System and computer-implemented method for detecting medical device errors by analyzing acoustic signals generated by components of medical device

Abstract

A system comprising a sensor element arranged to detect a noise spectrum or a vibration spectrum of a medical device to be supervised, and a supervising element, where the sensor element and the supervising element can communicate with each other and thereby enable the provision of data corresponding to an acoustic signal, the system further includes a localized or distributed detection engine that analyzes data corresponding to the acoustic signal such that a typical fault of each of the plurality of individual components of the medical device can be distinguished, where the system further includes a notification engine that notifies the medical device to communicate with the detection engine. The notification engine provides an indication of a maintenance status of the medical device and / or one or more of a plurality of individual components of the medical device. The disclosure also relates to a computer-implemented method for determining a maintenance state of a medical device.

Description

Technical Field

[0001] The present disclosure relates to systems and computer-implemented methods for detecting medical device errors by analyzing acoustic signals generated by components of the medical device. Background Art

[0002] In the medical field, and in particular in the dialysis field, many medical devices are used for a long time. Some of these machines may be used by individuals in a home environment, while others may be used by one or more individuals in a nursing home or a nursing center.

[0003] These medical devices may be of a life-sustaining nature and may sometimes be used regularly, such as for dialysis treatments. Failure of one or more components in a medical device may result in a life-threatening situation for one or more associated patients.

[0004] While newer equipment is equipped with functionality that enables maintenance status to be determined, either independently or in conjunction with a centralized infrastructure, older equipment typically does not include such functionality.

[0005] However, there are still many old devices in use. This is partly due to the fact that users may use these devices for a long time and the cost of these devices is quite high.

[0006] Therefore, there is a need to provide such functionality also for older medical devices.

[0007] While some of these medical devices may provide upgrades to their respective hardware and firmware, allowing collaboration with other components to share additional information, it should be noted that hardware and firmware upgrades may require a revalidation process due to regulatory requirements for upgraded devices to be used within the medical field.

[0008] However, such a validation process is both expensive and time-consuming. In addition, it is important to note that firmware upgrades usually involve high quality requirements in order to be successful and not render the medical device useless in the event of a malfunction.

[0009] Even so, it should be noted that a firmware upgrade may not provide the measured data. That is, although a firmware upgrade may provide data, such as data related to the speed of the motor, such information may not be sufficient to determine the type of error in the case of a reduced speed. Therefore, more complex changes to the hardware implementation of the medical device will be required. However, such changes are generally not feasible due to the long downtime of the medical device and the significant additional costs in the case of a comprehensive hardware change. Summary of the invention

[0010] The present disclosure proposes a system for detecting errors of a medical device by analyzing acoustic signals generated by components of the medical device, the system comprising:

[0011] a sensor element, which is arranged to detect a noise spectrum or a vibration spectrum of the medical device to be supervised and is arranged at the medical device to be supervised,

[0012] Monitoring components,

[0013] wherein the sensor element and the supervisory element can communicate with each other and thus enable data corresponding to the acoustic signal to be provided,

[0014] wherein the system further comprises a localized or distributed detection engine that analyzes data corresponding to the acoustic signal so that typical failures of each of a plurality of individual components of the medical device are distinguished,

[0015] Wherein the system further comprises a notification engine that provides an indication regarding a maintenance status of the medical device and / or one or more of the plurality of individual components of the medical device.

[0016] According to some embodiments, the present disclosure proposes: preprocessing data corresponding to the acoustic signal to reduce the amount of data corresponding to the acoustic signal.

[0017] According to certain embodiments, the present disclosure provides that the sensor element is remote from the supervisory element.

[0018] According to some embodiments, the present disclosure provides that the detection engine is based on at least one trained neural network.

[0019] In certain implementations, the notification engine provides the maintenance status via a network interface.

[0020] In some embodiments, the detection engine is capable of self-learning such that newly experienced faults of each of the plurality of individual components can be learned for future detection.

[0021] The present disclosure also proposes a computer-implemented method for determining a maintenance status of a medical device, the method comprising the following steps:

[0022] obtaining first acoustic data from acoustic signals generated by the medical device and / or a component of the medical device for use in (unsupervised / supervised) training the first machine learning model and / or the second machine learning model,

[0023] wherein the first machine learning model is configured to generate an operating state of the medical device based on the acoustic data, and

[0024] Wherein the first machine learning model or the second machine learning model is configured to determine a maintenance state based on the operating state and the acoustic data.

[0025] Acquire second acoustic data from acoustic signals generated by the medical device and / or a component of the medical device and generate a current operating state of the medical device and / or a component of the medical device via the first machine learning model.

[0026] in a further step, determining a maintenance status of the medical device and / or a component of the medical device via the first machine learning model and / or the second machine learning model,

[0027] • wherein the maintenance status is determined based at least in part on the second acoustic data and the generated operating status.

[0028] In some embodiments, the present disclosure provides that the computer-implemented method further comprises the step of providing the maintenance status via a network interface.

[0029] Other targeted implementations shall be subject to the description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Hereinafter, we will further describe the present invention with reference to the accompanying drawings.

[0031] Figure 1 shows a schematic overview of elements that may be involved in some embodiments,

[0032] Figure 2 A flowchart schematically illustrates steps that may be involved during training in certain embodiments,

[0033] Figure 3 A flowchart schematically illustrates steps that may be involved during operation of certain embodiments,

[0034] Figure 4 A schematic overview of the logic elements involved in some embodiments is shown. DETAILED DESCRIPTION

[0035] In the following we will refer to a medical device MD. A medical device may be any type of medical device for therapeutically treating a human or animal body. In particular, a medical device MD may be a device for removing certain components of a body fluid, such as blood. The medical device may be a therapeutic device, in particular a medical device for blood purification, such as a medical device for renal replacement therapy or for liver support, including medical devices suitable for one or more of hemodialysis, hemodiafiltration, peritoneal dialysis, hemoultrafiltration, etc. If not explicitly marked as required, any such reference should be understood as non-restrictive.

[0036] In addition, in the following, data transmission from one component to another component may be described. Data transmission can be achieved through any type of interface, including wired and wireless transmission. Therefore, any type of 1-wire, SPI, USB, RS-232, IP, optical or any other appropriate transmission scheme can be used as a means for wired transmission of data. Therefore, any type of Bluetooth, DECT, WLAN, ZigBee, mobile communication such as LTE-LTE-A and subsequent protocols can be used as a means for wireless transmission of data. If not explicitly marked as required, any such mention should be understood as non-restrictive. It should be noted that any such interface can be not only unidirectional, but also bidirectional, and can allow the update of firmware / software / computer-implemented methods, which itself will not change the firmware of the medical device MD, but can improve the analysis / function due to the learning process.

[0037] Furthermore, although not explained in detail below, it is assumed that data originating from one specific sensor element S can be distinguished from data from other sensor elements so that the data of a specific sensor element S can be evaluated per sensor element. For this purpose, the data transmission can be such that a unique identifier is used during the transmission or by associating a unique identifier with a specific data set to be transmitted.

[0038] According to the implementation method, see Figure 1 , providing a method for analyzing components C1, C2, C3, ... C of the medical device MD N A system 1 for generating an acoustic signal to detect errors of a medical device.

[0039] Components C1, C2, C3, ... C N It can be any type of component that can create a unique sound or produce a unique vibration when operated. Components C1, C2, C3, ... C N Preferably, they are components of the medical device MD. For the purpose of understanding, it should be noted that the components C1, C2, C3, ... C N The actual number may vary depending on the type of medical device MD.

[0040] The system 1 further comprises a sensor element S which is arranged to detect a noise spectrum or a vibration spectrum of the medical device MD to be supervised and which is arranged at the medical device MD to be supervised.

[0041] like Figure 4As shown, the sensor element S can be any type of vibration detection sensor and / or sound detection sensor SEN. For example, in order to provide more reliable measurement results, the sensor element S can be provided with a form-fitting housing so that the sensor element can be arranged at a pre-known position relative to a specific type of medical device MD, so that the components C1, C2, C3, ... C N Capable of generating reproducible sounds / vibrations during operation. It is also conceivable that the sensor element S can be fixed to the medical device MD, for example, by means of an (auxiliary) port arranged at a specific type of medical device MD. Assuming that the medical device MD is equipped with an RS-232 interface, it is conceivable that the sensor element can be attached to the RS-232 interface via a corresponding plug or socket. Although data exchange via the interface is not necessarily required during operation, data exchange can be used to collect other data to be transmitted. The sensor element S may also include: a memory MEM, such as a RAM; and a processing entity, such as a microcontroller or microprocessor, such as a RaspberryPI, ESP8266, ARM or DSP such as a DSP based on TMS 320; and some type of interface I / O for providing a wired or wireless connection to the supervisory element SE or the training device TD described in detail below. Therefore, the sensor element S can not only perform the acquisition of sensor data, but also store such data in the memory MEM directly or after processing. Similarly, processed or unprocessed data can also be provided to other devices via an interface. Although not described in detail here, the sensor element S may also pre-process and / or pre-evaluate the data before passing the (raw or processed) data to some other device.

[0042] like Figure 1 As shown, the system 1 further comprises a supervisory element SE. The supervisory element SE may be a workstation, such as a standard personal computer, a server or any other computing unit. The workstation may be started by software.

[0043] The sensor element S and the supervision element SE can communicate with each other, thereby enabling the provision of data corresponding to the acoustic signal. The data corresponding to the acoustic signal may also include vibrations detected by the vibration sensor. Furthermore, the data may be subjected to analog or digital filtering or any type of other processing.

[0044] Furthermore, the system 1 comprises localized or distributed detection engines DE1, DE2. That is, while the detection engine DE2 may be mainly present on the supervisory element SE, it is also possible that a part of the detection engine DE1 allowing some pre-processing may be located on the sensor element S. The detection engine is arranged to analyze data corresponding to the acoustic signal so that a plurality of individual components C1, C2, C3, ... C of the medical device MDN Typical failures of each individual component can be distinguished.

[0045] The system 1 further comprises a notification engine NE which provides information about the medical device and / or a plurality of individual components C1, C2, C3, ... C of the medical device MD. N An indication of the maintenance status of one or more individual components in a system.

[0046] For example, it is foreseeable that the detection engine will evaluate a specific operating condition indicating normal operation. For example, some medical devices may be used for different types of treatments, each of which causes unique sound and / or vibration characteristics. Therefore, analyzing the data corresponding to the acoustic signal can reveal a certain type of normal operation. Information about such operation can be provided to some databases DB storing, for example, operating data for determining maintenance intervals, and / or one or more remote devices RD1, RD2 (e.g., a handheld device such as a mobile phone with a dedicated application (app) or a web browser or enabled to receive push information) or a general network interface (e.g., Internet / Intranet access), and / or a training device TD via a notification engine NE, such as a WWW interface, a REST interface, a push notification. Specifically, the detection of the operating state can be performed by a detection engine DE1 located in the sensor element S. Similarly, a more complex analysis can be performed by a more powerful detection engine DE2, which provides further insight into whether a specific component is operating under normal conditions, and which error may be about to occur if it is not operating under normal conditions.

[0047] The database DB also enables post-training based on actually experienced faults that were not previously detected, thereby enabling detection to be enhanced.

[0048] For example, such analysis may be performed by a fully trained neural network, wherein the neural network may be part of a computer-implemented method for determining the maintenance status of a medical device.

[0049] A computer-implemented method for determining a maintenance status of a medical device may include certain steps for training a machine learning model.

[0050] For example, in Figure 2 In the example shown in FIG. 1 , the training comprises step S50 : obtaining from the medical device MD and / or the components C1 , C2 , C3 , . . . C N The first acoustic data is obtained from the generated acoustic / vibration signal for unsupervised / supervised training of the first machine learning model and / or the second machine learning model.

[0051] As described in detail above, the first machine learning model can be associated with the medical device or its individual components C1, C2, C3, ... C N The second machine learning model can be associated with any such components C1, C2, C3, ... C N of impending failure.

[0052] To this end, it is conceivable that acoustic / vibration signals are acquired during a training cycle, and corresponding indications relative to operating states are attributed to supervised training. This can also be used when repairing medical devices due to maintenance. In this regard, the raw data as well as the processed data can be forwarded to the training device TD. Upon detection of one or more components C1, C2, C3, ... C N When a specific fault is detected, the service technician can indicate the nature of the error to the training device TD. Once the respective first network and / or second network has been trained, it can be distributed to one or more detection engines DE1, DE2 for use in an operational setting.

[0053] For example, the first machine learning model may be configured to generate an operating state of the medical device MD based on acoustic / vibration data.

[0054] Depending on the type of machine learning, the first machine learning model and / or the second machine learning model may be configured to determine a maintenance status based on the operating status and the acoustic / vibration data.

[0055] That is, although different networks are conceivable, it may be sufficient to provide only a single network having both functionalities.

[0056] When training neural networks, they can Figure 3 The operating conditions shown are used.

[0057] In a first step S100, a medical device MD and / or components C1, C2, C3, ... C N Second acoustic / vibration data is obtained from the generated acoustic / vibration signal, and the current operating state of the medical device and / or components of the medical device is generated via the first machine learning model.

[0058] In another step S200, the medical device MD and / or the components C1, C2, C3, ... C of the medical device MD are determined via the first machine learning model and / or the second machine learning model. N In the maintenance state of step S100, this step can be started and / or ended before, after, or at the same time as step S100.

[0059] According to some embodiments, for example, in step S300, the notification engine NE provides the maintenance status via a network interface.

[0060] A maintenance status is determined based at least in part on the second acoustic data and the generated operating status.

[0061] The data corresponding to the acoustic / vibration signal may be pre-processed to reduce the amount of data corresponding to the acoustic / vibration signal.

[0062] Because the system 1 enables the sensor elements S to be remote from the supervisory element SE, multiple sensor elements S can provide data to a single supervisory element SE. Assume, for example, that a healthcare provider has medical equipment in different locations, such as in a nursing home, at home, in a hospital. It is now possible to monitor the operating status of multiple sensor elements S, and in particular the maintenance status of certain components thereof, by a single supervisory element SE. This enables, for example, the deployment of a technician when certain faults are about to occur.

[0063] For example, suppose that the motor C1 generates a specific noise / vibration due to being obstructed by dust, such as causing a reduction in rotation, or a whistling fan C2, a whistling pipe in the rotary pump C4, etc. Such suspicious motor noise / vibration may indicate that a serious condition will occur soon thereafter. In addition, the operating data may also indicate that a specific component should be replaced after a specific period of operation. Although this situation may not have been noticed before before the actual failure occurred, it will now be easier to monitor the operation and the impending or actual errors, so that a quick response can be made to resolve the problem before an undesirable condition can develop or before the medical device unexpectedly shuts down and the patient cannot receive the prescribed medical treatment.

[0064] For example, a fault of compressor C1 can be detected by an XGBoost model trained based on various features extracted from vibration / sound data. The most important features used by the model may be (but not limited to) root mean square, spectrum skew, and (larger) amplitude of low-frequency components, i.e., components in the range of 40 Hz to 100 Hz, for example. The frequency components can be obtained from the vibration / sound data by (short-time) Fourier transform. The detection of grinding of motor C1 can also be performed by another or the same XGBoost model trained based on various features extracted from vibration / sound data. Typically, motor grinding produces (larger) amplitude of high-frequency components, i.e., components in the range of 800 Hz to 1200 Hz, for example. Similarly, frequency components can be obtained from vibration / sound data by (short-time) Fourier transform. Similarly, motor C1 slippage is usually indicated by a whistling noise, which usually occurs intermittently and only lasts for a short duration (e.g., about 0.2 seconds). Various mid-range frequencies can help identify these patterns. Likewise, detection can be performed by another or the same XGBoost model, for example, using bidirectional lead and lagged feature inputs. While these faults represent more common faults, others, which may be batch-related, are less common. However, the use of a bidirectional long short-term memory neural network that learns normal operating patterns via self-supervised learning enables the labeling of abnormal sounds.

[0065] For this reason, XGBoost is an open source software library that implements an optimized distributed gradient boosting machine learning algorithm under the gradient boosting framework. XGBoost is the abbreviation of Extreme Gradient Boosting and is a scalable distributed gradient boosting decision tree (GBDT) machine learning library. It provides parallel tree boosting and is a leading machine learning library for regression, classification, and sorting problems. It should be noted that the use of XGBoost is exemplary and does not exclude other machine learning models for the purposes of the present disclosure. Specifically, the present disclosure contemplates the use of feature segmentation in self-supervised learning (such as Wave2Vec neural networks, as known in the field of speech recognition). Such a method enables data from healthy components and faulty components to be separated without manual labeling, i.e., unsupervised.

[0066] As indicated above, according to an embodiment, the detection engines DE1 , DE2 may be based on at least one trained neural network.

[0067] Such a neural network can be a convolutional network (e.g., as part of DE1) using, for example, short-time Fourier transform, root mean square and other techniques. In addition, the neural network can be a convolutional network. Training can be performed using a sliding window, for example with respect to a spectrogram (e.g., provided by a short-time Fourier transform). The data can be segmented into segments to create component-specific fingerprints and fault detection models, for example, for a compressor, a motor, or any other component C1, C2, C3, ... C N Specifically, the detection engines DE1 and DE2 are capable of handling concurrent component execution.

[0068] It is also contemplated to use an autoencoder to reconstruct the data. The autoencoder compresses (encodes) the input into a lower dimension (code) and uses it for the reconstruction of new samples (decoding). While healthy components produce lower reconstruction errors, it is noted that unhealthy components will produce higher reconstruction errors after reconstruction. Therefore, a simple threshold on the reconstruction error can indicate an impending failure. A more sophisticated approach can be based on a mean square error histogram, which, for example, in the case of unhealthy components will show a right-skewed error distribution relative to a similar histogram for the same healthy component.

[0069] The maintenance status is determined based at least in part on acoustic data in an acoustic signal emitted from the medical device and / or at least one of the medical device components.

[0070] In one example, the "raw" acoustic / vibration data, i.e., the acoustic / vibration data received from the sensing devices (microphones, vibration sensors) of the maintenance detection system, is first converted into a time-frequency representation. The time-frequency representation then enables the acoustic data to be processed as image data. In this case, the first machine learning model that generates the operating state based on the acoustic data may include a convolutional neural network (CNN) or at least a convolutional neural network layer of a more complex machine learning algorithm. CNNs are typically used for image data due to their high accuracy in classifying images compared to other types of neural networks.

[0071] The medical device and / or the components C1, C2, C3, ... C of the medical device are determined based at least in part on the second acoustic data and the generated operating state. N The maintenance state can be performed with the help of the first machine learning model or the second machine learning model. For example, the first machine learning model may include several modules / layers, wherein these modules / layers process the data "once", i.e., the generated operating state of the medical device / component remains "hidden", wherein the "hidden" operating state is further processed to determine the final maintenance state. In another example, the output of the first machine learning model (e.g., the operating state) is stored and transmitted to the second machine learning model.

[0072] The second machine learning model may include one or more machine learning algorithms to process (e.g., filter and / or classify) the acoustic data and the generated operating state of the medical device. In one aspect, the second machine learning model may include a supervised tree-based algorithm, such as an XGBoost model for extracting features from the acoustic data. For example, the results of the first machine learning model may generate an operating state indicating that the compressor of the PD device is currently active.

[0073] Then, a trained second machine learning model, such as an XGBoost model, can analyze the acoustic data of the compressor by comparing the actual acoustic data of the compressor with the trained acoustic patterns of the faulty compressor. Such a tree-based algorithm can also be used for any other medical device component to detect, for example, motor slippage or motor grinding.

[0074] The second machine learning model may also include any unsupervised / self-supervised machine learning algorithm, wherein the unsupervised / self-supervised machine learning algorithm can detect a faulty component based on any acoustic deviation from a "healthy" acoustic pattern of the medical device or medical device component. In one example, a bidirectional long short-term memory (LSTM) neural network model is employed, wherein abnormal sounds are labeled as unusual, i.e., sounds that would not occur during normal device operation. Other unsupervised / self-supervised machine learning algorithms may employ an autoencoder to determine a reconstruction error, wherein the reconstruction error can be constructed as a measure of a faulty medical device component.

[0075] According to some embodiments, the detection engines DE1 , DE2 are capable of self-learning so that newly experienced faults of each of the plurality of individual components can be learned for future detection.

[0076] The first machine learning model and / or the second machine learning model may be run on the sensor element S and / or a remote computing unit such as an external server (e.g. a supervisory element SE). Some calculations may also be performed on the local computing unit CPU of the sensor element S, for example, calculating acoustic / vibration data that has been short-time Fourier transformed. In the case of the supervisory element SE, the sensor element S may include a network gateway (modem, router) I / O that is capable of communicating with a local area network and / or a wide area network. In this way, calculations that require a large amount of computing resources may be performed quickly on suitable remote hardware.

[0077] The maintenance status may be determined for any medical device MD or medical device component, wherein the maintenance status includes information on the normal operation of the medical device and / or its components, for example, the medical device and its components C1, C2, C3, ... C NWork safely and perform effective and error-free treatment according to a given prescription. The maintenance status can be displayed on any external computing device (via computer, smartphone or tablet) RD1, RD2, for example via a web application, where the computing device communicates with the maintenance detection system using wireless / wired communication methods (Wi-Fi, Ethernet, ...).

[0078] The maintenance status can be depicted as, for example, any kind of data table, list, or graph including information about the normal operation of the medical device and components of the medical device. For example, components that are operating normally can be displayed as "OK", wherein components that are not operating normally can be marked as "NOT OK / NOK". The maintenance status can also include more detailed information, such as raw acoustic data in the acoustic signal of the medical device MD and / or intermediate results from the first machine learning model and / or the second machine learning model. In one example, the final result or intermediate result of any machine learning model can be depicted as a graph or any other type of visualization.

[0079] In addition, the maintenance status may include: the current maintenance status part, that is, the medical device MD and its components C1, C2, C3, ... C N Normal operation in the current examination; and / or a future maintenance status portion, i.e., the medical device and its components are operating normally at a later point in time. For example, the current maintenance status portion of the component may be displayed as "OK", but the future maintenance status portion, such as the predicted maintenance status within four weeks, may display "NOK". The prediction algorithm may include a neural network-based machine learning algorithm, wherein the machine learning algorithm may be included in the first machine learning model and / or the second machine learning model. In this way, faulty components can be detected in a timely manner, thereby avoiding uncontrolled shutdown of the medical device or danger to the patient.

[0080] As used herein with reference to quantities or values, the term "about" means "within ±10%."

[0081] The terms “comprises / comprising,” “includes / including,” “has / having” and their conjugations mean “including but not limited to.”

[0082] The term "consisting of" means "including and limited to."

[0083] The term "consisting essentially of" means that the composition, method, or structure may include additional ingredients, steps, and / or components, but only if such additional ingredients, steps, and / or components do not materially alter the basic and novel characteristics of the claimed composition, method, or structure.

[0084] As used herein, the singular forms "a", "an", and "the" include plural references unless the context clearly dictates otherwise. For example, the term "a compound" or "at least one compound" may include a plurality of compounds, including mixtures thereof.

[0085] In the present application, embodiments may be presented with reference to a range format. It should be understood that the description in a range format is merely for convenience and brevity, and should not be construed as an inflexible limitation on the scope of the present invention. Therefore, the description of a range should be considered to have explicitly disclosed all possible sub-ranges and individual numerical values ​​within the range. For example, a description of a range such as "from 1 to 6" should be considered to have explicitly disclosed sub-ranges such as "from 1 to 3", "from 1 to 4", "from 1 to 5", "from 2 to 4", "from 2 to 6", "from 3 to 6", etc. and individual numbers within the range, such as 1, 2, 3, 4, 5 and 6. This applies no matter how large the range is.

[0086] Whenever a numerical range is indicated herein (e.g., "10-15," "10 to 15," or any pair of numbers associated by other such range indications), the numerical range is meant to include any number (fractional or integer) within the indicated range limits, including the range limits, unless the context clearly dictates otherwise. The phrases "range / variable range / ranges between a first indicated number and a second indicated number" and "range / variable range / ranges from a first indicated number to / until / up to or to a second indicated number" (or other such range indicating terms) are used interchangeably herein and are meant to include the first indicated number and the second indicated number and all fractions and integers therebetween.

[0087] Unless otherwise indicated, the numbers used herein and any numerical ranges based thereon are approximations within reasonable measurement precision and rounding errors as understood by one skilled in the art.

[0088] As used herein, the term "method" refers to manners, means, techniques and procedures for accomplishing a given task, including but not limited to those manners, means, techniques and procedures known to practitioners in the fields of chemistry, pharmacology, biology, biochemistry and medicine or readily developable from known manners, means, techniques and procedures.

[0089] It should be understood that certain features described in the context of different embodiments for the sake of clarity may also be provided in combination in a single embodiment. Conversely, various features of the invention described in the context of a single embodiment for the sake of brevity may also be provided individually, or in any suitable sub-combination, or as appropriate in any other described embodiment of the invention. Certain features described in the context of various embodiments should not be considered essential features of those embodiments unless the embodiment is inoperative without those elements.

[0090] Although the present invention has been described in conjunction with specific embodiments of the present invention, it is apparent that many alternatives, modifications and variations will become apparent to those skilled in the art. Therefore, the present invention is intended to include all such alternatives, modifications and variations that fall within the spirit and broad scope of the appended claims.

Claims

1. A system for detecting an error of a medical device by analyzing an acoustic signal or a vibration signal generated by at least one component of the medical device, the system comprising: a sensor element arranged to detect a noise spectrum or a vibration spectrum of the medical device and arranged at the medical device; a supervisory element, wherein the sensor element and the supervisory element are adapted to communicate with each other such that data corresponding to the acoustic signal can be transmitted between the sensor element and the supervisory element; a localized or distributed detection engine adapted to analyze data corresponding to the acoustic signal so that typical failures of each of a plurality of individual components of the medical device are distinguished; and A notification engine adapted to provide an indication regarding a maintenance status of the medical device and / or one or more of a plurality of individual components of the medical device.

2. The system according to claim 1, wherein: Data corresponding to the acoustic signal or the vibration signal is preprocessed, thereby reducing the amount of data corresponding to the acoustic signal.

3. The system according to claim 1, wherein: The sensor element is remote from the supervisory element.

4. The system according to claim 1, wherein: The detection engine is based on at least one trained neural network.

5. The system according to claim 1, wherein: The notification engine is adapted to provide maintenance status via a network interface.

6. The system according to claim 1, wherein: The detection engine is capable of self-learning so that newly experienced faults of each individual component of the plurality of individual components can be learned for future detection.

7. A computer-implemented method for determining a maintenance status of a medical device and / or a component of the medical device, the method comprising: obtaining first acoustic data or first vibration data from a first acoustic signal or a first vibration signal generated by the medical device and / or a component of the medical device for training a first machine learning model and / or a second machine learning model, wherein the first machine learning model is configured to generate an operating state of the medical device and / or a component of the medical device based on the first acoustic data or the first vibration data, and wherein the first machine learning model or the second machine learning model is configured to determine the maintenance state based on the operating state and the first acoustic data or the first vibration data; Acquire second acoustic data or second vibration data from a second acoustic signal or a second vibration signal generated by the medical device and / or a component of the medical device, and generate a current operating state of the medical device and / or a component of the medical device via the first machine learning model; and determining a maintenance status of the medical device and / or a component of the medical device via the first machine learning model and / or the second machine learning model; Wherein, the maintenance status is determined at least in part based on: (i) the second acoustic data or the second vibration data and (ii) an operating status generated by the first machine learning model.

8. The computer implemented method of claim 7, further comprising providing the maintenance status via a network interface.

9. The computer-implemented method of claim 7, wherein: The training is unsupervised.

10. The computer-implemented method of claim 7, wherein: The first machine learning model is configured to determine the maintenance status based on the operating status and the second acoustic data or the second vibration data, and wherein determining the maintenance status of the medical device and / or a component of the medical device is performed via the first machine learning model.

Images