Method, medium, aircraft for detecting anomalies, and method for training anomaly detection systems

Abstract

The present application relates to a method, medium, aircraft, and method for training anomaly detection system for detecting anomalies. The method includes acquiring sensor data captured by a sensor of an aircraft during a power-on event. The sensor data includes a plurality of parameter values, each parameter value corresponding to a sampling period. The method also includes determining a set of incremental values, each incremental value indicating a difference between parameter values ​​of consecutive sampling periods of the sensor data. The method also includes determining a set of quantized incremental values ​​by assigning the incremental values ​​to quantization bins based on the size of the incremental values. The method also includes determining a normalized count of the incremental values ​​for each quantization bin. The method also includes comparing the normalized count of the incremental values ​​with an anomaly detection threshold. The method also includes generating an output indicating whether the sensor data indicates an operational anomaly based on the comparison.

Description

Technical Field

[0001] The present disclosure generally relates to anomaly prediction and detection of aircraft equipment. Background Art

[0002] Modern aircraft are extremely complex, with many integrated systems interacting to provide features such as increased efficiency, improved safety, and enhanced performance. When a failure occurs in any of these systems, a significant amount of time and resources can be expended to repair the failure and prepare the aircraft for further operation. In addition, performing unplanned maintenance to remedy a failure condition can reduce the operational availability of the aircraft. For example, the parts, equipment, or technicians required to perform the unplanned maintenance may not be at the aircraft location when a failure condition occurs. In this case, the parts, equipment, or technicians may have to be transported to the aircraft, resulting in significant delays.

[0003] Aircraft operators schedule many maintenance activities in an attempt to avoid such situations. Scheduled maintenance is performed according to a schedule that is targeted, at least in part, at avoiding failures. While scheduled maintenance can avoid many failure conditions, it is inefficient because components that appear to be operating normally (e.g., not experiencing a failure) are replaced. Summary of the invention

[0004] In a particular embodiment, an aircraft includes a sensor configured to capture sensor data during a power-on event. The sensor data includes a plurality of parameter values, and each parameter value corresponds to a corresponding sampling period during the power-on event. The aircraft also includes one or more processors and a storage device, the storage device being configured to store the sensor data and instructions executable by the one or more processors to detect an operational anomaly of aircraft equipment associated with the sensor. When the one or more processors execute the instructions, the instructions cause the one or more processors to perform operations including determining a set of incremental values. Each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value in a pair of parameter values ​​corresponding to a continuous sampling period of the sensor data. The operations also include determining a set of quantized incremental values ​​by assigning the incremental values ​​from the set of incremental values ​​to a quantization bin based on the size of the incremental values. The operations also include determining a normalized count of the incremental values ​​for each quantization bin. The operations also include performing a comparison of the normalized incremental value count of a particular quantization bin with an anomaly detection threshold of the particular quantization bin. The operations also include generating an output based on the comparison, the output indicating whether the sensor data indicates an operational anomaly of the aircraft equipment.

[0005] In another specific embodiment, a method includes acquiring sensor data captured by a sensor of an aircraft during a power-on event at a computing device. The sensor data includes a plurality of parameter values, and each parameter value corresponds to a corresponding sampling period during the power-on event. The method also includes determining, by the computing device, a set of incremental values. Each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value in a pair of parameter values ​​corresponding to consecutive sampling periods of the sensor data. The method also includes determining, by the computing device, a set of quantized incremental values ​​by assigning the incremental values ​​in the set of incremental values ​​to quantization bins based on the magnitude of the incremental values. The method also includes determining, by the computing device, a normalized count of the incremental values ​​for each quantization bin. The method also includes performing, by the computing device, a comparison of the normalized count of the incremental values ​​for a particular quantization bin with an anomaly detection threshold for the particular quantization bin. The method also includes generating, by the computing device, an output based on the comparison, the output indicating whether the sensor data indicates an operational anomaly of aircraft equipment associated with the sensor.

[0006] In another specific embodiment, a non-transitory computer-readable medium stores instructions that, when executed by a processor, cause the processor to initiate, perform, or control operations. The operations include acquiring sensor data captured by a sensor of an aircraft during a power-on event. The sensor data includes a plurality of parameter values, and each parameter value corresponds to a corresponding sampling period during the power-on event. The operations also include determining a set of incremental values. Each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value in a pair of parameter values ​​corresponding to consecutive sampling periods of the sensor data. The operations also include determining a set of quantized incremental values ​​by assigning the incremental values ​​in the set of incremental values ​​to quantization bins based on the magnitude of the incremental values. The operations also include determining a normalized count of the incremental values ​​for each quantization bin. The operations also include performing a comparison of the normalized incremental value count of a particular quantization bin with an anomaly detection threshold for the particular quantization bin. The operations also include generating an output based on the comparison, the output indicating whether the sensor data indicates an operational anomaly of aircraft equipment associated with the sensor.

[0007] In another specific embodiment, a method for training an anomaly detection system includes acquiring, at a computing device, a plurality of sets of sensor data captured by onboard sensors of one or more aircraft during a plurality of power-on events. Each set of sensor data corresponds to a respective power-on event of a respective aircraft and includes a plurality of parameter values. Each parameter value corresponds to a respective sampling period during the respective power-on event. The method further includes determining, by the computing device, a set of incremental values ​​for each set of sensor data. Each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value in a pair of parameter values ​​from consecutive sampling periods corresponding to the set of sensor data. The method further includes determining, by the computing device, a set of quantized incremental values ​​by assigning the incremental values ​​of the set of incremental values ​​to quantized bins based on the magnitude of the incremental values. The method further includes determining, by the computing device, a normalized count of the incremental values ​​for each quantized bin. The method further includes setting an anomaly detection threshold for each quantized bin based on the normalized count of the incremental values ​​and fault indication data associated with the one or more aircraft.

[0008] The features, functions, and advantages described herein can be achieved independently in various embodiments or may be combined in yet other embodiments, further details of which can be found with reference to the following description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 is a diagram illustrating an example of a system including one or more aircraft and an anomaly detection system;

[0010] Figure 2 It is shown by Figure 1 A diagram of details of operations performed by an anomaly detection system to detect anomalies based on sensor data;

[0011] Figure 3 Is shown as training Figure 1 A diagram of details of operations performed by an anomaly detection system;

[0012] Figure 4 is used Figure 1 A flowchart of an example of a method for detecting anomalies by an anomaly detection system;

[0013] Figure 5 Is for Figure 1 A flowchart of an example of a method for determining an anomaly detection threshold by an anomaly detection system;

[0014] Figure 6 is a block diagram of a computing environment including a computing device configured to support aspects of the computer-implemented method and computer-executable program instructions (or code) according to the present disclosure;

[0015] Figure 7 Yes Figure 1A flow chart of an example of an anomaly detection system for an aircraft life cycle;

[0016] Figure 8 Yes Figure 1 Block diagram of the anomaly detection system for an aircraft. DETAILED DESCRIPTION

[0017] The disclosed systems and methods are capable of detecting abnormal operation of aircraft equipment, which may occur before an actual fault condition is experienced. Detecting anomalies before a fault condition occurs may allow aircraft operators to schedule maintenance more effectively and efficiently. For example, rather than replacing components based on flight time (which may be wasteful if the component does not actually need to be replaced), a component may be replaced when an anomaly detection system indicates that the component will soon need to be replaced, but before the component experiences a fault. As a result, inefficiencies caused by replacing components that do not need to be replaced and inefficiencies caused by the aircraft experiencing a fault condition are avoided or reduced.

[0018] The accompanying drawings and the following description illustrate specific exemplary embodiments. It should be understood that those skilled in the art will be able to design various arrangements that, although not explicitly described or shown herein, embody the principles described herein and are included within the scope of the technical solutions that follow this specification. In addition, any examples described herein are intended to aid in understanding the principles of the present disclosure and are to be interpreted as not limiting. Therefore, the present disclosure is not limited to the specific embodiments or examples described below, but is defined by the technical solutions and their equivalents.

[0019] Specific embodiments are described herein with reference to the accompanying drawings. In the specification, common features are represented by common reference numerals in all drawings. In some figures, multiple instances of a particular type of feature are used. Although the features are physically and / or logically different, the same reference numeral is used for each feature, and different instances are distinguished by adding letters to the reference numerals. When features are referred to herein as a group or a type (for example, when a specific one is not mentioned), the reference numerals used are without distinguishing letters. However, when a specific feature of multiple features of the same type is mentioned herein, the reference numeral is used with a distinguishing letter. For example, with reference to Figure 3 , multiple examples of aircraft are also shown and associated with reference numerals 102A and 102K. When referring to one of these aircraft, a distinguishing letter is used. For illustration, when referring to the first aircraft 102A, the distinguishing letter "A" is used. However, when referring to any arbitrary aircraft or aircraft as a group, reference numeral 102 is used without a distinguishing letter.

[0020] As used herein, various terms are only used for the purpose of describing specific embodiments and are not intended to be limited. For example, the singular forms "a", "an" and "the" are intended to also include plural forms, unless the context clearly indicates otherwise. In addition, the terms "comprise", "comprises" and "comprising" can be used interchangeably with "include", "includes" or "including". In addition, the term "wherein" can be used interchangeably with the term "where". As used herein, "exemplary" represents examples, embodiments and / or aspects, and should not be interpreted as limiting or expressing preferences or preferred embodiments. As used herein, ordinal terms (for example, "first", "second", "third", "Kth", "Nth", "Pth", etc.) are used to modify elements, for example, structures, components, operations, etc., which themselves do not represent any priority or order of the element relative to another element, but only distinguish the element from another element with the same name (but using ordinal terms). As used herein, the term "group" refers to the grouping of one or more elements, and the term "multiple" refers to multiple elements.

[0021] As used herein, "generating," "calculating," "using," "selecting," "accessing," and "determining" are interchangeable unless the context indicates otherwise. For example, "generating," "calculating," or "determining" a parameter (or signal) may refer to actively generating, calculating, or determining a parameter (or signal), or may refer to using, selecting, or accessing a parameter (or signal) that has already been generated, for example, by another component or device.

[0022] As used herein, "coupling" may include "communicative coupling", "electrical coupling" or "physical coupling", and may also (or alternatively) include any combination thereof. Two devices (or components) may be directly or indirectly coupled (e.g., communicative coupling, electrical coupling or physical coupling) via one or more other devices, components, wires, buses, networks (e.g., wired networks, wireless networks or combinations thereof), etc. As illustrative, non-limiting examples, two electrically coupled devices (or components) may be included in the same device or in different devices, and may be connected via electronic devices, one or more connectors or inductive coupling. In some embodiments, two devices (or components) that are communicatively coupled, such as in electrical communication, may directly or indirectly send and receive electrical signals (digital signals or analog signals), for example, via one or more wires, buses, networks, etc. As used herein, "direct coupling" is used to describe two coupled (e.g., communicatively coupled, electrically coupled or physically coupled) devices without intermediate components.

[0023] Figure 11 is a diagram illustrating an example of a system 100 including one or more aircraft 102 and an anomaly detection system 104. Aircraft 102 may be a rotorcraft 102A, a fixed-wing aircraft 102B, or another aircraft. Aircraft 102 includes aircraft equipment 106, such as one or more components of a propulsion system, one or more components of a hydraulic system, one or more components of an electrical system, one or more components of an environmental system, and the like.

[0024] Aircraft 102 also includes one or more sensors 108 associated with aircraft equipment 106. Sensors 108 are configured to generate sensor data 114. Sensor data 114 includes a plurality of data samples of time series 116. Each data sample indicates a corresponding parameter value of time series 116. For example, in Figure 1 In the example, the time series 116 of data samples includes a first parameter value 118 (labeled as “PV1”), a second parameter value 120 (labeled as “PV2”), and a third parameter value 122 (labeled as “PV p ”). The first parameter value 118 corresponds to a value detected by the particular one of the sensors 108 during a first sampling period, the second parameter value 120 corresponds to a value detected by the particular one of the sensors 108 during a second sampling period, and the Pth parameter value 122 corresponds to a value detected by the particular one of the sensors 108 during a Pth sampling period (where P is an integer greater than 2). The sampling periods are consecutive. For example, the Pth sampling period follows the second sampling period, which follows the first sampling period. The time series of data samples 116 may also include one or more additional parameter values ​​(e.g., between the second parameter value 120 and the Pth parameter value 122), which are not shown for ease of illustration. The specific measurement value or information represented by each parameter value 118-122 depends on the type of sensor 108 that generated the sensor data 114. For example, the parameter values ​​118-122 may indicate temperature, pressure, rotation speed, velocity, position, humidity, vibration rate, or any other information associated with the operation or status of the aircraft equipment 106.

[0025] The sampling periods of time series 116 may be evenly spaced (e.g., periodic) or unevenly spaced (e.g., non-periodic). For example, one sensor 108 may collect one data sample every 10 milliseconds or at some other sampling rate to generate evenly spaced data samples. As another example, one sensor 108 may collect one data sample each time a state of aircraft equipment 106 changes (or changes by a certain amount) to generate unevenly spaced data samples of time series 116. As yet another example, one sensor 108 may collect data samples at different sampling rates at different times, e.g., at a first sampling rate while in flight and at a second sampling rate when powered on on the ground.

[0026] In some implementations, the sensor data 114 includes data samples from two or more sensors 108. In such implementations, the sensor data 114 may include two or more time series 116 (e.g., one for each sensor), and each of the two or more time series 116 may be evenly spaced or unevenly spaced. If the two or more time series 116 are evenly spaced, the sampling rates associated with the two or more time series 116 may be equal to each other (e.g., both sensors may have a sampling rate of 15 Hz), or the sampling rates associated with the two or more time series 116 may be unequal (e.g., one sensor may have a sampling rate of 15 Hz, while another sensor may have a sampling rate of 20 KHz).

[0027] In some embodiments, the sensor data 114 includes metadata in addition to the time series 116 of data samples indicating parameter values ​​118-122. For example, the sensor data 114 may include time index information (e.g., a timestamp or sequence number) associated with the time series 116, a sensor identifier indicating which sensor 108 generated the sensor data 114, or other information related to the sensor 108, the sensor data 114, the aircraft 102, the aircraft equipment, or the time series 116. The sensor data 114 may be represented via a digital signal or an analog signal. For example, the sensor data 114 may be sent as one or more digitally encoded packets transmitted via a digital communication protocol. As another example, the sensor data 114 may be encoded in one or more analog electrical signals, wherein a signal characteristic (e.g., amplitude, frequency, or phase) of the analog electrical signal represents a parameter value.

[0028] Sensor data 114 is provided to anomaly detection system 104, which is configured to analyze sensor data 114 to predict or detect operational anomalies associated with aircraft equipment 106. Figure 1 , anomaly detection system 104 is shown as being separate from aircraft 102. For example, in Figure 1 In the embodiment, the anomaly detection system 104 and the aircraft 102 communicate via the network 124. However, in some embodiments, the anomaly detection system 104 is integrated into the aircraft 102. For example, the anomaly detection system 104 may be included in the health monitoring system of the aircraft 102 (e.g., Figure 8 Health monitoring system 816 shown.

[0029] exist Figure 1In the example shown, the anomaly detection system 104 includes, is included in, or corresponds to a computing device. For example, the anomaly detection system 104 includes one or more processors 128, one or more storage devices 130, and an interface 132 (e.g., a data port, a wireless communication system, or another communication interface or an input / output interface). The interface 132 is configured to receive the sensor data 114 and send the output 168 generated by the anomaly detection system 104 to an onboard device 110 of the aircraft 102, an offboard device 126 (e.g., a maintenance computer system), or both. In other embodiments, the anomaly detection system 104 is implemented using a dedicated circuit (e.g., an application specific integrated circuit), a programmed field programmable gate array, or a combination of the above.

[0030] exist Figure 1 In the example shown, storage device 130 stores instructions 134 executable by processor 128 to analyze sensor data 114 and generate output 168. Figure 1 In the example shown, the instruction 134 includes various functional modules, including: a delta value calculator 138, a delta value quantizer 140, a delta value counter 142, a normalization factor calculator 144, a count normalizer 146, an anomaly detector 148, and an output generator 150. In the following description, the functional modules of the instruction 134 are described to illustrate a specific embodiment of the anomaly detection system 104. In other embodiments, the instruction 134 includes more functional modules, fewer functional modules, or different functional modules. For example, in some embodiments, two or more functional modules are combined in a single module. For illustration, the delta value quantizer 140 and the delta value counter 142 can be combined in a histogram generator (not shown). In another example, one or more functional modules can be divided into two or more different modules. For illustration, the anomaly detector 148 can be divided into two different anomaly detectors that use different detection thresholds, for example, a first detection threshold that generates an alert (e.g., a maintenance alert) sent to the offboard device 126 and a second detection threshold that generates an alert (e.g., a crew alert) sent to the onboard device 110. In yet another example, in some embodiments, dedicated hardware rather than software instructions can be used to perform one or more of the functions described below. For illustration, dedicated circuitry can be used to perform the operations of the delta value calculator 138 by delaying a sample and subtracting the sample from a subsequent sample.

[0031] exist Figure 1In the example shown, the delta value calculator 138 may be executed by the processor 128 to determine delta values ​​160 based on the sensor data 114. Each delta value 160 corresponds to a difference between parameter values ​​of two consecutive data samples of the time series 116. For illustration, the first delta value ("DV1") may be calculated using the equation DV1 = PV2 - PV1, and the P-1 delta value ("DV2") may be calculated using the equation DV1 = PV2 - PV1. P-1 ”) can be calculated as DV P-1 =PV P -PV P-1 In some implementations, the delta value calculator 138 ignores, discards, or eliminates the sign of each delta value 160 . For example, the first delta value ( DV1 ) may be calculated using the equation DV1 = |PV2 −PV1|. In this example, each delta value 160 represents the magnitude of the difference between adjacent parameter values ​​of the time series 116 .

[0032] The delta value quantizer 140 may be executed by the processor 128 to assign the delta values ​​160 into quantization bins (also referred to herein simply as “bins”) to generate quantized delta values ​​162. Each bin represents a range of delta values, and the specific range of delta values ​​represented by each bin and the number of bins for the sensor data 114 from each sensor 108 are indicated by the bin data 154 in the configuration data 136 stored in the storage device 130. By reference to Figure 3 and Figure 5 The training process described above determines the bin data 154. In the example of assigning the delta values ​​160 to bins, the bin data 154 can specify evenly spaced bins, each bin corresponding to a range of delta values ​​of 5 units. In this example, if a particular delta value (e.g., the first delta value DV1 from the example above) has a value of 4.5, then the particular delta value is assigned to the first bin (corresponding to a range from 0 to 5 units), and if the particular delta value has a value of 32.1, then the particular delta value is assigned to the seventh bin (corresponding to a range from 30 to 35 units).

[0033] In some embodiments, the bin data 154 implicitly or explicitly specifies a difference threshold, and only delta values ​​that satisfy the difference threshold are assigned to the quantization bins. To illustrate, the bin data 154 may indicate that the first bin will have a range from 0.5 to 6. In this illustrative example, delta values ​​160 less than 0.5 will not be assigned to bins; therefore, the difference threshold is 0.5. Alternatively, the delta value calculator 138 may discard or zero delta values ​​160 that do not satisfy a different threshold.

[0034] The delta value counter 142 may be executed by the processor 128 to determine a count of the number of delta values ​​160 assigned to each bin, thereby generating a count of quantized delta values ​​164. The count of quantized delta values ​​164 for each bin indicates the number of times the sensor data 114 transitioned a particular magnitude between two consecutive samples during a particular power-on event.

[0035] For power-on events of longer duration, the number of incremental values ​​of a particular size is expected to be higher. In this case, a "power-on event" refers to any situation in which the engine or auxiliary power unit of the aircraft 102 is running and providing power to the aircraft equipment 106. The power-on event ends when the engine or auxiliary power unit of the aircraft 102 stops running or stops providing power to the aircraft equipment 106. The duration of the power-on event of the aircraft 102 can vary significantly, for example, from a few minutes for a quick test to a few hours for a long-distance flight. Compared with power-on events of shorter duration, power-on events of longer duration are generally expected to have a higher count of quantized incremental values ​​164 for any particular bin. The anomaly detection system 104 normalizes the count of quantized incremental values ​​164 to a standard (e.g., normalization standard 152) to ensure that the comparison with the anomaly detection threshold 156 is meaningful. Normalization adjusts the count of quantized incremental values ​​164 for each bin to an estimate corresponding to the standardized power-on duration represented by the normalization standard 152 of the configuration data 136 (e.g., the normalized count of incremental values ​​166).

[0036] The normalization criteria 152 indicates a baseline duration of a normalized power-on event or the number of data samples (e.g., a count of parameter values) that the corresponding sensor 108 will send during the normalized power-on event. For example, if a particular sensor 108 has a sampling rate of 1 Hz and the duration of the normalized power-on event is 90 minutes, the normalization criteria 152 for the particular sensor 108 may indicate that the normalized power-on event has a duration of 90 minutes, or may indicate that the normalized power-on event corresponds to 5400 samples from the particular sensor 108 (i.e., 1 sample / second*60 seconds / minute*90 minutes). For sensors 108 that generate evenly spaced data samples, the power-on event duration is equal to the normalization criteria 152 divided by the sampling rate of the sensor 108; therefore, for sensors 108 that generate evenly spaced data samples, the normalization criteria 152 may indicate either value. For a particular sensor 108 that generates unevenly spaced data samples, the duration of the normalized power-on event is used as the normalization criteria 152. Different sensors 108 may have different sampling rates or different sampling patterns (e.g., periodic or non-periodic). Thus, each sensor 108 used for anomaly detection may be associated with a corresponding normalization criterion 152 .

[0037] The duration of the normalized power-on event is an arbitrarily specified value, such as selected by a user associated with anomaly detection system 104 during or prior to training anomaly detector 148. As an example, the duration of the normalized power-on event may be set based on an average duration of power-on events associated with a group of aircraft 102.

[0038] The normalization factor calculator 144 may be executed by the processor 128 to determine the normalization factor based on the normalization criteria 152 and information describing the power-on event (e.g., the power-on duration 158) associated with the sensor data 114. For example, the normalization factor may be determined by dividing the power-on duration 158 (or the count of sampling cycles during the power-on event) associated with the time series 116 by the duration of the normalized power-on event (or the count of sampling cycles during the normalized power-on event) indicated by the normalization criteria 152.

[0039] The count normalizer 146 may be executed by the processor 128 to normalize the count of the quantized delta values ​​164 based on a normalization factor to generate a normalized count of the quantized delta values ​​166 (also referred to herein as a normalized count of the quantized delta values ​​or a normalized count). The normalized count of the quantized delta values ​​166 for a particular power-on event includes one normalized count for each bin of each sensor 108 that provides sensor data 114 to the anomaly detection system 104. The normalized count for the particular bin represents an estimate (e.g., an extrapolation) of how many delta values ​​would be associated with the particular bin if the duration of the particular power-on event equaled the duration of the normalized power-on event.

[0040] The anomaly detector 148 may be executed by the processor 128 to compare each normalized count of the delta values ​​166 to a corresponding anomaly detection threshold 156. If the normalized count of the delta values ​​166 satisfies (e.g., is greater than or equal to) the corresponding anomaly detection threshold 156, the anomaly detector 148 causes the output generator 150 to generate an output 168 indicating that an anomaly has been detected. The anomaly detection threshold 156 is established during the training process, for example, with reference to Figure 3 and Figure 5 Describe the training process.

[0041] The anomaly detection thresholds 156 include one or more thresholds for each bin of each sensor 108 that provides sensor data 114 to the anomaly detection system 104. For example, in a specific embodiment where only one sensor 108 provides sensor data 114 to the anomaly detection system 104, the anomaly detection thresholds 156 include one threshold for each bin, and if the normalized count of the incremental values ​​166 of any bin satisfies the corresponding anomaly detection threshold 156, the anomaly detection system 104 detects an anomaly. As another example, in a specific embodiment where only one sensor 108 provides sensor data 114 to the anomaly detection system 104, the anomaly detection thresholds 156 include two thresholds (e.g., a first threshold and a second threshold) for each bin. In this example, if the normalized count of the incremental values ​​166 of any bin satisfies the corresponding first threshold, the anomaly detection system 104 detects a first type of anomaly, and if the normalized count of the incremental values ​​166 of any bin satisfies the corresponding second threshold, the second type of anomaly is detected. For illustration, the first threshold may be used to trigger a notification to the off-board device 126, and the second threshold may be used to trigger a notification to the on-board device 110.

[0042] As yet another example, in certain embodiments, a plurality of sensors 108 provide corresponding sensor data 114 to the anomaly detection system 104, and the sensor data 114 from each sensor 108 is used to generate a corresponding normalized count of delta values ​​166. In this example, the anomaly detection thresholds 156 include one or more anomaly detection thresholds 156 for each bin of the normalized count of delta values ​​166 for each sensor 108. Thus, if 12 sensors 108 provide corresponding sensor data 114 to the anomaly detection system 104, resulting in 12 corresponding normalized counts of delta values ​​166, each with 10 bins, the anomaly detection thresholds 156 specify at least 120 anomaly detection thresholds 156 (corresponding to 12 sensors x 10 bins per sensor). The number of bins used to quantize delta values ​​may be different for different sensors 108. For example, ten bins may be used to quantify the delta values ​​160 associated with the sensor data 114 from the first sensor, and seven bins (or another number greater or less than ten) may be used to quantify the delta values ​​160 associated with the sensor data 114 from the second sensor. Furthermore, some sensors 108 may be associated with a single anomaly detection threshold per bin, while other sensors 108 may be associated with two or more anomaly detection thresholds per bin.

[0043] In some embodiments, the anomaly detection thresholds 156 include more than two different thresholds for each bin of each sensor, and the specific thresholds used depend on the conditions present during the power-on event. For example, the anomaly detection system 104 may receive data identifying the conditions present during the power-on event from one or more sensors 108, from a user (e.g., via an onboard device 110), or from another device (e.g., an offboard device 126). In this example, the anomaly detection system 104 may select a specific anomaly detection threshold 156 for each bin and each sensor 108 based on the data identifying the conditions present during the power-on event. To illustrate, the environmental control system of the aircraft 102 may operate differently depending on environmental conditions, such as ambient temperature, ambient humidity, ambient pressure, etc. In this illustrative example, the anomaly detection thresholds 156 may include, for example, a first set of thresholds used when hotter weather conditions are present and a second set of thresholds used when colder weather conditions are present. In this illustrative example, anomaly detection system 104 may select to use the first set of thresholds or the second set of thresholds based on user input (e.g., via onboard device 110), based on weather data (e.g., from offboard device 126), or based on sensor data 114 from one of sensors 108 (e.g., an ambient temperature sensor).

[0044] In particular embodiments, some sensors 108 experience many relatively small changes under normal operating conditions (e.g., when no failure is imminent). In such embodiments, some bins of normalized counts of delta values ​​166 may not be associated with anomaly detection thresholds 156. For example, if normal operation of aircraft equipment 106 monitored by a particular sensor 108 is associated with sample-to-sample changes of up to 5-10 units (e.g., delta values ​​of 5-10 units), one or more bins associated with delta values ​​of 0-10 units are not associated with corresponding anomaly detection thresholds 156.

[0045] When the anomaly detector 148 detects an operational anomaly, the output generator 150 generates an output 168. The output 168 may include an anomaly notification message, a signal (e.g., a voltage or current) provided to the onboard indicator 112, etc. In some embodiments, the output 168 may include details of the detected anomaly, such as an identification of the aircraft equipment 106 associated with the anomaly, an identification of the sensor 108 that generated the sensor data 114, a sample of the sensor data 114, a summary of the sensor data 114, a tag or other identifier associated with the anomaly detection threshold 156 that was met, etc.

[0046] Output 168 is sent to onboard device 110, offboard device 126, or both. For example, output 168 may be sent to onboard device 110 to cause onboard indicator 112 to notify a crew member (e.g., flight crew or ground crew) of an operational anomaly. As another example, offboard device 126 may include or correspond to a maintenance computing device or a ground operations computing device. In this example, output 168 may cause offboard device 126 to notify ground or maintenance personnel of an operational anomaly, or to automatically schedule one or more maintenance activities based on the operational anomaly.

[0047] By detecting operational anomalies before a fault condition occurs, the anomaly detection system 104 enables the operator of the aircraft 102 to schedule maintenance for the aircraft 102 more effectively and efficiently. In addition, the anomaly detection system 104 detects anomalies using a computationally efficient process similar to generating a histogram. For many sensor data streams, large changes in a short period of time are unusual and can indicate an impending fault condition. To illustrate, in an environmental control system, some relatively small changes in compressor outlet pressure from time to time are typical and expected. However, over the same period of time, a similar number of large changes in compressor outlet pressure can indicate that the compressor (or pressure sensor) is beginning to experience conditions that typically lead to failures. To illustrate, such a large change can indicate that the compressor has difficulty starting the rotation of the impeller, which can indicate the initial stages of a bearing failure. Using the computationally efficient calculation and counting operations described above, the anomaly detection system 104 can generate histogram-like data (e.g., normalized counts of delta values ​​166), and use the histogram-like data to determine whether the sensor data 114 includes relevant changes, such as too many (e.g., at least a threshold number) of changes of a certain size (e.g., in a certain bin). The use of histogram-like data also allows the anomaly detection system 104 to consider many different ranges of variation in the sensor data 114 and corresponding thresholds. For example, while a small number of "large" delta values ​​160 may indicate an anomaly, in some cases an anomaly may also (or alternatively) be indicated by a large number of "medium" sized delta values ​​160. To illustrate, in some embodiments, 25 compressor outlet pressure delta values ​​in the 20-30 pounds per square inch (psi) range may be as significant as 4 compressor outlet pressure delta values ​​in the 40-50 psi range over the same time period. The use of the above-described histogram-like data enables the anomaly detection system 104 to effectively detect either condition as well as other conditions.

[0048] Figure 2 It is shown by Figure 1 1. An anomaly detection system 104 of FIG. 100 is a diagram of details of operations 200 performed to detect anomalies based on sensor data 114. In certain embodiments, reference Figure 2 The operation 200 described is performed by Figure 1The one or more processors 128 of the aircraft 102 execute the functional modules of the instructions 134 according to the configuration data 136. The operations 200 are performed during or after a power-on event of the aircraft 102.

[0049] exist Figure 2 , the delta value calculation operation 238 generates the delta value 160 based on the sensor data 114. For example, the delta value calculation operation 238 determines the delta value 160 by calculating the difference or the absolute value of the difference between the parameter values ​​indicated by each pair of consecutive data samples of the sensor data 114. Therefore, each delta value 160 indicates how much the parameter measured has changed from one sampling period of the sensor data 114 to the next sampling period.

[0050] exist Figure 2 In FIG. 1 , the incremental value quantization operation 240 assigns each incremental value 160 to a quantization bin 254 based on the size of the corresponding incremental value 160. For example, in Figure 2 , the first group of incremental values ​​160 have sizes within the range associated with the first bin 254A and are assigned to the first bin 254A as a first group of quantized incremental values ​​162A. Similarly, the Nth group of incremental values ​​160 have sizes within the range associated with the Nth bin 254N and are assigned to the Nth bin 254N as an Nth group of quantized incremental values ​​162N (where N is an integer greater than 1). The number of bins 254 used, the range associated with each bin 254, or other information describing the bins 254 is indicated in the bin data 154. For example, the first bin data 154A indicates the size range associated with the first bin 254A, and the Nth bin data 154N indicates the size range associated with the Nth bin 254N.

[0051] exist Figure 2 , the incremental value counting operation 242 counts the number of incremental values ​​160 assigned to each bin 254. For example, the incremental value counting operation 242 counts the number of incremental values ​​160 in the first group of quantized incremental values ​​162A to determine a first count 164A. Similarly, the incremental value counting operation 242 counts the number of incremental values ​​160 in the Nth group of quantized incremental values ​​162N to determine an Nth count 164N.

[0052] exist Figure 2, the normalized count operation 246 determines the normalized count of the quantized delta value 164 using the normalization factor 210. For example, the first count 164A associated with the first bin 254A is multiplied by the normalization factor 210 to determine the first normalized count 166A associated with the first bin 254A. Similarly, the Nth count 164N associated with the Nth bin 254N is multiplied by the normalization factor 210 to determine the Nth normalized count 166N associated with the Nth bin 254N. The normalization factor 210 is determined by the normalization factor operation 244 based on the power-on duration 158 and the normalization criterion 152. For example, in Figure 2 In the illustrated embodiment, the normalization factor 210 is equal to the power-on duration 158 divided by the normalization criterion 152 .

[0053] exist Figure 2 , the anomaly detection operation 248 compares the normalized counts of one or more delta values ​​166 to one or more corresponding thresholds 156 to determine whether an abnormal condition is indicated. For example, the first normalized count 166A is compared to the first threshold 156A associated with the first bin 254A. Similarly, the Nth normalized count 166N is compared to the Nth threshold 156N associated with the Nth bin 254N. As shown in FIG. Figure 1 As described above, in some embodiments, one or more bins 254 are not associated with a threshold 156. For example, if any number of delta values ​​160 within the delta value range associated with a particular bin is acceptable (e.g., does not indicate an impending fault condition), then the particular bin is not associated with a threshold 156. Figure 1 As described, in some embodiments, one or more bins 254 are associated with more than one threshold 156. For example, a particular bin 254 may be associated with a first threshold (e.g., a lower threshold) and a second threshold (e.g., an upper threshold), and the generated output 168 may depend on the particular threshold being met. As another example, a particular bin 254 may be associated with multiple thresholds, and the particular threshold 156 used by the anomaly detection operation 248 may depend on conditions associated with the power-on event of the aircraft 102, such as whether the aircraft was flying during the power-on event, ambient environmental conditions associated with the power-on event, or a particular function or device operating during the power-on event, etc.

[0054] exist Figure 2 , output generation operation 250 generates output 168 based on the comparison performed by anomaly detection operation 248. For example, the specific content of output 168, where output 168 is sent, or both may be selected by output generation operation 250 based on which one or more of thresholds 156 are met. In some implementations, in response to threshold 156 not being met, output 168 indicates that no anomaly was detected.

[0055] Figure 3 Is shown as training Figure 1 In certain embodiments, reference is made to the details of the operations performed by the anomaly detection system 104. Figure 3 The operation 300 described is performed by Figure 1 The one or more processors 128 of the anomaly detection system 104 are executed during training to generate at least a portion of the configuration data 136. In other embodiments, reference Figure 3 The operations 300 are described as being performed by another device, such as a different computing device, than the anomaly detection system 104. In such implementations, the other device sends at least a portion of the configuration data 136 to the anomaly detection system 104, or the anomaly detection system 104 retrieves at least a portion of the configuration data 136 from the other device.

[0056] Anomaly detection system 104 is trained based on sensor data 314 of multiple power-on events from one or more aircraft 102. Figure 1 and 2 In contrast to sensor data 114, sensor data 314 represents multiple historical power-on events. Figure 3 , a first set of sensor data 314A is generated during a power-on event of a first aircraft 102A, and a Kth set of sensor data 314K is generated during a power-on event of a Kth aircraft 102K, where K is an integer greater than 1. In some embodiments, one or more additional sets of sensor data 314 are generated during power-on events of one or more other aircraft 102. In some embodiments, multiple sets of sensor data 314 are generated by a single aircraft 102 at multiple power-on events.

[0057] exist Figure 3 , the incremental value calculation operation 338 generates a set of incremental values ​​360 for each set of sensor data 314. For example, the incremental value calculation operation 338 determines the first set of incremental values ​​360A by calculating the difference or the absolute value of the difference between the parameter values ​​indicated by each pair of consecutive data samples of the first set of sensor data 314A. Similarly, the incremental value calculation operation 338 determines the Kth set of incremental values ​​360K by calculating the difference or the absolute value of the difference between the parameter values ​​indicated by each pair of consecutive data samples of the Kth set of sensor data 314K.

[0058] exist Figure 3 In FIG. 3 , the incremental value quantization operation 340 assigns the incremental values ​​of each group of incremental values ​​360 to the quantization bins 254 based on the size of the corresponding incremental values. For example, in Figure 3, the first set of increment values ​​360A are assigned to bins 254 to generate a first set of quantized increment values ​​362A. Similarly, the Kth set of increment values ​​360K are assigned to bins 254 to generate a Kth set of quantized increment values ​​362K (where K is an integer greater than 1).

[0059] The number of bins 254 used, the range associated with each bin 254, or other information describing the bins 254 is determined based on the range of values ​​associated with the sets of incremental values ​​360 or based on the range of parameter values ​​indicated by a corresponding set of sensor data 314 or associated with a sensor type that generated a corresponding set of sensor data 314. For example, the number of bins 254 used for a particular set of sensor data 314 can be specified by a user, can be a default value, can be selected based on a heuristic, or can be optimized or determined based on a training process to provide specified detection characteristics to the anomaly detection system 104. Typically, the bins 254 represent ranges of equal size. For example, if the first bin 254 represents a range of 5 units, then the other bins 254 also represent a range of 5 units.

[0060] When the number of bins 254 to be used is known, the range of each bin 254 can be determined by dividing the size of the maximum increment value of the groups of increment values ​​360 by the number of bins 254. In some embodiments, a safety margin can be added to the size of the maximum increment value to ensure that the set range of bins 254 includes all possible increment values. Alternatively, if the sensor data 314 indicates a fault condition when an increment value of a certain size is detected, a size less than the certain size can be used as the maximum increment value to divide the bins 254.

[0061] If the number of bins 254 to be used is not known or specified in advance, the number of bins 254 may be arbitrarily or randomly selected during a first training iteration and may be updated (if necessary) during subsequent training iterations to improve the accuracy of the anomaly detection system 104. To illustrate, the anomaly detection system 104 may be trained using a set of training data (e.g., several sets of sensor data 314 and fault indication data 308 associated with the aircraft 102 that generated the sensor data 314) and a first number of bins 254. In this illustrative example, the fault indication data 308 indicates whether each particular aircraft 102 experienced a fault condition after the power-on event associated with the sensor data 314 (e.g., within a certain number of subsequent power-on events). Thus, the fault indication data 308 acts as a label (for supervised training) to indicate whether each set of sensor data 314 is associated with a future fault condition. Continuing with the illustrative example, after the anomaly detection system 104 is trained using the first set of training data and the first number of bins 254, the anomaly detection system 104 can be tested using a set of test data to determine whether the anomaly detection system 104 can accurately predict which sensor data from the set of test data indicates a future fault condition. If the test indicates that the anomaly detection system 104 meets the specified acceptance criteria (e.g., is sufficiently accurate, produces an acceptably low false alarm rate, produces an acceptably low false alarm rate, etc.), the anomaly detection system 104 can be released for use with the first number of bins 254. However, if the test indicates that the anomaly detection system 104 fails to meet the specified acceptance criteria, the anomaly detection system 104 can be retrained and retested using the second number of bins 254.

[0062] exist Figure 3 , the incremental value counting operation 342 counts the number of incremental values ​​assigned to each bin 254. For example, the incremental value counting operation 342 counts the number of incremental values ​​of the first group of quantized incremental values ​​362A assigned to each bin 254 to determine a first bin count 364A, and counts the number of incremental values ​​of the Kth group of quantized incremental values ​​362K assigned to each bin 254 to determine a Kth bin count 364K.

[0063] exist Figure 3 , a normalized count operation 346 determines a normalized count 366 based on the bin count 364 and the normalization factors 310. The normalization factors 310 include one normalization factor for each set of sensor data 314. In other words, there are K normalization factors 310, which correspond to the K sets of sensor data 314. Each normalization factor 310 is equal to the power-on duration associated with the set of sensor data 314 divided by a normalization criterion. The normalization criterion is any standardized power-on duration or representative power-on duration, for example, an average power-on duration associated with all power-on events that generated the set of sensor data 314.

[0064] exist Figure 3 In , the threshold setting operation 312 generates one or more representative counts for each bin 254 based on the normalized counts 366. Figure 3 In the illustrated example, the threshold setting operation 312 identifies a representative fault count 326 for each bin 254 based on the aggregate bin counts (or average aggregate bin counts) associated with the bin 254 and the fault indication data 308. For example, the threshold setting operation 312 identifies a first representative fault count 326A for the first bin 254A. The first representative fault count 326A indicates a value of the normalized count 366 of the first bin 254A that may indicate an impending fault condition. In this example, the representative fault count 326A may be determined by a machine learning process, such as a support vector machine, that uses several sets of normalized counts 366 and fault indications 310 as training data to identify a boundary between normalized counts 366 that do not indicate an impending fault and normalized counts 366 that indicate an impending fault for each bin 254. In other examples, a statistical or heuristic process may be used to identify a boundary between normalized counts 366 that do not indicate an impending fault and normalized counts 366 that indicate an impending fault for each bin. In some implementations, the threshold setting operation 312 also uses a similar process to identify a representative non-fault count 324 for each bin 254. The representative non-fault count 324 indicates a value of the normalized count 366 for the corresponding bin 254 that may not indicate an impending fault condition.

[0065] In some embodiments, the representative failure count 326 associated with each bin 254 is set as the anomaly detection threshold 156 of the corresponding bin 254. For example, the representative failure count 326A associated with the first bin 254A is stored in the configuration data 136 as the first anomaly detection threshold 156A associated with the first bin data 154A, and the representative failure count 326N associated with the Nth bin 254N is stored in the configuration data 136 as the Nth anomaly detection threshold 156N associated with the Nth bin data 154N.

[0066] As described above, the anomaly detection system 104 can use more than one anomaly detection threshold for each bin 254. In particular embodiments, the threshold setting operation 312 uses the representative fault count 326 associated with each bin 254 to set the upper anomaly detection threshold 156 for the corresponding bin 254, and uses the representative non-fault count 324 associated with the bin 254 to set the lower anomaly detection threshold 156 for the bin 254. Alternatively, the lower anomaly detection threshold for each bin 254 can be set based on an offset from the representative fault count 326 (e.g., the representative fault count 326 minus a particular value, which can be arbitrary or can be based on statistics associated with several sets of sensor data 314). In another alternative, the representative fault count 326 can be the average of the lowest normalized counts 366 associated with the fault condition, and the lower and upper anomaly detection thresholds can be set based on a confidence interval or statistical variance of the normalized counts 366.

[0067] After establishing the anomaly detection threshold 156, Figure 1 Anomaly detection system 104 may use configuration data 136 to detect abnormal conditions associated with power-on events of aircraft 102 .

[0068] Figure 4 is used Figure 1 Flow chart of an example of a method 400 for detecting anomalies by the anomaly detection system 104. In a particular embodiment, the method 400 comprises Figure 1 The processor 128 initiates, controls or executes the process.

[0069] Method 400 includes, at 402, acquiring, at a computing device, sensor data captured by sensors of an aircraft during a power-on event. As a specific example, Figure 1 The anomaly detection system 104 receives sensor data 114 generated by one or more sensors 108 of the aircraft 102. The sensor data 114 includes a plurality of parameter values ​​(eg, PV1 118, PV2 120, and PV P 122), and each parameter value corresponds to a corresponding sampling period during the power-on event.

[0070] The method 400 also includes determining, by the computing device, a set of incremental values ​​at 404. As a specific example, Figure 1 The incremental value calculator 138 calculates incremental values ​​160. Each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value in a pair of parameter values ​​corresponding to consecutive sampling periods of the sensor data.

[0071] The method 400 also includes determining, by the computing device, at 406, a set of quantized incremental values ​​by assigning the incremental values ​​of the set of incremental values ​​to quantization bins based on the magnitude (magnitude) of the incremental values. As a specific example, Figure 1 The delta value quantizer 140 assigns each delta value 160 to a quantization bin 254 based on the size of the delta value 160 to generate a count of quantized delta values ​​164 .

[0072] The method 400 also includes determining, by the computing device, a normalized count of increment values ​​for each quantization bin at 408. As a specific example, Figure 1 The count normalizer 146 determines the normalized count of the increment value for each quantization bin 254. To illustrate, the count normalizer 146 multiplies the count of each quantization increment value 164 for each quantization bin 254 by the normalization factor 210.

[0073] The method 400 also includes, at 410, performing, by the computing device, a comparison of the normalized count of increment values ​​for a particular quantization bin with an anomaly detection threshold for the particular quantization bin. As a specific example, Figure 1 The anomaly detector 148 of the embodiment compares the normalized count of the incremental values ​​166 for a particular quantization bin 254 to the anomaly detection threshold 156 for the particular quantization bin 254. In some embodiments, multiple anomaly detection thresholds 156 are associated with some of the quantization bins 254. In such embodiments, the normalized count of incremental values ​​166 is compared to each anomaly detection threshold 156. Alternatively, in such embodiments, one anomaly detection threshold 156 for a particular quantization bin 254 is selected (e.g., based on a condition associated with a power-on event), and the normalized count of the incremental values ​​166 for the particular quantization bin 254 is compared to the selected anomaly detection threshold 156.

[0074] Method 400 also includes, at 412, generating, by the computing device, an output based on the comparison, the output indicating whether the sensor data indicates an operational anomaly of aircraft equipment associated with the sensor. As a specific example, Figure 1 Output generator 150 generates output 168 that indicates whether an operational anomaly is detected in sensor data 114 .

[0075] The method 400 is a computationally efficient process for detecting operating anomalies before a fault condition occurs. Thus, the method 400 enables aircraft operators and maintainers to schedule maintenance more effectively and efficiently.

[0076] Figure 5 Is for Figure 1 Flow chart of an example of a method 500 for determining an anomaly detection threshold by the anomaly detection system 104 of FIG. 1 . In certain embodiments, the method 500 is composed of Figure 1In other embodiments, the method 500 is performed by a computing device other than the anomaly detection system 104 (e.g., the off-board device 126), and the anomaly detection threshold (and possibly other configuration data) is subsequently provided to the anomaly detection system 104.

[0077] The method 500 includes acquiring, at a computing device, a plurality of sets of sensor data captured by onboard sensors of one or more aircraft during a plurality of power-on events at 502. Each set of sensor data corresponds to a respective power-on event of a respective aircraft and includes a plurality of parameter values. Each parameter value corresponds to a respective sampling period during the respective power-on event.

[0078] The method 500 also includes determining, by the computing device, a set of incremental values ​​for each set of sensor data at 504. Each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value from a pair of parameter values ​​corresponding to consecutive sampling periods of the set of sensor data. Figure 3 As described above, the computing device may perform one or more delta value calculation operations 338 .

[0079] The method 500 also includes, at 506, determining, by the computing device, to quantize the set of incremental values ​​by assigning the incremental values ​​of the set of incremental values ​​to quantization bins based on the magnitude (amplitude) of the incremental values. Figure 3 As described above, the computing device may perform one or more delta value quantization operations 340 .

[0080] The method 500 also includes determining, by the computing device, a normalized count of the increment values ​​for each quantization bin at 508. For example, as shown in FIG. Figure 3 As described above, the computing device may perform one or more of the increment value counting operation 342 and the normalized counting operation 346 .

[0081] The method 500 also includes, at 510, setting an anomaly detection threshold for each quantization bin based on the normalized count of delta values ​​and fault indication data associated with one or more aircraft. Figure 3 As described, the computing device may perform one or more threshold setting operations 312 .

[0082] Figure 6 is a block diagram of a computing environment 600 including a computing device 610 configured to support various aspects of computer-implemented methods and computer-executable program instructions (or codes) according to the present disclosure. As a specific example, the computing device 610 includes, is included in, or corresponds to Figure 1 As another specific example, the computing device 610 includes, is included in, or corresponds to Figure 1The computing device 610 or a portion thereof is configured to execute instructions to initiate, execute or control a reference Figure 1-Figure 5 For example, in some embodiments, the computing device 610 includes a computer configured to perform Figure 4 Method 400, Figure 5 For illustration, in the method 500 or both of the components of the operation. Figure 6 In the example, the computing device 610 includes Figure 1 Instructions 134 are executable to perform anomaly detection based on sensor data. In another illustrative example, Figure 6 , computing device 610 includes an anomaly detection trainer 636 corresponding to instructions executable to perform operation 300 , method 500 , or both.

[0083] The computing device 610 includes one or more processors 620. The processor 620 is configured to communicate with a system memory 630, one or more storage devices 640, one or more input / output interfaces 650, one or more communication interfaces 660, or any combination thereof. The system memory 630 includes volatile storage devices (e.g., random access memory (RAM) devices), non-volatile storage devices (e.g., read-only memory (ROM) devices, programmable read-only memory, and flash memory), or both. The system memory 630 stores an operating system 632, which may include a basic input / output system for booting the computing device 610 and a complete operating system that enables the computing device 610 to interact with users, other programs, and other devices. The system memory 630 also stores application programs, such as instructions 134, anomaly detection trainer 636, or both.

[0084] The one or more storage devices 640 include non-volatile storage devices, such as magnetic disks, optical disks, or flash memory devices. In certain examples, the storage devices 640 include removable and non-removable storage devices. The storage devices 640 are configured to store operating systems, images of operating systems, applications (e.g., one or more applications 634), and program data (e.g., Figure 1-Figure 3 Configuration data 136). In certain aspects, system memory 630, storage device 640, or both include tangible computer-readable media. In certain aspects, one or more storage devices 640 are external to computing device 610.

[0085] The one or more input / output interfaces 650 enable the computing device 610 to communicate with one or more input / output devices 670 to facilitate user interaction. For example, the one or more input / output interfaces 650 may include a display interface, an input interface, or both. The processor 620 is configured to communicate with a device or controller 680 via one or more communication interfaces 660. For example, the one or more communication interfaces 660 may include a network interface. The device or controller 680 may include, for example, an aircraft 102, a device of the network 124, a sensor 108, an onboard device 110, an offboard device 126, or any combination thereof.

[0086] In some embodiments, when executed by one or more processors, the non-transitory computer-readable medium stores instructions that cause the one or more processors to initiate, execute, or control operations to perform some or all of the above functions. For example, the instructions may be executable to implement Figure 1-Figure 5 For illustration, when executed by processor 620, the instructions of application 634 may cause processor 620 to initiate, perform, or control operations to train anomaly detection system 104, detect anomalies based on sensor data 114, or both. In some embodiments, Figure 1-Figure 5 Part or all of one or more operations or methods are performed by one or more processors (e.g., one or more central processing units (CPUs), one or more graphics processing units (GPUs), one or more digital signal processors (DSPs)) that execute instructions, dedicated hardware circuits, or any combination thereof.

[0087] Figure 7 Yes Figure 1 Flowchart of an example life cycle 700 of an anomaly detection system for an aircraft. During pre-production, at 702, the example life cycle 700 includes specification and design of an aircraft (e.g., aircraft 102). During specification and design of the aircraft, the life cycle 700 may include specification and design of anomaly detection system 104. At 704, the life cycle 700 also includes material procurement, which may include procurement of materials for anomaly detection system 104.

[0088] During production, at 706, the life cycle 700 includes component and subassembly manufacturing and, at 708, system integration of the aircraft. For example, the life cycle 700 may include component and subassembly manufacturing of the anomaly detection system 104 and system integration of the anomaly detection system 104. At 710, the life cycle 700 includes certification and delivery of the aircraft and, at 712, placing the aircraft in service. Certification and delivery may include certification of the anomaly detection system 104 to place the anomaly detection system 104 in service. During customer service, the aircraft may be scheduled for routine maintenance and service (which may also include modification, reconfiguration, refurbishment, etc.). At 714, the life cycle 700 includes performing maintenance and service on the aircraft, which may include performing maintenance and service on the anomaly detection system 104. Alternatively, maintenance and service may be performed on the aircraft based on an operational anomaly detected by the anomaly detection system 104.

[0089] Each process of life cycle 700 may be performed or implemented by a system integrator, a third party, and / or an operator (e.g., a customer). For the purposes of this description, a system integrator may include, but is not limited to, any number of aircraft manufacturers and major system subcontractors; a third party may include, but is not limited to, any number of suppliers, subcontractors, and vendors; an operator may be an airline, a leasing company, a military entity, a service organization, etc.

[0090] Figure 8 is a block diagram of aircraft 102. Figure 8 In the example shown, Figure 1 The anomaly detection system is a component of aircraft 102. Figure 8 In FIG. 8 , aircraft 102 includes airframe 802 having a plurality of systems, such as aircraft equipment 106 and interior 804 . Examples of aircraft equipment 106 include one or more components of propulsion system 808 , electrical system 812 , environmental system 814 , hydraulic system 810 , and health monitoring system 816 . Figure 8 In the example shown, the sensors 108, the anomaly detection system 104, or both are part of the health monitoring system 816. Any number of other systems or other devices may also be included.

[0091] The illustrations of the examples described herein are intended to provide a general understanding of the structures of various embodiments. These illustrations are not intended to be a complete description of all elements and features of the equipment and systems utilizing the structures or methods described herein. Many other embodiments may be apparent to those skilled in the art after reading this disclosure. Other embodiments may be utilized and derived from this disclosure, so that structural and logical replacements and changes may be made without departing from the scope of this disclosure. For example, method operations may be performed in an order different from that shown in the figure, or one or more method operations may be omitted. Therefore, this disclosure and the accompanying drawings should be considered illustrative rather than restrictive.

[0092] In addition, although specific examples have been shown and described herein, it should be understood that any subsequent settings designed to achieve the same or similar results can replace the specific embodiments shown. This disclosure is intended to cover any and all subsequent modifications or changes of various embodiments. After reading this disclosure, combinations of the above embodiments and other embodiments not specifically described herein will be apparent to those skilled in the art.

[0093] Based on such understanding, the abstract of the present disclosure is submitted, that is, it will not be used to explain or limit the scope or meaning of the technical solution. In addition, in the above specific embodiments, in order to simplify the present disclosure, various features can be combined together or described in a single embodiment. The examples described above illustrate but do not limit the present disclosure. It should also be understood that according to the principles of the present disclosure, many modifications and variations are possible. As reflected in the following technical solutions, the subject matter claimed for protection may be directed to not all features of any disclosed example. Therefore, the scope of the present disclosure is limited by the following technical solutions and their equivalents.

Claims

1. A method for detecting anomalies, comprising: acquiring, at a computing device, sensor data captured by a sensor of the aircraft during a power-on event, wherein the sensor data includes a plurality of parameter values, and wherein each parameter value corresponds to a respective sampling period during the power-on event; determining, by the computing device, a set of incremental values, wherein each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value in a pair of parameter values ​​corresponding to consecutive sampling periods of the sensor data; determining, by the computing device, a set of quantized incremental values ​​by assigning incremental values ​​in the set of incremental values ​​to quantized bins based on the magnitude of the incremental values; The calculation device determines the normalized count of the incremental value of each quantization bin in the following manner: determining a power-on duration of the power-on event represented by the sensor data; determining a normalization factor based on the power-on duration of the power-on event and based on a normalization criterion; determining a count of incremental values ​​assigned to each quantized bin; and adjusting the count of the incremental values ​​assigned to each quantization bin based on the normalization factor to determine a normalized count of the incremental values; performing, by the computing device, a comparison of the normalized count of increment values ​​for a particular quantization bin with an anomaly detection threshold for the particular quantization bin; and An output is generated by the computing device based on the comparison, the output indicating whether the sensor data indicates an operational anomaly of aircraft equipment associated with the sensor.

2. The method of claim 1 , further comprising comparing, by the computing device, each incremental value with a difference threshold, wherein: Only delta values ​​that satisfy the difference threshold are assigned to the quantization bin. 3 . The method of claim 1 , further comprising selecting an anomaly detection threshold for each quantization bin based on an operating condition associated with the power-on event.

4. The method according to claim 3, wherein: The aircraft equipment corresponds to a component of an environmental system, and wherein the operating condition is associated with an ambient environmental condition.

5. The method according to claim 1, wherein: The anomaly detection threshold for each quantization bin includes a first threshold and a second threshold, and wherein the output includes a first indication in response to the comparison, the first indication indicating that a normalized count of incremental values ​​associated with the particular quantization bin is greater than or equal to the first threshold associated with the particular quantization bin and less than the second threshold associated with the particular quantization bin.

6. The method according to claim 5, wherein: The output includes a second indication responsive to the comparison, the second indication indicating that a normalized count of delta values ​​associated with the particular quantization bin is greater than or equal to the second threshold associated with the particular quantization bin.

7. A non-transitory computer-readable storage medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising: acquiring sensor data captured by a sensor of the aircraft during a power-on event, wherein the sensor data includes a plurality of parameter values, and wherein each parameter value corresponds to a respective sampling period during the power-on event; determining a set of incremental values, wherein each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value in a pair of parameter values ​​corresponding to consecutive sampling periods of the sensor data; determining a set of quantized incremental values ​​by assigning incremental values ​​in the set of incremental values ​​to quantized bins based on the magnitude of the incremental values; The normalized count of increment values ​​for each of the bins is determined by: determining a power-on duration of the power-on event represented by the sensor data; determining a normalization factor based on the power-on duration of the power-on event and based on a normalization criterion; determining a count of incremental values ​​assigned to each quantized bin; and adjusting the count of the incremental values ​​assigned to each quantization bin based on the normalization factor to determine a normalized count of the incremental values; performing a comparison of the normalized count of delta values ​​for a particular bin with an anomaly detection threshold for the particular bin; and An output is generated based on the comparison, the output indicating whether the sensor data indicates an operational anomaly of aircraft equipment associated with the sensor.

8. The non-transitory computer-readable storage medium of claim 7, wherein: The operations further include comparing each delta value to a difference threshold, and wherein only delta values ​​satisfying the difference threshold are assigned to quantization bins.

9. The non-transitory computer-readable storage medium of claim 7, wherein: The operations also include selecting the anomaly detection threshold for each quantization bin based on an operating condition associated with the power-on event.

10. The non-transitory computer-readable storage medium of claim 7, wherein: Performing the comparison includes: The normalized count of incremental values ​​is compared with a first threshold associated with the particular quantization bin and a second threshold associated with the particular quantization bin, wherein the output includes a first indication in response to the comparison, the first indication indicating that the normalized count of incremental values ​​associated with the particular quantization bin is greater than or equal to the first threshold and less than the second threshold.

11. The non-transitory computer-readable storage medium of claim 10, wherein: The output also includes a second indication responsive to the comparison, the second indication indicating that the normalized count of incremental values ​​is greater than or equal to the second threshold.

12. An aircraft comprising: a sensor configured to capture sensor data during a power-on event, wherein the sensor data includes a plurality of parameter values, and wherein each parameter value corresponds to a respective sampling period during the power-on event; one or more processors; and a storage device configured to store the sensor data and instructions executable by the one or more processors to detect an operational anomaly of aircraft equipment associated with the sensor, wherein when the instructions are executed by the one or more processors, the instructions cause the one or more processors to perform operations comprising: determining a set of incremental values, wherein each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value in a pair of parameter values ​​corresponding to consecutive sampling periods of the sensor data; determining a set of quantized incremental values ​​by assigning incremental values ​​in the set of incremental values ​​to quantized bins based on the magnitude of the incremental values; The normalized count of increment values ​​for each of the bins is determined by: determining a power-on duration of the power-on event represented by the sensor data; determining a normalization factor based on the power-on duration of the power-on event and based on a normalization criterion; determining a count of incremental values ​​assigned to each quantized bin; and adjusting the count of the incremental values ​​assigned to each quantization bin based on the normalization factor to determine a normalized count of the incremental values; performing a comparison of the normalized count of delta values ​​for a particular bin with an anomaly detection threshold for the particular bin; and An output is generated based on the comparison, the output indicating whether the sensor data indicates an operational anomaly of the aircraft equipment.

13. The aircraft of claim 12, further comprising an onboard indicator, wherein: The operations also include sending the output to the onboard indicator.

14. The aircraft of claim 12, wherein: The operations also include transmitting the output to an off-board device.

15. A method for training an anomaly detection system, the method comprising: Acquiring, at the computing device, a plurality of sets of sensor data captured by onboard sensors of one or more aircraft during a plurality of power-on events, wherein each set of sensor data corresponds to a respective power-on event of a respective aircraft and includes a plurality of parameter values, and wherein each parameter value corresponds to a respective sampling period during the respective power-on event; determining, by the computing device, from each set of sensor data, a set of incremental values, wherein each incremental value in the set of incremental values ​​indicates a difference between a first parameter value and a second parameter value in a pair of parameter values ​​from consecutive sampling periods corresponding to the set of sensor data; determining, by the computing device, a set of quantized incremental values ​​by assigning incremental values ​​of the set of incremental values ​​to quantized bins based on the magnitude of the incremental values; The calculation device determines the normalized count of the incremental value of each quantization bin in the following manner: determining a power-on duration of the power-on event represented by the sensor data; determining a normalization factor based on the power-on duration of the power-on event and based on a normalization criterion; determining a count of incremental values ​​assigned to each quantized bin; and adjusting the count of the incremental values ​​assigned to each quantization bin based on the normalization factor to determine a normalized count of the incremental values; and An anomaly detection threshold is set for each quantization bin based on the normalized count of delta values ​​and fault indication data associated with one or more aircraft.

16. The method according to claim 15, wherein: The plurality of sets of sensor data include a first set of sensor data associated with a fault condition, and wherein an anomaly detection threshold associated with a particular quantization bin is set further based on a first normalized count of delta values ​​associated with the first set of sensor data.

17. The method of claim 15, further comprising comparing, by the computing device, each incremental value with a difference threshold, wherein: Only delta values ​​that satisfy the difference threshold are assigned to the quantization bin.

Images