Anomaly detection and fault isolation for vehicle subsystems

Abstract

The invention relates to anomaly detection and fault isolation for vehicle subsystems. Techniques for anomaly detection are disclosed. The techniques include identifying a plurality of redundant subsystems in a system, the plurality of redundant subsystems being three or more redundant subsystems. The technique further includes forming a plurality of combinations of redundant subsystems, each combination associated with a subset of the plurality of redundant subsystems. The technique also includes a first combination from the plurality of combinations of redundant subsystems that has a minimum variability in the identifying subsystem. The technique further includes calculating one or more nominal values for one or more parameters of a first subsystem of the plurality of subsystems using the first combination, and detecting an anomaly in the first subsystem based on the calculated one or more nominal values.

Description

Technical Field

[0001] Aspects of the present disclosure relate to anomaly detection for complex systems including vehicles. Background Art

[0002] Detecting degradation and isolating faults in complex systems (e.g., aircraft, other vehicles, or other complex systems) is very challenging. For example, an anomaly in a specific aircraft subsystem can be detected by identifying correlated degradation signatures within a feature prior to failure. But these degradation signatures may not be reflected in direct sensor measurements. In addition, identifying anomalies often requires characterizing or deriving the nominal behavior of the component under varying operating conditions, which is very challenging because there may be wide variations in sensor data under dynamic operating conditions and the associated controllers may attempt to compensate for the degradation. Limited sensor data and low sample data also add to the challenge. Summary of the invention

[0003] Aspects include a method. The method includes identifying a plurality of redundant subsystems in a system, the plurality of redundant subsystems being three or more redundant subsystems. The method also includes forming a plurality of combinations of the redundant subsystems, each combination being associated with a subset of the plurality of redundant subsystems. The method also includes identifying a first combination having the least variability among the subsystems from the plurality of combinations of the redundant subsystems. The method also includes calculating one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of subsystems using the first combination, and detecting an anomaly in the first subsystem based on the calculated one or more nominal values.

[0004] Aspects also include a system comprising one or more processors and one or more memories storing a program that performs an operation when executed on any combination of the one or more processors. The operation includes identifying a plurality of redundant subsystems in the system, wherein the plurality of redundant subsystems are three or more redundant subsystems. The operation also includes forming a plurality of combinations of redundant subsystems, each combination being associated with a subset of the plurality of redundant subsystems. The operation also includes identifying a first combination having the least variability among the subsystems from the plurality of combinations of redundant subsystems. The operation also includes using the first combination to calculate one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of subsystems, and detecting an anomaly in the first subsystem based on the calculated one or more nominal values.

[0005] Aspects also include a computer program product including a non-transitory computer-readable storage medium having a computer-readable program code implemented therewith, the computer-readable program code being executable by one or more computer processors to perform an operation. The operation includes identifying a plurality of redundant subsystems in a system, the plurality of redundant subsystems being three or more redundant subsystems. The operation also includes forming a plurality of combinations of redundant subsystems, each combination being associated with a subset of the plurality of redundant subsystems. The operation also includes identifying a first combination having the least variability among the subsystems from the plurality of combinations of redundant subsystems. The operation also includes calculating one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of subsystems using the first combination, and detecting an anomaly in the first subsystem based on the calculated one or more nominal values.

[0006] Aspects also include any of the aspects discussed herein wherein the system comprises an aircraft and the three or more redundant subsystems comprise redundant subsystems in the aircraft.

[0007] Aspects also include any aspects discussed herein wherein identifying a first combination having minimal variability is based on measured sensor values ​​of one or more subsystems and at least one of: (i) calculated physically-based parameters of one or more subsystems, or (ii) parameters from a digital twin associated with the one or more subsystems.

[0008] Aspects also include any aspects discussed herein wherein identifying the first combination having the least variability is based on both calculated physics-based parameters of the one or more subsystems and parameters from the digital twin related to the one or more subsystems.

[0009] Aspects also include any aspects discussed herein wherein calculating one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of subsystems using the first combination includes averaging each of the one or more parameters across the subsystems in the first combination.

[0010] Aspects also include any of the aspects discussed herein, wherein identifying a first combination from a plurality of combinations of redundant subsystems having a minimum variability among the subsystems further identifies one or more outliers that are excluded from the first combination.

[0011] Aspects also include any aspects discussed herein, wherein detecting an anomaly in a first subsystem based on one or more calculated nominal values ​​comprises determining a health index of the first subsystem based on calculating at least one of a cosine similarity and a Euclidean distance of multiple parameters in the subsystem compared to the calculated one or more nominal values.

[0012] Aspects also include any of the aspects discussed herein, wherein determining the health index of the first subsystem is based on calculating both cosine similarities and Euclidean distances of a plurality of parameters in the subsystem compared to calculated one or more nominal values.

[0013] Aspects also include any of the aspects discussed herein, the method or operations further comprising identifying a first component at least partially responsible for the anomaly in the first subsystem.

[0014] Aspects also include any of the aspects discussed herein, wherein identifying a first component at least partially responsible for the anomaly in the first subsystem further comprises identifying the first component using an inference algorithm based on the anomaly.

[0015] Aspects also include any of the aspects discussed herein wherein the inference algorithm is a Bayesian network trained based on historical data related to operation of the aircraft. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order that the manner in which the above-described features may be understood in detail, a more particular description, briefly summarized above, may be had by reference to example aspects, some of which are illustrated in the accompanying drawings.

[0017] Figure 1 Anomaly detection and fault isolation of subsystems with different redundancy levels according to one aspect are shown.

[0018] Figure 2 A controller for anomaly detection and fault isolation of a subsystem according to one aspect is shown.

[0019] Figure 3 Anomaly detection and fault isolation using redundancy according to one aspect is shown.

[0020] Figure 4 is a flow diagram illustrating anomaly detection and fault isolation using redundancy according to one aspect.

[0021] Figure 5 An example of creation of labeled data for inference network-based fault isolation according to one aspect is shown.

[0022] Figure 6 is a flow chart illustrating an initial set of absolute features according to one aspect.

[0023] Figure 7 is a flow chart illustrating anomaly detection and fault isolation of a subsystem in a particular aircraft tail section according to one aspect.

[0024] Figure 8 Further illustrated is anomaly detection and fault isolation of subsystems in a particular aircraft tail section according to one aspect.

[0025] Fig. 9 Anomaly detection and fault isolation of subsystems for all aircraft in a fleet according to one aspect is shown.

[0026] Fig.10 Further illustrated is anomaly detection and fault isolation of subsystems for all aircraft in a fleet according to one aspect. DETAILED DESCRIPTION

[0027] One or more aspects disclosed herein provide anomaly detection and enhanced fault isolation for complex systems (e.g., aircraft or other vehicles). In one aspect, the complex system has subsystems with multiple redundancies. Relative or comparative features can be calculated relative to the nominal values ​​of the corresponding features, and the nominal values ​​can be dynamically estimated for each data sample (e.g., aircraft flight segment or flight segment or each time sample). The method is particularly suitable for subsystems with more than dual redundancy (e.g., three or more redundancies), which is very common for aircraft.

[0028] In one aspect, initial features are calculated from direct sensor measurements, physics-based features are calculated, and parameters are estimated from the digital twin. For each operational segment (e.g., an aircraft flight segment), a nominal value for each feature is dynamically calculated by averaging the corresponding features of two or more subsystems that behave similarly. Similarity of different subsystems can be calculated using similarity scores or using any other suitable technique.

[0029] On the one hand, high-level features are then calculated based on the differences of the initial features compared to the corresponding nominal values, and statistical features of those differences are derived relative to time series data across operating segments (e.g., flight segments). Based on calculating Euclidean distance and cosine similarity (or any other suitable technique), the health index of each subsystem can be estimated. By comparing the health indices over consecutive operating segments (e.g., flight segments), subsystems with degradation are isolated. Then, before a fault event is triggered (e.g., based on a maintenance message), the data of degradation across operating segments (e.g., flight segments) is mapped with corresponding fault labels based on historical data obtained in fleet operations. The historical labeled data is used to train an inference algorithm (e.g., a Bayesian network or a suitable machine learning (ML) model) that is used to detect the location of degradation or isolate degraded subsystem components. This is described below with respect to Figure 1 and Figures 3 to 6 Further discussion.

[0030] Furthermore, although the above and below Figures 3 to 6The techniques discussed are applicable to systems with multiple redundant subsystems, but additional techniques can be used to identify nominal values ​​for anomaly detection without the need for redundant subsystems. For example, historical data for a particular subsystem (e.g., a subsystem in the tail section of a particular aircraft) can be used to identify nominal values ​​for that subsystem, and then the nominal values ​​can be used to detect anomalies and isolate faults. This is discussed below with respect to additional Figure 1 as well as Figures 7 and 8 As another example, historical data for a fleet of systems (e.g., a fleet of aircraft) can be used to identify nominal values ​​for subsystems in the fleet, which can then be used to detect anomalies and isolate faults. This is discussed below with respect to Figure 1 and Figures 7 to 10 Further discussion.

[0031] In one aspect, one or more of these techniques have significant advantages over existing solutions. For example, degradation is often so slow that identifying changes over time is very difficult and computationally expensive. In addition, automated systems can compensate for degradation, making it more difficult to find approaching anomalies. One or more of the techniques discussed herein can be used to detect degradation and approaching anomalies in an accurate and computationally efficient manner.

[0032] Figure 1 Anomaly detection and fault isolation of subsystems with different redundancy levels according to one aspect are shown. The computing environment 100 uses multiple input values ​​to detect anomalies, isolate faults, or both of subsystems of a complex system (e.g., an aircraft, another vehicle, or any other suitable complex system). As shown, the input includes sensor values ​​102, physics-based features 104, and digital twin values ​​106. These are only examples, and any suitable subset of these values ​​and additional values ​​(e.g., instead of sensor values ​​102, physics-based features 104, and any combination of digital twin values ​​106 or in addition to sensor values ​​102, physics-based features 104, and digital twin values ​​106) can be used. For example, sufficiently robust physics-based features 104 can eliminate the need for digital twin values ​​106 (and vice versa).

[0033] In one aspect, the sensor values ​​102 are measurements (e.g., direct measurements) related to the operation of the associated subsystem. Physics-based features 104 are calculated from the underlying data (e.g., based on the measured sensor values ​​102). For example, the heat transfer coefficient for the operation of the condenser subsystem may be calculated using sensor data but not directly measured. This is merely an example, and any suitable physics-based features may be used. Digital twin values ​​106 are generated using a modeled version of the physical subsystem in question. For example, a given physical subsystem may be modeled as a digital twin, and various digital twin values ​​106 may be generated using that model. Figure 6Each of these values ​​is discussed further.

[0034] In one aspect, the sensor values ​​102, physics-based features 104, and digital twin values ​​are combined at block 112. These values ​​are compared to the nominal feature values ​​122 generated at block 120 at comparator 150 to generate comparison features and feature errors 154. The comparison features and feature errors 154 are then used to identify detected anomalies and isolated faults 156.

[0035] In one aspect, the nominal characteristic value 122 may be generated at block 120 using any of a number of different techniques. For example, a given system may have multiple redundant subsystems. As one example, an aircraft may include multiple redundant cabin air compressor subsystems. These may be monitored using an aircraft condition monitoring system (ACMS). At block 130, if the subsystem in question has more than two redundant subsystems (e.g., three or more), the redundancy is used to calculate the nominal value at block 132. This is discussed below with respect to Figures 3 to 6 Further discussion.

[0036] In one aspect, if the subsystem in question has a redundancy of 2 or less, a nominal value may be calculated at block 142 or 144. For example, at block 142, the nominal value of a particular subsystem (e.g., a subsystem in the tail section of a particular aircraft) is calculated using historical data for the aircraft. This is discussed below with respect to the attached Figures 7 and 8 As another example, at block 144, nominal values ​​are calculated for subsystems of all aircraft in a given fleet at a fleet level (e.g., for a fleet of aircraft). This is discussed below with respect to the attached Figures 9 and 10 Further discussion.

[0037] Figure 2 2 shows a controller 200 for abnormal detection and fault isolation of a subsystem according to one aspect. In one aspect, the controller 200 can be used as follows: Figure 1 One or more aspects of anomaly detection and fault isolation are shown in the illustrated computing environment 100. The controller 200 includes a processor 202, a memory 210, and a network component 220. The processor 202 generally retrieves and executes programming instructions stored in the memory 210. The processor 202 is included to represent a single central processing unit (CPU), multiple CPUs, a single CPU with multiple processing cores, a graphics processing unit (GPU) with multiple execution paths, etc.

[0038] Network components 220 include components required for controller 200 to interface with components via a network (e.g., Figure 1For example, controller 200 can use network component 220 to interface with remote storage and computing nodes (eg, within computing environment 100 or external to computing environment 100).

[0039] The controller 200 can interface with other elements in the system via a local area network (LAN) (e.g., an enterprise network, a wide area network (WAN), the Internet, or any other suitable network). The network component 220 may include a wired, WiFi, or cellular network interface component and associated software to facilitate communication between the controller 200 and the communication network.

[0040] Although the memory 210 is shown as a single entity, the memory 210 may include one or more memory devices having memory blocks associated with physical addresses, such as random access memory (RAM), read-only memory (ROM), flash memory, or other types of volatile and / or non-volatile memory. The memory 210 typically includes program code for performing various functions related to the use of the controller 200. The program code is typically described as various functional "applications" or "services" within the memory 210, but alternative implementations may have different functions and / or combinations of functions. Within the memory 210, anomaly detection services 212 facilitate anomaly detection of subsystems (e.g., within an aircraft, another vehicle, or any other suitable system). Fault isolation services 214 facilitate isolation of faults within a subsystem (e.g., a subsystem in which an anomaly is detected using anomaly detection services 212). This is further discussed below with respect to subsequent figures.

[0041] although Figure 2 Anomaly detection service 212 and fault isolation service 214 are depicted as being located in memory 210, but for clarity, this representation is provided only as an illustration. More generally, controller 200 may include one or more computing platforms, such as, for example, computer servers, which may be co-located, or may form an interactively linked but distributed system, such as a cloud-based system (e.g., a public cloud, a private cloud, a hybrid cloud, or any other suitable cloud-based system). Thus, processor 202 and memory 210 may correspond to distributed processor and memory resources within a computing environment.

[0042] Figure 3Anomaly detection and fault isolation using redundancy according to one aspect is shown. In one aspect, system 300 (e.g., an aircraft, another vehicle, or any other suitable system) includes multiple redundant subsystems. For example, the aircraft may include a left cabin air compressor 310 and a right cabin air compressor 320. Each of the left cabin air compressor 310 and the right cabin air compressor 320 may itself include a redundant air compressor subsystem. For example, the left cabin air compressor 310 may include an air compressor subsystem 312 and a redundant air compressor subsystem 314. The right cabin air compressor 320 may include an air compressor subsystem 322 and a redundant air compressor subsystem 324.

[0043] In one aspect, each of the air compressor subsystems 312, 314, 322, and 324 can function redundantly with one another. Thus, the system 300 includes four redundant subsystems, allowing the use of redundant subsystems to calculate nominal values ​​for anomaly detection (e.g., as described above with respect to Figure 1 For example, a combination of subsystems with the least variability can be identified. In addition, a combination of redundant subsystems can be formed and used to detect nominal values. Local nominal values ​​can be calculated for each parameter by averaging the identified combinations, and a subsystem health index can be determined for each subsystem. The health index can be used to detect anomalies and identify subsystems with anomalies, and then the inference algorithm can be trained and used to identify the components that cause degradation. This is discussed below with respect to Figure 4 Further discussion.

[0044] Figure 4 4 is a flowchart illustrating anomaly detection and fault isolation using redundancy according to one aspect. At block 402, an anomaly detection service (e.g., Figure 2 The anomaly detection service 212 shown in FIG. 2 forms a combination of redundant subsystems, wherein the system includes more than two redundant subsystems (e.g., Figure 3 In one aspect, the anomaly detection service forms a combination by selecting a combination of a subset of redundant subsystems. For example, if a given system has three redundant subsystems A, B, and C, the anomaly detection service will form the pair {[A, B], [A, C], [B, C]}. As another example, Figure 3The illustrated system 300 includes four redundant air compressor subsystems 312, 314, 322, and 324. The anomaly detection service may form a combination of two subsystems (e.g., {[312, 314], [312, 322], [312, 324], [314, 322], [314, 324], [322, 324]}) or a combination of three subsystems (e.g., {[312, 314, 322], [312, 314, 314], [312, 322, 324], [314, 322, 324]}).

[0045] At block 404, the anomaly detection service identifies the combination of subsystems that has the least variability (e.g., the least average variability across the combination). As described above, in one aspect, the system includes more than two redundant subsystems. The anomaly detection service may calculate the variability of each of the combinations of subsystems (e.g., as discussed above with respect to block 402). The combination with the least variability may be used to identify anomalies (e.g., outliers). For example, suppose the system includes Figure 3 Four redundant air compressor subsystems are shown 312, 314, 322, and 324. If the combination [312, 314, 322] has the least variability (e.g., between the subsystems in the combination), this indicates that the remaining subsystems (e.g., air compressor subsystem 324) are outliers and abnormal.

[0046] In one aspect, an anomaly detection service uses sensor values ​​(e.g., Figure 1 sensor values ​​102 shown), calculated physics-based features (e.g., Figure 1 The physically based features 104 shown) and digital twin values ​​(e.g., Figure 1 Any combination of the digital twin values ​​106 shown in the figure can be used to identify the combination of subsystems with the least variability. For example, sensor values, calculated physics-based features, and digital twin values ​​can all be used to identify the variability of operation across subsystems. Alternatively, any suitable subset of sensor values, calculated physics-based features, and digital twin values ​​can be used. This is discussed below with respect to Figure 6 Further discussion.

[0047] At block 406, the anomaly detection service calculates a local nominal (e.g., average) for each parameter by averaging the identified combinations. In one aspect, the anomaly detection service uses the combination of redundant subsystems identified above at block 404. The anomaly detection service averages the various parameters (e.g., detected sensor values) associated with each of these subsystems across the subsystems in the combination. This produces a local nominal value while avoiding including outliers.

[0048] For example, as described above, it is assumed that the combination [312, 314, 322] has the least variability across the subsystems in the combination and is identified at block 404. The anomaly detection service can average the sensor data points across the three subsystems [312, 314, 322] to identify the local nominal for each parameter. This avoids including the air compressor subsystem 324, which may be an outlier.

[0049] At block 408, the anomaly detection service determines a health index for each redundant subsystem. For example, the anomaly detection service may calculate cosine similarity, Euclidean distance, or both for all parameters in the subsystem using the local nominal values ​​calculated at block 406. This may be used to calculate a health index for each subsystem.

[0050] At block 410, the anomaly detection service detects the anomaly and identifies the subsystem that includes the anomaly. For example, the anomaly detection service may compare subsystem behavior across multiple operating cycles (e.g., multiple flight segments of an aircraft) by calculating a health index for the subsystem for each operating cycle. If a given subsystem consistently has a low health index (e.g., the lowest health index among redundant subsystems present in the system), then the subsystem may have an anomaly.

[0051] For example, suppose the aircraft includes Figure 3 Redundant air compressor subsystems [312, 314, 322, 324] are shown. The health index of each of these subsystems can be calculated across the three flight segments using the techniques discussed above in block 408 for each flight segment. If the air compressor subsystem 314 has the lowest health index across all three flight segments, it can be identified as having an anomaly.

[0052] In one aspect, a number of actions can be taken based on the detected anomaly. For example, an alert can be created to indicate the anomaly and alert maintenance personnel to investigate or repair the subsystem (e.g., before a failure). As another example, automatic actions can be taken to disable a subsystem or repair a subsystem (e.g., automatic software or hardware correction). These are merely examples, and any suitable action can be taken.

[0053] At block 412, the anomaly detection service trains an inference algorithm (e.g., a Bayesian network). For example, historical data across flight segments may be used to train a Bayesian network to identify components that cause failures within a subsystem with an anomaly. Training may include creating a discretized feature set map with degradation labels. The Bayesian network is only one example, and any suitable inference algorithm (e.g., a suitable machine learning (ML) model) may be used. This is discussed below with respect to Figure 5 Further discussion.

[0054] At block 414, the anomaly detection service uses an inference algorithm to isolate the component causing the failure. For example, the anomaly detection service may use an inference algorithm to infer the component causing the failure using the operational data of the subsystem having the anomaly.

[0055] In one aspect, isolating components can also be used for a number of actions. For example, an alert can be created to indicate that a component caused a failure for maintenance personnel to investigate or repair the component (e.g., before the failure). As another example, automatic actions can be taken to disable the component or repair the component (e.g., automatic software or hardware correction). These are merely examples, and any suitable action can be taken.

[0056] Figure 5 An example of the creation of labeled data for fault isolation based on an inference network according to one aspect is shown. View 510 shows regions of sensor data, including a fault region 512 and a degraded region 514. In addition, view 510 shows region A and region B. On the one hand, region A shows a nominal region, while region B shows a degraded region. For example, view 510 can show sequential flight segments from left to right. In the initial A region, labeled 516A, the sensor values ​​are nominal. This indicates that the subsystem may be operating normally. In the degraded region B, labeled 514, the subsystem sensor values ​​indicate degradation before the final failure in the fault region 512. The subsystem is then repaired (e.g., the faulty component or aspect is repaired or replaced), and it operates normally again in the second A region, labeled 516B.

[0057] Table 520 shows a discretized feature set mapping used by an inference network (e.g., a Bayesian network). In one aspect, table 520 shows an example feature mapping of sensor data across multiple flight segments, as shown in view 510. Each of columns 522A-N includes degradation data, and each row corresponds to a flight segment (e.g., for a given aircraft). The table illustrates historical degradation data and can be used to train an inference algorithm for fault isolation.

[0058] In one aspect, for each feature in a flight phase (e.g., taxi, takeoff, cruise, or any other suitable flight phase) of a flight segment, an aggregate feature (e.g., a mean, standard deviation, Z-score, or any other suitable aggregate feature) is calculated. The calculated aggregate features can be discretized into various classification states (e.g., "high," "low," "healthy," "degraded," or any other suitable classification state) relative to the corresponding nominal value of each aggregate feature. A table (e.g., table 520) is prepared for all flight segments in the fleet. During the degradation phase, the component failure mode is mapped to the adjacent historical ground truth replacement for each row of the table. Once this data is prepared for the entire fleet, various multi-class classifiers, inference algorithms (e.g., Bayesian networks), or pattern-based methods can be applied to isolate degraded components.

[0059] Figure 6 is a flow chart illustrating an initial set of absolute features according to one aspect. In one aspect, Figure 6 Corresponds to Figure 4 As described above, in one aspect, an anomaly detection service (e.g., Figure 2 The anomaly detection service 212 shown uses sensor values ​​(e.g., Figure 1 sensor values ​​102 shown), calculated physics-based features (e.g., Figure 1 The physically based features 104 shown) and digital twin values ​​(e.g., Figure 1 The selected subsystem combination (e.g., as described above with respect to Figure 4 The average variability of the selected combination discussed in block 402 shown.

[0060] At block 602, the anomaly detection service identifies measured sensor parameters. In one aspect, one or more sensors may be used to measure various sensor values ​​for a given subsystem. These sensor parameters represent directly measured states of the subsystem.

[0061] At block 604, the anomaly detection service calculates one or more physics-based parameters. In one aspect, the physics-based parameters are not directly measured. Instead, the physics-based parameters are calculated using sensor data. For example, the air compressor subsystem (e.g., Figure 3 314, 322, and 324). The heat transfer coefficient is not measured directly, but calculated from measured sensor values. This is just one example, and any suitable physics-based parameter may be used.

[0062] At box 606, the anomaly detection service determines parameters from the digital twin. In one aspect, a digital twin is a virtual representation of a physical asset (e.g., enabled by data and simulators). A digital twin can be used for real-time prediction, optimization, monitoring, control, and improved decision making (among other uses). For example, a digital can be a properly synchronized host of useful information (e.g., structure, function, and behavior) of a physical entity in a virtual space, where the information flow achieves convergence between the physical and virtual states. In one aspect, a digital twin is a digital replica of an operating physical asset, having one or more of the following properties: separate, adaptable, continuous, and scalable.

[0063] In one aspect, an anomaly detection service uses (or generates) the subsystem in question (e.g., Figure 3 The Anomaly Detection service uses the digital twin to identify the parameters of the subsystem under various operating conditions and what-if scenarios. The Anomaly Detection service can use these parameters to identify the average variability across redundant subsystems, as described above for Figure 4 discussed.

[0064] Figure 7 700 is a flowchart illustrating anomaly detection and fault isolation of a subsystem in a tail section of a particular aircraft according to one aspect. In one aspect, Figures 3 to 6 Multiple redundant subsystems (e.g., more than two redundant subsystems, such as Figure 3 This corresponds to anomaly detection and fault isolation Figure 1 132 shown in FIG.

[0065] Figure 7 Corresponding to block 142, and illustrating anomaly detection of a subsystem in a particular aircraft tail section using historical data of the system (eg, multiple redundant subsystems are not required). At block 702, an anomaly detection service (eg, Figure 2 ) identifies a chain of operations for a particular system (e.g., the tail of a particular aircraft). For example, the anomaly detection service may identify a chain of flight segments (e.g., 40 flight segments) for the tail of a particular aircraft. The anomaly detection service may use historical warnings for the aircraft to identify the chain of flight segments. For example, the anomaly detection service may identify historical warnings for a particular aircraft and then identify a chain of flight segments away from the warning (e.g., before or after the warning).

[0066] At block 704, the anomaly detection service downloads sensor data for the operational chain. For example, at block 702, the anomaly detection service may identify a historical flight segment chain (e.g., 40 flight segments) for the aircraft. The anomaly detection service may download sensor data for one or more specific subsystems in the flight segment chain.

[0067] At block 706, the anomaly detection service clusters the operating phases based on the operating conditions. For example, the anomaly detection service may cluster the flight phases based on the operating conditions during the flight. The operating conditions may include altitude, speed (e.g., Mach number), throttle position, flight phase, and any other suitable operating conditions.

[0068] At block 708, the anomaly detection service computes the average, maximum, and minimum values ​​of the selected parameter across the operational phase clusters. For example, the anomaly detection service may compute the average, maximum, and minimum values ​​(or any combination thereof) of one or more parameters of the subsystem in question across various operational phase clusters.

[0069] At block 710, the anomaly detection service identifies outliers (if any) and removes them from the data. For example, the anomaly detection service may identify outlier parameter values ​​using the mean, maximum, and minimum values ​​calculated at block 708. This may include parameter values ​​that differ from the mean based on statistical analysis.

[0070] At block 712, the anomaly detection service determines if a local nominal exists. In one aspect, a local nominal is a configuration parameter defined by a user. If yes, the process proceeds to block 716. If no, the process proceeds to block 714. At block 714, the anomaly detection service considers sensor data in all chains of the system to identify anomalies. At block 716, the anomaly detection service considers sensor data on the most recent chain (e.g., as opposed to all chains). At block 718, the anomaly detection service recalculates the average, maximum, and minimum values ​​(or any combination thereof) of the parameter across all operational clusters, considering only the selected most recent chain.

[0071] On the one hand, Figure 7 The technique shown is used to identify the nominal value of the subsystem in question (e.g., based on the historical operation of the particular system, rather than on the Figures 3 to 6 After identifying the local nominals, the same techniques discussed above with respect to blocks 408 to 414 can be used for anomaly detection and fault isolation. For example, the anomaly detection service can use the nominal values ​​to determine a health index for the subsystem in question, detect anomalies and identify subsystems with anomalies, train an inference algorithm for fault isolation, and use the inference algorithm to isolate the component that caused the fault. Figure 4 These are discussed further in blocks 408, 410, 412, and 414 shown in FIG.

[0072] Figure 8 Further illustrated is anomaly detection and fault isolation of a subsystem in a tail section of a particular aircraft according to one aspect. In one aspect, Figure 8The characteristic (e.g., parameter or combination of parameters) of a subsystem across multiple flight segments is shown. The y-axis represents the characteristic value for a given flight segment, while the x-axis represents the number of flight segments. Line 814 represents the nominal value of the system (e.g., using the above description of Figure 7 The variation of the graph from the nominal value represented by line 814 is used to identify anomalies in the operation of the subsystem.

[0073] For example, at 810, a subsystem degradation or failure is corrected. A maintenance message may be triggered, a part may be replaced, or any other suitable correction may be made. After 810, the plotted characteristic (e.g., a subsystem parameter or combination of parameters) increases, indicating improved operation. At 812, a degradation alert is triggered (e.g., when the graphical characteristic decreases). In one aspect, at 816, an average (e.g., plus or minus three standard deviations) of the graphical characteristic based on the number of flight segments (e.g., based on 15 flight segments) is calculated.

[0074] Fig. 9 A flowchart 900 is shown for anomaly detection and fault isolation of subsystems of all aircraft in a fleet according to one aspect. In one aspect, Figures 3 to 6 Multiple redundant subsystems (e.g., more than two redundant subsystems, such as Figure 3 This corresponds to anomaly detection and fault isolation Figure 1 132 shown in FIG.

[0075] Fig. 9 Corresponding to block 144, and illustrating anomaly detection of subsystems of all aircraft in a fleet using historical data of the fleet (e.g., without requiring multiple redundant subsystems). At block 902, an anomaly detection service (e.g., Figure 2 ) identifies chains of operations for a system of a fleet (e.g., aircraft of a fleet). For example, the anomaly detection service may identify chains of flight segments for a fleet of aircraft. The anomaly detection service may use historical warnings for the fleet to identify chains of flight segments. For example, the anomaly detection service may identify historical warnings for aircraft in the fleet and then identify chains of flight segments away from the warning (e.g., before or after the warning).

[0076] At block 904, the anomaly detection service downloads sensor data for the operational chain. For example, at block 902, the anomaly detection service may identify a historical flight segment chain for a fleet of aircraft. The anomaly detection service may download sensor data for one or more specific subsystems in the flight segment chain.

[0077] At block 906, the anomaly detection service clusters the operating phases based on the operating conditions. For example, the anomaly detection service may cluster the flight phases based on the operating conditions during the flight. The operating conditions may include altitude, speed (e.g., Mach number), throttle position, flight phase, and any other suitable operating conditions.

[0078] At block 908, the anomaly detection service computes the average, maximum, and minimum values ​​of the selected parameter across the operational phase clusters. For example, the anomaly detection service may compute the average, maximum, and minimum values ​​(or any combination thereof) of one or more parameters of the subsystem in question across various operational phase clusters.

[0079] At block 910, the anomaly detection service identifies outliers (if any) and removes them from the data. For example, the anomaly detection service may identify outlier parameter values ​​using the mean, maximum, and minimum values ​​calculated at block 908. This may include parameter values ​​that differ from the mean based on statistical analysis.

[0080] At block 912, the anomaly detection service recalculates the average, maximum, and minimum values ​​(or any combination thereof) of the parameter across all operational clusters. For example, at block 910, the anomaly detection service removes outliers from the data. The anomaly detection service can then recalculate the average, maximum, and minimum values ​​(e.g., without the outlier data).

[0081] On the one hand, Fig. 9 The technique shown is used to identify nominal values ​​for the subsystem in question (e.g., based on historical operation of the system for a fleet, rather than Figures 3 to 6 After the nominal values ​​are calculated at the fleet level, the same techniques discussed above with respect to blocks 408 to 414 can be used for anomaly detection and fault isolation. For example, the anomaly detection service can use the nominal values ​​to determine a health index for the subsystem in question, detect anomalies and identify subsystems with anomalies, train an inference algorithm for fault isolation, and use the inference algorithm to isolate the component that caused the fault. Figure 4 These are discussed further in blocks 408, 410, 412, and 414 shown in FIG.

[0082] Fig.10 Further illustrated is anomaly detection and fault isolation of subsystems for all aircraft in a fleet according to one aspect. View 1010 shows statistics (e.g., average, maximum, minimum) for each parameter of each nominal chain for historical flight segments across aircraft in the fleet. In one aspect, entry 1012 illustrates a nominal value (e.g., using a Fig. 9), while entry 1014 illustrates a value that is different from the nominal value and can be used for anomaly detection. View 1020 shows a flight segment chain for an aircraft representing the most likely nominal flight segment, as described above with respect to Fig. 9 As discussed in block 902 shown in .

[0083] This application involves the following terms:

[0084] Clause 1. A method comprising:

[0085] identifying a plurality of redundant subsystems in the system (130), the plurality of redundant subsystems being three or more redundant subsystems;

[0086] forming a plurality of combinations (402) of the plurality of redundant subsystems, each combination being associated with a subset of the plurality of redundant subsystems;

[0087] identifying, from the plurality of combinations of redundant subsystems, a first combination having a minimum variability among the subsystems (404);

[0088] calculating one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of redundant subsystems using the first combination (406); and

[0089] An anomaly in the first subsystem is detected based on the calculated one or more nominal values ​​(410).

[0090] Clause 2. The method of clause 1, wherein the system comprises an aircraft, and the three or more redundant subsystems comprise redundant subsystems (130) in the aircraft.

[0091] Clause 3. A method according to clause 2, wherein identifying the first combination having the minimum variability is based on measured sensor values ​​(102) of one or more subsystems and at least one of: (i) calculated physically-based parameters (104) of the one or more subsystems; and (ii) parameters from a digital twin (106) associated with the one or more subsystems.

[0092] Clause 4. A method according to clause 3, wherein identifying the first combination having the minimum variability is based on both the calculated physically-based parameters (104) of the one or more subsystems and parameters from the digital twin (106) associated with the one or more subsystems.

[0093] Clause 5. The method of clause 2, wherein calculating the one or more nominal values ​​of one or more parameters of the first subsystem of the plurality of redundant subsystems using the first combination comprises:

[0094] Each of the one or more parameters is averaged across subsystems in the first combination (406).

[0095] Clause 6. The method of clause 5, wherein identifying the first combination having the minimum variability among the subsystems from the plurality of combinations of redundant subsystems further identifies one or more outliers excluded from the first combination (404).

[0096] Clause 7. The method of clause 2, wherein detecting the anomaly in the first subsystem based on the calculated one or more nominal values ​​comprises:

[0097] A health index of the first subsystem is determined based on calculating at least one of a cosine similarity and a Euclidean distance of a plurality of parameters in the subsystem compared to the calculated one or more nominal values ​​(408).

[0098] Clause 8. The method of clause 7, wherein determining the health index of the first subsystem is based on calculating both the cosine similarity and the Euclidean distance of the plurality of parameters in the subsystem compared to one or more calculated nominal values ​​(408).

[0099] Clause 9. The method according to clause 2, further comprising:

[0100] A first component at least partially responsible for an anomaly in the first subsystem is identified (414).

[0101] Clause 10. The method of clause 9, wherein identifying the first component at least partially responsible for the anomaly in the first subsystem further comprises:

[0102] The first component is identified using an inference algorithm based on the anomaly (412, 414).

[0103] Clause 11. The method of clause 10, wherein the reasoning algorithm is a Bayesian network (412, 414) trained based on historical data related to operation of the aircraft.

[0104] Clause 12. A system, comprising:

[0105] one or more processors (202); and

[0106] One or more memories (210) storing a program that, when executed on any combination of the one or more processors, performs operations including:

[0107] identifying a plurality of redundant subsystems in the system (130), the plurality of redundant subsystems being three or more redundant subsystems;

[0108] forming a plurality of combinations (402) of the plurality of redundant subsystems, each combination being associated with a subset of the plurality of redundant subsystems;

[0109] identifying, from the plurality of combinations of redundant subsystems, a first combination having a minimum variability among the subsystems (404);

[0110] calculating one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of redundant subsystems using the first combination (406); and

[0111] An anomaly in the first subsystem is detected based on the calculated one or more nominal values ​​(410).

[0112] Clause 13. The system of clause 12, wherein the system comprises an aircraft, and the three or more redundant subsystems comprise redundant subsystems (130) in the aircraft.

[0113] Clause 14. A system according to clause 13, wherein identifying the first combination having the minimum variability is based on measured sensor values ​​(102) of one or more subsystems and at least one of: (i) calculated physically-based parameters (104) of the one or more subsystems; and (ii) parameters from a digital twin (106) associated with the one or more subsystems.

[0114] Clause 15. The system of clause 13, wherein:

[0115] Wherein, using the first combination to calculate the one or more nominal values ​​of one or more parameters of the first subsystem in the plurality of redundant subsystems comprises:

[0116] averaging each of the one or more parameters across subsystems in the first combination (406); and

[0117] Wherein, detecting the abnormality in the first subsystem based on the calculated one or more nominal values ​​comprises:

[0118] A health index of the first subsystem is determined based on calculating at least one of a cosine similarity and a Euclidean distance of a plurality of parameters in the subsystem compared to the calculated one or more nominal values ​​(408).

[0119] Clause 16. The system according to clause 13, further comprising:

[0120] Identifying a first component (414) at least partially responsible for an anomaly in the first subsystem includes:

[0121] The first component is identified using an inference algorithm based on the anomaly (412, 414).

[0122] Clause 17. A computer program product, the computer program product comprising:

[0123] A non-transitory computer-readable storage medium (210) having computer-readable program code embodied thereon, the computer-readable program code being executable by one or more computer processors to perform operations comprising:

[0124] identifying a plurality of redundant subsystems in the system (130), the plurality of redundant subsystems being three or more redundant subsystems;

[0125] forming a plurality of combinations (402) of the plurality of redundant subsystems, each combination being associated with a subset of the plurality of redundant subsystems;

[0126] identifying, from the plurality of combinations of redundant subsystems, a first combination having a minimum variability among the subsystems (404);

[0127] calculating one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of redundant subsystems using the first combination (406); and

[0128] An anomaly in the first subsystem is detected based on the calculated one or more nominal values ​​(410).

[0129] Clause 18. The computer program product of Clause 17, wherein the system comprises an aircraft, and the three or more redundant subsystems comprise redundant subsystems (130) in the aircraft.

[0130] Clause 19. A computer program product according to clause 18,

[0131] Wherein, using the first combination to calculate the one or more nominal values ​​of one or more parameters of the first subsystem in the plurality of redundant subsystems comprises:

[0132] averaging each of the one or more parameters across subsystems in the first combination (406); and

[0133] Wherein, detecting the abnormality in the first subsystem based on the calculated one or more nominal values ​​comprises:

[0134] A health index of the first subsystem is determined based on calculating at least one of a cosine similarity and a Euclidean distance of a plurality of parameters in the subsystem compared to the calculated one or more nominal values ​​(408).

[0135] Clause 20. The computer program product of Clause 18, further comprising:

[0136] Identifying a first component (414) at least partially responsible for an anomaly in the first subsystem includes:

[0137] The first component is identified using an inference algorithm based on the anomaly (412, 414).

[0138] In the present disclosure, reference is made to various aspects. However, it should be understood that the present disclosure is not limited to the specifically described aspects. On the contrary, it is expected that any combination of the following features and elements (whether or not related to different aspects) is implemented and practiced to provide the teachings provided herein. In addition, when describing the elements of various aspects in the form of "at least one of A and B", it will be understood that only element A, only element B, and all aspects including elements A and B are expected. In addition, although some aspects can achieve advantages that are superior to other possible solutions and / or superior to the prior art, whether a given aspect achieves a specific advantage does not limit the present disclosure. Therefore, the aspects, features, aspects, and advantages disclosed herein are merely illustrative and are not considered to be elements or limitations of the appended claims unless clearly stated in the claims. Similarly, reference to "the present invention" should not be interpreted as a summary of any inventive subject matter disclosed herein, and should not be considered to be an element or limitation of the appended claims unless clearly stated in the claims.

[0139] As will be appreciated by those skilled in the art, the aspects described herein may be implemented as a system, method, or computer program product. Therefore, the various aspects may take the form of complete hardware aspects, complete software aspects (including firmware, resident software, microcode, etc.), or combined software and hardware aspects, which may all be generally referred to herein as "circuits," "modules," or "systems." In addition, the aspects described herein may take the form of a computer program product, which is implemented in one or more computer-readable storage media, on which a computer-readable program code is implemented.

[0140] Program code embodied on a computer readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0141] The computer program code for performing the operations of various aspects of the present disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​(such as Java, Smalltalk, C++, etc.) and conventional procedural programming languages ​​(such as "C" programming language or similar programming languages). The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer or entirely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any type of network (the network includes a local area network (LAN) or a wide area network (WAN)), or can be connected to an external computer (e.g., by using the Internet of an Internet service provider).

[0142] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products of various aspects of the present disclosure. It should be understood that each frame of the flowchart and / or block diagram and the combination of frames in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device create a device for implementing the function / action specified in the frame of the flowchart and / or block diagram.

[0143] These computer program instructions may also be stored in a computer-readable medium, which may direct a computer, other programmable data processing apparatus, or other device to work in a specific manner so that the instructions stored in the computer-readable medium produce an article of manufacture including instructions for implementing the functions / actions specified in the boxes of the flowchart illustrations and / or block diagrams.

[0144] The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operating steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, so that the instructions executed on the computer, other programmable data processing apparatus, or other device provide a process for implementing the functions / actions specified in the boxes of the flowchart illustrations and / or block diagrams.

[0145] The flowchart illustration and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various aspects of the present disclosure. In this regard, each frame in the flow chart or block diagram can represent a module, segment or part of a code, which includes one or more executable instructions for realizing a specified logical function. It should also be noted that in some alternative embodiments, the function pointed out in the frame may not occur in the order pointed out in the figure. For example, according to the function involved, the two frames shown in succession can actually be performed substantially simultaneously, or the frame can sometimes be performed in reverse order or out of order. It should also be noted that each frame in the block diagram and / or flow chart and the combination of frames in the block diagram and / or flow chart can be realized by a combination of a system based on special-purpose hardware or special-purpose hardware and computer instructions that performs a specified function or action.

[0146] While the foregoing is directed to various aspects of the present disclosure, other and further aspects of the disclosure may be devised without departing from the basic scope thereof and the scope of the disclosure is determined by the appended claims.

Claims

1. A method, comprising: identifying a plurality of redundant subsystems in the system (130), the plurality of redundant subsystems being three or more redundant subsystems; forming a plurality of combinations (402) of the plurality of redundant subsystems, each combination being associated with a subset of the plurality of redundant subsystems; identifying, from the plurality of combinations of redundant subsystems, a first combination having a minimum variability among the subsystems (404); calculating one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of redundant subsystems using the first combination (406); as well as An anomaly in the first subsystem is detected based on the calculated one or more nominal values ​​(410).

2. The method according to claim 1, wherein: The system includes an aircraft, and the three or more redundant subsystems include redundant subsystems (130) in the aircraft.

3. The method according to claim 2, wherein: Identifying the first combination having the minimum variability is based on measured sensor values ​​(102) of one or more subsystems and at least one of: (i) calculated physics-based parameters (104) of the one or more subsystems; and (ii) parameters from a digital twin (106) associated with the one or more subsystems.

4. The method according to claim 3, wherein: Identifying the first combination having the minimum variability is based on both the calculated physics-based parameters (104) of the one or more subsystems and parameters from the digital twin (106) associated with the one or more subsystems.

5. The method according to claim 2, wherein: Calculating the one or more nominal values ​​of one or more parameters of the first subsystem of the plurality of redundant subsystems using the first combination includes: Each of the one or more parameters is averaged across subsystems in the first combination (406).

6. The method according to claim 5, wherein: Identifying the first combination having the minimum variability among the subsystems from the plurality of combinations of redundant subsystems further identifies one or more outliers excluded from the first combination (404).

7. The method according to claim 2, wherein: Detecting the anomaly in the first subsystem based on the calculated one or more nominal values ​​comprises: A health index of the first subsystem is determined based on calculating at least one of a cosine similarity and a Euclidean distance of a plurality of parameters in the subsystem compared to the calculated one or more nominal values ​​(408).

8. The method according to claim 7, wherein: Determining the health index of the first subsystem is based on calculating both the cosine similarity and the Euclidean distance of the plurality of parameters in the subsystem compared to the calculated one or more nominal values ​​(408).

9. A system, comprising: one or more processors (202); as well as One or more memories (210) storing programs that, when executed on any combination of the one or more processors, perform operations including: identifying a plurality of redundant subsystems in the system (130), the plurality of redundant subsystems being three or more redundant subsystems; forming a plurality of combinations (402) of the plurality of redundant subsystems, each combination being associated with a subset of the plurality of redundant subsystems; identifying, from the plurality of combinations of redundant subsystems, a first combination having a minimum variability among the subsystems (404); calculating one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of redundant subsystems using the first combination (406); and An anomaly in the first subsystem is detected based on the calculated one or more nominal values ​​(410).

10. A computer program product, the computer program product comprising: A non-transitory computer-readable storage medium (210) having computer-readable program code embodied thereon, the computer-readable program code being executable by one or more computer processors to perform operations comprising: identifying a plurality of redundant subsystems in the system (130), the plurality of redundant subsystems being three or more redundant subsystems; forming a plurality of combinations (402) of the plurality of redundant subsystems, each combination being associated with a subset of the plurality of redundant subsystems; identifying, from the plurality of combinations of redundant subsystems, a first combination having a minimum variability among the subsystems (404); calculating one or more nominal values ​​of one or more parameters of a first subsystem of the plurality of redundant subsystems using the first combination (406); and An anomaly in the first subsystem is detected based on the calculated one or more nominal values ​​(410).

Images