Maintenance computing system and method for an aircraft utilizing a predictive classifier

Abstract

A computing system (10) includes a processor (12) and a non-volatile memory (20) storing executable instructions that, in response to execution by the processor (12), cause the processor (12) to execute an inspection classifier (22) including at least a first artificial intelligence model (22a), the inspection classifier (22) configured to receive runtime event input data (28A-C) from a plurality of data sources associated with an aircraft, the data sources including structural health monitoring sensors equipped on the aircraft; extract features (22f) of the runtime event input data (28A-C); decide a predicted inspection classification (54A) based on the extracted features (22f), the predicted inspection classification (54A) being one of a plurality of candidate inspection classifications (30Aa-c); and output the predicted inspection classification (54A).

Description

Technical Field

[0001] This invention generally relates to machine learning and artificial intelligence processes and systems, and more specifically to determining appropriate classifications for maintenance events, repairs, and maintenance tracking. For example, the techniques described herein can be applied to aircraft-related data. Background Technology

[0002] Aircraft maintenance, including post-flight inspections, initial maintenance, and subsequent inspections and repairs, is essential for keeping aircraft flight-ready and represents a significant cost to vehicle operators. This maintenance typically involves a non-destructive assessment of a portion of the aircraft, followed by initial maintenance. The use of non-destructive assessment techniques can be labor-intensive and subject to human judgment regarding appropriate inspection techniques, maintenance, and follow-up plans. Summary of the Invention

[0003] This overview is not an extensive summary of this specification. It is not intended to identify key or essential elements of the specification, nor is it intended to depict any scope of the specification, any particular embodiment, or any scope of the claims. Its sole purpose is to present some concepts of this specification in a simplified form as a prelude to the more detailed descriptions presented in this disclosure.

[0004] A computing system is disclosed, comprising a processor and a non-volatile memory storing executable instructions, which, in response to execution by the processor, cause the processor to execute an inspection classifier including at least a first artificial intelligence (AI) model. The inspection classifier is configured to receive runtime event input data from multiple data sources associated with an aircraft. The data sources include structural health monitoring sensors mounted on the aircraft. The inspection classifier extracts features from the runtime event input data, determines a predicted inspection classification based on the extracted features, and outputs the predicted inspection classification. The predicted inspection classification is one of multiple candidate inspection classifications.

[0005] The features, functions, and advantages already discussed can be implemented independently in various embodiments or combined in other embodiments, further details of which can be seen in the following description and figures. Attached Figure Description

[0006] Figure 1 An illustration is shown depicting a maintenance system based on an example embodiment of the disclosure in this subject matter.

[0007] Figure 2 The depiction is based on Figure 1 The diagram shows the inspection classifier, maintenance classifier, and monitoring classifier of the maintenance system.

[0008] Figure 3A The depiction is based on Figure 1 The diagram illustrates the training phases of the classifier, including the initial training phase and the feedback training phase.

[0009] Figure 3B The depiction is based on Figure 1 The diagram illustrates the training phases of the maintenance classifier, including the initial training phase and the feedback training phase.

[0010] Figure 3C The depiction is based on Figure 1 The diagram illustrates the training phases of the monitoring classifier, including the initial training phase and the feedback training phase.

[0011] Figure 4A Examples of the information disclosed in this subject matter illustrate possible inspection classifications, maintenance classifications, and monitoring classifications that can be output by the inspection classifier, maintenance classifier, and monitoring classifier, respectively.

[0012] Figure 4B Another example of the disclosure in this subject matter shows a diagram depicting possible inspection classifications, maintenance classifications, and monitoring classifications that can be output by the inspection classifier, maintenance classifier, and monitoring classifier, respectively.

[0013] Figure 5 The depiction is based on Figure 1 A diagram illustrating the multidimensional feature space of the examiner.

[0014] Figure 6 The content disclosed in accordance with this topic is shown. Figure 4A An illustration of an exemplary graphical user interface for a check classifier in an example embodiment.

[0015] Figure 7 A specific example of the publicly available content in this topic is shown below. Figure 1 The flowchart shows the process of feedback training during the operation of the maintenance system.

[0016] Figure 8A An example of the content disclosed in this topic shows a flowchart of the training phase for a maintenance calculation method used in conjunction with aircraft maintenance.

[0017] Figure 8B yes Figure 8A The flowchart continues, showing the runtime phases of the method, illustrating the processing steps involved in using an inspection classifier that includes feedback training.

[0018] Figure 8C yes Figure 8B The flowchart continues, showing the runtime phases of the method, illustrating the processing steps involved in using a maintenance classifier that includes feedback training.

[0019] Figure 8D yes Figure 8C The flowchart continues, showing the runtime phases of the method, illustrating the processing steps involved in using a monitoring classifier that includes feedback training.

[0020] Figure 9 A schematic diagram of an example computing environment that can be used according to the systems and methods described herein is shown. Detailed Implementation

[0021] In summary, applying machine learning and artificial intelligence to the maintenance process to improve the accuracy, reliability, and efficiency of inspection and diagnosis presents both challenges and opportunities. Therefore, a maintenance system is provided that uses separate AI models to perform inspection, maintenance, and monitoring predictions based on runtime input. Each model can be trained with feedback based on technician input to improve performance over time.

[0022] refer to Figure 1 A maintenance system 10 is provided for the maintenance of vehicles (e.g., aircraft, ships, spacecraft, automobiles, etc.). System 10 includes a maintenance computing device 11 comprising a processor 12, an input / output module 16, volatile memory 14, and non-volatile memory 20 storing an application program 32 and three classifiers: an inspection classifier 22, a repair classifier 24, and a monitoring classifier 26, which respectively include a first artificial intelligence model 22a, a second artificial intelligence model 24a, and a third artificial intelligence model 26a. A bus 18 operatively couples the processor 12, the input / output module 16, and the volatile memory 14 to the non-volatile memory 20. Although application 32 and classifiers 22, 24, 26 are described as being hosted on a single computing device 11, it should be understood that application 32 and classifiers 22, 24, 26 may alternatively be hosted on multiple computing devices to which computing device 11 is communicatively coupled via network 15, including client computing devices 36 operatively coupled to maintain computing device 11. In some examples, network 15 may take the form of a local area network (LAN), a wide area network (WAN), a wired network, a wireless network, a personal area network (PAN), or a combination thereof, and may include the Internet.

[0023] System 10 includes a processor 12 configured to store application program 32 and classifiers 22, 24, 26 in non-volatile memory 20, which retains data containing stored instructions even without externally applied power. This non-volatile memory may be, for example, flash memory, hard disk, read-only memory (ROM), electrically erasable programmable memory (EEPROM), etc. Instructions include one or more programs, including application program 32, and data used by these programs sufficient to perform the operations described herein. In response to execution by processor 12, the instructions cause processor 12 to execute at least a check classifier 22 comprising a first artificial intelligence model 22a, a repair classifier 24 comprising at least a second artificial intelligence model 24a, and / or at least a monitoring classifier 26 comprising a third artificial intelligence model 26a.

[0024] Processor 12 is a microprocessor that includes a central processing unit (CPU), graphics processing unit (GPU), application-specific integrated circuit (ASIC), system-on-a-chip (SoC), field-programmable gate array (FPGA), logic circuitry, or other suitable type of microprocessor configured to perform the functions described herein. System 10 further includes volatile memory 14, such as random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), etc., which temporarily stores data only when power is applied during program execution.

[0025] In one example, a user operating client computing device 36 can send a maintenance-related query 35 to maintenance computing device 11. (See reference...) Figure 2Further described, maintenance-related query 35 may include input data 28A-C related to inspections by inspection classifier 22, inspection-related input data 48A-C related to maintenance by repair classifier 24, or maintenance-related input data 58A-C related to maintenance monitoring actions by monitoring classifier 26. The processor 12 of maintenance computing device 11 is configured to receive maintenance-related query 35 from the user and execute one of the classifiers 22, 24, and 26 of application 32 to determine the most appropriate inspection classification via inspection classifier 22, the most appropriate maintenance classification via repair classifier 24, and / or the most appropriate monitoring classification via monitoring classifier 26. Therefore, as discussed below, based on the maintenance design indicated by maintenance-related input data 58A-58C, the monitoring classification generated by monitoring classifier 26 may indicate, for example, the predicted lifecycle of a maintenance, a monitoring plan for a maintenance, and / or monitoring maintenance actions, which are subsequent maintenance performed on the initial maintenance for which maintenance-related input data has been collected during monitoring. Then, the processor 12 returns a response 37 (query result) to the maintenance-related query 35 based on the decisions made by the classifiers 22, 24, and 26 of the application 32. The response 37 contains the output as the result of the decisions made by the classifiers 22, 24, and 26.

[0026] The client computing device 36 can execute an application client 32A to send a query 35 to the maintenance computing device 11 upon detecting user input 38, and subsequently receive the query result 37 from the maintenance computing device 11. The application client 32A can be coupled to the graphical user interface 34 of the client computing device 36 to display the graphical output 40 of the received query result 37.

[0027] refer to Figure 2 The diagram depicts a view of an inspection classifier 22, a maintenance classifier 24, and a monitoring classifier 26. The computing system 10 includes a processor 12 and a non-volatile memory 20 storing executable instructions that, in response to execution by the processor 12, cause the processor 12 to execute an inspection classifier 22 including at least a first artificial intelligence model 22a. The inspection classifier 22 is configured to receive runtime event input data 28A-C from multiple data sources associated with a vehicle (e.g., an aircraft), including structural health monitoring sensors mounted on the vehicle. The structural health monitoring sensors can be selected from a group consisting of inertial accelerometers, inertial gyroscopes, strain gauges, displacement transducers, air velocity sensors, temperature sensors, etc. Although... Figure 2 Three sets of runtime event input data 28A-C are shown, but it should be understood that there is no particular limitation on the number of sets of runtime event input data, and they can be alternatively numbered as, for example, four or more sets of input data.

[0028] The inspection classifier 22 extracts features 22f from the runtime event input data 28A-C, which may include at least one of camera images, audio data, or dimensionality measurements. Based on the extracted features 22f, it determines a predicted inspection classification 54A and outputs the predicted inspection classification 54A. The predicted inspection classification 54A is one of several candidate inspection classifications. The processor 12 receives user input 38 for the adopted inspection classification 30A for the runtime event input data 28A-C and uses the runtime event input data 28A-C and the adopted inspection classification 30A as feedback training data pairs to perform feedback training of the first artificial intelligence model 22a. The first artificial intelligence model 22a may include an input layer 22b connected to one or more convolutional layers 22c; and an output layer 22d including multiple nodes 22e, each node 22e indicating the value of the extracted feature vector of the extracted features 22f.

[0029] refer to Figure 3A The inspection classifier 22 has been trained on inspection classifier training data 27, which includes inspection training input data 29A and associated inspection ground reality labels 29B. The inspection classifier training data 27 may further include at least one of camera images, audio data, or dimensional measurements. The inspection training input data 29A includes structural health data from structural health monitoring sensors equipped on the vehicle. The processor 12 is configured to pair the inspection ground reality labels 29B with the inspection training input data 29A and perform training of the first artificial intelligence model 22a using each pair of inspection ground reality labels 29B and inspection training input data 29A from the inspection classifier training data 27.

[0030] Similarly, processor 12 is configured to receive user input 38, which includes an adopted check classification 30A for runtime input data 28A-C, pair the adopted check classification 30A with the runtime input data 28A-C to create check feedback training data pairs 31A, and use the check feedback training data pairs 31A to run feedback training of check classifier 22. Figure 4A As shown, the inspection category 30A input by the user can be selected from multiple candidate inspection categories 30Aa-c. For example, it can include event A: non-destructive testing 30Aa is not recommended, event B: simple non-destructive testing 30Ab is recommended, and event C: complex non-destructive testing 30Ac is recommended.

[0031] Return to reference Figure 2The processor is further configured to run a maintenance classifier 24, which includes at least a second artificial intelligence model 24a. The maintenance classifier 24 is configured to receive runtime inspection input data, including inspection-related input data 48A-C and the adopted inspection classification 30A, extract features from the runtime inspection input data, determine a predicted maintenance classification 54B based on the extracted features 24f, and output the predicted maintenance classification 54B. Although Figure 2 Three sets of inspection-related input data 48A-C are shown, but there is no particular limitation on the number of sets of inspection-related input data, and they can be alternatively numbered as, for example, four or more sets of input data. The predicted repair classification 54B is one of several candidate repair classifications. The processor 12 receives user input 38 for the adopted repair classification 30B for the runtime inspection-related input data 48A-C, and uses the runtime inspection-related input data 48A-C and the adopted repair classification 30B as feedback training data pairs to perform feedback training of the second artificial intelligence model 24a. The second artificial intelligence model 24a may include an input layer 24b, which is connected to one or more convolutional layers 24c; and an output layer 24d, which includes multiple nodes 24e, each node 24e indicating the value of the extracted feature vector of the extracted features 24f.

[0032] refer to Figure 3B The maintenance classifier 24 has been trained on maintenance classifier training data 47, which includes maintenance training input data 49A and associated maintenance ground condition labels 49B. Maintenance classifier training data 47 includes imaging learning and electrical measurements. Processor 12 is configured to pair maintenance ground condition labels 49B with maintenance training input data 49A and to perform training of the second artificial intelligence model 24a using each pair of ground condition labels 49B and maintenance training input data 49A from maintenance classifier training data 47.

[0033] Similarly, processor 12 is configured to receive user input 38, including a maintenance classification 30B adopted for inspection-related input data 48A-C for runtime inspection input data, pair the adopted maintenance classification 30B with the inspection-related input data 48A-C to create a maintenance feedback training data pair 31B, and use the maintenance feedback training data pair 31B to run feedback training of maintenance classifier 24. Figure 4A As shown, the repair category 30B entered by the user can be selected from multiple candidate repair categories 30Ba-d. Candidate repair categories 30Ba-d can include simple repair 30Ba, complex repair 30Bb, monitoring without repair 30Bc, and neither monitoring nor repair 30Bd.

[0034] Return to reference Figure 2The processor 12 further executes a monitoring classifier 26 including at least a third artificial intelligence model 26a. The monitoring classifier 26 is configured to receive runtime maintenance input data, including maintenance-related input data 58A-C and the adopted maintenance classification 30B, extract features 26f from the runtime maintenance input data, determine a predicted monitoring classification 54C based on the extracted features 26f, and output the predicted monitoring classification 54C. Although Figure 2 Three sets of maintenance-related input data 58A-C are shown, but it should be understood that there is no particular limitation on the number of sets of maintenance-related input data, and they can alternatively be numbered, for example, four or more sets of input data. The predicted monitoring classification 54C is one of several candidate monitoring classifications. The processor 12 receives user input 38 for the adopted monitoring classification 30C of the maintenance-related input data 58A-C for runtime maintenance input data, and uses the maintenance-related input data 58A-C and the adopted monitoring classification 30C as feedback training data to train the execution of the third artificial intelligence model 26a. The third artificial intelligence model 26a may include an input layer 26b, which is connected to one or more convolutional layers 26c; and an output layer 26d, which includes a plurality of nodes 26e, each node 26e indicating the value of the extracted feature vector of the extracted features 26f.

[0035] refer to Figure 3C The monitoring classifier 26 has been trained on monitoring classifier training data 57, which includes monitoring training input data 59A and associated monitoring ground condition labels 59B. Monitoring training input data 59A includes at least one of maintenance materials or maintenance types. The processor 12 is configured to pair monitoring ground condition labels 59B with monitoring training input data 59A and to perform training of the third artificial intelligence model 26a using each pair of monitoring ground condition labels 59B and monitoring training input data 59A in the monitoring classifier training data 57.

[0036] Similarly, processor 12 is configured to receive user input 38, which includes a monitoring classification 30C applied to runtime input data 58A-C, pairing the applied monitoring classification 30C with the runtime input data 58A-C to create a monitoring feedback training data pair 31C, and using the monitoring feedback training data pair 31C to run feedback training of the monitoring classifier 26. Figure 4A As shown, the monitoring category 30C input by the user can be selected from multiple candidate predicted monitoring categories 30Ca-c. The candidate predicted monitoring categories 30Ca-c can include the predicted maintenance lifecycle 30Ca, the maintenance monitoring plan 30Cb, and the monitoring maintenance actions 30Cc.

[0037] refer to Figure 4BAnother example of ground condition labels 130 associated with training input data is shown. The user-input inspection category 130A can be selected from: burned structure (composite material) 130Aa, burned structure (metal) 130Ab, delamination (composite material) 130Ac, fastener damage 130Ad, missing fastener 130Ae, metal melting 130Af, missing sealant 130Ag, paint chips 130Ah, paint peeling 130Ai, sealant tear 130Aj, crack 130Ak, dent 130Am, hole 130An, scratch 130Ao, and tear 130Ap. The user-input repair category 130B can be selected from: mixing 130Ba, plug placement 130Bb, fastener replacement 130Bc, skin replacement 130Bd, sanding 130Be, patching 130Bf, and skin patch 130Bg. The user-inputted predicted monitoring category 130C may include the predicted maintenance lifecycle 130Ca, the monitoring plan for maintenance 130Cb, and the monitoring of maintenance actions 130Cc.

[0038] For example, for lightning strike damage leaving molten metal on the skin surface, the user can enter the repair category 130Ba to use a metal-removing grinding technique, leaving a slight dent where the metal was ground away. For damage in unstructured areas behind the skin, the user can enter the plug placement 130Bb repair category to use a frozen plug, where the hole is inspected and prepared, and then a skin "plug" is fitted. This skin "plug" is frozen to shrink in size and then frictionally fitted into the hole at a warmer operating temperature. For fastener damage, the user can enter the fastener replacement 130Bc repair category to remove the fastener, check if the hole might be cracked, if the hole size is too large, and fit a replacement fastener. In some complex damage cases, when the skin patch is unsuitable, the user can enter the skin replacement 130Bd repair category. For minor skin area damage, such as slight damage, if grinding can remove the damage while maintaining a certain skin thickness, surface smoothness within specified limits, and the resulting surface unevenness within permissible limits (especially in the areas of the Pitot probe and attachment angle sensor), the user can enter the grinding 130Be repair category. For lightning burns on composite material surfaces, the user can enter the patching 130Bf repair category to remove the damage in a ring-like manner, ensuring that the number of composite layers is calculated when patching removes all damage. The number of layers affected by patching can be used to determine the type of repair required. For field area skin, where damage is difficult to repair with simple methods, the user can enter the skin patching 130Bg repair category to eliminate damage such as tears or large holes by cutting the material and placing a patch in its place.

[0039] refer to Figure 5A detailed view of the inspection classifier 22 is shown using multidimensional space 44. It should be understood that the maintenance classifier 24 and the monitoring classifier 26 can be similarly configured with multidimensional spaces adapted to their respective inputs. Multimodal information from multiple sensors is organized into multidimensional space 44, which contains the sum of inputs 28A-C to define the category of the event. In multidimensional space 44, a data point 45A is created based on a first dimension 46A corresponding to the first runtime event input data 28A, a second dimension 46B corresponding to the second runtime event input data 28B, and a third dimension 46C corresponding to the third runtime event input data 28C. Over time, multiple data points 45A, 45B, and 45C are created in this multidimensional space 44, such that multidimensional space 44 is associated with the resulting output: the predicted inspection classification 54A.

[0040] In this example, events are categorized into one of three classes: Event A 42A (minor impact event, corresponding to "no non-destructive testing"), Event B 42B (moderate impact event, corresponding to "simple non-destructive testing"), and Event C 42C (major impact event, "complex non-destructive testing"). By appropriately summing and decomposing the inputs 42A-C, and adjusting the event factors based on the runtime event input data 28A-C and the adopted inspection classification 30A, an algorithm is used within the multidimensional space 44 to guide the next action.

[0041] Over time, the accuracy of the predicted classification 54A of the multidimensional space 44 output increases by continuously adjusting the input data 28A-C and the added user input 30A, which continue to serve as input to the first artificial intelligence model 22a.

[0042] Therefore, in aviation applications, relative inputs from various sensors located on and off the aircraft being compared, along with predicted steps for aircraft maintenance or monitoring, can be transmitted to continue aircraft operations. Using machine learning in networked design allows for the summarization of action processes using a multidimensional view of various sensor inputs, surpassing the capabilities of spreadsheets and database relational processing with three or more linear tools.

[0043] A value is given by the correlation in the multidimensional space and then used as a response to lifecycle actions in manufacturing, design, or service. Machine learning is used to link the data to decisions made in the past, helping to define the multidimensional space by obtaining, for example, inputs that have a smaller impact on the output. Then, because many data points 45A-C are created in the multidimensional space 44, the algorithm used in the multidimensional space 44 can be adjusted over time to provide a smaller impact on the output classification 54A.

[0044] Reference Figure 6The illustration shows the data based on... Figure 1 The example execution application client 32A of the client computing device 36 is an exemplary maintenance system graphical user interface 34. In this example, the inspection classifier has output a predicted inspection classification 54A for a burnt structure (metal) based on runtime event input data of an aircraft with identifiable damage to its tail. A user prompt 56 is displayed on the graphical user interface 34 for the user to indicate whether they accept the predicted inspection classification 54A. If the user does not accept the predicted inspection classification 54A, a text box 55 is provided for the user to enter a new inspection classification for the tail damage. In other embodiments, for example, a multi-selector or drop-down menu may be alternatively configured for the user to enter new alternative inspection classifications. The newly entered inspection classification is then used for feedback training of the inspection classifier. If the user accepts the predicted inspection classification 54A, then the predicted inspection classification is then used for feedback training of the inspection classifier.

[0045] refer to Figure 7 A flowchart of a method 300 for training an inspection classifier, a maintenance classifier, and a monitoring classifier is shown below, based on an example. The following description of method 300 is based on the description above and... Figure 1-6 The software and hardware components shown in Figure 9 are used to provide this. It should be understood that method 300 can also be executed in other environments using other suitable hardware and software components.

[0046] In step 302, the inspection classifier receives pilot reports as runtime event input data. In step 304, the inspection classifier receives structural health monitoring sensor data from structural health monitoring sensors mounted on the aircraft as runtime event input data. In step 306, the inspection classifier receives visual data from the aircraft as runtime event input data. In step 308, the inspection classifier extracts features from the runtime event input data. In step 310, the inspection classifier determines the predicted inspection classification (applicable inspection) based on the extracted features and outputs the predicted inspection classification. In step 312, the inspection classifier receives the adopted inspection classification input by the user for feedback training. In step 314, feedback training is performed on the inspection classifier using the runtime event input data and the adopted inspection classification input by the user.

[0047] In step 316, the maintenance classifier receives the inspection classification adopted by the user. In step 318, the maintenance classifier receives eddy current detection as inspection-related input data. In step 320, the maintenance classifier receives infrared detection as inspection-related input data. In step 322, the maintenance classifier receives ultrasonic detection as inspection-related input data. In step 324, the maintenance classifier extracts features from the inspection-related input data. In step 326, the maintenance classifier determines the predicted maintenance classification (applicable maintenance) based on the extracted features and outputs the predicted maintenance classification. In step 328, the maintenance classifier receives the adopted maintenance classification adopted by the user for feedback training. In step 330, feedback training is run on the maintenance classifier using the inspection-related input data and the adopted maintenance classification adopted by the user.

[0048] In step 332, the monitoring classifier receives the maintenance classification adopted by the user. In step 334, the monitoring classifier receives sealing data as runtime maintenance input data. In step 336, the monitoring classifier receives speed belt data as runtime maintenance input data. In step 338, the monitoring classifier receives polishing data as runtime maintenance input data. In step 340, the monitoring classifier extracts features from the runtime maintenance input data. In step 342, the monitoring classifier determines the predicted monitoring classification (applicable monitoring action) based on the extracted features and outputs the predicted monitoring classification. In step 344, the monitoring classifier receives the monitoring classification adopted by the user. In step 346, the monitoring classifier is trained using runtime maintenance input data and the adopted monitoring classification adopted by the user.

[0049] refer to Figures 8A-8D A flowchart of a maintenance calculation method 400 for the maintenance of vehicles (e.g., aircraft) is shown. Figure 8A The diagram illustrates the processing steps of method 400 related to training the inspection classifier, maintenance classifier, and monitoring classifier. Refer to the description above and... Figure 1-6 The software and hardware components shown in Figure 9 are used to provide the following description of method 400. It should be understood that method 400 can also be executed in other environments using other suitable hardware and software components.

[0050] In step 402, an inspection classifier is trained on inspection classifier training data, which includes inspection training input data and associated inspection ground condition labels. The training input data includes structural health data from structural health monitoring sensors mounted on the aircraft, and the ground condition labels are inspection classifications associated with the training input data and input by the user, selected from a plurality of candidate inspection classifications. Step 402a may include receiving inspection classifier training input data in step 402, including at least one of camera images, audio data, or dimensionality measurements, and the runtime event input data further includes at least one of camera images, audio data, or dimensionality measurements.

[0051] In step 404, a maintenance classifier is trained on inspection training data, which includes inspection training input data and associated ground condition labels. The inspection training input data includes imaging learning and electrical measurements, and the ground condition labels are maintenance categories input by the user and associated with the inspection training input data, selected from multiple candidate maintenance categories. In step 406, a monitoring classifier is trained on monitoring training data, which includes monitoring training input data and associated ground condition labels. The monitoring training input data includes maintenance-related data, and the ground condition labels are monitoring categories input by the user and associated with the monitoring training input data, selected from multiple candidate monitoring categories.

[0052] refer to Figure 8B The flowchart is shown, which is Figure 8A The method 400 continues, and the operation and execution of feedback training of the inspection classifier are shown. In step 502, the inspection classifier, which includes at least a first artificial intelligence model, is run using a processor and associated memory. Step 502 includes step 504 receiving runtime event input data from multiple data sources associated with the aircraft, including structural health monitoring sensors mounted on the aircraft; step 506 extracting features from the runtime event input data; step 508 determining a predicted inspection classification based on the extracted features, the predicted inspection classification being one of multiple candidate inspection classifications; and step 510 outputting the predicted inspection classification. Step 504 may include step 504a selecting a structural health monitoring sensor from a group consisting of inertial accelerometers, inertial gyroscopes, strain gauges, displacement transducers, air velocity sensors, and temperature sensors. Step 508 may include step 508a configuring the candidate inspection classifications to include not recommending non-destructive testing, recommending simple non-destructive testing, and recommending complex non-destructive testing.

[0053] After step 502 of running the inspection classifier, in step 512, user input regarding the inspection classification adopted for the runtime event input data is received. In step 514, the runtime event input data and the adopted inspection classification are used as feedback training data pairs to perform feedback training of the first artificial intelligence model.

[0054] refer to Figure 8C The flowchart is shown, which is Figure 8B The method 400 continues, and the operation and execution of feedback training of the repair classifier are shown. In step 602, the repair classifier, which includes at least a second artificial intelligence model, is run using a processor and associated memory. Step 602 includes receiving runtime inspection input data (604) including inputs associated with the inspection and the adopted inspection classification, extracting features from the runtime inspection input data (606), determining a predicted repair classification based on the extracted features (608) as one of a plurality of candidate repair classifications, and outputting the predicted repair classification (610). Step 608 may include configuring candidate repair classifications to include simple repair, complex repair, monitored but not repaired, and neither monitored nor repaired (608a).

[0055] After step 602 of running the maintenance classifier, in step 612, user input regarding the maintenance classification adopted for the inspection-related input data is received. In step 614, the inspection-related input data and the adopted maintenance classification are used as feedback training data to train the execution of the second artificial intelligence model.

[0056] refer to Figure 8D The flowchart is shown, which is Figure 8C The method 400 continues, and illustrates the operation and feedback training of the monitoring classifier. In step 702, the monitoring classifier, which includes at least a third artificial intelligence model, is run using a processor and associated memory.

[0057] Step 702 includes a step 704 of receiving runtime maintenance input data, which includes maintenance-related input data and the adopted maintenance classification; a step 706 of extracting features from the runtime maintenance input data; a step 708 of determining a predicted monitoring classification based on the extracted features, the predicted monitoring classification being selected from one of a plurality of candidate monitoring classifications; and a step 710 of outputting the predicted monitoring classification. Step 704 may include a step 704a of receiving maintenance-related input data including at least one of maintenance materials or maintenance type. Step 708 may include a step 708a of determining the predicted monitoring classification, the predicted monitoring classification including a predicted maintenance lifecycle, a monitoring plan for maintenance, and monitored maintenance actions.

[0058] After step 702 of running the monitoring classifier, in step 712, user input regarding the adopted monitoring classification for the maintenance-related input data is received. In step 714, the maintenance-related input data and the adopted monitoring classification are used as feedback training data to train the execution of the third artificial intelligence model.

[0059] The aforementioned systems and methods offer the technological advantage of enabling machine learning techniques to predict inspection, repair, and follow-up categories associated with the maintenance of aircraft components. This assists technicians in their inspections, repairs, and subsequent actions, while providing them with control over decisions made at each stage of the maintenance process. The systems and methods are configured to learn and improve the accuracy of their predictions over time, as each AI model is trained based on real-world feedback from technicians. Utilizing such systems, efficient and high-quality inspection, repair, and follow-up procedures can be reliably maintained.

[0060] Figure 9 A non-limiting embodiment of a computing system 800 that can implement one or more of the above-described processes is illustrated schematically. The computing system 800 is shown in a simplified form. The computing system 800 can embody the above-described and... Figure 1 and Figure 2 The maintenance computing device 11 or the client computing device 36 is shown. (See reference.) Figure 1 and 2 The computing system 800 may take the form of one or more personal computers, server computers, tablet computers, home entertainment computers, network computing devices, gaming devices, mobile computing devices, mobile communication devices (e.g., smartphones), and / or other computing devices, as well as wearable computing devices such as smartwatches and head-mounted augmented reality devices.

[0061] The computing system 800 includes a logic processor 802, volatile memory 804, and non-volatile storage device 806. The computing system 800 may optionally include a display subsystem 808, an input subsystem 810, a communication subsystem 812, and / or... Figure 9 Other components not shown.

[0062] The logic processor 802 includes one or more physical devices configured to execute instructions. For example, the logic processor may be configured to execute instructions as part of one or more application programs, programs, routines, libraries, objects, components, data structures, or other logical constructs. Such instructions can be used to perform tasks, implement data types, transition the state of one or more components, achieve technical effects, or otherwise achieve desired results.

[0063] A logic processor may include one or more physical processors (hardware) configured to execute software instructions. Additionally or alternatively, a logic processor may include one or more hardware logic circuits or firmware devices configured to run hardware-implemented logic or firmware instructions. The processor of logic processor 802 may be single-core or multi-core, and the instructions running on it may be configured for sequential, parallel, and / or distributed processing. The various components of the logic processor may optionally be distributed across two or more separate devices that can be remotely located and / or configured for coordinated processing. Aspects of the logic processor may be virtualized and run by remotely accessible, networked computing devices configured in a cloud computing configuration. In this context, it will be understood that these virtualized aspects run on different physical logic processors on various different machines.

[0064] The non-volatile storage device 806 includes one or more physical devices configured to store instructions executable by a logic processor to implement the methods and processes described herein. When such methods and processes are implemented, the state of the non-volatile storage device 806 can be transformed—for example, to store different data.

[0065] Non-volatile storage device 806 may include removable and / or built-in physical devices. Non-volatile storage device 806 may include optical memory (e.g., CD, DVD, HD-DVD, Blu-ray disc, etc.), semiconductor memory (e.g., ROM, EPROM, EEPROM, FLASH memory, etc.), and / or magnetic memory (e.g., hard disk drive, floppy disk drive, magnetic tape drive, MRAM, etc.) or other high-capacity storage technologies. Non-volatile storage device 806 may include non-volatile, dynamic, static, read / write, read-only, sequential access, location-addressable, file-addressable, and / or content-addressable devices. It should be understood that non-volatile storage device 806 is configured to retain instructions even when power to non-volatile storage device 806 is cut off.

[0066] Volatile memory 804 may include a physical device containing random access memory. Volatile memory 804 is typically used by logic processor 802 to temporarily store information during the processing of software instructions. It should be understood that volatile memory 804 typically does not continue storing instructions when power is lost.

[0067] The logic processor 802, volatile memory 804, and non-volatile storage device 806 can be integrated together into one or more hardware logic components. Such hardware logic components may include, for example, field-programmable gate arrays (FPGAs), application-specific integrated circuits (PASICs / ASICs), application-specific standard products (PSSPs / ASSPs), system-on-a-chip (SoCs), and complex programmable logic devices (CPLDs).

[0068] The terms "module," "program," and "engine" can be used to describe an aspect of computing system 800, typically implemented in software by a processor to perform a specific function using a portion of volatile memory. This function involves translation processing, specifically configuring the processor to perform that function. Therefore, a module, program, or engine can be instantiated by logic processor 802 using a portion of volatile memory 804 to execute instructions held by non-volatile storage device 806. It should be understood that different modules, programs, and / or engines can be instantiated from the same application, service, code block, object, library, routine, API, function, etc. Similarly, the same module, program, and / or engine can be instantiated by different applications, services, code blocks, objects, routines, APIs, functions, etc. The terms "module," "program," and "engine" can contain a single or a group of executable files, data files, libraries, drivers, scripts, database records, etc.

[0069] When included, the display subsystem 808 can be used to present a visual representation of the data held by the non-volatile storage device 806. The visual representation can take the form of a graphical user interface (GUI). Since the methods and processes described herein change the data held by the non-volatile storage device, thereby changing the state of the non-volatile storage device, the state of the display subsystem 808 can also be transformed to visually represent the changes in the underlying data. The display subsystem 808 may include one or more display devices using virtually any type of technology. Such display devices may be combined with the logic processor 802, the volatile memory 804, and / or the non-volatile storage device 806 in a shared housing, or such display devices may be peripheral display devices.

[0070] When included, the input subsystem 810 may include or interface with one or more user input devices, such as a keyboard, mouse, touchscreen, or game controller. In some embodiments, the input subsystem may include or interface with selected Natural User Input (NUI) components. Such components may be integrated or peripheral, and the translation and / or processing of input actions may be performed on-machine or off-machine. Example NUI components may include microphones for speech and / or speech recognition; infrared, color, stereo, and / or depth cameras for machine vision and / or gesture recognition; head trackers, eye trackers, accelerometers, and / or gyroscopes for motion detection and / or intent recognition; and electric field sensing components and / or any other suitable sensors for assessing brain activity.

[0071] When included, the communication subsystem 812 can be configured to communicatively couple the various computing devices described herein to each other and to other devices. The communication subsystem 812 may include wired and / or wireless communication devices compatible with one or more different communication protocols. As a non-limiting example, the communication subsystem may be configured to communicate via a wireless telephone network, or a wired or wireless local area network or wide area network (e.g., an HDMI connection via Wi-Fi). In some embodiments, the communication subsystem may allow the computing system 800 to send and / or receive messages to and / or from other devices via a network such as the Internet.

[0072] By leveraging the multidimensional space of multiple corresponding input data streams across multiple artificial intelligence models, the system and process described in this paper have the potential benefit of improving the reliability of AI and machine learning decision systems, and utilize past human learning experience to create a baseline for the prediction and classification of AI model outputs. The simple sensing of a single parameter as an indication of aircraft damage, multiplexed with sensor inputs from other single sources, distinguishes it from systems that measure other parameters and compare them with historical values ​​of the same parameter.

[0073] It should be understood that the “and / or” used in this article refers to logically separated operations, therefore A and / or B have the following truth table.

[0074] A B A and / or B real real real real Fake real Fake real real Fake Fake Fake

[0075] Furthermore, it should be understood that the terms “comprising,” “having,” “including,” variations thereof, and other similar words used in the detailed description or claims are intended to be included in a manner similar to the term “comprising.” The term “comprising” serves as an open transitional word without excluding any additional or other elements.

[0076] It should be understood that the configurations and / or methods described herein are exemplary in nature, and these specific embodiments or examples should not be considered limiting, as many variations are possible. The specific procedures or methods described herein may represent one or more of any number of processing strategies. Therefore, the various actions shown and / or described may be performed in the order shown and / or described, in another order, in parallel, or omitted. Similarly, the order of the above processes may be changed.

[0077] In addition, this disclosure includes configurations based on the following terms.

[0078] Clause 1. A maintenance computing system comprising: a processor and a non-volatile memory storing executable instructions, the executable instructions being responsive to execution by the processor to cause the processor to: run an inspection classifier including at least a first artificial intelligence model, the inspection classifier being configured to: receive runtime event input data from a plurality of data sources associated with a vehicle, the data sources including structural health monitoring sensors mounted on the vehicle; extract features from the runtime event input data; determine a predicted inspection classification based on the extracted features, the predicted inspection classification being one of a plurality of candidate inspection classifications; and output the predicted inspection classification.

[0079] Clause 2. The maintenance computing system according to Clause 1, wherein the inspection classifier has been trained on inspection classifier training data including inspection training input data and associated inspection ground condition labels, the inspection training input data including structural health data from one or more structural health monitoring sensors equipped on the vehicle, and the inspection ground condition labels being user-input inspection classifications associated with the inspection training input data, the user-input inspection classifications being selected from the plurality of candidate inspection classifications.

[0080] Clause 3. The maintenance computing system as described in Clauses 1 and 2, wherein the inspection classification training data further comprises at least one of camera images, audio data, or dimensionality measurements; and wherein the runtime event input data further comprises at least one of camera images, audio data, or dimensionality measurements.

[0081] Clause 4. The maintenance computing system according to any one of Clauses 1-3, wherein the processor is configured to: receive a user input of an adopted inspection classification in response to runtime event input data; and use the runtime event input data and the adopted inspection classification as feedback training data to train a first artificial intelligence model.

[0082] Clause 5. The maintenance calculation system according to any one of Clauses 1 to 4, wherein the one or more structural health monitoring sensors are selected from the group consisting of inertial accelerometers, inertial gyroscopes, strain gauges, displacement transducers, air velocity sensors, and temperature sensors.

[0083] Clause 6. A maintenance calculation system pursuant to any one of Clauses 1 to 5, wherein the candidate inspection classification includes recommending no non-destructive testing, recommending simple non-destructive testing, and recommending complex non-destructive testing.

[0084] Clause 7. The maintenance computing system according to Clause 4, wherein the processor is further configured to run a maintenance classifier including at least a second artificial intelligence model, the maintenance classifier being configured to: receive runtime inspection input data including inspection-related input data and an adopted inspection classification; extract inspection features from the runtime inspection input data; determine a predicted maintenance classification based on the extracted inspection features, the predicted maintenance classification being one of a plurality of candidate maintenance classifications; and output the predicted maintenance classification.

[0085] Clause 8. The maintenance computing system according to Clause 7, wherein the maintenance classifier has been trained on maintenance classifier training data, which includes maintenance training input data and associated ground condition labels, the maintenance training input data including imaging learning and electrical measurements, and the ground condition labels are user-input maintenance categories associated with the maintenance training input data, the user-input maintenance categories being selected from a plurality of candidate maintenance categories.

[0086] Clause 9. A maintenance calculation system as described in Clause 7 or 8, wherein the candidate maintenance categories include simple maintenance, complex maintenance, monitored but not maintained, and neither monitored nor maintained.

[0087] Clause 10. The maintenance computing system according to any one of Clauses 7-9, wherein the processor is configured to: receive user input of a maintenance classification adopted for the runtime inspection input data; and use the inspection-related input data and the adopted maintenance classification as feedback training data to train a second artificial intelligence model.

[0088] Clause 11. The maintenance computing system according to Clause 7, wherein the processor further runs a monitoring classifier including at least a third artificial intelligence model, the monitoring classifier being configured to: receive runtime maintenance input data, the runtime maintenance input data including maintenance-related input data and the maintenance classification adopted; extract maintenance features from the runtime maintenance input data; determine a predicted monitoring classification based on the extracted maintenance features, the predicted monitoring classification being one of a plurality of candidate monitoring classifications; and output the predicted monitoring classification.

[0089] Clause 12. The maintenance calculation system as described in Clause 11, wherein the maintenance-related input data includes at least one of maintenance materials or maintenance type.

[0090] Clause 13. The maintenance computing system as described in Clause 11 or 12, wherein the predicted monitoring categories include predicted maintenance lifecycles, monitoring plans for maintenance, and monitoring maintenance actions.

[0091] Clause 14. The maintenance computing system pursuant to Clause 11, wherein the processor is configured to: receive user input for a monitoring classification adopted for runtime maintenance input data; and use the runtime maintenance input data and the adopted monitoring classification as feedback training data to train a third artificial intelligence model.

[0092] Clause 15. A maintenance computation method comprising: running an inspection classifier using a processor and associated memory, the inspection classifier including at least a first artificial intelligence model, the running of the inspection classifier comprising: receiving runtime event input data from a plurality of data sources associated with a vehicle, the data sources including structural health monitoring sensors mounted on the vehicle; extracting features from the runtime event input data; determining a predicted inspection classification based on the extracted features, the predicted inspection classification being one of a plurality of candidate inspection classifications; and outputting the predicted inspection classification.

[0093] Clause 16. The maintenance calculation method according to Clause 15 further includes: training an inspection classifier on inspection classifier training data prior to running the inspection classifier, the inspection classifier training data including training input data and associated ground condition labels, the training input data including structural health data from structural health monitoring sensors mounted on the vehicle, and the ground condition labels being user-input inspection classifications associated with the training input data, the user-input inspection classifications being selected from a plurality of candidate inspection classifications.

[0094] Clause 17. The maintenance computation method according to Clause 15 or 16, wherein the inspection classifier training data further includes at least one of camera images, audio data, or dimensionality measurements; and wherein the runtime event input data further includes at least one of camera images, audio data, or dimensionality measurements.

[0095] Clause 18. The maintenance calculation method according to any one of Clauses 15 to 17 further includes: receiving user input of an inspection classification adopted for runtime event input data; and using the runtime event input data and the adopted inspection classification as feedback training data to perform feedback training on executing a first artificial intelligence model.

[0096] Clause 19. The maintenance calculation method according to any one of Clauses 15 to 18, wherein the structural health monitoring sensor is selected from the group consisting of an inertial accelerometer, an inertial gyroscope, a strain gauge, a displacement transducer, an air velocity sensor, and a temperature sensor.

[0097] Clause 20. The maintenance calculation method according to any one of Clauses 15 to 19, wherein the candidate inspection is divided into non-destructive testing not recommended, simple non-destructive testing recommended, and complex non-destructive testing recommended.

[0098] Clause 21. The maintenance calculation method according to Clause 18 further includes: running a maintenance classifier that includes at least a second artificial intelligence model, wherein running the inspection classifier includes: receiving runtime inspection input data, the data including inspection-related input data and an inspection classification adopted; extracting inspection features from the runtime inspection input data; determining a predicted maintenance classification based on the extracted inspection features, the predicted maintenance classification being one of a plurality of candidate maintenance classifications; and outputting the predicted maintenance classification.

[0099] Clause 22. The maintenance calculation method according to Clause 21 further includes: training a maintenance classifier on maintenance classifier training data before running the maintenance classifier, the maintenance classifier training data including maintenance classifier training input data and associated ground condition labels, the maintenance classifier training input data including imaging learning and electrical measurements, and the ground condition labels being user-input maintenance classifications associated with the maintenance classifier training input data, the user-input maintenance classifications being selected from a plurality of candidate maintenance classifications.

[0100] Clause 23. The maintenance calculation method as described in Clause 21 or 22, wherein the candidate maintenance categories include simple maintenance, complex maintenance, monitored but not maintained, and neither monitored nor maintained.

[0101] Clause 24. The maintenance calculation method according to any one of Clauses 21 to 23, wherein the processor is configured to: receive user input of a maintenance classification adopted for runtime inspection input data; and use the runtime inspection input data and the adopted maintenance classification as feedback training data to perform feedback training on a second artificial intelligence model.

[0102] Clause 25. The maintenance calculation method according to Clause 21 further includes: running a monitoring classifier comprising at least a third artificial intelligence model, the running monitoring classifier comprising: receiving runtime maintenance input data, the data including maintenance-related input data and a maintenance classification adopted; extracting maintenance features from the runtime maintenance input data; determining a predicted monitoring classification based on the extracted maintenance features, the predicted monitoring classification being one of a plurality of candidate monitoring classifications; and outputting the predicted monitoring classification.

[0103] Clause 26. The maintenance calculation method according to any one of Clause 25, wherein the maintenance-related input data includes at least one of maintenance materials or maintenance type.

[0104] Clause 27. The maintenance calculation method as described in Clause 25 or 26, wherein the plurality of candidate monitoring categories includes predicted maintenance lifecycle, monitoring plans for maintenance, and monitoring maintenance actions.

[0105] Clause 28. The maintenance calculation method according to any one of Clauses 25-27 further comprises: training the monitoring classifier on monitoring training data prior to running the monitoring classifier, the monitoring training data including monitoring training input data and associated ground condition labels, the monitoring training input data including imaging learning and electrical measurements, and the ground condition labels being user-input maintenance categories associated with the inspection training input data, the user-input maintenance categories being selected from a plurality of candidate maintenance categories.

[0106] Clause 29. The maintenance calculation method according to any one of Clauses 25 to 28 further includes: receiving user input of a monitoring classification adopted for runtime maintenance input data; and using the runtime maintenance input data and the adopted monitoring classification as feedback training data to perform feedback training on a third artificial intelligence model.

[0107] Clause 30. A maintenance computing system comprising: a processor and a non-volatile memory storing executable instructions, the executable instructions being responsive to execution of the processor to cause the processor to: run an inspection classifier configured to determine a predicted inspection classification based on runtime event input data from structural health monitoring sensors mounted on a vehicle; output the predicted inspection classification; receive user input for the adopted inspection classification on the runtime event input data; perform feedback training on the execution of the inspection classification using the runtime event input data and the adopted inspection classification as feedback training data; run a maintenance classifier to determine a predicted maintenance classification based on runtime inspection input data, the runtime inspection input data including inspection-related input data and the adopted inspection classification; output the predicted maintenance classification; receive user input for the adopted maintenance classification on the runtime inspection input data; and perform feedback training on the execution of the maintenance classifier using the runtime inspection input data and the adopted maintenance classification as feedback training data.

[0108] Clause 31. The maintenance computing system according to Clause 31, wherein the processor is further configured to: run a monitoring classifier to determine a predicted monitoring classification based on runtime maintenance input data, the runtime maintenance input data including maintenance-related input data and the adopted maintenance classification; output the predicted monitoring classification; receive user input for the adopted monitoring classification of the runtime maintenance input data; and use the runtime maintenance input data and the adopted monitoring classification as feedback training data to perform feedback training on the execution monitoring classifier.

[0109] This subject matter disclosure includes all novel and non-obvious combinations and sub-combinations of the various features and techniques disclosed herein. The various features and techniques disclosed herein are not necessarily necessary for all examples disclosed herein. Furthermore, the various features and techniques disclosed herein may define patentable subject matter beyond the disclosed examples and may find use in other embodiments not expressly disclosed herein.

Claims

1. A maintenance computing system (10), comprising: A processor (12) and a non-volatile memory (20) storing executable instructions, which, in response to execution by the processor (12), cause the processor (12) to: Running an inspection classifier (22) that includes at least a first artificial intelligence model (22a), said inspection classifier (22) being configured to: Runtime event input data (28A-C) is received from multiple data sources associated with the vehicle, including structural health monitoring sensors equipped on the vehicle; Extract the features (22f) of the runtime event input data (28A-C); Based on the extracted features (22f), a predicted inspection category (54A) is determined, wherein the predicted inspection category (54A) is one of a plurality of candidate inspection categories (30Aa-c); and Output the predicted inspection classification (54A); The processor (12) is configured to: Receive user input (38) for the adopted inspection classification (30A) of the runtime event input data (28A-C); and Using the runtime event input data (28A-C) and the adopted inspection classification (30A) as feedback training data pairs (31A), feedback training of the first artificial intelligence model (22A) is performed; and A maintenance classifier (24) comprising at least a second artificial intelligence model (24a) is run, the maintenance classifier (24) being configured to: Receive runtime inspection input data (30A, 48A-C), which includes inspection-related input data (48A-C) and the inspection classification adopted (30A); Extract the inspection features (24f) from the runtime inspection input data (30A, 48A-C); The predicted maintenance category (54B) is determined based on the extracted inspection features (24f), and the predicted maintenance category (54B) is one of a plurality of candidate maintenance categories (30Ba-d); Output the predicted repair classification (54B); Receive user input (38) for the maintenance category (30B) adopted for the runtime check input data (30A, 48A-C); and Using the inspection-related input data (48A-C) and the adopted maintenance classification (30B) as feedback training data pairs (31B), the feedback training of the second artificial intelligence model (24a) is performed.

2. The maintenance computing system (10) according to claim 1, wherein, The inspection classifier (22) has been trained on inspection classifier training data (27) including inspection training input data (29A) and associated inspection ground condition labels (29B), the inspection training input data (29A) including structural health data from one or more structural health monitoring sensors equipped on the vehicle, and the inspection ground condition labels (29B) are inspection classifications (130A) of user inputs (38) associated with the inspection training input data (29A), the inspection classifications (130A) of the user inputs (38) being selected from the plurality of candidate inspection classifications (30Aa-c).

3. The maintenance computing system (10) according to claim 2: The examination of the classifier training data (27) further includes at least one of camera images, audio data, or dimensionality measurements; and The runtime event input data (28A-C) further includes at least one of camera images, audio data, or dimensional measurements.

4. The maintenance calculation system (10) according to any one of claims 1-3, wherein the one or more structural health monitoring sensors are selected from the group consisting of an inertial accelerometer, an inertial gyroscope, a strain gauge, a displacement transducer, an air velocity sensor and a temperature sensor.

5. The maintenance computing system (10) according to claim 1, wherein the maintenance classifier (24) has been trained on maintenance classifier training data (47), the maintenance classifier training data (47) including maintenance training input data (49A) and associated ground condition labels (49B), the maintenance training input data (49A) including imaging learning and electrical measurements, and the ground condition labels (49B) being user-input maintenance categories (30B) associated with the maintenance training input data (49A), the user-input maintenance categories (30B) being selected from the plurality of candidate maintenance categories (30Ba-d).

6. The maintenance computing system (10) according to claim 1, wherein the processor (12) further runs a monitoring classifier (26), which includes at least a third artificial intelligence model (26a), the monitoring classifier (26) being configured to: Receive runtime maintenance input data (30B, 58A-C), which includes maintenance-related input data (58A-C) and the maintenance category used (30B); Extract the maintenance features (26f) from the runtime maintenance input data (30B, 58A-C); The predicted monitoring category (54C) is determined based on the extracted maintenance features (26f), and the predicted monitoring category (54C) is one of a plurality of candidate monitoring categories (30Ca-c); Output the predicted monitoring category (54C); Receive user input (38) for the monitoring classification (30C) adopted for the runtime maintenance input data (30B, 58A-C); and Using the runtime maintenance input data (30B, 58A-C) and the adopted monitoring classification (30C) as feedback training data pairs (31C), the feedback training of the third artificial intelligence model (26a) is performed.

7. The maintenance calculation system (10) according to claim 6, wherein the maintenance-related input data (58A-C) includes at least one of maintenance materials or maintenance type.

8. A maintenance calculation method (300), comprising: Running an inspection classifier (22) using a processor (12) and associated memory (20), the inspection classifier (22) comprising at least a first artificial intelligence model (22a), and running the inspection classifier (22) includes: Runtime event input data (28A-C) is received from multiple data sources associated with the vehicle, the data sources including structural health monitoring sensors equipped on the vehicle; Extract the features (22f) of the runtime event input data (28A-C); The predicted inspection category (54A) is determined based on the extracted features (22f), and the predicted inspection category (54A) is one of a plurality of candidate inspection categories (30Aa-c); Output the predicted inspection classification (54A); Receive user input (38) for the adopted inspection classification (30A) of the runtime event input data (28A-C); Using the runtime event input data (28A-C) and the adopted inspection classification (30A) as feedback training data pairs (31A), feedback training of the first artificial intelligence model (22a) is performed; and Running a maintenance classifier that includes at least a second artificial intelligence model, wherein running the maintenance classifier includes: Receive runtime inspection input data, which includes inspection-related input data and the inspection classification used; Extract the inspection features from the runtime inspection input data; The predicted maintenance category is determined based on the extracted inspection features, and the predicted maintenance category is one of multiple candidate maintenance categories. Output the predicted repair category; Receive user input regarding the maintenance category used for the runtime check input data; and The second artificial intelligence model is trained using the runtime inspection input data and the adopted maintenance classification as feedback training data pairs.

9. The maintenance calculation method according to claim 8, further comprising: Before running the inspection classifier, the inspection classifier is trained on inspection classifier training data, which includes training input data and associated ground condition labels. The training input data includes structural health data from the structural health monitoring sensors equipped on the vehicle, and the ground condition labels are user-input inspection classifications associated with the training input data, which are selected from the plurality of candidate inspection classifications.

10. The maintenance calculation method according to claim 9, The classifier training data further includes at least one of camera images, audio data, or dimensionality measurements; and The runtime event input data further includes at least one of camera images, audio data, or dimensional measurements.

11. The maintenance calculation method according to claim 8, wherein the structural health monitoring sensor is selected from the group consisting of an inertial accelerometer, an inertial gyroscope, a strain gauge, a displacement transducer, an air velocity sensor, and a temperature sensor.

12. The maintenance calculation method according to claim 8, further comprising: Before running the maintenance classifier, the maintenance classifier is trained on maintenance classifier training data, which includes maintenance classifier training input data and associated ground condition labels. The maintenance classifier training input data includes imaging learning and electrical measurements, and the ground condition labels are user-input maintenance categories associated with the maintenance classifier training input data, which are selected from the plurality of candidate maintenance categories.

13. The maintenance calculation method according to claim 8, further comprising: Run a monitoring classifier that includes at least a third artificial intelligence model, wherein running the monitoring classifier includes: Receive runtime maintenance input data, which includes maintenance-related input data and the maintenance classification used; Extract the maintenance features from the runtime maintenance input data; The predicted monitoring category is determined based on the extracted maintenance features, and the predicted monitoring category is one of multiple candidate monitoring categories. Output the predicted monitoring category; Receive user input regarding the monitoring classification adopted for the runtime maintenance input data; and The third artificial intelligence model is trained using the runtime maintenance input data and the adopted monitoring classification as feedback training data pairs.

14. The maintenance calculation method according to claim 13, wherein the maintenance-related input data includes at least one of maintenance materials or maintenance type.

15. The maintenance calculation method according to claim 13, further comprising: Before running the monitoring classifier, the monitoring classifier is trained on monitoring training data, which includes monitoring training input data and associated ground condition labels. The monitoring training input data includes imaging learning and electrical measurements, and the ground condition labels are user-inputted maintenance categories associated with maintenance training input data, which are selected from the plurality of candidate maintenance categories.

16. A maintenance computing system, comprising: A processor and non-volatile memory storing executable instructions, which, in response to execution by the processor, cause the processor to: Run an inspection classifier, which is configured to determine the predicted inspection classification based on runtime event input data from structural health monitoring sensors equipped on the vehicle; Output the predicted inspection classification; Receive user input regarding the inspection and classification of the runtime event input data; The runtime event input data and the adopted inspection classification are used as feedback training data pairs to perform feedback training of the inspection classifier; The maintenance classifier is run to determine the predicted maintenance classification based on runtime inspection input data, which includes inspection-related input data and the inspection classification used. Output the predicted repair category; Receive user input regarding the maintenance category used for the runtime check input data; as well as The maintenance classifier is trained using the runtime inspection input data and the adopted maintenance classification as feedback training data pairs.

17. The maintenance computing system of claim 16, wherein the processor is further configured to: A monitoring classifier is run to determine the predicted monitoring classification based on runtime maintenance input data, which includes maintenance-related input data and the maintenance classification used. Output the predicted monitoring category; Receive user input regarding the monitoring classification adopted for the runtime maintenance input data; as well as The monitoring classifier is trained using the runtime maintenance input data and the adopted monitoring classification as feedback training data pairs.

Images