Method for extracting technical information from technical manual

Abstract

The invention discloses a method for extracting information from a technical manual. The method comprises the following steps: extracting selected keywords from a technical manual; appending the selected keyword to the table to provide master data; receiving a portion of the master data and a first tag table corresponding to the portion of the master data; applying a machine learning model or a natural language processing (NLP) model to generate a second tag table corresponding to the entire master data; processing the tags in the second tag table to correspond to the master data; and outputting the updated second tag table to match the master data to extract information from the technical manual so that the user can review or use, or review and use, the information contained in the technical manual without reading or searching throughout the technical manual.

Description

Technical Field

[0001] The present disclosure relates generally to technical information in technical manuals, and more particularly to methods of extracting information from technical manuals. Background Art

[0002] The technical manual for an aircraft engine or other aircraft system may be of a relatively large size. Therefore, manually extracting master inspection characteristic (MIC) information and / or other information (such as notes and precautions) from the aircraft engine manual consumes a significant amount of time and effort. In addition, the process of extracting the MIC information and / or other information may be prone to human error. Given the complexity of aircraft engines, MIC information is very valuable to engineers working on aircraft engines. However, errors in extracting MIC information may slow down or pause the work being performed. In addition, it may not be practical to find and extract the desired technical information from the relatively large content size of the technical manual as it may take time. BRIEF DESCRIPTION OF THE DRAWINGS

[0003] The foregoing and other features and advantages will be apparent from the following more particular description of various exemplary embodiments, as illustrated in the drawings, in which like reference numerals generally indicate identical, functionally similar, and / or structurally similar elements.

[0004] Figure 1 is a diagram of a system for extracting technical information from a technical manual according to an embodiment of the present disclosure.

[0005] Figure 2A is sample text found in a technical manual according to an embodiment of the present disclosure.

[0006] Figure 2B is a sample of master inspection characteristic (MIC) information in master data extracted from a technical manual according to an embodiment of the present disclosure.

[0007] Figure 3A An example of input of MIC information from an engine manual according to an embodiment of the present disclosure is shown.

[0008] Figure 3B An example of predicted output from a machine learning (ML) model according to an embodiment of the present disclosure is shown.

[0009] Figure 4 is a diagram of a system for extracting technical information from a technical manual according to another embodiment of the present disclosure.

[0010] Figure 5 is a schematic diagram of a computer system for implementing a method for extracting information from a technical manual according to an embodiment of the present invention.

[0011] Figure 6 is a flowchart of a method for extracting technical information from a technical manual according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0012] Additional features, advantages, and embodiments of the present disclosure are set forth or apparent by consideration of the following detailed description, drawings, and claims. Furthermore, the foregoing overview and the following detailed description of the present disclosure are exemplary and are intended to provide further explanation without limiting the scope of the present disclosure as claimed.

[0013] Various embodiments of the present disclosure are discussed in detail below. Although specific embodiments are discussed, this is for illustrative purposes only. Those skilled in the relevant art will recognize that other components and configurations may be used without departing from the spirit and scope of the present disclosure.

[0014] Embodiments of the present disclosure attempt to provide a method for extracting master inspection characteristic (MIC) information from a technical manual (e.g., an engine manual) in an efficient manner. The method uses a scalable end-to-end architecture with a machine learning (ML) backbone to fully automate the process of extracting information from a technical manual in a very short time compared to manual work. The method also compensates for the accuracy of the extraction, computing power, or time consumed by the process. The method significantly reduces the time and labor spent on accurately extracting information from a technical manual, and organizes the extracted MIC information in an engineer-friendly manner.

[0015] Thus, primary inspection characteristic (MIC) information and / or other information, such as notes and precautions, can be extracted from relatively large text-based manuals. In some embodiments, the method can be based on an ML architecture with a linear support vector machine (SVM) classifier coupled with a Chi Square correlation. The method can be customized to extract information from a technical manual and organize the information in a desired manner that a technical user can quickly understand. The method can be used to extract not only MIC information, but also any other information.

[0016] Figure 1 is a diagram of a system for extracting technical information from a technical manual according to an embodiment of the present disclosure. Figure 6 is a flow chart of a method for extracting technical information from a technical manual according to an embodiment of the present disclosure. Figure 1 and Figure 6, the method includes extracting selected keywords from the technical manual 100 using, for example, a Regex-based extraction program 102 by a computer system at 602. Regex is an abbreviation for "regular or rational expression". Regex is a character sequence that specifies a search pattern in a text. Typically, such a pattern is used by a string search algorithm to perform a "find" or "find and replace" operation on a string, or for input validation. In an embodiment, before extracting keywords by performing the Regex-based extraction program 102, at 103, the technical manual 100 is converted from a first file type (such as Standard General Markup (SGM), Portable Document Format (PDF), Text (TXT), Document (DOC / DOCX), Rich Text Format (RTF), Hypertext Markup Language (HTML or HTM), EXCEL (XLSX), LaTeX (LATEX), etc.) to a second file type (such as Extensible Markup Language (XML)).

[0017] The Regex-based extraction program 102 is used to extract certain relevant keywords from the technical manual 100. The method includes appending the extracted relevant keywords to a table (e.g., a spreadsheet file, such as an EXCEL (XLSX) table file 104) by a computer system at 604 to form master data 105. The master data 105 in the XLSX table file 104 is an organized document with all the details about the master inspection characteristic (MIC) information 106 from the technical manual 100. The MIC information 106 can be used by engineers to quickly identify the MIC information 106 of their needs. The master data 105 containing the MIC information 106 is used as a data set for the entire machine learning (ML) model.

[0018] The method includes, at 606, if the MIC information 106 is completely new, performing manual labeling 108 of a portion 105A (e.g., about 10%) of the master data 105 to generate a first label table 109 corresponding to the portion 105A of the master data 105. At 606, the manual labeling 108 to generate the first label table 109 is performed to train the ML model to predict and generate the MIC information 106 based on the master data 105. The method includes inputting the portion 105A of the master data 105 and the first label table 109 corresponding to the portion 105A of the master data 105 to a computer system implementing the method. The method includes receiving the portion 105A of the master data 105 and the first label table 109 corresponding to the portion 105A by the computer system at 608.

[0019] At 610, the method can continue to perform text preprocessing 110 of the master data 105 by a computer system to provide preprocessed data 111. In an embodiment, the text preprocessing 110 includes using a Pandas data frame 112 to provide PANDAS preprocessed data 113. PANDAS (derived from the terms "panel data" and "python data analysis") is a software library for data manipulation and data analysis written for the Python programming language. PANDAS provides data structures and operations for manipulating numerical tables and time series. PANDAS allows various data manipulation operations, such as merging, shaping, selection, and data cleaning. Examples of cleaning data include removing empty characters, hyperlinks, punctuation marks, etc. For example, a specific library from Pandas can be used for data cleaning purposes. Text preprocessing 110 can be performed to clean text data. The text preprocessing 110 also includes removing empty elements and stop words from the Pandas preprocessed data 113 at 114 to provide cleaned data 115. Text preprocessing includes classifying the cleaned data 115 at 116 to obtain preprocessed data 111. Although PANDAS is used in a python environment, other libraries similar to PANDAS can also be used when implemented in other programming language environments (e.g., C, C++, R, or Java languages).

[0020] The method also includes generating an ML model 118 at 612 and applying the ML model 118 to the preprocessed data 111 to generate a second label table 120 corresponding to the entirety of the main data 105. In an embodiment, the preprocessed data 111 is converted into a term frequency-inverse document frequency (TF-IDF) 122 feature using TFidVectorizer to obtain TF-IDF data 123. Any machine learning process can be used for numerical data, but not for text data. Therefore, a process for converting text to numerical data is used. In this case, TF-IDF can be used because it uses term frequency and inverse document frequency, thereby negating the existence of widely used words. TF-IDF is an algorithm that uses word frequency to determine how relevant a word is to a given document. TF-IDF measures the originality of a word by comparing the number of times a word appears in a document (i.e., frequency) to the number of documents in which the word appears. TF-IDF converts text into a meaningful numerical representation to accommodate machine algorithms for prediction. For example, you can use CountVectorizer and TFidfTransformer from sklearn library to convert text into TF-IDF sparse matrix.

[0021] The method includes applying a chi-squared test 124 at 614 to determine correlations in the TF-IDF data 123. The chi-squared test is a statistical hypothesis test used for contingency table analysis when the sample size is relatively large. The chi-squared test is mainly used to check whether two categorical variables (two dimensions of a contingency table) are independent, for example, in terms of affecting a test statistic (a value in the contingency table). There are many types of chi-squared tests, including Pearson's test, Fisher's test, binomial test, McNemar's test, etc. The chi-squared test is performed to remove correlations in the input parameters. This test is very useful if the input data also has categorical variables.

[0022] The method includes inputting the result 125 of the chi-square test 124 to a linear support vector machine (SVM) classifier 126 at 616 to obtain a second label table 120. The SVM classifier 126 is a supervised machine learning model for analyzing data for classification and regression analysis and outlier detection. In addition to performing linear classification, the SVM classifier can also perform nonlinear classification as needed. However, in an example embodiment, the SVM classifier 126 is a linear SVM classifier. Instead of or in addition to the SVM classifier, other models, such as tree-based models (e.g., random forest methods, decision trees, XGBoost models, etc.) and / or Logistic regression can also be used. In this case, a support vector machine (SVM) method is used for classification, wherein data is separated by creating a hyperplane to appropriately classify data into different classes.

[0023] The method also includes post-processing the tags in the second tag table 120 at 130, 618 to match the MIC information 106 in the master data 105. The method includes outputting the updated second tag table at 620 to match the master data, thereby extracting information from the technical manual so that the user can review or use, or review and use the information contained in the technical manual by reading the tags in the updated second tag table without reading or searching in the entire technical manual. In an embodiment, the tags in the updated second tag table are organized according to the maintenance checklist or the recommendation for parameter adjustment. The data size of the technical manual is at least five times larger than the data size of the updated second tag table. In an embodiment, post-processing 130 includes training with Distil BERT to match the MIC information 106 with the tags in the second tag table 120. Distil BERT is a transformer model trained by extracting the Bidirectional Encoder Representations from Transformers (BERT) library. The BERT library is an encoder stack of the transformer architecture. The BERT library has 12 layers in the encoder stack. BERT is pre-trained in language modeling and next sentence prediction. BERT learns latent representations of words and sentences in context. After pre-training, BERT can be fine-tuned on smaller datasets with fewer resources to optimize performance for specific tasks such as language reasoning and text classification, and sequence-to-sequence based language generation tasks. As described in the above paragraphs, first, the SVM model is adjusted on the training data and labels are generated, which reduces the time to annotate labels. Using the SVM model, labels are generated, and this becomes the input data for the Distil BERT model. The Distil BERT model captures domain keywords. The Distil Bert model can be fine-tuned with domain keywords.

[0024] Figure 2A is a sample text 200 found in a technical manual 100 according to an embodiment of the present disclosure. For example, Figure 2A As shown, the technical manual 100 includes a checklist 202, such as measuring and recording diameters, and measuring balance in different planes and at two different engine speeds per minute (RPM). The checklist 202 (including, for example, diameters and measuring balance in different planes and at two different RPM) may contain information that the user may want to highlight and extract from the technical manual 100. Figure 6 As described at 602 and 604 in the technical manual 100, the above method is applied to the technical manual 100 to extract the information in the checklist 202 and organize the information into a table (e.g., Figure 1 In the XLSX file 105 shown in FIG. Figure 2Bshown.

[0025] Figure 2B is the master data 105 ( Figure 1 ) in the MIC information 106. Figure 2B In the illustrated embodiment, the MIC information 106 is organized in an XLSX file. A sample 204 of the MIC information 106 includes, for example, lines that specify "record diameter D" and "measure diameter D if its value is between 10 inches and 12 inches." The sample 204 of the MIC information 106 also includes, for example, lines regarding "record balance values ​​of the compressor rotor at different planes and different RPMs." For example, the sample 204 of the MIC 106 regarding measuring diameter D and recording balance values ​​of the compressor rotor at different planes and different RPMs can be used as part of a training data set (e.g., part of the portion 105A of the MIC information 106). The use of the sample 204 of the MIC 106 (in this example, with the recording of the measurement and balance values ​​of the diameter) can be used, for example, as a training data set. The training data set is obtained by manually labeling 108 ( Figure 1 ) and created in Figure 6 606 in the resulting label table 109 to generate a first label table corresponding to that portion of the master data (in this example, the sample 204 contains records of diameter measurements and balance values). This data can be used to train the ML model 118 to create a correspondence between the words "measure diameter" and "record at various planes and different RPMs" in the technical manual 100 and the specific checklist items "record diameter" and "record plane A balance at A1 RPM" in the resulting label table 109, such as Figure 1 shown.

[0026] Figure 3A 1 shows an example of input from MIC information 106 from an engine manual according to an embodiment of the present disclosure. In this embodiment, MIC information 106 from master data 105 in a technical manual 100 is input to an ML model 118. For example, Figure 3A As shown, “dimension A of the type A seal measured at six equally spaced locations” is input to the ML model 118. The ML model 118 is then calculated based on the portion 105A of the MIC information 106 as a training data set to predict a second label table 120 corresponding to the entirety of the master data 105. Thus, Figure 6 As shown at 608 to 612 in FIG. 1 , the computer system receives the portion 105A of the master data 105 and the first tag table 109 corresponding to the portion 105A of the master data to then predict the second tag table 120, such as Figure 1 shown.

[0027] Figure 3BAn example of a predicted output 300 from an ML model 118 is shown in accordance with an embodiment of the present disclosure. In this example, Figure 6 As shown at 612 in the MIC information 106, the ML model 118 is calculated based on a portion of the sample 204 of the MIC information 106 regarding the measured diameter D and the balance values ​​of the compressor rotor recorded at different planes and different RPMs to predict the organization of the statement "measure the dimension A of the A-type seal at six equally spaced positions" into the predicted MIC 302 including "recorded dimension A1, recorded dimension A2, ..., recorded dimension A6". By specifically listing each measured value of the six dimensions at the six positions, a technician can easily notice whether the measurement data is missing due to negligence, etc., and thus human errors can be significantly reduced or prevented. The predicted output 300 of the ML model 118 "recorded dimension A1, recorded dimension A2, ..., recorded dimension A6" can be compared with the extracted MIC 304 "recorded dimension L1, recorded dimension K1, ..., recorded id dimension J4, ..., recorded average dimension K1, recorded average dimension J1". In this example, "record size A1, record size A2, ..., record size A6" can be compared with the extracted MIC 304 "record size L1, record size K1, ..., record id size J4, ..., record average size K1, record average size J1" corresponding to the tags "L1, L2, ..." in the second tag table 120. As described above, in this example, the manually extracted MIC 304 corresponding to the first tag table 109 and the manually extracted MIC 304 corresponding to the first tag table 109 are compared with the manually extracted MIC 304 corresponding to the second tag table 120. Figure 1 There is good agreement between the predicted MICs 302 of the second label table 120 shown. In general, the present ML model 118 has an initial accuracy of at least 93%. This accuracy can be improved by training with more data (e.g., more than 10% from the same technical manual) and / or using additional data from different engine manuals.

[0028] Additionally, the method is computationally light due to a relatively simple and efficient architecture. The method is modular in that it can be implemented keeping in mind new MICs / features that can be added in the future. Furthermore, new modules can be easily integrated into the current architecture without hampering the functionality of any older modules.

[0029] The present method significantly reduces human labor because it takes only a fraction of the time (less than a minute) to extract, clean, and label the technical manual data compared to manual extraction. However, only approximately a small fraction (approximately 10%) of the data is initially labeled manually to train the ML model. This small fraction of data is only required if there is new MIC information to be extracted. If there is no new MIC information to be extracted, manual training may not be required. This feature not only helps reduce labor but also saves valuable time.

[0030] The method can also be scaled up to accurately extract data from any technical manual (e.g., an engine manual). The method can also be scaled up to extract any information (except the MIC) from a technical manual. The method also provides an output that can be stored as an organized file that a user can easily read. The method enables a user to quickly review or use, or review and use, the information contained in the technical manual without having to read or search through the entire technical manual. The extracted data can be organized according to a maintenance checklist or parameter adjustment recommendations, and the data size of the technical manual is at least five times larger than the data size of the extracted data.

[0031] In addition, although the technical manual 100 (e.g., an engine manual) may have a large data size (between 10MB and 100MB), the resulting extracted second tag table 120 has a smaller data size (between 1MB and 2MB). Therefore, the amount of data to be analyzed by the engineer is greatly reduced. In an embodiment, the data size of the technical manual 100 is at least five times larger than the data size of the second tag table. Therefore, a user (e.g., an engineer) can quickly find a specific tag (e.g., a specific parameter in the technical manual). The result of the method is a comprehensive tabular check list (updated second tag table) that encapsulates all possible information without losing quality information, such as Figure 3B In addition to saving time, this table view eliminates the possibility of human error when creating checklists.

[0032] The method was tested on an aircraft engine manual (e.g., SGM file) with more than 64,000 lines of text. More than 1,400 MICs were extracted from the 64,000 lines (e.g., organized as an XLSX file). Initially, 150 MICs (approximately 10% of the data) were manually labeled to train the ML model. The method was then tested on completely unseen data from the more than 1,400 MICs. Various ML models (including random forests, logistic regression, multinomial naive Bayes, and linear support vector machines) can be used to extract labels on all MIC information. An initial accuracy greater than or equal to 93.7% (about 94%) can be achieved. This accuracy can be further improved by initially manually labeling more data.

[0033] Figure 44 is a flow chart of a method for extracting technical information from a technical manual according to another embodiment of the present disclosure. The method includes parsing the text of a technical manual 400 (e.g., an engine manual) at 402, and extracting standard terms from the technical manual 400 using, for example, a Regex-based extraction procedure at 404. In an embodiment, the technical manual 400 is converted from a first file type (such as Standard Generalized Markup (SGM) or Portable Document Format (PDF)) to a second file type (such as Extensible Markup Language (XML)) before the Regex-based extraction procedure is performed.

[0034] The method includes creating a domain-based dictionary 406. In an embodiment, creating the domain-based dictionary 406 includes selecting keywords depending on the field of interest of the technical manual 400. For example, the technical manual 400 may be an engine manual, and the field of interest may be the electrical construction of the engine or the mechanical construction of the engine. For example, if the field of interest is the electrical construction of the engine, the dictionary focuses on terms such as voltage, current, electrical connection, electricity, etc. For example, if the field of interest is the mechanical construction of the engine, the dictionary focuses on terms such as burner, air flow, rotation, diameter, etc. The dictionary is a domain term built for adding to the vocabulary of the deep learning model. The manual may have both mechanical terms and electrical terms, however, if the dictionary is created for the mechanical domain, only MIC data can be captured for the mechanical domain.

[0035] The method includes, after creating a domain-based dictionary 406, extracting MIC sentences or MIC data 408 from a technical manual 400 based on the domain-based dictionary 406. The method also includes sampling the MIC data 408 and manually labeling a portion 410 of the MIC data 408. Labeled annotation may be the most "painful" task in machine learning model building. In this process, only a portion of the data is used for labeling, and the model has been trained on the partially labeled data and used to make inferences on the remaining data, so the time to annotate the remaining data is quite short.

[0036] The method includes performing text preprocessing and normalization of the MIC data 408 to provide preprocessed data 412. In an embodiment, the text preprocessing includes using a Pandas data frame to provide Pandas preprocessed data, as described above with respect to Figure 1 described in the paragraph.

[0037] The method includes applying an artificial intelligence (AI) model, such as a natural language processing (NLP) model 414, to the preprocessed data 412 to provide a label table for predicting the MIC data 408. NLP models, machine learning (ML) models, and / or deep learning (DL) models may be used as AI models. ML models include, for example, any of the linear support vector machine (SVM) classifier models, as described above with respect to Figure 1 The method includes adjusting the NLP model 414 based on the label table of the predicted MIC data 408 416. The machine learning model is used to eliminate the annotation process. Initially, only a portion of the training data is labeled, which is a time-consuming process, and the ML model is used to generate labels for the remaining data.

[0038] In the above paragraphs, although the method is described herein for a technical manual, the method can be applied to any type of manual, and more broadly to any type of text or manuscript that requires extraction of specific keywords. For example, although in the above paragraphs, the method is applied to an engine manual (such as an aircraft engine manual) for illustration, the method can also be applied to any type of technical manual, such as a medical device manual, a mechanical device manual, an electronic device manual, etc. In addition, the method can also be applied to an engine or system of an automobile (e.g., a car, a van, a sports utility vehicle, a truck), or a manual of an engine or system of a boat / ship, a vessel, or a train, etc. Creating a table based on the engine manual eliminates the possibility of human error in recording and checking. This also allows the automatic creation of a task summary from the engine manual.

[0039] Currently, the MIC creation process is a manual process and requires a subject matter expert (SME) to read the manual and create a task list. The present system and method creates a task list from the engine manual in a faster manner without human error of missing task items. The process of acquiring text using NLP and creating task items from the text acquired using NLP that is specific to the desired task is novel.

[0040] Figure 5 is a schematic diagram of a computer system 500 for implementing a method for extracting information from a technical manual according to an embodiment of the present invention. Figure 5, the computer system 500 includes a general purpose computing device including a central processing unit (CPU or processor) 520 and a system bus 510, which connects various system components including memory 530 (such as read-only memory (ROM) 540 and random access memory (RAM) 550) to the processor 520. The computer system 500 may include a cache of high-speed memory directly connected to the processor 520, adjacent to the processor 520, or integrated as a part of the processor 520. The computer system 500 copies data from the memory 530 and / or storage device 560 to the cache for rapid access by the processor 520. In this way, the cache provides a performance improvement that avoids delays in the processor 520 when waiting for data. These modules and other modules can control or be configured to control the processor 520 to perform various actions. Other memories 530 may also be available for use. The memory 530 may include a variety of different types of memory with different performance characteristics. The present disclosure may operate on a computer system 500 having more than one processor 920, or on a group or cluster of computing devices networked together to provide greater processing power. The processor 520 may include any general purpose processor and hardware modules or software modules (such as module 1 562, module 2 564, and module 3 566 stored in the storage device 560 and configured to control the processor 520), as well as a dedicated processor in which software instructions are incorporated into the actual processor design. The processor 520 may essentially be a completely independent computer system, including multiple cores or processors, buses, memory controllers, caches, etc. Multi-core processors may be symmetric or asymmetric.

[0041] The system bus 510 can be any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, and a local bus using any of a variety of bus architectures. For example, during startup, a basic input / output (BIOS) stored in a memory ROM 540 or the like can provide basic routines that facilitate the transfer of information between elements within the computer system 500. The computer system 500 also includes a storage device 560, such as a hard disk drive, a disk drive, an optical drive, a tape drive, etc. The storage device 560 may include software modules 562, 564, and 566 for controlling the processor 520. Other hardware or software modules may be envisioned. The storage device 560 is connected to the system bus 510 via a drive interface. The drive and associated computer-readable storage media provide non-volatile storage of computer-readable instructions, data structures, program modules, and other data for the computer system 500. In one aspect, a hardware module that performs a specific function includes a software component stored in a tangible computer-readable storage medium, which is connected to necessary hardware components (such as the processor 520, the system bus 510, the output device 570 (e.g., a display), etc.) to perform the function. In another aspect, the system may use a processor and a computer-readable storage medium to store instructions that, when executed by a processor (e.g., one or more processors), cause the processor to perform a method or other specific actions. Basic components and appropriate variations are contemplated depending on the type of device (such as whether the computer system 500 is a small handheld computing device, a desktop computer, or a computer server).

[0042] Although the exemplary embodiments described herein employ a storage device 560 such as a hard disk, other types of computer-readable media that can store data accessible by a computer such as magnetic cassettes, flash memory cards, digital versatile disks, magnetic cassettes, random access memory (RAM) 550, and read-only memory (ROM) 540 may also be used in the exemplary operating environment. Tangible computer-readable storage media, computer-readable storage devices, or computer-readable memory devices specifically exclude media such as transient waves, energy, carrier signals, electromagnetic waves, and signals themselves.

[0043] To enable user interaction with computer system 500, input device 590 represents any number of input mechanisms, such as a microphone for voice, a touch-sensitive screen for gesture or graphic input, a keyboard, a mouse, motion input, voice, etc. Output device 570 may also be one or more of many output mechanisms known to those skilled in the art. In some cases, a multimodal system enables a user to provide multiple types of input to communicate with computer system 500. Communication interface 580 typically manages and controls user input and system output. There is no limitation to operating on any particular hardware arrangement, and therefore, the basic features herein can easily replace the improved hardware or firmware arrangement developed.

[0044] The technology discussed herein refers to computer-based systems, actions taken by computer-based systems, information sent to computer-based systems, and information sent from computer-based systems. Those of ordinary skill in the art will recognize that the inherent flexibility of computer-based systems allows for various possible configurations, combinations, and divisions of tasks and functions between and within components. For example, the processes discussed herein can be implemented using a single computing device or multiple computing devices working in combination. Databases, memories, instructions, and applications can be implemented on a single system, or distributed on multiple systems. Distributed components can operate sequentially or in parallel.

[0045] Further aspects of the disclosure are provided by the subject matter of the following clauses.

[0046] A method for extracting information from a technical manual, the method being implemented on a computer system including one or more processors. The method comprises: (a) extracting selected keywords from the technical manual by a computer system; (b) appending the selected keywords to a table by the computer system to provide master data; (c) receiving a portion of the master data and a first tag table corresponding to the portion of the master data by the computer system; (d) applying a machine learning model or a natural language processing (NLP) model to preprocessed data by the computer system to generate a second tag table corresponding to the entire master data; and (e) processing tags in the second tag table by the computer system to correspond to the master data, so that a user can review or use, or review and use the information contained in the technical manual without having to read or search in the entire technical manual. The data size of the technical manual is at least five times larger than the data size of the second tag table.

[0047] The method according to the preceding clause, wherein extracting keywords from the technical manual comprises using an extraction procedure based on regular expressions or rational expressions.

[0048] A method as in any preceding clause, wherein the master data comprises master inspection characteristic (MIC) information from the technical manual.

[0049] A method as in any preceding clause, wherein appending the selected keywords to the table to provide the master data comprises appending the selected keywords to a spreadsheet file.

[0050] The method of any preceding clause, further comprising converting the technical manual from a first file type to a second file type prior to extracting the selected keywords.

[0051] A method as in any preceding clause, wherein the first file type is Standard Generalized Markup (SGM) or Portable Document Format (PDF), and the second file type is Extensible Markup Language (XML).

[0052] A method according to any preceding clause, wherein the portion of the master data is manually tagged to generate the first tag table prior to receiving the portion of the master data and the first tag table corresponding to the portion of the master data by the computer system.

[0053] The method according to any of the preceding clauses further includes performing text preprocessing of the master data by the computer system to format the master data to provide preprocessed data, wherein the text preprocessing of the master data by the computer system includes performing the text preprocessing of the master data using PANDAS to provide PANDAS preprocessed data.

[0054] A method as in any preceding clause, wherein said text pre-processing of said master data by said computer system comprises removing empty elements and stop words in said PANDAS pre-processed data to provide cleaned data.

[0055] A method as in any preceding clause, wherein said textual pre-processing of said master data by said computer system comprises classifying said cleansed data to provide said pre-processed data.

[0056] A method according to any preceding clause, wherein applying the machine learning model to the preprocessed data by the computer system to generate the second label table corresponding to the entire master data includes converting the preprocessed data into term frequency-inverse document frequency (TF-IDF) features to obtain TF-IDF data.

[0057] A method as described in any preceding clause, wherein applying, by the computer system, the machine learning model to the preprocessed data to generate the second label table corresponding to the entire master data includes applying a chi-square test to determine correlations in the TF-IDF data.

[0058] The method according to any of the preceding clauses further includes inputting the results of the chi-square test into the machine learning model in the computer system to obtain the second label table.

[0059] A method as described in any preceding clause, wherein the machine learning model comprises a linear support vector machine (SVM) classifier.

[0060] A method as in any preceding clause, wherein the master data comprises master inspection characteristic (MIC) information.

[0061] The method of any preceding clause, further comprising, before appending, by the computer system, the selected keywords to a table to provide the master data, creating a domain-based dictionary and appending, by the computer system, the selected keywords to the table to provide the master data based on the domain-based dictionary.

[0062] A method as in any preceding clause, wherein the data size of the technical manual is between 10MB and 100MB, and the data size of the second tag table is between 1MB and 2MB.

[0063] A method as in any preceding clause, wherein the technical manual is a manual for an aircraft engine, a manual for an aircraft system, or both.

[0064] The method of any preceding clause further comprises self-adjusting, by the computer system, the machine learning model based on the second label table corresponding to the master data.

[0065] A non-transitory computer-readable medium storing computer executable code, the computer executable code, when executed by a computer system, causes the computer system to perform a method of extracting information from a technical manual, the method comprising: (a) extracting selected keywords from the technical manual by the computer system; (b) appending the selected keywords to a table to provide master data by the computer system; (c) receiving a portion of the master data and a first tag table corresponding to the portion of the master data by the computer system; (d) applying a machine learning model or a natural language processing (NLP) model to the pre-processed data by the computer system to generate a second tag table corresponding to the entire master data; and (e) processing tags in the second tag table to correspond to the master data by the computer system, so that a user can review or use, or review and use information contained in the technical manual without reading or searching in the entire technical manual. The data size of the technical manual is at least five times larger than the data size of the second tag table.

[0066] The non-transitory computer-readable medium according to the preceding clause further comprises performing text preprocessing of the master data to format the master data to provide preprocessed data, wherein the text preprocessing of the master data performed by the computer system comprises performing the text preprocessing of the master data using PANDAS to provide PANDAS preprocessed data.

[0067] Further aspects of the disclosure are provided by the subject matter of the following clauses.

[0068] A method for extracting information from a technical manual, the method being implemented on a computer system including one or more processors, the method comprising: extracting selected keywords from the technical manual by the computer system; appending the selected keywords to a table by the computer system to provide master data; receiving a portion of the master data and a first tag table corresponding to the portion of the master data by the computer system; applying a machine learning model or a natural language processing (NLP) model by the computer system to generate a second tag table corresponding to the entire master data; processing tags in the second tag table by the computer system to correspond to the master data; and outputting an updated second tag table by the computer system to match the master data, thereby extracting information from the technical manual so as to enable a user to review or use, or review and use, the information contained in the technical manual by reading the tags in the updated second tag table without having to read or search in the entire technical manual. The tags in the updated second tag table are organized according to a maintenance checklist or parameter adjustment suggestion. The data size of the technical manual is at least five times larger than the data size of the updated second tag table.

[0069] The method according to the preceding clause, extracting keywords from the technical manual comprises using an extraction procedure based on regular expressions or rational expressions.

[0070] A method as in any preceding clause, the master data comprising master inspection characteristic (MIC) information from the technical manual.

[0071] A method as in any preceding clause, appending the selected keywords to the table to provide the master data comprises appending the selected keywords to a spreadsheet file.

[0072] A method as in any preceding clause, manually tagging the portion of the master data to generate the first tag table prior to receiving the portion of the master data and the first tag table corresponding to the portion of the master data by the computer system.

[0073] The method according to any of the preceding clauses further includes inputting the results of the chi-square test into the machine learning model in the computer system to obtain the second label table.

[0074] A method according to any preceding clause, wherein the machine learning model comprises a linear support vector machine (SVM) classifier.

[0075] A method as in any preceding clause, the master data comprising master inspection characteristic (MIC) information.

[0076] The method of any preceding clause, further comprising, before appending, by the computer system, the selected keywords to a table to provide the master data, creating a domain-based dictionary and appending, by the computer system, the selected keywords to the table to provide the master data based on the domain-based dictionary.

[0077] A method as in any preceding clause, the data size of the technical manual being between 10MB and 100MB, and the data size of the second tag table being between 1MB and 2MB.

[0078] A method as in any preceding clause, the technical manual being a manual for an aircraft engine, a manual for an aircraft system, or both.

[0079] The method of any preceding clause further comprises self-adjusting, by the computer system, the machine learning model based on the second label table corresponding to the master data.

[0080] The method of any preceding clause, further comprising converting the technical manual from a first file type to a second file type prior to extracting the selected keywords.

[0081] A method as in any preceding clause, the first file type being Standard Generalized Markup (SGM) or Portable Document Format (PDF), and the second file type being Extensible Markup Language (XML).

[0082] The method according to any of the preceding clauses further includes performing text preprocessing of the master data by the computer system to format the master data to provide preprocessed data, and the text preprocessing of the master data by the computer system includes using PANDAS to perform text preprocessing of the master data to provide PANDAS preprocessed data.

[0083] A method as in any preceding clause, wherein said text pre-processing of said master data by said computer system comprises removing empty elements and stop words in said PANDAS pre-processed data to provide cleaned data.

[0084] A method as claimed in any preceding clause, wherein said textual pre-processing of said master data by said computer system comprises classifying said cleansed data to provide said pre-processed data.

[0085] According to the method of any preceding clause, applying the machine learning model to the preprocessed data by the computer system to generate the second label table corresponding to the entire master data includes converting the preprocessed data into term frequency-inverse document frequency (TF-IDF) features to obtain TF-IDF data.

[0086] The method of any preceding clause, applying, by the computer system, the machine learning model to the preprocessed data to generate the second label table corresponding to the entire master data comprises applying a chi-square test to determine correlations in the TF-IDF data.

[0087] A non-transitory computer-readable medium storing computer executable code, the computer executable code, when executed by a computer system, causes the computer system to perform a method of extracting information from a technical manual, the method comprising: extracting selected keywords from the technical manual by the computer system; appending the selected keywords to a table by the computer system to provide master data; receiving a portion of the master data and a first tag table corresponding to the portion of the master data by the computer system; performing text preprocessing of the master data by the computer system to format the master data to provide preprocessed data; applying a machine learning model or a natural language processing (NLP) model to the preprocessed data by the computer system to generate a second tag table corresponding to the entire master data; processing tags in the second tag table by the computer system to correspond to the master data; and outputting an updated second tag table by the computer system to match the master data, thereby extracting information from the technical manual so that a user can review or use, or review and use, the information contained in the technical manual by reading the tags in the updated second tag table without having to read or search in the entire technical manual. The tags in the updated second tag table are organized according to a maintenance checklist or parameter adjustment suggestion. The data size of the technical manual is at least five times larger than the data size of the updated second tag table.

[0088] Although the above description is directed to the preferred embodiments of the present disclosure, other changes and modifications are obvious to those skilled in the art and can be made without departing from the spirit or scope of the present disclosure. In addition, the features described in conjunction with one embodiment of the present disclosure can be used in conjunction with other embodiments, even if not explicitly stated above.

Claims

1. A method for extracting information from a technical manual, the method being implemented on a computer system comprising one or more processors, characterized in that: The method comprises: extracting selected keywords from the technical manual by a computer system; appending, by the computer system, the selected keywords to a table to provide master data; receiving, by the computer system, a portion of the master data and a first tag table corresponding to the portion of the master data; Applying a machine learning model or a natural language processing (NLP) model by the computer system to generate a second label table corresponding to the entire master data; Processing, by the computer system, tags in the second tag table to correspond to the master data; and outputting, by the computer system, an updated second tag table to match the master data, thereby extracting information from the technical manual so as to enable a user to review or use, or review and use, the information contained in the technical manual by reading the tags in the updated second tag table without having to read or search through the entire technical manual, The tags in the updated second tag table are organized according to a maintenance checklist or parameter adjustment suggestions, and the data size of the technical manual is at least five times larger than the data size of the updated second tag table.

2. The method according to claim 1, characterized in that in, Extracting keywords from the technical manual includes using an extraction procedure based on regular expressions or rational expressions.

3. The method according to claim 1, characterized in that: in, The master data includes master inspection characteristic (MIC) information from the technical manual.

4. The method according to claim 1, characterized in that in, Appending the selected keywords to the table to provide the master data includes appending the selected keywords to a spreadsheet file.

5. The method according to claim 1, characterized in that in, Prior to receiving the portion of the master data and the first tag table corresponding to the portion of the master data by the computer system, the portion of the master data is manually tagged to generate the first tag table.

6. The method according to claim 1, characterized in that Further comprising inputting the result of the chi-square test into the machine learning model in the computer system to obtain the second label table.

7. The method according to claim 1, characterized in that in, The machine learning model includes a linear support vector machine (SVM) classifier.

8. The method according to claim 1, characterized in that in, The master data includes master inspection characteristic (MIC) information.

9. The method according to claim 1, characterized in that: Further comprising creating a domain-based dictionary and appending the selected keywords to the table through the computer system to provide the master data based on the domain-based dictionary before appending the selected keywords to the table through the computer system to provide the master data.

10. The method according to claim 1, characterized in that in, The data size of the technical manual is between 10MB and 100MB, and the data size of the second tag table is between 1MB and 2MB.

Images