observed signal or observed sequence of signals.

8. The method of claim 7, comprising determining that at least one observed signal or observed sequence of signals is not indicative of regular behavior of the entity if at least one predicted future signal or predicted future sequence of signals differs from at least one corresponding observed signal or observed sequence of signals by over a threshold amount.

9. The method of claim 7 or 8, wherein an efficiency of the analysis is evaluated based on determining at least one difference between at least one predicted future signal or predicted future sequence of signals and at least one corresponding signal or sequence of signals known to be regular behavior of the entity, preferably wherein a plurality of predicted future signals or predicted future sequences of signals are determined comprising an order of certainty, and an average difference between a selected number of predicted future signals and observed signals or a difference between a selected number of predicted future sequences of signals and observed sequences of signals are determined.

10. The method of any previous claim, wherein the method comprises initiating at least one action, optionally comprising providing a notification, if an anomaly in the state of the entity is determined.

11. The method of any previous claim, wherein the signals are obtained from at least one piezoelectric sensor, preferably wherein the signals comprise information relating to a magnitude of vibrations in six vibrational modes.

12. The method of any previous claim, wherein the signals indicative of vibrations occurring in the entity are obtained and analyzed as essentially unfiltered signals.

13. A computer-implemented method for training at least one artificial intelligence model, the method comprising- obtaining signals indicative of vibrations occurring in an entity,- training of the one or more artificial intelligence models with obtained training signals indicative of vibrations occurring in the entity to classify signals and / or sequences of signals as being signals that are indicative of regular behavior, wherein the at least one artificial intelligence model comprises at least one transformer model, optionally a performer model.

14. An arrangement for analyzing the state of an entity, the arrangement comprising at least one conversion unit configured to receive a plurality of first signals indicative of vibrations occurring in the entity, wherein the plurality of first electrical signals comprise at least a plurality of analog electrical signals which are obtained via at least one piezoelectric sensor, wherein the conversion unit is configured to deliver at least one second signal as an output, wherein said second signal is a digital signal indicative of the first signals, and at least one remote processing unit, wherein the remote processing unit is configured to receive the second signals, wherein the at least one of the conversion unit or remote processing unit is further configured to- utilize at least one artificial intelligence model to analyze the obtained first and / or second signals,- determine, based on the analysis, if the signals are indicative of regular behavior of the entity, and- determine an anomaly in the state of the entity based on predetermined criteria if at least one obtained signal or sequence of signals is not indicative of regular behavior of the entity.

15. The arrangement of claim 14, wherein the arrangement is configured to utilize a plurality of first and / or second signals as training data for the at least one artificial intelligence model to determine a first class of signals and / or signal sequences which are indicative of regular behavior of the entity.

16. The arrangement of claim 14 or 15, wherein the arrangement is configured to transmit the second signals to the remote processing unit wirelessly.

17. A system for analyzing the state of an entity, the system comprising at least one arrangement according to claim 14 and at least one piezoelectric sensor configured to provide signals indicative of a magnitude of vibrations in six vibrational modes, wherein the piezoelectric sensor preferably is an elongated cable-like sensor.

Images