Method and system for monitoring a device
The method uses digital fingerprints to detect device anomalies, addressing the need for expert intervention in monitoring, enabling proactive maintenance and reducing downtime and costs.
Patent Information
- Application Number
- JP2024531123
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2021-11-24
- Filing Date
- 2022-11-24
- Publication Date
- 2024-11-13
AI Technical Summary
Existing monitoring methods for device operational anomalies require advanced expertise, are cumbersome, and often fail to detect issues in a timely manner, leading to costly downtime due to equipment failures.
A method and system for monitoring device operation using unique identifiers, or 'digital fingerprints', derived from operating parameters, which compare deviations from reference identifiers to detect anomalies without requiring expert intervention, enabling proactive maintenance.
Facilitates early detection of device anomalies, reducing downtime and maintenance costs by providing efficient, less complex, and less intrusive monitoring, allowing for automated alerts and notifications.
Smart Images

Figure 2024542265000001_ABST
Abstract
Description
[Technical field]
[0001] The various aspects and variations relate to methods and systems for monitoring a device. Further, the various aspects and variations relate to a device comprising a system for monitoring its operating state. The various aspects and variations also relate to a non-transitory computer-readable medium having software instructions and a computer program product. [Background technology]
[0002] It is becoming more and more important to monitor the operating conditions of devices. For example, predictive maintenance is widely adopted to detect operational anomalies of devices, machines, or systems by performing data analysis. The goal is to be able to detect possible defects of equipment and processes and correct them before they lead to failures or outages. Failure of a critical machine component can result in hours or days of production downtime, since some parts are often not readily available. For example, machines that are essential to the operation of a brewery, such as filling machines or vacuum pumps, can break down unexpectedly. This can cause a significant amount of downtime in large parts of the production process, since the resources and spare parts inventory required for repairs may not be available.
[0003] If abnormal operation of a device is caught early, the associated costs can be significantly lower. As time progresses, it can become more expensive to repair the device, eventually reaching a peak in costs when the machine is not operational. Early detection symptoms are often unnoticeable without the aid of tools and sensors, so constant vigilance is required to be able to detect such problems before they occur. This often involves performing regular inspections, which can be a complex task, especially when multiple machines or devices work together, such as on a production line.
[0004] Anomalies can be seen as parts of a signal that deviate significantly from the normally observed signal, i.e., the signal when the device is operating normally. Thus, given that the relevant signals are being examined, anomalies can be an effective indicator of incipient equipment malfunction.
[0005] In its simplest form, anomalies are detected by a domain expert (e.g., a trained operator) by visual inspection of sensor measurement data. The expert observes large peaks / dips, large fluctuations, or other large-scale trends and determines whether they are within the expected range of the process. Such an approach requires a contextual understanding of what is being measured and what can be expected from that measurement, based on an understanding of the physical principles behind the machine components. Obviously, observing multiple measurements on a regular basis can be very time-consuming for an expert. In addition, scalability is also limited, as only a selected group of experts can interpret the signals. As a result, problems are often not detected in a timely manner. Summary of the Invention [Means for solving the problem]
[0006] Preferably, there should be a reduction in equipment downtime due to unexpected outages or equipment failures by better indicating when maintenance should be performed in advance, without requiring a high level of expertise to understand the available data.
[0007] Furthermore, existing methods can be tedious and difficult to implement in the field and often require a specialized human operator to precisely adjust or fine-tune one or more operating parameters. It is desirable to improve the monitoring of device functionality in various processes. Furthermore, it is desirable to have such monitoring systems and methods that are less complex, less intrusive, and / or less expensive.
[0008] It would be desirable to provide a method and system that overcomes at least one of the above disadvantages.
[0009] Additionally or alternatively, it may be desirable to provide improved monitoring of the internal operation of the device.
[0010] Additionally or alternatively, it may be preferable to provide improved detection of irregular operation of one or more components of the device.
[0011] Additionally or alternatively, it may be desirable to provide methods that can improve the maintenance of a device or part thereof.
[0012] Additionally or alternatively, it may be desirable to improve the efficiency of monitoring the internal functions of a device or part thereof.
[0013] Thus, a first aspect provides, for execution in an electronic processing or electronic computing device, a method of monitoring the device, the method comprising: receiving a set of data indicative of at least one operating parameter of the device within a predefined time interval; calculating an identifier based on the set of data that uniquely identifies the set of data; and determining deviation data comprising data regarding the occurrence of at least one deviation between the identifier and at least one reference identifier for the at least one operating parameter of the device.
[0014] This provides an advantageous way of detecting when the behavior of a device deviates in some way from normal operation. Anomaly detection is performed by computing a unique identifier for the measurement data and comparing the unique identifier to a reference identifier determined, for example, under verified normal conditions of the device.
[0015] Based on at least one deviation between the identifier and at least one reference identifier for the at least one operating parameter of the device, it can be determined whether the at least one monitored operating parameter of the device provides a desired underlying performance of the device. The method may involve systematically comparing the calculated identifier to one or more associated reference identifiers.
[0016] The identifier may be calculated to provide a nearly unique marker of identification of a set of data. Optionally, the identifier is calculated by an identification unit configured to group identifiers according to their characteristics, thus aiding in matching one identifier fingerprint to one or more reference identifiers. In some embodiments, a database of reference identifiers is used, such that multiple deviations are calculated to find a match. A query identifier that needs to be matched can be compared to a subset of reference identifiers in an existing database. This works in the same way as identifying matching fingerprints in a fingerprint classification system. The identifier may capture a particular pattern in a set of data, including the presence or absence of a particular characteristic pattern in a signal.
[0017] The identifier may be a numerical structure derived from the content signal. The identifier may serve as a statistically unique identifier of a signal that resembles a fingerprint from the content signal, rather than a different registered content signal. The identifier may be calculated based on a transformation rule. In some embodiments, the transformation rule is configured to take into account only some of the identified patterns. In some embodiments, the identified patterns may be further decomposed before calculating the identifier.
[0018] Optionally, the set of data within a predefined time interval is part of a data stream. The calculated identifier may be a compact and unique data stream identifier that includes components that may be derived or extracted from or underlie a monitored signal associated with at least one operating parameter of the device. The calculated identifier may correspond to a form of data fingerprint. The identifier or data fingerprint may be robust to signal modifications such as distortion, noise, etc. The data fingerprint may be efficiently stored in a database and searched to enable signal identification.
[0019] Optionally, the identifier forms a digital fingerprint of the set of data within a predefined time interval, the digital fingerprint representing a unique identification of the work regime associated with the at least one operational parameter.
[0020] A set of data indicative of at least one operating parameter of the device within a predefined time interval may have a unique profile that can be easily captured and compared using the identifier. Similar to a fingerprint, the data print characteristics contained in each successive set of data of the same operating parameter of the device may be nearly unique under normal operating conditions. The identifier allows the identification of the set of data to be recognized by using the characteristics of the data prints in the set of data. Optionally, the monitored signal is pre-processed to obtain a set of data, the pre-processing involving, for example, a feature extraction step in which only relevant data associated with a particular pattern is kept.
[0021] The identifier may depend on databased features within the set of data.
[0022] Optionally, the digital fingerprint is calculated using a statistical model configured to receive as input at least a portion of the set of data at a predefined time interval.
[0023] The statistical model may be tuned to provide an identifier for a particular data set type. Optionally, the identifier is a set of multiple unique feature values extracted from the set of data. The set of data may be obtained by monitoring a signal indicative of at least one operating parameter of the device using a sensor.
[0024] The identifier may be generated using a processor module. In some embodiments, the processor module is integrated into the device, and the device can perform automatic self-diagnosis and communicate the results to other devices or systems. The device may have a controller communicatively coupled to other devices or systems. For example, the device may have a wireless unit that allows wireless communication with remote devices or systems. Wired connections are also envisioned. A combination of wired and wireless communication is also possible.
[0025] The identifier may uniquely correlate, correspond, and / or associate with a particular unique identifying characteristic of the monitored signal associated with the received set of data. In some embodiments, the set of data may relate to one or more properties of the signal at one or more instances in time within a predefined time interval. Based on the set of data, the identifier may be determined to analyze whether the identifier uniquely correlates, corresponds, and / or associates (or does not) with a reference identifier. In some embodiments, the deviation is determined by comparing the identifier with one or more past identifiers employed as reference identifiers. Based on the deviation, a degree of similarity may be determined, and if there is a match, meaning that the set of data correlates, corresponds, and / or associates with the reference data (e.g., related to past historical data), and therefore may confirm that the device's operating parameters provide normal / regular device operating conditions. Identification of the deviation between the identifier and the reference identifier may, for example, enable the remote device or system to provide an output or instruction to the user, allowing the user to initiate maintenance or repair of the device. Many variations of the concept and use of the identifier are within the scope of this disclosure.
[0026] Optionally, the identifier is determined based on a subset of features in the set of data, and one or more robust features are extracted from the set of data, and a robust hash signature is determined from the extracted one or more robust features.
[0027] The transformation rule may involve applying a hashing algorithm. The hashing algorithm may be applied to selected portions of the content signal (e.g., several portions of the content signal) to generate identifiers. In some embodiments, the transformation rule is applied repeatedly to generate a robust set of identifiers (e.g., a hash sequence). This may improve accuracy and reduce the risk of false positives. Content may be identified in a specified subset of the sequence. For example, the sequence may use sub-identifiers from each predefined portion (e.g., 1 / 8th) of the data set and may require a predefined number of sub-identifiers (e.g., 16) to identify the set of data. Using more sub-identifiers may improve accuracy.
[0028] The identifier may be a content identifier extracted from a set of data. Various kinds of identifier resolvers may be used, employing specific rules governing the computation of the identifier.
[0029] In some embodiments, each trusted identifier is a digital hash extracted from its respective set of data, the digital hash being generated following a step of digitally reducing the size of the data set through a digital reduction algorithm.
[0030] In some embodiments, the identifier captures one or more identifying patterns in the data. Based on comparing the calculated identifier with the reference identifier, it can be determined whether at least a portion of a particular identifying pattern in the monitored data set appears in the reference data set. In this way, the monitored data set can be linked to a particular operating mode of the device. Based on the deviation between the identifier and the reference identifier, a similarity measure can be generated.
[0031] Optionally, the at least one deviation between the identifier and the at least one reference identifier is repeatedly calculated for subsequent time intervals.
[0032] Optionally, the deviation is used to generate a match percentage of the calculated identifier and the reference identifier. In response to the match percentage being below a predefined match level, an alert, alarm, and / or notification of abnormal behavior may be generated.
[0033] Optionally, at least one reference identifier is adjusted over time using more recent reference data.
[0034] Optionally, a sliding time window is used to select the reference data from which the reference identifier is determined. In this way, it is advantageously possible to gradually adapt to long-term changes that occur. Such dynamic thresholding may provide significant advantages, since the conditions under which the device operates may change slightly over time. The model can adaptively adjust to learn more from recent data.
[0035] Optionally, as time progresses, the predefined time interval is shifted so that newer data is used to determine the at least one deviation.
[0036] In some embodiments, a plurality of reliable identifiers from the respective plurality of sets of data for successive predefined time intervals are generated. The successive predefined time intervals may have a partial overlap. However, it is also possible that a time delay is provided between the successive predefined time intervals. It is also possible that the time intervals follow each other immediately. The plurality of reliable identifiers may be compared to one or more reference identifiers to determine respective deviations.
[0037] Optionally, the set of data corresponds to a univariate signal associated with a single operational parameter of the device.
[0038] The method may monitor univariate signals to detect anomalies. Anomalous behavior of univariate time series signals may be detected using an identifier (see digital fingerprint). A univariate signal may indicate that a signal indicative of at least one operating parameter is measured only along a single dimension. Interactions between multiple measurements are not taken into account. Each measurement may be analyzed independently to determine a reference identifier, and various statistics may be calculated based on a reference window. This may be considered as a digital fingerprint of normal behavior. The same process may be used to calculate a fingerprint of the data to be tested, i.e., an identifier may be calculated and compared to the reference fingerprint. Deviations between the calculated identifier, calculated based on the monitored data, and the reference identifier, determined under verified normal device operation, are easily interpretable by an expert. This provides the ability to automatically generate human-readable alerts or notifications.
[0039] Optionally, the set of data corresponds to a multivariate signal associated with a plurality of operational parameters of the device.
[0040] Not only single measurements but multiple measurements can be used. The method may work with more advanced multivariate time series models. For this purpose, signal processing techniques (e.g., Fourier analysis, wavelet analysis, etc.) can be used. For example, if the data has a high granularity and / or at least one operating parameter of the device is frequency-dependent, it may be beneficial to work with a frequency domain analysis, e.g., operating parameters related to the number of revolutions per minute, interactions between gears in a machine, etc.
[0041] Advantageously, multiple measurements and their interactions are taken into account, which can increase accuracy and robustness. Furthermore, by employing multiple measurements, less domain expertise may also be required for successful implementation. In this way, the method can be applied more efficiently. Furthermore, this approach may better handle seasonal trends of the monitored signals from which the data sets are generated. Additionally or alternatively, this approach may offer more flexibility since it is easy to include additional features if necessary.
[0042] Optionally, a trained machine learning model is employed that is configured to identify interactions and / or dependencies between multiple operational parameters in the multivariate signal.
[0043] Using the reference data, the model can be trained to recognize interactions and dependencies between measurements. The trained model can be employed to reconstruct the test data. For normal / regular data, the reconstructed test data may resemble the real data and the reconstruction error may be low. For abnormal / irregular data, the reconstruction error may be high, so that it can be distinguished from the normal / regular data. The model may be a regression model. The model may be used based on multiple measurements, as some models may be more suitable than others based on the type of interactions between the signals captured by the multiple measurements. Exemplary models are tree-based methods, neural networks, etc. However, other models such as Bayesian models may also be used.
[0044] Optionally, an index of multiple reference identifiers for at least one operational parameter of the device is used.
[0045] An index may be formed in the database, where it is envisioned that the index includes multiple unique reference identifiers correlating to multiple operating conditions, e.g., irregular operating conditions, different types or levels of irregular operating conditions, etc. Each of the identifiers calculated based on the received set of data may be compared to the multiple unique reference identifiers in the database.
[0046] Optionally, multiple sensors are used to collect different data sets indicative of different operational parameters of the device, and identifiers calculated from the different data sets are compared to respective reference identifiers to determine multiple deviations used to monitor the operation of the device.
[0047] The statistical identifier (see fingerprint) can be calculated in different ways. The two "fingerprints" can be compared by determining the deviation between the calculated identifier and the reference identifier. In this way, it can be determined whether the identifier of the test window is outside a specified threshold (i.e. absolute or relative to the reference). The identifier serves as a fingerprint of the signal. The reference identifier (see reference fingerprint) can be adjusted. For this purpose, the reference period can be changed. For example, the reference period can include two weeks of complete data. As time progresses, the time window for extracting such data is adjusted so that the model compares with more recent machine behavior. Therefore, the time window can be shifted as time progresses. Advantageously, in this way, the accuracy and robustness of the monitoring of the operation of the device can be significantly improved.
[0048] Optionally, detection of an operation anomaly is triggered when a value indicating a size of the at least one deviation is greater than a preset threshold.
[0049] Optionally, notifications are sent wirelessly to one or more operator mobile devices. In this way, operators can be more efficiently notified when a particular part or device needs to be repaired, replaced, or maintained to prevent an outage from occurring. Different types of warning messages can be sent to the operator, providing additional information about the anomaly so that an initial assessment can be made by the operator remotely.
[0050] Optionally, detection of a behavior anomaly is triggered only when a value indicating the size of the deviation exceeds a preset threshold within a predefined time span.
[0051] Optionally, the thresholds are adjustable. In this way, the thresholds can be changed without requiring redeployment of the model. Advantageously, an operator can set thresholds on measured and monitored parameters to understand patterns and / or co-relationships between out-of-threshold indications and incidents occurring on the same device. Under static conditions (e.g., a filling device operating at nominal speed), the value of the signal may tend to fluctuate around some central value. The operator can define how much the calculated identifier must change relative to the reference identifier in order for the model to generate a flag or alarm (see user notification).
[0052] Optionally, calculating the identifier is performed using an algorithm, and the reference identifier is calculated using the same algorithm based on a set of reference data under verified normal operation of the device.
[0053] Optionally, the method includes collecting reference data. This may include determining measurements used for anomaly detection and determining conditions under which the signal has normal and / or stationary behavior. For this purpose, data may be collected over several periods of time, such as minutes, hours, or weeks. In some embodiments, at least several months of data may be collected. The minimum amount of data may depend on the time window desired to calculate the reference information. The reference data is assumed to be relatively free of anomalies.
[0054] Optionally, the reference identifier is repeatedly recalculated over time.
[0055] Optionally, the transient signal values associated with the at least one operating parameter are measured at multiple time steps, and the set of data indicative of the at least one operating parameter of the device within a predefined time interval includes multiple measurement points at multiple consecutive time steps within the predefined time interval.
[0056] Optionally, the behavior anomaly is classified into an anomaly family and an anomaly type based on at least one deviation between the identifier and at least one reference identifier.
[0057] Parameters reflecting signal characteristics (mean, median, volatility, etc.) can be stored in a database and accessed via the platform, allowing for the study of the machine's signal behavior over time. Good scalability can be obtained when the device signal behavior is known and / or default parameters are used.
[0058] The reference identifier may, for example, easily capture the condition where the monitored signal should fluctuate around an expected value (see fixed signal). Additionally, in some embodiments, the reference identifier may capture the condition where the signal is expected to exhibit static behavior. A variety of other characteristics may be easily captured using the unique identifier.
[0059] Optionally, the device is a machine or part thereof.
[0060] The method may be used to monitor a coupled system. The coupled system may have multiple devices. Alternatively or additionally, each device may generate multiple sets of data indicative of at least one operating parameter of the device. It is also envisaged that one or more devices of the system are provided with multiple sensors for generating multiple sets of data. A user interface may be provided that provides a general overview of the operation of the coupled system. For example, the user interface may show the operating status across all devices, machines, subsystems, lines, components, etc. of the coupled system in which the monitoring method is employed.
[0061] Optionally, the method is performed for proactive maintenance.
[0062] Optionally, the method includes performing anomaly detection, which is provided to detect patterns. Such anomaly detection may capture domain knowledge of experts and help experts find problems faster. Furthermore, anomaly detection may detect hidden patterns or non-linear behaviors that are not easily detected by human experts. This approach leads to condition-based maintenance, which helps maintenance experts do their jobs more efficiently.
[0063] Optionally, a prediction is made of when the device will fail based on the detected deviation. The time taken to react may depend on the type of measurement and its suitability for capturing the underlying physical process.
[0064] Optionally, a human readable explanation is provided as to why the deviation is flagged.
[0065] In some embodiments, the identifier is generated by extracting features of the set of data using a spectrogram. The generated identifier can be compared to one or more reference identifiers to determine deviations. In some embodiments, a database is searched for reference identifiers similar to the identifier generated based on the set of data. The method can include recognizing an operational behavior of the device based on results of the search.
[0066] In some embodiments, a pre-processing step is performed in which the monitored signal is re-sampled such that the data set is acquired at a sampling rate that matches the sampling rate of the set of reference data used to determine the reference identifier.
[0067] According to a second aspect, there is provided a system for monitoring a device, the system comprising a controller configured to receive a set of data indicative of at least one operating parameter of the device within a predefined time interval, calculate an identifier based on the set of data, the identifier uniquely identifying the set of data, and determine at least one deviation between the identifier and at least one reference identifier for the at least one operating parameter of the device.
[0068] Extracting an identifier from the underlying data content in the set of data may include the extraction of robust features of the data content. For example, a robust hash signature may be generated from the extracted features. Advantageously, extracting such robust features allows them to be invariant under various processing operations and to compactly represent the underlying perceptual content. Thus, relatively small changes in the feature values may not result in drastic changes in the extracted hash bits, conferring robustness to the fingerprint. Furthermore, robust hash extraction allows the extracted fingerprint bits to be unique. This may significantly affect the required search time, i.e. the time it takes to find a match with a reference identifier or in a database of reference identifiers (see reference fingerprints). Advantageously, in this way, a very efficient algorithm may be obtained.
[0069] The system may provide an improved way to understand where a problem or issue occurs or where it most impacts a process. The method may be applied to multiple devices or components within a larger system. A high level overview of critical machines, components and their common failure modes may be provided. The method may evaluate the frequency of occurrence of failure modes, response time for repairs, and comparison with manufacturer set benchmarks based on historical data. The method further provides a way to better gain knowledge of what has happened and what is currently going on in order to evaluate how the situation is evolving. Without the need for advanced expertise, one may better determine how much the operation of the device is deteriorating and evaluate how quickly the situation is evolving. Furthermore, by comparing the identifiers with each other, the method may allow to better determine the cause of the component malfunctioning. In this way, it may be easier to determine whether to shut down the device and intervene or wait until the next maintenance window. Based on the determined deviation, one or more failure modes may be determined. Improved warning time windows may be provided. Through the analysis of past examples, the relationship between the failure modes and the selected measurements may be more easily evaluated.
[0070] For example, a database of reference identifiers may be searchable for comparison with the calculated identifier based on the set of data, and the best match may be found to be the one with the smallest distance from the query identifier.
[0071] Matching an identifier conveying characteristic information relating to a set of data indicative of at least one operating parameter of a device within a predefined time interval with one or more pre-configured reference identifiers may enable efficient evaluation of the behavior of the at least one operating parameter.
[0072] Actions may be performed in response to a failure to match due to a high deviation between the identifier and the reference identifier, e.g., the deviation exceeds a preset threshold. Actions may include notifying a user, triggering an alert, performing more diagnostic activity, scheduling maintenance on the device, scheduling repairs on the device, adjusting the operating state of the device (e.g., reducing load until the problem is resolved), etc.
[0073] Advantageously, the identifier and the reference identifier can be easily compared to quickly and efficiently determine whether the received set of data is sufficiently similar to the reference data set from which the reference identifier was determined.
[0074] It will be appreciated that data sets associated with a variety of signals can be monitored and received to calculate the identifier. For example, current, flow rate, consumption indicators, power, pressure, acceleration, vibration, frequency, temperature, voltage. A variety of other signals can be used. Combinations of different signals are also envisioned.
[0075] According to a third aspect there is provided a computer program product configured to perform the steps of the method according to the first aspect when executed on a controller.
[0076] When abnormal device behavior is detected, an action can be alerted to check the device. A computational model can be employed that is configured to compare the measured signal values with a reference value, and a relative amount of difference between the measured signal values and the reference value can be specified to achieve a balance between alerting on significant events and avoiding alarm saturation.
[0077] Anomalous behavior may be detected in a set of data indicative of at least one operating parameter of a device within a predefined time interval. The set of data may be obtained by performing measurements to obtain a time series signal. The time series signal may be analyzed to detect anomalies by calculating an identifier and comparing the identifier to one or more reference identifiers.
[0078] A user interface may be provided in which the most recent events are displayed. The method may include indicating the devices (e.g., labels, tag names) on which the most anomalies were detected in a selected time window. A total number of events in the selected time window may be displayed. The user interface may also be configured to indicate which portions of the set of data have been flagged as anomalies. In some examples, an anomaly detection dashboard may be provided.
[0079] The received set of data, the calculated identifier, the determined deviation, and / or the reference identifier may be stored in a database. The historical data may be used to fine-tune or tune the model to more accurately determine which deviations should be flagged.
[0080] According to a fourth aspect, a non-transitory computer readable medium is provided having software instructions for performing the computer implemented method according to the first aspect.
[0081] According to a further aspect, a method for monitoring a device is provided, the method including receiving reference data, generating a reference fingerprint based thereon, calculating the operational fingerprint based on the monitored data linked to at least one operational parameter of the device within a predefined time interval, and comparing the calculated operational fingerprint to the reference fingerprint.
[0082] Optionally, the reference fingerprint may be updated during monitoring, for example, a periodic update of the reference fingerprint may be performed.
[0083] The operational fingerprint and the reference fingerprint may be calculated for different durations and over different time intervals. If the operational fingerprint differs from the reference fingerprint in a certain way, this may indicate a fault or another reason for maintenance. Such anomaly detection may be employed for preventive maintenance.
[0084] Different types of signals can be monitored from which a set of data indicative of at least one operating parameter of the device is determined. The signals can be, for example, motor current, vibration, pressure, flow rate, speed, etc.
[0085] A device may have multiple signals monitored at a particular component of the device. A user may be informed of which device, or part of the device, an anomaly has been detected. For example, a bottle filling machine in a brewery process may have multiple servo drives. For each of the drives, the current and phase lag may be monitored. Based thereon, a set of data within a predefined time interval may be used to determine an identifier (see fingerprint) that can be compared to a reference identifier to determine deviations. When an anomaly is detected, an action may be initiated, for example, a user may be notified of the signal for which this anomaly was detected. In some examples, the operation of the device is adjusted based on the detection of the anomaly.
[0086] In some examples, the predefined time interval for collecting data does not change over time. Thus, the predefined time interval may remain fixed. Optionally, multiple time intervals are checked for the same signal.
[0087] According to another aspect, there is provided a method of monitoring a device, the method including: receiving a first set of data values of a signal indicative of an operational characteristic of a machine within a first reference time window; determining reference signal data based on the first set according to a first algorithm; receiving a second set of data values of the signal indicative of an operational characteristic of the machine within the operational time window; determining the operational signal data based on the second set according to the first algorithm; comparing the reference signal data to the operational signal data; and registering, if the operational signal data and the reference signal data differ by more than a first predefined threshold, detecting the difference as an anomaly.
[0088] Optionally, the reference signal data is recalculated over a second reference time window. Optionally, the recalculation of the reference signal data is performed after a first predefined period of time has elapsed after the first reference time window. Additionally or alternatively, the recalculation of the reference signal data is performed if the motion data differs from the reference data by less than a second predefined threshold for a second predefined period of time.
[0089] Optionally, the first threshold data is recalculated after calculating the new reference data.
[0090] Optionally, the signal represents at least one of vibration, rotational motion, linear motion, stress, relative position, or an entity derived therefrom.
[0091] Optionally, the first algorithm includes at least one of determining a mean, a median, a standard deviation, a hash value, a fingerprint, a checksum of at least a portion of the received data.
[0092] Optionally, an anomaly is detected when the operational data differs from the reference data by more than a second predefined threshold for a second predefined period of time.
[0093] Optionally, the data values are associated with time-sequential timestamps.
[0094] Optionally, an anomaly is detected if the operational signal data differs from the signal reference data more than a predefined number of times during a predefined period of time.
[0095] Although various aspects and further embodiments may involve computing devices and processes executed on computing devices, the disclosure also extends to computer programs, in particular computer programs on or in a carrier, adapted to carry out the disclosure. The program may be in the form of source or object code or in any other form suitable for use in the implementation of the disclosed processes. The carrier may be any entity or device capable of carrying a program.
[0096] "Optional" or "optionally" means that the subsequently described event or circumstance may or may not occur, and that the description includes instances where the event or circumstance occurs and instances where it does not occur.
[0097] The terms "coupled," "associated," and "connected," along with their derivatives, may be used herein to describe a functional or structural relationship between components. The terms may be used to indicate that two or more elements are in direct or indirect (with other elements between them) physical, optical, communication, and / or electrical contact with each other and / or that two or more elements cooperate or interact with each other (e.g., in a cause and effect relationship).
[0098] It will be understood that any of the aspects, features, and options described in consideration of the present method apply equally to the systems and devices and computer program products described. It will also be apparent that any one or more of the above aspects, features, and options may be combined. In particular, the options provided above may be combined with the general inventive concept with any one or more options, in any permutation.
[0099] Various aspects and variants thereof will be further elucidated on the basis of exemplary embodiments represented in the drawings, which are given by way of non-limiting illustration, and it should be noted that these figures are merely schematic representations of the aspects and variants thereof shown by way of non-limiting examples. [Brief description of the drawings]
[0100] [Figure 1] 1 shows a schematic diagram of an embodiment of a system. [Diagram 2] 1 shows a schematic diagram of exemplary data. [Diagram 3] 1 shows a schematic diagram of exemplary data. [Figure 4] 1 shows a schematic diagram of exemplary data. [Figure 5A] 1 shows a schematic diagram of exemplary successive predefined time intervals; [Figure 5B] 1 shows a schematic diagram of exemplary successive predefined time intervals; [Figure 5C] 1 shows a schematic diagram of exemplary successive predefined time intervals; [Figure 6] A schematic diagram of the method is shown. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0101] 1 shows a schematic diagram of an embodiment of a system 1 for monitoring a device 3. The system comprises a controller 5 configured to receive a set of data 7 indicative of at least one operating parameter of the device 3 within a predefined time interval T, to calculate, based on the set of data 7, an identifier 9 that uniquely identifies the set of data 7, and to determine at least one deviation 11 between the identifier 9 and at least one reference identifier 13 for the at least one operating parameter of the device 3.
[0102] To this end, the controller 5 comprises a processing unit, a storage unit, a data input unit and a communication unit as a data output unit. The processing unit is configured to execute the first aspect and any variants thereof, as will be discussed in more detail below. Thus, the processing unit may be configured to be programmed by a computer program product. The data input unit is configured to receive data from a sensor configured to monitor an operating parameter of the device 3. The sensor may be configured by the device 3, and the controller 5 or the sensor may be located external to both. The data input is configured to provide the signal received from the sensor to the processing unit, either after processing or as raw data.
[0103] The storage module is configured to store the received data, the data processed by the data input unit, and / or the data processed by the processing unit. Additionally or alternatively, the storage module may preferably include a non-transitory medium configured to store a computer program product for programming the processing unit to perform the first aspect and / or variants thereof.
[0104] The communication unit may be configured to transmit data received from the sensor, data received from the data input unit, data processed by the processing unit, other data, or a combination thereof, to a data processing device located remotely from the device 3. The data processing device may be configured to perform one or more steps of the first aspect or its variants. Thus, some or all parts of the first aspect and its variants may be performed jointly, in parallel or in series by the controller 5 and by the data processing device.
[0105] The identifier 9 and the reference identifier 13 are illustrated as digital fingerprints, however, it will be appreciated that such an identifier or digital fingerprint can have various forms, for example it could be a hash value.
[0106] The reason for flagging a particular behavior of a device may be easy for an end user (e.g., an operator) to understand. The results are easy to interpret, since detection of deviations of an identifier against a reference identifier does not require complex model knowledge and / or further analysis. Information on which device and / or which part of a device an anomaly was identified can be displayed to the user, e.g., resulting in a notification or alarm.
[0107] Depending on the algorithm used, the computation of the identifier may be sensitive to irregularities in the set of data. In some embodiments, the computation of the identifier (see digital fingerprint) may be particularly sensitive to irregular rises and / or falls of the data values relative to a plateau level, during which the monitored data values spike momentarily before returning to its plateau level, and during which the monitored data values drop momentarily before returning to its plateau level. The identifier may be used to robustly detect irregularities in the set of data that are linked to one or more operational irregularities of the device. For example, when the set of data is graphed as a function of time, unexpected spike or dip variations are visible. Such spike or dip variations of the data may indicate irregular operation of one or more components of the device associated with at least one operational parameter. Such sudden positive or negative peaks (valleys in the graph) may deviate from the plateau level and thus may result in significant deviations between the calculated identifier and the reference identifier.
[0108] A plateau level can be a range (see band / channel / envelope) with a baseline, an upper limit, and a lower limit. Irregularities in a set of data can be seen as a situation where the data values suddenly fall outside the range of the plateau level. The data values may have positive and / or negative peaks that fall within the plateau level without causing a significant effect on the calculated identifier, and positive and / or negative peaks that fall outside the plateau level that cause a significant effect on the calculated identifier.
[0109] It will be appreciated that a spike or peak in a data value in a set of data can be considered as a sharp rise (i.e., a sudden, rapid increase) followed by a sharp decrease. Thus, a momentary, rapid increase in a data value or a value indicative of at least one operating parameter of the device can result in a significant deviation of the calculated identifier from a reference identifier determined using a reference data set (a device operating under normal conditions). The reverse also applies to a dip (see negative peak). A dip can be considered as a temporary decrease in the level of a data value before returning to a plateau level. A dip can include a short downward slope followed by an upward slope.
[0110] The controller 5 may be communicatively coupled to the device 3. This may be done wired and / or wirelessly. However, it is also possible to have the controller integrated within the device or within a component of the device. The controller 5 may be configured to communicate with other devices of the coupled system of devices, for example other controllers associated with a brewery production line having multiple devices and sub-components. The device may also be communicatively coupled to one or more central controllers.
[0111] Device 3 may be a subcomponent of another device, machine, system, process, etc. For example, an electric servo drive may be critical for the operation of a filling machine in a brewery process. A failure would result in the brewery process having production downtime. Further, as an example, a vacuum pump in a filling machine may have the purpose of removing residual oxygen. A failure could result in product being wasted, resulting in huge costs and production delays. In these examples, the electric servo drive, the vacuum pump, or any other component or machine may be monitored to detect abnormal operation.
[0112] Various operating parameters may be monitored. Examples include, but are not limited to, data indicative of engine operation, data indicative of motor current, data indicative of vibration, data indicative of acceleration, data indicative of speed of moving parts, data indicative of phase lag of electrical signals, data indicative of pressure, data indicative of fluid flow. Combinations of different operating parameters are also possible. Additionally, monitoring of more complex operating parameters, for example involving electromechanical parameters, sound, audio, video, etc., is also envisioned.
[0113] Optionally, the reference identifier is determined periodically. This may involve recalculating the reference identifier to obtain a more recent reference identifier. In other words, the "reference fingerprint" may be changed over time. The updated reference identifier may provide improved monitoring. The updated reference identifier may be used for comparison with an identifier calculated based on a set of received data indicative of at least one operational parameter of the device within a predefined time interval. The reference identifier is preferably calculated using the same or an equivalent algorithm as that used to calculate the operational identifier whose behavior is monitored.
[0114] In an embodiment, the device 3 is a pressure control system pump. A sudden decrease in pressure may indicate a leak. If the decrease continues for a longer period of time, it may indicate a fault in the impeller or pump. If the pressure increases suddenly, there may be an obstruction in the piping that needs to be removed. Such possible technical issues may be better determined based on the calculated identifier, the reference identifier, and the deviation between the calculated identifier and the reference identifier. In some cases, multiple reference identifiers are used. For example, an index of reference identifiers may be employed, and a small deviation between the calculated identifier and one of the multiple reference identifiers in the index of reference identifiers may identify a particular anomaly. This provides an efficient and robust way of determining anomalies. Furthermore, anomalies may be detected faster.
[0115] In an embodiment, the device 3 is a beer filtration installation comprising multiple subcomponents. For example, when a filter of the filtration installation becomes clogged, the pressure in the piping increases. If the pressure increases too much, it leads to damage of pumps, filters and gaskets. The device may be monitored using data from a pressure sensor. The calculated identifier may be compared to one or more reference identifiers to determine deviations. Based on the deviations, anomalous operation may be detected in a robust manner.
[0116] In an embodiment, the device 3 is a flow control shaft. For example, the device may be configured to monitor data indicative of flow.
[0117] In an embodiment, the device 3 is a reservoir. For example, the device may be configured to monitor data indicative of the level of the medium in the reservoir. For example, a level transmitter sensor may be provided. In some cases, by calculating an identifier and determining the deviation between the calculated identifier and a reference identifier, electrical interference, electrical earth problems, a faulty level in the transmitter, dirt in the tank, etc. may be detected.
[0118] FIG. 2 is a schematic diagram illustrating an exemplary set of data 7 showing operational parameters of a device within a predefined time interval T. In the set of data 7, it can be observed that a portion 15 of the set of data 7 is irregular. This results in a calculated identifier having a noticeable deviation from a reference identifier. A reference identifier can be calculated for data that does not have such irregularities in the set of reference data. The deviation can thus result in this received set of data being flagged. An operational issue with the device can be diagnosed. As a result, one or more actions can be triggered before a critical outage occurs, such as, for example, notifying a user, ordering spare parts, scheduling a repair, etc.
[0119] In the illustrated embodiment, plateau levels can be observed. The plateau levels form a band or range within which data values are expected under normal conditions during normal operation of the device. Any sudden anomalies in the data values may cause the data values to fall outside the range of the plateau levels, resulting in significant changes in the calculated identifier. Comparing the resulting calculated identifier to a reference identifier will result in significant deviations, which may be an indication of abnormal operation of the device.
[0120] FIG. 3 shows a schematic diagram of an exemplary set of data 7 showing the operating parameters of a device within a predefined time interval T. Similar to the embodiment of FIG. 2, a portion 15 of the set of data 7 is irregular. As a result, based on deviations in the calculated identifiers and reference identifiers, irregularities can be effectively and accurately identified without the need for complex data processing models. The determined deviations may exceed a threshold such that the set of data is flagged. This may entail one or more actions so that the operational issue can be corrected before a critical outage.
[0121] Fig. 4 shows a schematic diagram of an exemplary data set 7 showing the operating parameters of a device within a predefined time interval T. Similar to the illustrated embodiment of Figs. 2 and 3, a portion 15 of the data set 7 is irregular. Although the portion 15 of the data set 7 is relatively narrow, it results in a large deviation when the calculated identifier is compared with the reference identifier. In this way, problems can be flagged and diagnosed in a timely manner. Advantageously, necessary measures can be taken to effect repairs before device outages and loss of functionality occur.
[0122] 5a, 5b, and 5c show schematic diagrams of exemplary successive predefined time intervals. A signal associated with at least one operating parameter of the device can be monitored, and a set of data is collected within a predefined time interval for analysis. For each set of data within a plurality of successive predefined time intervals, a corresponding identifier can be calculated to determine the deviation between the corresponding identifier and at least one reference identifier for at least one operating parameter of the device. This can be performed in different ways. As illustrated in FIG. 5a, in some cases, the successive predefined time intervals can have a time gap G between them. Depending on the device, the operating parameters, and the application, among others, it can vary from a few seconds to a few weeks or months. As illustrated in FIG. 5b, the successive predefined time intervals can have an overlapping interval O. In some cases, this can significantly improve the robustness of the system. As illustrated in FIG. 5c, the successive predefined time intervals can be sequential without a time interval in between.
[0123] Optionally, the method is performed periodically, with each iteration using a more recent set of data indicative of at least one operational parameter of the device within a predefined time interval, in some embodiments the method steps are performed at a particular frequency, e.g., every 10 minutes, every 15 minutes, etc.
[0124] FIG. 6 shows a schematic diagram of a method 100 for monitoring a device. In a first step 101, a set of data indicative of at least one operating parameter of the device within a predefined time interval is received. This may be obtained, for example, by one or more sensors. However, the data may also be readily available. In a second step 102, an identifier that uniquely identifies the set of data is calculated on the basis of the set of data. For this purpose, a transformation algorithm may be used, for example a hash algorithm for determining a digital fingerprint associated with the set of data. In a third step 103, at least one deviation between the identifier and at least one reference identifier for at least one operating parameter of the device is determined.
[0125] Various aspects and variations thereof may reduce equipment downtime caused by unexpected outages or equipment failures by providing better advance indication of when maintenance should be performed. Advantageously, the subject matter of the present disclosure may do this without requiring a high level of expertise from the operator.
[0126] The user interface may provide the most recent flagged anomalies, the time that a particular anomaly was flagged (i.e., flag time), and other information may be shown, such as the unit name (e.g., packaging line), machine / device name, tag name where the anomaly was detected, a description detailing why a particular flag was generated, etc.
[0127] In some embodiments, the user interface may show the number of anomalies by tag name. In this way, the most troublesome components can be easily identified.
[0128] In some examples, the user interface may provide a number of anomalies per device. In this way, it can be determined whether the anomaly detection scales horizontally to other components, e.g., other subcomponents, devices, machines, lines, etc. Advantageously, the most troublesome devices or subcomponents of devices can be easily recognized.
[0129] In some examples, the user interface may provide a number of flags that have been raised over a period of time for a device. Advantageously, this allows tracking of the number of flagged anomalies over time for measurements associated with a particular device (e.g., servo drive, pump, etc.). A threshold may be defined such that an alarm, alert, and / or notification is generated when the number of flags per hour exceeds the threshold.
[0130] In some embodiments, the user interface is configured to allow signal exploration. The user interface may allow filtering of the data. This may provide a suitable way to explore measured signals, reference signals, calculated identifiers, reference identifiers, flag summaries over a period of time, etc. This effectively provides an opportunity to dig deeper into the behavior of a particular signal. Advantageously, the user may explore correlations between signals and fault events and understand if thresholds / alarm criteria require further adjustments. Thus, a highly dynamically tunable system is obtained that can be adjusted to the nature of the device.
[0131] Optionally, the time window of the data used to determine the reference identifier is adjustable: a user may be allowed to select and / or adapt the time window of the reference data used to determine the reference identifier, for example a particular time window is one in which the device is known to have been operating under normal conditions.
[0132] Various parameters of the device can be used, such as the power supplied to the engine, the current provided to the engine, the speed of the device, the phase lag of the current in the motor, the voltage across the electrical circuitry of the device, etc.
[0133] Advantageously, the effect of the failure mode on the change in the signal behavior can be better captured using the identifier.
[0134] In an embodiment, the device being monitored is a servo drive. The monitored signals are indicative of the current and phase lag provided to the servo drive. The type of failure mode can be defined as misalignment of the servo drive, voltage surge, bearing issues within the servo drive, etc. The signals are typically expected to provide an early indication of the issue in the case of expected failure modes, leaving sufficient time (days to weeks) to react.
[0135] In an example, the device being monitored is a vacuum pump. The monitored signal can be vibration. This can be done by an accelerometer, but other means for monitoring vibrations can also be used (e.g., laser vibrometer, strain gauges, etc.). The type of failure mode can be defined as a bearing failure. The signal is expected to show irregular operation of the vacuum pump at least several weeks before serious failure.
[0136] It will be understood that the method may include computer-implemented steps. All steps above may be computer-implemented steps. The embodiments may include a computer device, and the process is executed on a computer device. The disclosure also extends to computer programs, in particular computer programs on or in a carrier, adapted to practice the various aspects and variations thereof. The programs may be in the form of source code or object code, or in any other form suitable for use in implementing the processes according to the disclosure. The carrier may be any entity or device capable of carrying a program. For example, the carrier may comprise a storage medium such as a ROM, for example a semiconductor ROM or a hard disk. Furthermore, the carrier may be a transmissible carrier, such as an electric or optical signal, which may be conveyed via an electric or optical cable, or by radio or other means, for example via the Internet or the cloud.
[0137] Some embodiments may be implemented, for example, using a machine or tangible computer-readable medium or article capable of storing instructions or sets of instructions that, when executed by the machine, can cause the machine to perform methods and / or operations in accordance with the embodiments.
[0138] Various embodiments may be implemented using hardware elements, software elements, or a combination of both. Examples of hardware elements may include processors, microprocessors, circuits, application specific integrated circuits (ASICs), programmable logic devices (PLDs), digital signal processors (DSPs), field programmable gate arrays (FPGAs), logic gates, registers, semiconductor devices, microchips, chipsets, etc. Examples of software may include software components, programs, applications, computer programs, application programs, system programs, machine programs, operating system software, mobile apps, middleware, firmware, software modules, routines, subroutines, functions, computer implemented methods, procedures, software interfaces, application program interfaces (APIs), methods, instruction sets, computing code, computer code, etc.
[0139] In this specification, various aspects and variations thereof are described with reference to specific examples of embodiments of the various aspects. However, it will be apparent that various modifications, variations, substitutions, and changes can be made without departing from the essence of the present disclosure. For the purpose of clarity and concise description, features are described herein as part of the same or separate embodiments, but alternative embodiments having all or some combinations of the features described in these separate embodiments are also assumed and understood to be included within the framework of the present disclosure outlined by the claims. Therefore, the specification, figures, and examples should be regarded in an illustrative sense, not a restrictive sense. The present disclosure is intended to embrace all alternatives, modifications, and variations that fall within the spirit and scope of the appended claims. Moreover, many of the described elements are functional entities that can be implemented as separate or distributed components, or in combination with other components, in any suitable combination and location.
[0140] In the claims, any reference signs placed between parentheses shall not be interpreted as limiting the scope of the claims. The word "comprises" does not exclude the presence of other features or steps than those listed in the claims. Moreover, the words "a" and "an" should not be interpreted as limiting to "only one", but are used in the sense of "at least one", without excluding a plurality. The mere fact that certain measures are recited in mutually different claims does not indicate that a combination of these measures cannot be used to advantage. [Explanation of symbols]
[0141] 1 System 3 Devices 5. Controller 7 sets 9 Identifiers 11 deviation 13 Reference Identifiers 15 part 100 ways
Claims
1. 1. A method for monitoring a device in an electronic computing device, comprising: receiving a set of data indicative of at least one operational parameter of the device within a predefined time interval; calculating an identifier based on the set of data using a first algorithm, the identifier uniquely identifying the set of data; and determining deviation data comprising data regarding an occurrence of at least one deviation between said identifier and at least one reference identifier for said at least one operational parameter of said device.
2. The method of claim 1 , wherein the identifier forms a digital fingerprint of the set of data within the predefined time interval, the digital fingerprint representing a unique identification of a work regime associated with the at least one operational parameter.
3. The method of claim 1 or 2, wherein the digital fingerprint is calculated using a statistical model configured to receive as input at least a portion of the set of data over the predefined time interval.
4. 4. The method of claim 1, wherein the identifier is determined based on a subset of features in the set of data, one or more robust features are extracted from the set of data, and a robust hash signature is determined from the extracted one or more robust features.
5. The method of claim 1 , wherein the set of data corresponds to a multivariate signal associated with a plurality of operational parameters of the device.
6. The method of claim 5 , wherein a trained machine learning model is employed that is configured to identify interactions and / or dependencies between the plurality of operational parameters in the multivariate signal.
7. The method according to claim 1 , wherein detection of an operational anomaly is triggered if a value indicative of a size of the at least one deviation is greater than a preset threshold value.
8. The method of claim 7 , wherein the detection of a behavior anomaly is triggered only when the value indicative of the size of the deviation exceeds the pre-set threshold within a predefined time span.
9. 9. The method according to claim 1, wherein the calculating of the identifier is performed using an algorithm, and the reference identifier is calculated using the same algorithm based on a set of reference data under verified normal operation of the device.
10. 10. The method of claim 1, wherein transient signal values associated with at least one operational parameter are measured at a plurality of time steps, and the set of data indicative of the at least one operational parameter of the device within the predefined time interval includes a plurality of measurement points at a plurality of consecutive time steps within the predefined time interval.
11. receiving device status data; The method of claim 1 , further comprising: determining a relationship between the device status data and the deviation.
12. 12. The method of claim 11, wherein the device status data is provided with status timing information and the deviation data is provided with deviation timing information, the method further comprising determining a relationship between the status data and the deviation data over time.
13. determining a device event associated with the device based on the device status data; The method of claim 12 , further comprising: determining a time interval between the device event and an occurrence of the deviation based on the status timing information and the deviation timing information.
14. determining a number of device events associated with the device based on the device status data; determining a time interval between the device event and an occurrence of one or more deviations based on the status timing information and the deviation timing information; identifying, from the number of determined device events, a first device event associated with a particular portion of the device and a second device event associated with the particular portion of the device; identifying a first occurrence of a first deviation and a second occurrence of a second deviation from the deviation data; 14. The method of claim 13, wherein a first time interval between the first device event and the first occurrence is less than a predefined relationship time interval from a second time interval between the second device event and the second occurrence.
15. 1. A system for monitoring a device, comprising: receiving a set of data indicative of at least one operational parameter of the device within a predefined time interval; calculating an identifier based on the set of data, the identifier uniquely identifying the set of data; determining at least one deviation between the identifier and at least one reference identifier for the at least one operational parameter of the device.
