Data-Driven Machine Learning for Modeling Aircraft Sensors
A data-driven machine learning method models aircraft components' relationships to predict abnormal behaviors, enhancing detection accuracy and reducing maintenance costs by considering the entire system context.
Patent Information
- Application Number
- JP2020092347
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2019-06-06
- Filing Date
- 2020-05-27
- Publication Date
- 2024-11-13
- Estimated Expiration
- 2040-05-27
AI Technical Summary
Current machine learning techniques for detecting component and subsystem deterioration in engineered systems, such as aircraft, are limited by their focus on individual subsystems, failing to utilize data from the entire system, leading to unreliable predictions and significant costs due to unscheduled maintenance and many false positives.
A data-driven machine learning approach that models the relationship between components and subsystems within a system, using supervised machine learning to generate predictive models based on training data from a fleet of aircraft, enabling real-time detection of abnormal operational behaviors and generating fault diagnosis notifications.
Accurately detects degraded components and subsystems by considering the entire system context, reducing unscheduled maintenance costs and improving prediction accuracy by identifying abnormalities early.
Smart Images

Figure 0007582794000001 
Figure 0007582794000002 
Figure 0007582794000003
Abstract
Description
[Technical field]
[0001] The present disclosure relates generally to state and / or fault sensing, and more specifically to techniques for modeling aircraft sensors using data-driven machine learning. [Background technology]
[0002] Engineered systems, such as, for example, plant equipment and machinery, and especially aircraft and other vehicles, typically contain many components and subsystems. For the system to function properly, these components must be periodically maintained or replaced. Sometimes, outside of routine maintenance, unexpected degradation of components and / or subsystems may occur. Degraded components may be identified by logs, noticeable effects, troubleshooting, routine inspections, and / or other fault detection methods. When this occurs, unscheduled maintenance and significant costs may be required.
[0003] Current approaches are typically manual-based and require the support of experts in the specific subsystem being inspected. These approaches are time-consuming and costly, and may not be sufficient to reliably predict problems and prevent unplanned downtime. Current methods for detecting component and subsystem degradation are also prone to many false positives and oversights.
[0004] Machine learning techniques are used to detect failures in specific subsystems. However, currently used machine learning techniques focus on individual subsystems and do not include data collected from other components and subsystems in the same designed system. As a result, current approaches may miss potentially useful information from sensors capturing environmental and external influences of the system as a whole. Thus, currently used machine learning techniques cannot reliably detect degradation of components and / or subsystems in the context of the designed system as a whole. Other drawbacks may also be present. Summary of the Invention
[0005] According to the present disclosure, a system and method are provided for detecting degradation of components and / or subsystems in a system being designed based on their relationship to other components and / or subsystems. The system and method of the present disclosure allows accurate detection of degraded components, thereby saving time and costs associated with unscheduled maintenance. For example, detection of degraded components allows for prediction of failures or anomalies, since degradation of a component may be a precursor to failure. According to one embodiment, the method includes receiving training data representing a set of operational behaviors associated with each component of a set of components that make up the system being designed. The method further includes generating a predictive model based on the training data, the predictive model being configured to predict a normal operational behavior associated with a given component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components. The method further includes receiving real-time data representing a set of real-time operational behaviors associated with each component of the set of components, including a real-time operational behavior associated with the given component. The method includes classifying the real-time operational behavior associated with the given component as either normal or abnormal based on the predictive model. The method further includes generating a fault diagnosis notification if the real-time operational behavior is classified as abnormal.
[0006] In some embodiments, the set of components includes vehicle components and the designed system is a vehicle. In some embodiments, the set of components includes aircraft components and the designed system is an aircraft. In some embodiments, the training data is received from a plurality of aircraft in a fleet of aircraft. In some embodiments, the classification of the real-time operational behavior associated with the given component is performed during flight of the aircraft by a processor located in an avionics bay of the aircraft. In some embodiments, generating the predictive model includes training the predictive model by a supervised machine learning process using the training data. In some embodiments, the set of real-time operational behaviors includes user inputs, equipment states, measurements, or a combination thereof.
[0007] In some embodiments, the method further includes generating a second prediction model based on the training data, the second prediction model configured to predict a second normal operational behavior associated with a second component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components. The set of real-time operational behaviors includes a second real-time operational behavior of the second component. In some embodiments, the method includes classifying the second real-time operational behavior associated with the second component as either normal or abnormal, and generating a fault diagnosis notification if the second real-time operational behavior is classified as abnormal.
[0008] In some embodiments, the fault diagnosis notification includes an identification of the given component.In some embodiments, the method includes sending the fault diagnosis notification to an output device.
[0009] A system according to some embodiments includes a set of components in a system being designed configured to generate real-time data representative of a set of real-time operational behaviors associated with each component in the set of components. The system further includes a computer-implemented predictive model to predict a normal operational behavior associated with a given component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components. The system further includes a processor configured to receive the set of real-time operational behaviors including the real-time operational behavior associated with the given component and classify the real-time operational behavior associated with the given component as either normal or abnormal based on the predictive model. The system further includes an output device configured to output a fault diagnosis notification if the processor classifies the real-time operational behavior as abnormal.
[0010] In some embodiments, the system includes a second processor configured to receive training data representative of a set of operational behaviors associated with each component in the set of components and generate the predictive model based on the training data. In some embodiments, the set of components includes vehicle components and the designed system is the vehicle. In some embodiments, the set of components includes aircraft components and the designed system is the aircraft. In some embodiments, the processor is located in an avionics bay of an aircraft. In some embodiments, the set of components includes user input devices, instruments, measurement sensors, or combinations thereof.
[0011] In some embodiments, a method includes receiving real-time data representative of a set of real-time operational behaviors associated with each component of a set of components that comprise a system being designed, the set of real-time operational behaviors including a real-time operational behavior associated with a given component in the set of components, the method further includes classifying the real-time operational behavior associated with the given component as either normal or abnormal based on a predictive model configured to predict a normal operational behavior associated with the given component in relation to other normal operational behaviors associated with each of the other components in the set of components, the method further includes generating a fault diagnosis notification if the real-time operational behavior is classified as abnormal.
[0012] In some embodiments, the method includes receiving training data representing a set of operational behaviors associated with each component of the set of components, and generating the predictive model based on the training data. In some embodiments, generating the predictive model includes training the predictive model by a supervised machine learning process using the training data. In some embodiments, the set of real-time operational behaviors includes user inputs, equipment states, measurements, or a combination thereof. [Brief description of the drawings]
[0013] [Figure 1] FIG. 1 is a block diagram illustrating an embodiment of a system for training a predictive model for detecting degraded components in a system being designed. [Diagram 2] FIG. 1 is a block diagram illustrating an embodiment of a system for predicting normal operational behavior associated with components of a system being designed using a predictive model. [Diagram 3]1 is a block diagram illustrating an embodiment of a system for detecting degraded components in a system under design. [Figure 4] FIG. 1 is a flow diagram illustrating an embodiment of a method for training a predictive model. [Diagram 5] FIG. 1 is a flow diagram illustrating an embodiment of a method for detecting degraded components in a system being designed. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0014] While the present disclosure is susceptible to various modifications and alternative forms, specific embodiments will be described below with reference to the drawings. However, it will be understood that this description is not intended to limit the present disclosure to these specific forms. Rather, all modifications, equivalent arrangements, and alternative arrangements are intended to be included within the scope of the present disclosure.
[0015] 1, a block diagram of an embodiment of a system 100 for training a predictive model 138 for detecting degraded components in a designed system 112 is shown. The designed system 112 has a component set 114 including a first component 116, a second component 118, and a third component 120. Although three components are shown in the figure as an example, in practice the component set 114 may include more or less than three components. In most practical applications, the component set 114 will include a large number of components.
[0016] The designed system 112 may correspond to a vehicle such as an aircraft. For example, FIG. 1 illustrates a fleet of aircraft 122 including a first aircraft 122A, a second aircraft 122B, and a third aircraft 122C. The aircraft 122A-C may be the same type of aircraft and may include the same components. Thus, each of the aircraft 122A-C may correspond to the designed system 112.
[0017] For example, first aircraft 122A includes a first component 116A corresponding to first component 116, a second component 118A corresponding to second component 118, and a third component 120A corresponding to third component 120. Similarly, second aircraft 122B includes a first component 116B corresponding to both first component 116A and first component 116, a second component 118B corresponding to both second component 118A and second component 118, and a third component 120B corresponding to both third component 120A and third component 120. The third aircraft 122C also includes a first component 116C corresponding to the first component 116A, the first component 116B, and the first component 116, a second component 118A, the second component 118B, and the second component 118C corresponding to the second component 118, and a third component 120A, the third component 120B, and the third component 120C corresponding to the third component 120.
[0018] Each of the components 116, 118, 120 generates data continuously during flight. For example, the set of components 114 may include user input devices (e.g., flight controls, user prompts, audio and video recording devices, etc.), instruments (e.g., control surfaces, engine systems, motors, etc.), measurement sensors (e.g., pressure sensors, temperature sensors, etc.), or a combination thereof. Given the volume of data generated by the set of components 114, this generated data may be used as the training data 102.
[0019] The training data 102 may represent a set of operational behaviors 104 associated with each component of a set of components 114 that constitute the system 112 being designed. For example, the training data 102 represents a first operational behavior 106 associated with a first component 116, a second operational behavior 108 associated with a second component 118, and a third operational behavior 110 associated with a third component 120. The training data 102 is collected continuously while the system 112 is operating normally, so that the set of operational behaviors 104 includes, with some exceptions, nominal behaviors that are suitable for training the predictive model 138. It should be noted that the training data stream may include some amount of off-nominal data. However, due to the large amount of data received, the off-nominal data does not reach a level that has a substantial impact. In other words, the system 100 of the present disclosure is a robust solution that can tolerate such "noisy" training data. Furthermore, because the training data 102 is collected from the entire fleet 122 , the set of operational behaviors 104 is a description of the nominal behavior based on the entire fleet 122 , rather than any individual aircraft in the fleet 122 .
[0020] The system 100 includes a computing device 130 configured to train a predictive model 138. The computing device 130 includes a processor 132 and a memory 134. The processor 132 may include a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a peripheral interface controller (PIC), or other type of microprocessor. It may also be implemented as an integrated circuit, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a combination of logic gates, other types of digital or analog electrical components, or the like, or a combination thereof. In some embodiments, the processor 132 may be distributed across multiple processing elements and configured for distributed processing.
[0021] Additionally, computing device 130 includes memory 134. Examples of memory include random access memory (RAM), read only memory (ROM), magnetic disk memory, optical disk memory, flash memory, other types of memory capable of storing data and processor instructions, or combinations thereof. In some embodiments, memory or portions thereof may be located external to computing device 130 or may be located remotely from other portions of the computing device. Memory 134 may store instructions that, when executed by processor 132, cause processor 132 to perform operations. These operations may correspond to any of the operations described herein, and specifically, these operations may correspond to training predictive model 138.
[0022] The processor 132 and memory 134 are used together to implement a supervised machine learning process 136 that generates a predictive model 138. The predictive model 138 may include any artificial intelligence model that may be utilized to classify operational behavior as either normal or abnormal. For example, the predictive model 138 may include a decision tree, association rules, or other type of machine learning classification process, or a combination thereof, and may be implemented as a support vector machine network, a Bayesian network, a neural network, or other type of machine learning classification network system, or a combination thereof.
[0023] The training data 102 is received from multiple system implementations during operation. For illustrative purposes, in FIG. 1, the training data 102 is received from a fleet of aircraft 122 or from any one of the individual aircraft 122A, 122B, 122C in the fleet of aircraft 122, depending on whether the system under design 112 is analyzed at the fleet level or at the individual aircraft level. It is noted that while FIG. 1 shows an aircraft, the system under design may be any type of equipment or electrical system, and is not limited to an aircraft. A predictive model 138 is then trained by a supervised machine learning process 136 using the training data 102. Since the set of operational behaviors 104 is largely the nominal behavior of the component collection 114, the predictive model 138 is configured to distinguish between normal and abnormal operational behaviors of the first component 116. Advantageously, these predictions are based not only on the first operational behavior 106 associated with the first component 116, but also on the other operational behaviors 108, 110. These other operational behaviors may not be associated with component 116, but may be relevant because second component 118 and third component 120 interact with first component 116 in system under design 112. Similarly, second predictive model 139 is configured to distinguish between normal and abnormal operational behaviors of second component 118, and third predictive model 140 is configured to distinguish between normal and abnormal operational behaviors of third component 120.
[0024] Since the predictive model 138 is trained based on the training data 102 that represents the first operation behavior 106 of the first component 116 as well as other behaviors 108, 110 of the other components 118, 120, the predictive model 138 is configured to determine whether the first component 116 is operating normally or whether an anomaly has occurred, taking into account the entirety of the designed system 112. The second predictive model 139 and the third predictive model 140 provide similar benefits, as well as other benefits.
[0025] Referring to FIG. 2, a block diagram of an embodiment of a system 200 for predicting a first normal operation behavior 202 associated with a component 116 of a system 112 being designed using a predictive model 138 is shown. As described with reference to FIG. 1, the trained predictive model 138 may be utilized to predict the first normal operation behavior 202 of the first component 116. For example, the predictive model 138 may be input with a second normal operation behavior 204 of the second component 118 and a third normal operation behavior 206 of the third component 120. The predictive model 138 may identify or predict the first normal operation behavior 202 of the first component 116 based on the second normal operation behavior 204 and the third normal operation behavior 206. As will be further described with reference to FIG. 3, the predicted first normal operation behavior 202 may be compared to an actual behavior of the component 116 to determine whether the behavior of the component 116 is normal or whether an anomaly has occurred. 2 illustrates the case where the first normal operation behavior 202 of the first component 116 is predicted, the second prediction model 139 can be used to predict the second normal operation behavior 204 of the second component 118 based on the first normal operation behavior 202 and the third normal operation behavior 206. Similarly, the third prediction model 140 can be used to predict the third normal operation behavior 206 of the third component 120 based on the first normal operation behavior 202 and the second normal operation behavior 204.
[0026] 3, which is a block diagram illustrating an embodiment of a system 300 for detecting degraded components in a designed system 112. The system 300 includes the designed system 112 having a collection of components 114 and a computing device 322. If the designed system 112 is an aircraft, the computing device 322 may be located in an avionics bay 320 of the aircraft. Alternatively, the computing device 322 may be a ground-based device used for post-flight processing.
[0027] The computing device 322 includes a processor 324, a memory 326, and an output device 328. The processor 324 may include a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a peripheral interface controller (PIC), or other type of microprocessor, and may be implemented as an integrated circuit, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a combination of logic gate circuits, other types of digital or analog electrical components, or the like, or a combination thereof. In some embodiments, the processor 324 may be distributed across multiple processing elements and configured for distributed processing.
[0028] Examples of memory 326 include random access memory (RAM), read only memory (ROM), magnetic disk memory, optical disk memory, flash memory, other types of memory capable of storing data and processor instructions, or combinations thereof. In some embodiments, memory 326 or portions thereof may be located external to computing device 322 or may be located remotely from other portions of the computing device. Memory 326 may store instructions that, when executed by processor 324, cause processor 324 to perform operations, such as any of the operations described herein, and in particular, to detect a failure in one of components 116, 118, 120 using predictive model 138.
[0029] Output device(s) 328 may include any device capable of communicating with a user or with a device external to computing device 322. For example, output device(s) 328 may include user output devices such as indicator lights, display screens, speakers, etc. Output device(s) 328 may also include network output devices such as serial communications equipment, network cards, etc.
[0030] The predictive models 138, 139, 140 may be stored in memory 326 or in other forms on the computing device 322. An indication 330 of a fault diagnosis (e.g., fault detection) may be generated as described herein based on results obtained using the predictive models 138, 139, 140. The output device 328 is configured to communicate the fault diagnosis indication 330 to a user or other device.
[0031] During operation, the designed system 112 generates real-time data 310 that includes a set of real-time operational behaviors 308. As used herein, the term "real-time" means that the real-time data 310 corresponds to a currently ongoing process (e.g., a flight currently in flight) or a recent process (e.g., a flight recently flown by an aircraft). In particular, the concept of "real-time" takes into account delays in data access. The set of real-time operational behaviors 308 may include, for example, a first real-time operational behavior associated with the first component 116, a second real-time operational behavior associated with the second component 118, and a third real-time operational behavior associated with the third component 120.
[0032] The computing device 322 receives the real-time data 310. Based on the real-time data 310, the processor 324 uses the predictive model 138 to classify a first real-time operational behavior 302 associated with the component 116 as either normal or abnormal. For example, the predictive model 138 calculates the normal operational behavior 202 of the first component 116. The first normal operational behavior 202 is compared to the first real-time operational behavior 302 to determine whether the first real-time operational behavior 302 is abnormal. If the first real-time operational behavior 302 is classified as abnormal, the processor 324 generates a fault diagnosis notification 330. In some embodiments, the classification of the first real-time operational behavior 302 is performed in-flight.
[0033] By performing similar calculations, it can be determined whether the second real-time operational behavior 304 and / or the third real-time operational behavior 306 are abnormal. For example, the second prediction model 139 can calculate the second normal operational behavior 204 associated with the second component 118, and the third prediction model 140 can calculate the third normal operational behavior 206 associated with the third component 120. Thus, the normal operational behavior 202, 204, 206 of each of the components 116, 118, 120 can be calculated based on the other operational behaviors in the set of real-time operational behaviors 308. If any operational behavior in the set of real-time operational behaviors 308 is abnormal, a fault diagnosis notification 330 can be generated that indicates which of the components 116, 118, 120 is operating abnormally.
[0034] By using predictive models 138-140 that classify real-time data 310 based on the real-time operational behavior 302, 304, 306 of the components 116, 118, 120, respectively, the system 300 is able to detect degraded components and / or subsystems based on the entire system being designed, rather than based on the behavior of a single component. Thus, the system 300 is able to detect anomalous behavior that would be missed by other detection systems. Other advantages may also be realized.
[0035] 4, which is a flow diagram illustrating an embodiment of a method 400 for training a predictive model. The method 400 receives 402 training data representing a set of operational behaviors associated with each component of a collection of components that make up a system being designed. For example, the computing device 130 receives the training data 102.
[0036] The method 400 further includes generating 404 a predictive model based on the training data, the predictive model configured to predict normal operational behavior associated with a given component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components. For example, the predictive models 138, 139, 140 are configured to predict normal operational behavior associated with the components 116, 118, 120 in relation to the entire system being designed 112. This allows for diagnosing impairments in the system being designed 112 that are associated with components other than those modeled in the predictive models 138, 139, 140. The method 400 also includes performing 406 a pattern analysis based on the first predictive model and the second predictive model to identify the relevant components. By analyzing the patterns associated with which of the predictive models 138, 139, 140 are abnormal and which are normal, the subsystem that is the root cause of the abnormal behavior can be identified, and in some cases, which components of the subsystem are the culprits. In other words, parts that have deteriorated and need replacing can be discovered.
[0037] Thus, by constructing a predictive model that predicts normal operational behavior associated with a given component in a system being designed relative to other normal operational behaviors associated with each of the other components in the system, method 400 can determine whether a given component is operating normally or has an anomaly while taking into account the entire system being designed.
[0038] 5, an embodiment of a method 500 for detecting degraded components in a designed system is illustrated. The method 500 includes receiving 502 real-time data representative of a set of real-time operational behaviors associated with each component of a set of components that make up the designed system. The set of real-time operational behaviors includes a real-time operational behavior associated with a given component in the set of components. For example, the computing device 322 receives the real-time data 310.
[0039] The method 500 further includes classifying 504 a real-time operational behavior associated with the given component as either normal or abnormal based on a predictive model configured to predict the normal operational behavior associated with the given component in relation to other normal operational behaviors associated with each of the other components in the collection of components. For example, the first real-time operational behavior 302 is classified by the predictive model 138.
[0040] The method 500 further includes generating a fault diagnosis notification if the real-time operational behavior is classified as abnormal, at 506. For example, the processor 324 generates the fault diagnosis notification 330.
[0041] The method 500 also includes sending the fault diagnosis notification to an output device, at 508. For example, the fault diagnosis notification 330 is sent to the output device 328 for output.
[0042] The present disclosure further includes examples according to the following appendices.
[0043] Supplementary Note 1. A method comprising: receiving training data representing a set of operational behaviors associated with each component of a set of components constituting a system being designed; generating a predictive model based on the training data, the predictive model being configured to predict a normal operational behavior associated with a given component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components; receiving real-time data including a real-time operational behavior associated with the given component and representing a set of real-time operational behaviors associated with each component of the set of components; classifying the real-time operational behavior associated with the given component as either normal or abnormal based on the predictive model; and generating a fault diagnosis notification if the real-time operational behavior is classified as abnormal.
[0044] Addendum 2. The method of Addendum 1, wherein the set of components includes a vehicle component and the system being designed is a vehicle.
[0045] Addendum 3. The method of any one of Addendums 1 to 2, wherein the set of components includes aircraft components and the system being designed is an aircraft.
[0046] Clause 4. The method of any of clauses 1 to 3, wherein the training data is received from a plurality of aircraft in a fleet.
[0047] Addendum 5. The method of any of Addendums 1-4, wherein the classification of the real-time operational behavior associated with the given component is performed during flight of the aircraft by a processor located in an avionics bay of the aircraft.
[0048] Addendum 6. The method of any of Addendums 1-5, wherein generating the predictive model includes training the predictive model by a supervised machine learning process using the training data.
[0049] Addendum 7. The method of any of Addendums 1-6, wherein the set of operational behaviors of the training data includes both normal and non-nominal operational behaviors.
[0050] Addendum 8. The method of any one of Addendums 1 to 7, wherein the set of real-time operational behaviors includes user inputs, equipment states, measurements, or a combination thereof.
[0051] Appendix 9. The method of any of appendices 1-8, further comprising: generating a second prediction model based on the training data, the second prediction model configured to predict a second normal operational behavior associated with a second component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components, the set of real-time operational behaviors including a second real-time operational behavior for the second component, classifying the second real-time operational behavior associated with the second component as either normal or abnormal, and generating a fault diagnosis notification if the second real-time operational behavior is classified as abnormal.
[0052] Addendum 10. The method of any one of Addendums 1 to 9, further comprising performing a pattern analysis based on the first prediction model and the second prediction model to identify the given component, and the fault diagnosis notification includes an identification of the given component.
[0053] Addendum 11. The method of any one of Addendums 1 to 10, further comprising sending the fault diagnosis notification to an output device.
[0054] Addendum 12. A system comprising: a set of components in a designed system configured to generate real-time data representative of a set of real-time operational behaviors associated with each component in the set of components; a computerized predictive model for predicting normal operational behavior associated with a given component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components; a processor configured to receive the set of real-time operational behaviors including the real-time operational behavior associated with the given component and classify the real-time operational behavior associated with the given component as either normal or abnormal based on the predictive model; and an output device configured to output a fault diagnosis notification if the processor classifies the real-time operational behavior as abnormal.
[0055] Addendum 13. The system of Addendum 12, further comprising a second processor configured to receive training data representing a set of operational behaviors associated with each component of the set of components and generate the predictive model based on the training data.
[0056] Addendum 14. The system of Addendum 12 or 13, wherein the set of components includes a vehicle component and the system to be designed is a vehicle.
[0057] Addendum 15. The system of any one of Addendums 12 to 14, wherein the set of components includes aircraft components and the system to be designed is an aircraft.
[0058] Addendum 16. The system of any one of Addendums 12 to 15, wherein the set of components includes a group of user input devices, a group of instruments, a group of measurement sensors, or a combination thereof.
[0059] Addendum 17. A method comprising: receiving real-time data representing a set of real-time operational behaviors associated with each component of a set of components constituting a system being designed, the set including a real-time operational behavior associated with a given component in the set of components; classifying the real-time operational behavior associated with the given component as either normal or abnormal based on a predictive model configured to predict normal operational behavior associated with the given component in relation to other normal operational behaviors associated with each of the other components in the set of components; and generating a fault diagnosis notification if the real-time operational behavior is classified as abnormal.
[0060] Addendum 18. The method of Addendum 17, further comprising receiving training data representing a set of operational behaviors associated with each component of the set of components, and generating the predictive model based on the training data.
[0061] Addendum 19. The method of Addendum 17 or 18, wherein generating the predictive model includes training the predictive model by a supervised machine learning process using the training data.
[0062] Addendum 20. The method of any one of Addendums 17-19, wherein the set of real-time operational behaviors includes user inputs, equipment states, measurements, or a combination thereof.
[0063] The method 400 classifies real-time operational behavior associated with a given component as either normal or abnormal based on a predictive model configured to predict normal operational behavior associated with the given component relative to other normal operational behaviors associated with each of the other components in a set of components, thereby enabling the method to detect degraded components and / or subsystems in the context of the entire system being designed, rather than the behavior of a single component in the system. Other advantages may also be included.
[0064] Although various embodiments have been described in the above description, the disclosure is not limited thereto, but should be construed as a whole to encompass all modifications and variations apparent to those skilled in the art.
Claims
1. receiving training data representing a set of operational behaviors associated with each component of a collection of components that make up a system being designed; generating a predictive model based on the training data, the predictive model being configured to predict a normal operational behavior associated with a given component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components; receiving real-time data representing a set of real-time operational behaviors associated with each component of the collection of components, the real-time data including a real-time operational behavior associated with the given component; classifying the real-time operational behavior associated with the given component as either normal or abnormal based on the predictive model; generating a fault diagnosis notification if the real-time operational behavior is classified as abnormal.
2. The method of claim 1 , wherein the collection of components includes a vehicle component and the system under design is a vehicle.
3. The method of claim 1 or 2, wherein the collection of components includes aircraft components and the designed system is an aircraft.
4. The method of any preceding claim, wherein the training data is received from a plurality of aircraft in a fleet.
5. 5. The method of claim 3 or 4, wherein the classification of the real-time operational behavior associated with the given component is performed during flight of the aircraft by a processor located in an avionics bay of the aircraft.
6. The method of any of claims 1 to 5, wherein generating the predictive model comprises training the predictive model by a supervised machine learning process using the training data.
7. The method of any of claims 1 to 6, wherein the set of operational behaviors of the training data includes both nominal and non-nominal operational behaviors.
8. The method of any of claims 1 to 7, wherein the set of real-time operational behaviors includes user inputs, equipment states, measurements, or a combination thereof.
9. generating, based on the training data, a second prediction model configured to predict a second normal operational behavior associated with a second component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components, wherein the set of real-time operational behaviors includes a second real-time operational behavior for the second component; classifying the second real-time operational behavior associated with the second component as either normal or abnormal; The method of any of claims 1 to 8, further comprising: generating a fault diagnosis notification if the second real-time operational behavior is classified as abnormal.
10. 10. The method of claim 9, further comprising: performing a pattern analysis based on the first-mentioned predictive model and the second predictive model to identify the given component, and wherein the fault diagnosis notification includes an identification of the given component.
11. The method of any preceding claim, further comprising sending the fault diagnosis notification to an output device.
12. a collection of components in a system being designed configured to generate real-time data representative of a set of real-time operational behaviors associated with each component in the collection of components; a predictive model, the predictive model being computed to predict a normal operational behavior associated with a given component in the set of components in relation to other normal operational behaviors associated with each of the other components in the set of components; a processor configured to receive the set of real-time operational behaviors including a real-time operational behavior associated with the given component, and to classify the real-time operational behavior associated with the given component as either normal or abnormal based on the predictive model; and an output device configured to output a fault diagnosis notification if the processor classifies the real-time operational behavior as abnormal.
13. 13. The system of claim 12, further comprising a second processor configured to receive training data representing a set of operational behaviors associated with each component in the set of components and to generate the predictive model based on the training data.
14. the set of components includes a vehicle component, and the system being designed is a vehicle; or the set of components includes aircraft components, and the system being designed is an aircraft; or The system of claim 12 or 13, wherein the set of components comprises or satisfies at least one of: user input devices, instruments, measurement sensors, or a combination thereof.
Citation Information
Patent Citations
Reporting and prioritizing faults for aircraft downtime reduction
EP3413246A1
Artificial intelligence-based health management of host system
JP2016224913A
Pre-processor and abnormality sign diagnostic system
JP2019028929A
Hybrid propulsion vertical take-off and landing aircraft
JP2019501830A
Detection of anomalies in multivariate data
WO2018045241A2
