System and method for classifying maintenance issues for aircraft

Abstract

To provide an effective, efficient, and accurate system and method for allowing accurate retrieval of a record indicating an arbitrary issue.SOLUTION: In a system 100, an input device 108 selects one or more seed records 116 regarding one or more maintenance issues of one or more vehicles. Further, a user interface 104 outputs one or more first electronic signals that include the one or more seed records 116. A records database 110 includes maintenance records 112 for the one or more vehicles. A similarity engine control unit 102 communicates with the user interface 104 and the records database 110. The similarity engine control unit 102 receives the one or more first electronic signals including the one or more seed records 116, searches the maintenance records 112 within the records database 110, and finds one or more first return records 118 including a subset of the maintenance records 112 that are similar to the one or more seed records 116.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority to U.S. Provisional Patent Application No. 63 / 648,220, filed May 16, 2024, which is incorporated herein by reference in its entirety.

[0002]

[0002] Embodiments of the present disclosure generally relate to systems and methods for classifying aircraft maintenance issues. [Background technology]

[0003]

[0003] Aircraft are used to transport passengers and cargo between various locations. Numerous aircraft depart and arrive at a typical airport each day.

[0004]

[0004] Various aircraft within a fleet typically undergo periodic maintenance procedures. As can be appreciated, large commercial aircraft include numerous systems, components, devices, etc. that require periodic maintenance. Maintenance personnel often search the maintenance records of numerous aircraft to identify potential maintenance issues.

[0005] Known methods for searching maintenance records (particularly in military organizations) include relying on category codes embedded in the maintenance records. However, such methods may not be useful when the area of ​​interest is not neatly categorized or when the categories address multiple issues. Furthermore, at least some of the maintenance records may contain at least one significant coding error that substantially alters the meaning of the maintenance record.

[0006]

[0006] Because of these limitations, and knowing that the written narrative of a maintenance log ultimately represents the mechanic's intent, certain known methods have been developed that allow for keyword searching, for example. However, for many tasks, to be effective, keyword searching must be able to accommodate a large number of keywords, synonyms, acronyms, alternative spellings, and misspellings (e.g., terms such as handset, hand microphone, PA system, intercom, intercom, FA, PA, F / AP / A, etc.). Keyword searching also requires additional logic to filter out similar but unrelated material. For example, note the distinct difference in meaning between "FOUND LEAK IN CHECK VALVE" and "FOUND IN VALVE LEAK CHECK."

[0007]

[0007] Other known methods utilize natural language processing (NLP), such as named entity recognition (NER), to identify words / tokens used in a particular grammatical sense and attempt to identify the components, conditions, and activities contained therein. However, while these methods represent a slight improvement over keyword search, such methods still fall short of fully capturing hierarchy and context. As an example, such methods cannot distinguish whether a hinge issue categorically belongs to any particular door.

[0008] Another known method creates a multi-label classifier. Multi-label classifiers can be useful in classifying records by component type (e.g., grouping together all records related to doors, whether they are hinge or latch problems), and can classify records according to a finite set of subjective labels for activities and conditions, for example. However, multi-label classifiers require extensive human input to provide the labeling. Summary of the Invention

[0009] What is needed is an effective, efficient, and accurate system and method that allows for accurate searching of records that represent any problem. Additionally, what is needed is such a system and method for searching maintenance records for vehicles, such as aircraft.

[0010] With these needs in mind, certain embodiments of the present disclosure provide a system including a user interface having a display and an input device. The input device is configured to select one or more seed records for one or more maintenance issues of one or more vehicles. The user interface is further configured to output one or more first electronic signals including the one or more seed records. A record database includes maintenance records for the one or more vehicles. A similarity engine control unit is in communication with the user interface and the record database. The similarity engine control unit is configured to (a) receive the one or more first electronic signals including the one or more seed records, (b) search the record database for maintenance records to find one or more first return records including a subset of maintenance records similar to the one or more seed records, and (c) output one or more second electronic signals including the one or more first return records to the user interface. The user interface is configured to display the one or more first return records on a display. The user interface is further configured to select relevance of the one or more first return records to provide one or more first relevance selections. The user interface is further configured to output one or more third electronic signals including the one or more first relevance selections.

[0011] In at least one embodiment, the similarity engine control unit is further configured to receive one or more third electronic signals having the one or more first relevance selections, search the records database for service records to find one or more second return records similar to the one or more first relevance selections, and output one or more fourth electronic signals to a user interface including the one or more second return records. The display is configured to display the one or more second return records. The user interface is further configured to select a relevance of the one or more second return records to provide the one or more second relevance selections. The user interface is further configured to output one or more fifth electronic signals including the one or more second relevance selections.

[0012] In at least one embodiment, the similarity engine control unit is further configured to establish a problem definition set based on the one or more first relevance selections. In a further embodiment, the similarity engine control unit is further configured to automatically label one or more of the maintenance records in the record database based on the problem definition set.

[0013] In at least one embodiment, the one or more vehicles include one or more aircraft.

[0014] In at least one embodiment, the one or more relevance selections include one or more true positives and one or more false positives. The one or more relevance selections may also include one or more near-miss negatives.

[0015] The affinity engine control unit may be further configured to, at least in part, select one or more seed records.

[0016] In at least one embodiment, the affinity engine control unit is further configured to generate one or more semantic affinity charts and display the one or more semantic affinity charts on a display, hi at least one embodiment, the one or more semantic affinity charts indicate clusters of data.

[0017] In at least one embodiment, the similarity engine control unit is an artificial intelligence (AI) or machine learning system.

[0018]

[0018] The system may also include one or more robots (or other such autonomous systems, devices, or components) configured to automatically perform one or more maintenance procedures on one or more vehicles based at least in part on one or more first relevance selections.

[0019] Certain embodiments of the present disclosure provide a method that includes receiving, by a similarity engine control unit, one or more first electronic signals including one or more seed records, searching, by the similarity engine control unit, for maintenance records in a record database to find one or more first return records that include a subset of maintenance records similar to the one or more seed records, outputting, by the similarity engine control unit, one or more second electronic signals including the one or more first return records to a user interface, displaying the one or more first return records on a display, selecting, via the user interface, relevance of the one or more first return records to provide one or more first relevance selections, and outputting, by the user interface, one or more third electronic signals including the one or more first relevance selections. [Brief explanation of the drawings]

[0020] [Figure 1]

[0020] A block diagram of a system according to one embodiment of the present disclosure is shown. [Figure 2]

[0021] 1 illustrates a front view of a display according to one embodiment of the present disclosure. [Figure 3]

[0022] 1 illustrates a front view of a display according to one embodiment of the present disclosure. [Figure 4]

[0023] 1 illustrates a front view of a display according to one embodiment of the present disclosure. [Figure 5]

[0024] 1 shows a flowchart of a method according to one embodiment of the present disclosure. [Figure 6]

[0025] FIG. 2 illustrates a schematic block diagram of an affinity engine control unit according to one embodiment of the present disclosure. [Figure 7]

[0026] 1 illustrates a perspective front view of an aircraft according to one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0021]

[0027] The foregoing summary, as well as the following detailed description of specific embodiments, will be better understood when read in conjunction with the accompanying drawings. As used herein, the use of the singular form "a" or "an" preceding an element or step does not necessarily exclude a plurality of such elements or steps. Furthermore, references to "one embodiment" are not intended to be interpreted as excluding the existence of additional embodiments that incorporate features described herein. Furthermore, embodiments that "comprising" or "having" one or more elements having certain conditions may include additional elements that do not have those conditions (unless expressly stated otherwise).

[0022]

[0028] 1 illustrates a block diagram of a system 100 according to one embodiment of the present disclosure. The system 100 includes an affinity engine control unit 102 in communication with a user interface 104, such as via one or more wired or wireless connections. In at least one embodiment, the affinity engine control unit 102 is an artificial intelligence or machine learning system.

[0023]

[0029] In at least one embodiment, a set of predefined classes of records is developed in the similarity engine control unit 102. One or more control units may use machine learning or artificial intelligence to develop the set of predefined classes of records.

[0024]

[0030] The user interface 104 includes a display 106 and an input device 108. In at least one embodiment, the display 106 is an electronic device configured to electronically display images, videos, text, etc. The display 106 may be a monitor, screen, television, touch screen, etc. The input device 108 may include a keyboard, a mouse, a stylus, a touch screen interface (i.e., the input device 108 may be integrated with the display 106), etc. The display 106 is configured to display visual graphics, videos, text, etc. The user interface 104 may be or be part of a computer workstation. As another example, the user interface 104 may be a handheld device such as a smartphone, tablet, etc. In at least one embodiment, the similarity engine control unit 102 and the user interface 104 are part of a common computing system. As another example, the similarity engine control unit 102 may be located remotely from the user interface 104.

[0025]

[0031] The affinity control unit 102 also communicates with a records database 110, such as via one or more wired or wireless connections. The records database 110 stores problem records, such as maintenance records 112, for aircraft 114. Optionally, the records database 110 may store problem records for various other types of vehicles. As another example, the records database 110 may store problem records for various other systems, devices, etc., other than vehicles. In at least one example, the records database 110 stores thousands, millions, or more maintenance records 112 for hundreds, thousands, or more aircraft 114.

[0026]

[0032] In operation, an individual interacts with the user interface 104 to provide one or more seed records 116. For example, the individual uses the input device 108 to select one or more seed records 116. Each seed record 116 includes information related to a given reliability issue. For example, each seed record 116 may be a maintenance record for a maintenance issue on the aircraft 114. The seed records 116 are initially selected and define classes and / or clusters of records. In at least one embodiment, the seed records 116 are not keywords, but rather are records in the form of maintenance records 112.

[0027]

[0033] The similarity engine control unit 102 receives seed record(s) 116 from the user interface 104. That is, the similarity engine control unit 102 receives one or more electronic signals including information about the seed record(s) 116. In response to receiving the seed record(s) 116, the similarity engine control unit 102 searches the maintenance records 112 in the record database 110 to find any maintenance records that appear to match the seed record(s) 116. For example, the similarity engine control unit 102 scans a large sample set interspersed with labeled problem records. In response to identifying similar maintenance records 112 in the record database 110, the similarity engine control unit 102 outputs the similar maintenance records 112 to the user interface as return records 118 (i.e., one or more electronic signals including information about the return records 118). The return records 118 are displayed on the display 106 of the user interface 104. The user then reviews the return records 118 to determine whether they are relevant to the issue(s) provided in the seed record(s) 116. The user then operates the input device 108 to indicate which return records 118 are relevant and which are not, outputting one or more relevance selections 120 (i.e., one or more electronic signals containing information about the relevance selections 120). The one or more relevance selections 120 are selections regarding the relevance of the return records 118. The user reviews the return records 118 and identifies relevant return records 118 (“true positives”) and non-relevant return records 118 (“false positives”). The similarity engine control unit 102 receives the relevance selection(s) 120 from the user interface 104 and again searches the record database 110 to refine the similarity search for maintenance records 112. Based on the relevance selection 120, the similarity engine control unit 102 finds similar maintenance records 112 and outputs return records 118 to the user interface 104, and the process can repeat.

[0028]

[0034] This process may be repeated as many times as desired to refine the issue definition for the maintenance record 112. That is, this process may continue until an individual is satisfied with the return record(s) 118, which may then provide an issue definition set for the maintenance record 112. The issue definition set defines classes and / or clusters of one or more issues of interest (e.g., maintenance issues related to the aircraft 114) to provide an issue model. The similarity engine control unit 102 may supervise the process (e.g., through the use of a binary label classifier based on a threshold confidence score) or provide an unsupervised process, such as through a cluster classifier based on a threshold similarity score. In at least one embodiment, the similarity engine control unit 102 uses the issue model (defined by the issue definition set for the maintenance record 112) to automatically identify and automatically label (without human intervention) issues within the maintenance record 112.

[0029]

[0035] As one example, the maintenance issue is a no-start event. The no-start event relates to the replacement of an engine starter due to a failure to start the engine. Numerous records exist regarding the maintenance and inspection of engine starters, as well as similar records addressing auxiliary power unit starters. There are three prominent reasons for the unscheduled removal of an engine starter, all of which could be described as "failures" (metal chips / debris, oil leaks, and inability to start). However, the no-start event is of particular interest and does not directly refer to the engine starter. As another example, the maintenance issue is an insect-related issue. This is rare. However, such an occurrence could occur in the galley, but could occur anywhere on the aircraft. As another example, the maintenance issue is corrosion on an exterior surface. Corrosion can occur anywhere on the aircraft, but is most common on flight surfaces, exterior panels, and exterior structures. As another example, the maintenance issue is cracked brakes. The aforementioned issues are merely examples and are non-limiting.

[0030]

[0036] As described herein, the similarity engine control unit 102 operates to identify topically related documents (e.g., maintenance records 112). In at least one embodiment, the similarity engine control unit 102 defines arbitrary clusters or classes of maintenance records 112 based on a defined set of records that exemplify an issue of interest (e.g., a maintenance or reliability issue). The issue may be very specific (e.g., a crack in a main landing gear disc brake, an engine starter replacement due to a no-start event), or it may be equally broad in every respect (e.g., all replacements on the aircraft, all soiled upholstery in the interior cabin, all electrical shorts on the aircraft, all corrosion incidents on the aircraft, all insect infestations on the aircraft, etc.). In at least one embodiment, the similarity engine control unit 102 operates according to an iterative process of refinement, where a user successively marks relevant / unrelated instances from a list (e.g., return records 118) provided by the similarity engine control unit 102 via document similarity analysis (a form of unsupervised machine learning). In at least one embodiment, the similarity engine control unit 102 then presents the user with the prescribed set of models as binary classifiers to continue using.

[0031]

[0037] The system 100 and method described herein provides a middle ground between pre-existing categorical information (component codes, arbitrary condition / activity categories) and undefined, ad-hoc queries, while simultaneously allowing a user analyst to define any resulting issues for continuous monitoring, forecasting, predictive maintenance alert development, and inventory optimization. In at least one embodiment, the similarity engine control unit 102 identifies topically related documents within the maintenance records 112, with particular consideration given to the fundamental problem of identifying maintenance records related to common reliability issues. Downstream processes such as reliability analysis, repair effectiveness analysis, predictive maintenance, and maintenance / inventory optimization depend on accurately collecting relevant, representative records of issues, regardless of the issue's specificity, type, or component.

[0032]

[0038] In at least one embodiment, the similarity engine control unit 102 utilizes a large-scale language model that is pre-trained on a relevant aviation maintenance corpus (including maintenance records and maintenance reference manuals). The similarity engine control unit 102 can operate as an unsupervised similarity engine. Its performance depends on the choice of embedding (character, token, word, sentence, graph) and similarity metric (e.g., cosine similarity) and implementation (e.g., distance from centroid).

[0033]

[0039] As described herein, a user first seeks at least some relevant records to form a starting set. In particular, the user first identifies seed records 116. The seed records 116 are then input to the similarity engine control unit 102 via the user interface 104. In at least one embodiment, the similarity engine control unit 102 can assist in the selection of seed records 116, such as through artificial intelligence, returning records by category or through conversation (e.g., "Find all records that describe failed autostarts").

[0034]

[0040] After receiving the seed records 116, the similarity engine control unit 102 searches the maintenance records 112 to find an initial set of similar maintenance records 112. This initial set is provided on the display 106 of the user interface 104 as return records 118. The user reviews the list of return records 118 and indicates which are relevant (e.g., which records exemplify the issue and which records do not exemplify the issue). The similarity engine control unit 102 receives the list of relevant records (and optionally irrelevant records) as relevance selections 120. The similarity engine control unit 102 then operates in an unsupervised manner to further search the record database 110 based on the relevance selections 120, and the process is repeated until the user is satisfied that most (if not all) of the return records 118 adequately represent the issue of interest. As the similarity iterations progress, the near-miss false positives become increasingly specific and valuable. As one example, a user can provide input (along with modifiers) via the user interface 104 that a particular return record is close to the subject of interest, which can be provided in the relevance selection 120 for the similarity engine control unit 102 to further refine its search of the record database 110. The near-miss negatives allow the similarity engine control unit 102 to learn not only from the positive set but also from the curated negative set, defining classes with much greater accuracy than a randomly selected labeled training set would allow.

[0035]

[0041] In at least one embodiment, once satisfied, the user provides a final selection of relevance selections 120 that form the authoritative challenge set. The similarity engine control unit 102 then uses the authoritative challenge set to automatically (without human intervention) label each of the maintenance records 112 in the record database 110 that appear to belong to the authoritative challenge set. This is a supervised learning step, allowing the performance of the classifier to be explicitly measured. In this manner, any customized challenge class can be defined as desired.

[0036]

[0042] As described herein, the similarity engine control unit 102 maximizes or otherwise increases the efficiency of human expert attention and avoids the inherent limitations of predetermined, discrete, multi-label classification schemes. It is noteworthy that traditional supervised machine learning methods typically require defining multiple discrete classes, where all possibilities of interest (and, indeed, multiple possibilities of no interest) must be labeled. When document labeling requires subjective judgment and human technical knowledge, humans typically manually label documents. Subtle linguistic differences between classes require more human labeling. More classes require more human labeling. However, many individuals ultimately do not care about the majority of classes. Furthermore, it is impossible to predict all the semantic patterns or distinctions that an individual will ultimately care about.

[0037]

[0043] In at least one embodiment, the similarity engine control unit 102 provides or otherwise includes a set of defined classes. This set of defined classes can be used to train the similarity engine control unit 102 and / or a supervised model classifier. The similarity engine control unit 102 may include a supervised model classifier. In at least one embodiment, the similarity engine control unit 102 utilizes an unsupervised learning process responsible for searching documents in a database (e.g., hundreds, thousands, millions, or more of maintenance records 112 in a record database 110). For example, given a sample input of N documents, such as N seed records 116 (where N is any non-zero natural number), the similarity engine control unit 102 efficiently identifies similar documents in the record database 110 and outputs return records 118 (“positives”) for an individual to manually label, which the individual can binary confirm (“true positives”) or reject (“false positives”), thereby providing a set of relevance selections 120. The initial batch of N seed records 116 may itself be generated from simple methods (e.g., multiple keyword searches or human review of randomly selected documents), or may be generated from more sophisticated preliminary methods, such as through artificial intelligence assistance by the affinity engine control unit 102.

[0038]

[0044] In at least one embodiment, the similarity engine control unit 102 can be tuned or otherwise programmed to explore neighboring meanings and thus thoroughly establish the boundaries of the patterns underlying the intended challenge. A high true positive rate (“precision”) is desirable and is an indicator of the ability to recognize patterns of topical meaning embedded within the user-supplied seed records 116. However, too high a precision may indicate that the similarity engine control unit 102 is returning only a narrow subset of the universe relevant to the intended challenge, and the resulting definition set from the refinement process will be skewed, biased, or over-fit to that subset. Therefore, the similarity engine control unit 102 is configured to return positives that are reasonable extrapolations or neighbors to the patterns supplied in the initial rounds. Otherwise, the similarity engine control unit 102 may risk over-fitting to a narrow definition of the challenge (e.g., “rust on the ailerons” instead of “rust on all wing surfaces”). Counterintuitively, this means that the goal of the similarity engine control unit 102 is not to maximize accuracy, but to intentionally introduce test positives from the periphery so that the average accuracy is between 70% and 90%, thereby increasing the efficient use of human attention by testing the semantic pattern breadth and range and their topic boundaries while confirming the core patterns of the prescribed set.

[0039]

[0045] In at least one embodiment, the similarity engine control unit 102 includes or otherwise utilizes a discriminative large-scale language model (which may also be used in the subsequent supervised machine learning phase of classifier training), such as one based on an encoder-only transformer large-scale language model architecture. Multiple embeddings (character embeddings, word embeddings, and sentence embeddings) may be tested when tuning the similarity engine, or may be optionally selected by the user. Document similarity may be assessed from multiple similarity measures (e.g., vector cosine similarity). Similarity may also be measured with respect to patterns common to an existing set (e.g., a common element among multiple records is the mention of a particular term) by determining similarity for each record at a time through mechanisms such as vector averaging, or in terms of exploring the neighborhood. Optionally, similarity may be measured at the request of a user (e.g., an individual may note in the return record, "Look for more like this").

[0040]

[0046] In at least one other embodiment, the similarity engine control unit 102 can include a generative large-scale language model that can prompt for documents similar to the provided set via canned prompts or through conversational refinement with a user via the user interface 104. In such an embodiment, the similarity engine control unit 102 can delegate some responsibility for the effectiveness of the search to individuals to create queries that best represent the desired problem (e.g., "Find all maintenance records similar to the following, but ignore any records that occurred during routine maintenance").

[0041]

[0047] In at least one embodiment, the affinity engine control unit 102 stores the seed records 116, the return records 118, and / or the relevance selections 120 in vector format. The affinity engine control unit 102 may include a search-augmented generative (RAG) model, or the like. In this embodiment, a user may rely entirely on prompt engineering to generate an initial batch of seed records 116 (e.g., "Find N instances of records describing unscheduled maintenance to nose landing gear" or "Find all records that mention insects inside an aircraft"). In at least one embodiment, the affinity engine control unit 102 utilizes large-scale language models (which may be preprogrammed, tuned, or trained from scratch) associated with the maintenance records 112 and adjacent documents (e.g., aircraft maintenance manuals) to improve semantic inference and contextual performance.

[0042]

[0048] In at least one embodiment, maintenance may be performed on the aircraft 114 based on the maintenance records 112. For example, maintenance procedures may be performed on one or more of the aircraft 114 based on the maintenance records 112 as automatically labeled by the similarity engine control unit 102 through use of a challenge definition set. After the maintenance records 112 for the aircraft 114 have been labeled by the similarity engine control unit 102, an alert may be sent by the similarity engine control unit 102 to the user interface 104 that a maintenance procedure will be performed on the aircraft 114. The maintenance procedures may be performed automatically, such as by one or more robots 115.

[0043]

[0049] As described herein, the system 100 includes a user interface 104 including a display 106 and an input device 108. The input device 108 is configured to select one or more seed records 116 related to one or more maintenance issues for one or more aircraft 114. The user interface 104 is further configured to output one or more first electronic signals including the one or more seed records 116. A records database 110 includes (e.g., stores) maintenance records 112 for the one or more aircraft 114. The similarity engine control unit 102 is in communication with the user interface 104 and the records database 110. The similarity engine control unit 102 is configured to receive the one or more first electronic signals including the one or more seed records 116. The similarity engine control unit 102 is further configured to search the maintenance records 112 in the records database 110 to find one or more first return records 118 including a subset of the maintenance records 112 similar to the one or more seed records 116. The affinity engine control unit 102 is further configured to output one or more second electronic signals including the one or more first return records 118 to the user interface 104. The user interface 104 is configured to display the one or more first return records 118 on the display 106. The user interface 104 is further configured to select a relevance of the one or more first return records 118 to provide one or more first relevance selections 120 and to output one or more third electronic signals including the one or more first relevance selections 120.

[0044]

[0050] In at least one embodiment, the similarity engine control unit 102 is further configured to receive a third electronic signal(s) having the one or more first relevance selections 120. The similarity engine control unit 102 may then search the service records 112 in the records database 110 to find one or more second return records 118 that are similar to the one or more first relevance selections. The similarity engine control unit 102 may then output one or more fourth electronic signals including the one or more second return records 118 to the user interface 104. The user interface 104 is configured to display the one or more second return records on the display 106. The user interface 104 is further configured to select a relevance of the one or more second return records 118 to provide the one or more second relevance selections 120. The one or more second relevance selections 120 are output as one or more fifth electronic signals.

[0045]

[0051] In at least one embodiment, the similarity engine control unit 102 is further configured to establish a set of problem definitions based on the one or more relevance selections 120. In a further embodiment, the similarity engine control unit 102 is further configured to automatically label one or more of the maintenance records 112 in the records database 110 based on the set of problem definitions.

[0046]

[0052] 2 shows a front view of the display 106, according to one embodiment of the present disclosure. Referring to FIGS. 1 and 2, as noted, in at least one embodiment, the similarity engine control unit 102 is tuned or otherwise programmed to explore neighboring meanings. For example, the similarity engine control unit 102 is configured to return positives that are plausible extrapolations or neighbors to patterns provided in initial rounds. Accordingly, the similarity engine control unit 102 may be configured to output one or more semantic similarity charts to allow an individual to visualize the clustering of labeled records, thereby providing an information feedback mechanism and allowing an individual to estimate the value and effectiveness of the resulting classifier.

[0047]

[0053] As shown in FIG. 2, the affinity engine control unit 102 generates semantic affinity charts 122a-d. It should be understood that the affinity charts 122a-d shown in FIG. 2 are merely exemplary and non-limiting. The affinity charts 122a-d provide easily identifiable electronic graphic representations on the display 106. The affinity engine control unit 102 provides the semantic affinity charts 122a-d to narrow search results. By generating and providing the semantic affinity charts 122a-d, the affinity engine control unit 102 reduces complexity by organizing and sorting search results, thereby allowing individuals to easily distinguish valuable information from irrelevant information.

[0048]

[0054] For example, the affinity engine control unit 102 views documents (e.g., electronic documents stored in the database of record 110) as data points in space. As can be appreciated, some of the documents are of interest, while others are uninteresting or of low relevance. Relevant documents may be intermixed with less relevant documents in the database of record 110. Thus, the affinity engine control unit 102 operates to cluster the data, allowing less relevant documents to be easily identified from relevant documents. The affinity engine control unit 102 separates the documents (e.g., search results) into distinct clusters, thereby allowing an individual to easily determine the field of relevant documents.

[0049]

[0055] Conversely, there is also the problem of finding a needle in a haystack. For example, the difference between relevant and unrelated may be negligible. One such example is shown in FIG. 2. As generated and illustrated by the affinity engine control unit 102, four semantic similarity charts 122a-d (e.g., plots of search results shown in clusters) show a single sample of documents represented as data points in two-dimensional space. The four plots are used as an approximation of the distribution of points in a high-dimensional space (greater than two dimensions). This approximation compresses the spatial distribution of data points into a two-dimensional representation. The varying parameter, or perplexity, used to create each plot is analogous to the zoom of a camera lens. It focuses on large- and small-scale differences in location between points in the high-dimensional space. Clear, distinct groups do not appear to form within plots 122a-d. Instead, there appears to be noisy overlap between documents of interest (represented by black dots) and documents of no interest (represented by gray dots). Thus, through the semantic similarity charts 122a-d presented by the affinity engine control unit 102, an individual can discern that the resulting data may be high in entropy and not easy to classify, extract, and / or identify.

[0050]

[0056] FIG. 3 illustrates a front view of the display 106 according to one embodiment of the present disclosure. As shown in FIG. 3, the affinity engine control unit 102 generates semantic affinity charts 124a-d. It should be understood that the affinity charts 124a-d illustrated in FIG. 3 are merely exemplary and non-limiting. As shown in FIG. 3, the data points illustrated in the plots (i.e., 124a-d) represent a sample of documents in a high-dimensional space. This sample has also been compressed into a two-dimensional electronic representation. The charts 124a-d in FIG. 3 provide a clearer boundary between related (black dots) and unrelated (gray dots). Referring to FIGS. 1-3, the affinity engine control unit 102 generates semantic affinity charts and displays them electronically on the display 106. The semantic affinity charts allow individuals to quickly and easily discern the relevance of document search results through clustering. When clustering is not clearly defined (e.g., as shown in Figure 2), an individual may decide that a more focused search of documents may be necessary.

[0051]

[0057] 4 illustrates a front view of the display 106 according to one embodiment of the present disclosure. The similarity engine control unit 102 may generate a chart 126 and electronically display the chart 126 on the display 106. In this embodiment, the similarity engine control unit 102 compares semi-supervised guided learning of embodiments of the present disclosure with traditional random learning. As shown in FIG. 4, embodiments of the present disclosure provide a method that is substantially more accurate than traditional approaches.

[0052]

[0058] 5 shows a flowchart of a method according to one embodiment of the present disclosure. Referring to FIG. 1 and FIG. 5, at 200, an individual operates a user interface to select a seed record 116 associated with a maintenance issue of interest. The individual can manually select the seed record 116, or the affinity engine control unit 102 can assist the individual in the selection, such as via a prompt.

[0053]

[0059] At 202, the similarity engine control unit 102 receives the seed record 116. The seed record 116 is then used to search the maintenance records 112 in the record database 110 to find maintenance records 112 that are similar to the seed record 116. At 204, the similarity engine control unit 102 provides the similar maintenance records to the user interface 104 as return records 118 (i.e., a subset of the maintenance records 112 that were determined to be similar to the seed record 116).

[0054]

[0060] At 206, an individual reviews the return records 118, such as on the display 106 of the user interface 104, to identify the relevance of each of the return records, thereby providing relevance selections 120. The relevance selections 120 include true positives. A true positive indicates that a particular record is relevant. In at least one embodiment, the relevance selections 120 may also include false positives. A false positive indicates that a particular record is not relevant. The similarity engine control unit 102 uses the true positives and false positives to further refine the search for maintenance records 112. In at least one example, the relevance selections 120 may also include notes entered by the user via an input device. The notes may include further instructions for the search (e.g., "Find more like this," "Like this, but not including X").

[0055]

[0061] At 208, the user determines whether further refinement of the relevance selection 120 is required, i.e., the user determines whether further searching based on the relevance selection 120 is required. If so, the method returns to 202.

[0056]

[0062] However, if the user is satisfied with the set of relevance selections, the method proceeds from 208 to 210, where the set of relevance selections establishes a problem definition set. At 212, the similarity engine control unit 102 can then use the problem definition set to automatically label related maintenance records 112 in the records database 110.

[0057]

[0063] 6 illustrates a schematic block diagram of the affinity engine control unit 102, according to one embodiment of the present disclosure. In at least one embodiment, the affinity engine control unit 102 includes at least one processor 300 in communication with a memory 302. The memory 302 stores instructions 304, received data 306, and generated data 308. The affinity engine control unit 102 illustrated in FIG. 6 is merely exemplary and non-limiting.

[0058]

[0064] As used herein, terms such as "control unit," "central processing unit," "CPU," "computer," and the like may include any processor-based or microprocessor-based system, including systems using microcontrollers, reduced instruction set computers (RISC), application specific integrated circuits (ASIC), logic circuits, and any other circuits or processors, including hardware, software, or a combination thereof, capable of performing the functions described herein. The above examples are illustrative only and thus are not intended to limit in any way the definition and / or meaning of the above terms. For example, the affinity engine control unit 102 may be or include one or more processors configured to control operations as described herein.

[0059]

[0065] The affinity engine control unit 102 is configured to execute sets of instructions stored in one or more data storage units or elements (such as one or more memories) to process data. For example, the affinity engine control unit 102 may include one or more memories or be coupled to one or more memories. The data storage units may also store data or other information as desired or needed. The data storage units may take the form of an information source or a physical memory element within a processing machine.

[0060]

[0066] The set of instructions may include various commands that instruct the affinity engine control unit 102 as a processing machine to perform particular operations (e.g., the methods and processes of various embodiments of the subject matter described herein). The set of instructions may take the form of a software program. The software may take various forms such as system software or application software. Furthermore, the software may take the form of a collection of separate programs, a program subset within a larger program, or a portion of a program. The software may also include modular programming in the form of object-oriented programming. The processing of input data by the processing machine may be in response to user commands, in response to results of previous processing, or in response to a request made by another processing machine.

[0061]

[0067] The diagrams of the examples herein may depict one or more control or processing units, such as the affinity engine control unit 102. It should be understood that this processing or control unit may represent a circuit, circuitry, or portion thereof, that may be implemented as hardware having associated instructions (e.g., software stored on a tangible, non-transitory computer-readable storage medium such as a computer hard drive, ROM, RAM, etc.) that perform the operations described herein. The hardware may include state machine circuitry hardwired to perform the functions described herein. Optionally, the hardware may include electronic circuitry including and / or connected to one or more logic-based devices, such as a microprocessor, processor, controller, etc. Optionally, the affinity engine control unit 102 may represent processing circuitry, such as one or more of a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), microprocessor(s), etc. The circuitry in various examples may be configured to execute one or more algorithms to perform the functions described herein. Such one or more algorithms, whether or not explicitly identified in a flowchart or method, may include aspects of the embodiments disclosed herein.

[0062]

[0068] As used herein, the terms "software" and "firmware" are used interchangeably and include any computer program stored in a data storage unit (e.g., one or more memories) for execution by a computer, including RAM memory, ROM memory, EPROM memory, EEPROM memory, and non-volatile RAM (NVRAM) memory. The types of data storage units listed above are merely exemplary and thus not limiting as to the types of memory that may be used for storing computer programs.

[0063]

[0069] 1-6 , embodiments of the present disclosure provide methods and systems that enable a computing device to quickly and efficiently analyze large amounts of data. For example, a similarity engine control unit 102 may receive and analyze hundreds, thousands, millions, or more maintenance records 112 from hundreds, thousands, or more aircraft 114. Thus, a large amount of data that a human cannot easily distinguish is tracked and analyzed. As described herein, the vast amount of data is efficiently organized and / or analyzed by the similarity engine control unit 102. The similarity engine control unit 102 analyzes the data in a relatively short amount of time to quickly and efficiently identify maintenance records 112 that have similar issues to, for example, a seed record 116. Humans are unable to analyze such large amounts of data in an effective and efficient manner. Therefore, embodiments of the present disclosure provide improved and efficient functionality and performance that vastly outperforms humans in analyzing vast amounts of data.

[0064]

[0070] In at least one embodiment, components of system 100, such as affinity engine control unit 102, provide and / or enable a computer system to operate as a dedicated computer system for classifying issues, such as maintenance issues. The affinity engine control unit 102 improves upon standard computing devices by identifying and automatically communicating such information to individuals (such as aircraft operators) in an efficient and effective manner.

[0065]

[0071] In at least one embodiment, the similarity engine control unit 102 uses a machine learning algorithm that automatically analyzes maintenance records 112 to find records similar to the seed record 116 and automatically labels the particular maintenance record 112 based on a set of challenge specifications. In at least one embodiment, all or a portion of the systems and methods described herein are or otherwise include artificial intelligence (AI) or machine learning systems that can automatically perform the operations of the methods also described herein. In at least one embodiment, the similarity engine control unit 102 can be or otherwise include a deterministic or rule-based rating system. In at least one embodiment, the similarity engine control unit 102 can be an artificial intelligence or machine learning system. These types of systems may be trained from outside information and / or self-trained, and can iteratively improve the accuracy of how data is analyzed, such as to identify similar issues and automatically label maintenance records. Over time, these systems improve by identifying and communicating with increasing accuracy and speed, thereby significantly reducing the likelihood of any potential errors. For example, an AI or machine learning system may learn and identify models, associate such models with received data, and identify potential conflicts. The AI ​​or machine learning systems described herein may include techniques enabled by adaptive predictive capabilities. The techniques exhibit at least some degree of autonomous learning to automate and / or enhance pattern detection (e.g., recognizing irregularities or regularities in data), customization (e.g., generating or modifying rules to optimize record matching), and the like. The systems may be trained and retrained using feedback from one or more prior analyses of data, ensemble data, and / or other such data. Based on this feedback, the systems may be trained by adjusting one or more parameters, weights, rules, criteria, etc. used in the same analysis.This process may be performed using data or ensemble data instead of training data and may be repeated multiple times to iteratively improve the identification and communication described herein. Training minimizes conflicts and interference by running an iterative training algorithm, in which the system is retrained with an updated set of data based on feedback examined prior to the most recent training of the system. This provides a robust analytical model that can better determine validation challenges, identify when to select records as similar, automatically label records, and so on.

[0066]

[0072] In one embodiment, the similarity engine control unit 102 includes or represents an artificial neural network (ANN) that identifies patterns within a visual, vector, or vector space representation of data, classifies the patterns based on their content (e.g., assigns Class #1, Class #2, etc.), and identifies records (e.g., documents) based on the classification. Using a specially trained ANN to identify records in this manner can provide improvements over conventional methods of document identification, including more accurate identification of related information, identifying related documents on a much larger scale than humans can identify documents, and identifying documents much faster and / or at a much more rapid frequency than humans are capable of. ANNs can be implemented using software, hardware, or a combination of software and hardware. The structure of an ANN can be a series of layers, each layer including one or more artificial neurons arranged as one or more neuron arrays. Each of these neurons can include or represent a register, a microprocessor, and at least one input. Each neuron can generate an output or activation based on an activation function that uses the output of the previous layer and a set of weights as input. Each neuron in the neuron array may be connected to another neuron in the same layer or another layer via one or more synaptic circuits. The synaptic circuits may include memory for storing synaptic weights. One example of this ANN may be a deep neural network having an input layer, an output layer, and multiple fully connected hidden layers. In some embodiments, the ANN (e.g., the similarity engine control unit 102) may be implemented by an application-specific integrated circuit (ASIC) specially customized for the particular artificial intelligence application described herein, which may provide superior computing power and reduced power consumption compared to conventional computers.

[0067]

[0073] The training data may be generated by receiving continuous data at the similarity engine control unit 102 and discretizing the continuous data using the similarity engine control unit 102. Optionally, the similarity engine control unit 102 may be trained with a pre-trained model. The training data or the pre-trained model may be received remotely by the similarity engine control unit 102 via one or more networks. The training data may be historical data. The neural network may use this historical data to learn patterns in the visual representation of the data to identify or detect the same or similar patterns in other data. The trained ANN monitors additional visual representations to identify and classify the patterns. If the trained ANN detects one or more patterns, the trained ANN may classify the pattern(s) to generate classification data that may be output to a user and / or used to retrain the ANN.

[0068]

[0074] The ANN of the similarity engine control unit 102 may continue to learn to improve the identification of patterns in the data visualization as well as the classification of identified patterns. This continuous learning may be performed, for example, by modifying the output generated by one or more of the neurons in response to receiving the same input (e.g., the neuron generates a different output after the modification), modifying the activation function of one or more neurons, modifying one or more of the weights, and / or modifying one or more of the connections between neurons (or modifying which neurons are connected to each other). By modifying one or more of these factors, the ANN may generate a different output (e.g., a different pattern is identified and / or a different classification is selected) than before the modification.

[0069]

[0075] FIG. 7 illustrates a perspective front view of an aircraft 114 according to one embodiment of the present disclosure. The aircraft 114 includes a propulsion system 412, including, for example, engines 414. Optionally, the propulsion system 412 may include more engines 414 than shown. The engines 414 are supported by wings 416 of the aircraft 114. In other embodiments, the engines 414 may be supported by a fuselage 418 and / or a tail section 420. The tail section 420 may also support a horizontal stabilizer 422 and a vertical stabilizer 424. The fuselage 418 of the aircraft 114 defines an interior cabin 430, which may include a cockpit or flight deck, one or more work sections (e.g., a galley, a crew baggage area, etc.), one or more passenger sections (e.g., first class, business class, and economy class), one or more restrooms, etc. FIG. 4 illustrates an example of an aircraft 114. It should be understood that the aircraft 114 may be sized, shaped, and configured differently than that shown in FIG.

[0070]

[0076] Optionally, embodiments of the present disclosure may be used with various other types of vehicles, such as automobiles, trains, ships, spacecraft, etc. Also, optionally, embodiments of the present disclosure may be used with various other devices, systems, components, etc. other than vehicles. For example, embodiments of the present disclosure may be used with consumer electronics products.

[0071]

[0077] Furthermore, the present disclosure includes embodiments according to the following clauses.

[0072]

[0078] Article 1. 1. A system comprising: a user interface including a display and an input device, the input device configured to select one or more seed records related to one or more maintenance issues for one or more vehicles, the user interface further configured to output one or more first electronic signals including the one or more seed records; a records database containing maintenance records for said one or more vehicles; and an affinity engine control unit in communication with the user interface and the records database, the affinity engine control unit comprising: receiving the one or more first electronic signals containing the one or more seed records; searching the maintenance records in the record database to find one or more first return records that include a subset of the maintenance records that are similar to the one or more seed records; and and outputting one or more second electronic signals including the one or more first return records to the user interface, wherein the user interface is configured to display the one or more first return records on the display, the user interface is further configured to select relevance of the one or more first return records to provide one or more first relevance selections, and the user interface is further configured to output one or more third electronic signals including the one or more first relevance selections.

[0073]

[0079] Article 2. The similarity engine control unit receiving the one or more third electronic signals having the one or more first relevance selections; searching the maintenance records in the records database to find one or more second return records similar to the one or more first relevance selections; and The system described in clause 1, further configured to perform the steps of: outputting one or more fourth electronic signals including the one or more second return records to the user interface, wherein the user interface is configured to display the one or more second return records; the user interface is further configured to select relevance of the one or more second return records to provide one or more second relevance selections; and the user interface is further configured to output one or more fifth electronic signals including the one or more second relevance selections.

[0074]

[0080] Article 3. 3. The system of claim 1 or 2, wherein the affinity engine control unit is further configured to establish a challenge definition set based on the one or more first relevance selections.

[0075]

[0081] Article 4. 4. The system of clause 3, wherein the similarity engine control unit is further configured to automatically label one or more of the maintenance records in the record database based on the set of challenge specifications.

[0076]

[0082] Article 5. 5. The system of any one of clauses 1 to 4, wherein the one or more vehicles include one or more aircraft.

[0077]

[0083] Article 6. 6. The system of any one of clauses 1 to 5, wherein the one or more first relevance selections include one or more true positives and one or more false positives.

[0078]

[0084] Article 7. 7. The system of clause 6, wherein the one or more first relevance selections further include one or more near miss negatives.

[0079]

[0085] Article 8. 8. The system of any one of clauses 1 to 7, wherein the similarity engine control unit is further configured, at least in part, to select the one or more seed records.

[0080]

[0086] Article 9. The similarity engine control unit generating one or more semantic similarity charts; and 9. The system of any one of clauses 1 to 8, further configured to: display the one or more semantic similarity charts on the display.

[0081]

[0087] Article 10. 10. The system of clause 9, wherein the one or more semantic similarity charts indicate clusters of data.

[0082]

[0088] Article 11. 11. The system of any one of clauses 1 to 10, wherein the similarity engine control unit is an artificial intelligence (AI) or machine learning system.

[0083]

[0089] Article 12. The system of any one of clauses 1 to 11, further comprising one or more robots configured to automatically perform one or more maintenance procedures on the one or more vehicles based at least in part on the one or more first relevance selections.

[0084]

[0090] Article 13. 1. A method for a system, the system comprising: a user interface including a display and an input device, the input device configured to select one or more seed records related to one or more maintenance issues for one or more vehicles, the user interface further configured to output one or more first electronic signals including the one or more seed records; a records database containing maintenance records for said one or more vehicles; and a similarity engine control unit in communication with the user interface and the records database; The method comprises: receiving, by the similarity engine control unit, the one or more first electronic signals including the one or more seed records; searching, by the similarity engine control unit, for the maintenance records in the record database to find one or more first return records that include a subset of the maintenance records that are similar to the one or more seed records; outputting, by the similarity engine control unit, one or more second electronic signals including the one or more first return records to the user interface; displaying the one or more first return records on the display; selecting, via the user interface, relevance of the one or more first return records to provide one or more first relevance selections; and outputting, by the user interface, one or more third electronic signals including the one or more first relevance selections.

[0085]

[0091] Article 14. receiving, by the affinity engine control unit, the one or more third electronic signals having the one or more first relevance selections; searching, by the similarity engine control unit, the maintenance records in the records database to find one or more second return records similar to the one or more first relevance selections; outputting, by the similarity engine control unit, one or more fourth electronic signals including the one or more second return records to the user interface; displaying the one or more second return records on the display; selecting, via the user interface, relevance of the one or more second return records to provide one or more second relevance selections; and 14. The method of clause 13, further comprising outputting, by the user interface, one or more fifth electronic signals including the one or more second relevance selections.

[0086]

[0092] Article 15. 15. The method of clause 13 or 14, further comprising establishing, by the affinity engine control unit, a challenge definition set based on the one or more first relevance selections.

[0087]

[0093] Article 16. 16. The method of clause 15, further comprising automatically labeling, by the similarity engine control unit, one or more of the maintenance records in the record database based on the set of challenge specifications.

[0088]

[0094] Article 17. 17. The method of any one of clauses 13 to 16, wherein the one or more vehicles include one or more aircraft.

[0089]

[0095] Article 18. 18. The method of any one of clauses 13 to 17, wherein the one or more relevance selections comprise one or more true positives and one or more false positives.

[0090]

[0096] Article 19. 19. The system of clause 18, wherein the one or more first relevance selections further include one or more near miss negatives.

[0091]

[0097] Article 20. 20. The method of any one of clauses 13 to 19, further comprising selecting, at least in part, the one or more seed records by the affinity engine control unit.

[0092]

[0098] Article 21.

[0099] generating one or more semantic similarity charts; and

[0100] 21. The method of any one of clauses 13 to 20, further comprising displaying the one or more semantic similarity charts on a display, the one or more semantic similarity charts indicating clusters of data.

[0093]

[0101] Article 22. 22. The method of any one of clauses 13 to 21, wherein the similarity engine control unit is an artificial intelligence (AI) or machine learning system.

[0094]

[0102] Article 23. 23. The method of any one of clauses 13 to 22, further comprising automatically performing one or more maintenance procedures on the one or more vehicles based at least in part on the one or more first relevance selections.

[0095]

[0103] As described herein, embodiments of the present disclosure provide effective, efficient, and accurate systems and methods that enable accurate searching of records that represent any issue. Further, embodiments of the present disclosure provide such systems and methods for searching maintenance records for vehicles, such as aircraft.

[0096]

[0104] For purposes of describing the embodiments of the present disclosure, various spatial and directional terms may be used, such as top, bottom, lower, center, sideways, horizontal, vertical, front, etc., but it should be understood that such terms are used solely with reference to the orientations shown in the drawings. These orientations may be flipped, rotated, or otherwise changed so that top becomes bottom, bottom becomes top, horizontal becomes vertical, etc.

[0097]

[0105] As used herein, a structure, limitation, or element that is "configured to" perform an task or operation is structurally shaped, configured, or adapted specifically to correspond to the task or operation. For clarity and to avoid doubt, an object that can merely be modified to perform a task or operation is not "configured / set up to" perform a task or operation as used herein.

[0098]

[0106] It should be understood that the above description is intended to be illustrative, not limiting. For example, the above-described examples (and / or aspects thereof) can be used in combination with each other. In addition, many modifications can be made to adapt a particular situation or material to the teachings of the various embodiments of the present disclosure without departing from the scope of the present disclosure. While the dimensions and types of materials described herein are intended to define aspects of the various embodiments of the present disclosure, the examples are by no means limiting, but are illustrative examples. Many other examples will be apparent to those skilled in the art upon reviewing the above description. The scope of the various embodiments of the present disclosure should be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled. In the accompanying claims and the detailed description herein, the words "including" and "in which" are used as the plain English equivalents of the words "comprising" and "wherein," respectively. Furthermore, terms such as "first," "second," and "third" are used merely as labels and are not intended to impose numerical requirements on their objects. Moreover, the limitations of the following claims are not written in means-plus-function form, and are not intended to be construed under 35 U.S.C. §112(f) unless such claim limitations expressly use the phrase "means for," followed by a statement of function lacking further structure.

[0099]

[0107] The description herein uses examples to disclose various embodiments of the present disclosure, including the best mode, and to enable any person skilled in the art to practice various embodiments of the present disclosure, including making and using any device or system and practicing any methods incorporated therein. The patentable scope of various examples of the present disclosure is defined by the claims, and may include other examples that occur to those skilled in the art. Such other examples are intended to be within the scope of the claims if the examples have structural elements that do not differ from the literal language of the claims, or if the examples include equivalent structural elements that differ only insignificantly from the literal language of the claims.

Claims

1. A system (100), comprising: a user interface (104) including a display (106) and an input device (108), the input device (108) configured to select one or more seed records (116) related to one or more maintenance issues for one or more vehicles, the user interface (104) further configured to output one or more first electronic signals including the one or more seed records (116); a records database (110) containing maintenance records (112) for said one or more vehicles; and a similarity engine control unit (102) in communication with the user interface (104) and the record database (110), the similarity engine control unit (102) comprising: receiving the one or more first electronic signals containing the one or more seed records (116); searching the maintenance records (112) in the records database (110) to find one or more first return records (118) that include a subset of the maintenance records (112) that are similar to the one or more seed records (116); and and outputting one or more second electronic signals including the one or more first return records (118) to the user interface (104), the user interface (104) configured to display the one or more first return records (118) on the display (106), the user interface (104) further configured to select relevance of the one or more first return records (118) to provide one or more first relevance selections, and the user interface (104) further configured to output one or more third electronic signals including the one or more first relevance selections.

2. The similarity engine control unit (102) receiving the one or more third electronic signals having the one or more first relevance selections; searching the maintenance records (112) in the records database (110) to find one or more second return records (118) similar to the one or more first relevance selections; and 2. The system (100) of claim 1, further configured to: output one or more fourth electronic signals including the one or more second return records (118) to the user interface (104), the user interface (104) configured to display the one or more second return records (118), the user interface (104) further configured to select relevance of the one or more second return records (118) to provide one or more second relevance selections, and the user interface (104) further configured to output one or more fifth electronic signals including the one or more second relevance selections.

3. The system (100) of claim 1, wherein the affinity engine control unit (102) is further configured to establish a challenge definition set based on the one or more first relevance selections.

4. 4. The system (100) of claim 3, wherein the similarity engine control unit (102) is further configured to automatically label one or more of the maintenance records (112) in the record database (110) based on the set of challenge definitions.

5. The system (100) of claim 1, wherein the one or more vehicles include one or more aircraft (114).

6. The system of claim 1 , wherein the one or more first relevance selections include one or more true positives and one or more false positives.

7. The system (100) of claim 6, wherein the one or more first relevance selections further comprise one or more near miss negatives.

8. The system (100) of claim 1, wherein the affinity engine control unit (102) is further configured to, at least in part, select the one or more seed records (116).

9. The similarity engine control unit (102) generating one or more semantic similarity charts (124a-d); and The system (100) of claim 1, further configured to: display the one or more semantic similarity charts (124a-d) on the display (106).

10. The system (100) of claim 9, wherein the one or more semantic similarity charts (124a-d) represent clusters of data.

11. The system (100) of claim 1 , wherein the similarity engine control unit (102) is an artificial intelligence (AI) or machine learning system (100).

12. 10. The system (100) of claim 1, further comprising one or more robots configured to automatically perform one or more maintenance procedures on the one or more vehicles based at least in part on the one or more first relevance selections.

13. A method for a system (100), comprising: a user interface (104) including a display (106) and an input device (108), the input device (108) configured to select one or more seed records (116) related to one or more maintenance issues for one or more vehicles, the user interface (104) further configured to output one or more first electronic signals including the one or more seed records (116); a records database (110) containing maintenance records (112) for said one or more vehicles; and an affinity engine control unit (102) in communication with the user interface (104) and the record database (110); The method comprises: receiving, by the similarity engine control unit (102), the one or more first electronic signals including the one or more seed records (116); searching, by the similarity engine control unit (102), the maintenance records (112) in the record database (110) to find one or more first return records (118) that include a subset of the maintenance records (112) that are similar to the one or more seed records (116); outputting, by the similarity engine control unit (102), one or more second electronic signals including the one or more first return records (118) to the user interface (104); displaying the one or more first return records (118) on the display (106); selecting, via the user interface (104), the relevance of the one or more first return records (118) to provide one or more first relevance selections; and outputting, by the user interface (104), one or more third electronic signals including the one or more first relevance selections.

14. receiving, by the affinity engine control unit (102), the one or more third electronic signals having the one or more first relevance selections; searching, by the similarity engine control unit (102), the maintenance records (112) in the records database (110) to find one or more second return records (118) that are similar to the one or more first relevance selections; outputting, by the similarity engine control unit (102), one or more fourth electronic signals including the one or more second return records (118) to the user interface (104); displaying the one or more second return records (118) on the display (106); selecting, via the user interface (104), the relevance of the one or more second return records (118) to provide one or more second relevance selections; and The method of claim 13 , further comprising outputting, by the user interface (104), one or more fifth electronic signals including the one or more second relevance selections.

15. 14. The method of claim 13, further comprising establishing, by the affinity engine control unit (102), a challenge definition set based on the one or more first relevance selections.

16. 16. The method of claim 15, further comprising automatically labeling, by the similarity engine control unit (102), one or more of the maintenance records (112) in the record database (110) based on the challenge definition set.

17. The method of claim 13 , wherein the one or more vehicles include one or more aircraft (114).

18. The method of claim 13 , wherein the one or more first relevance selections include one or more true positives and one or more false positives.

19. The method of claim 18 , wherein the one or more first relevance selections further include one or more near-miss negatives.

20. The method of claim 13, further comprising selecting the one or more seed records (116) at least in part by the affinity engine control unit (102).

21. generating one or more semantic similarity charts (124a-d); and 14. The method of claim 13, further comprising displaying the one or more semantic similarity charts (124a-d) on the display (106), the one or more semantic similarity charts (124a-d) representing clusters of data.

22. The method of claim 13 , wherein the similarity engine control unit (102) is an artificial intelligence (AI) or machine learning system (100).

23. The method of claim 13 , further comprising automatically performing one or more maintenance procedures on the one or more vehicles based at least in part on the one or more first relevance selections.

Images