NLP TOOLS USED IN CIVIL AVIATION: A SURVEY

: In this paper we have a look at the various application tools of NLP applied in the field of Civil Aviation. Civil aviation flights around the world are in huge demand. Order books of major Aircraft manufacturers such as Boeing, Airbus, Embraer etc are overflowing and delivery schedules for the next few years are full. In WINGS INDIA 2018 held at Hyderabad, Airbus reported that for the next 10 years India will be receiving one Airbus aircraft every week[1][2][3]. More aircrafts leads to more pressure on maintenance teams, ATC (Air Traffic Controller), ground marshals / crew and CNS (Communication Navigation Surveillance) handlers. MRO (Maintenance Repair and Overhaul) units by Boeing, Airbus, ATR apart from simulator Training units are coming up fast in countries like India, China, Srilanka and other countries, which calls for more automation of the day to day technical and maintenance works . NLP (Natural Language Processing) and AI (Artificial Intelligence) tools have been playing just the required role in a remarkable manner in managing incident reports, CNS assistance etc. The supporting role of NLP in maintenance, real time situation awareness and other areas are going to be a necessity in the foreseeable future.


INTRODUCTION
AIRBUS has forecasted that in between 2017 and 2036 the Asia-Pacific region will account for 41 percent of passenger aircrafts above 100 seats and freighters over 10 tons that is 14,280 aircrafts, with total demand of 34,900 aircrafts around the world [3] . In addition to it are already in-service crafts and those which are joining the fleets around the world from the long list of older order queues.
Maintenance of such a large number of aircrafts to keep them flying will be a huge challenge and automated tools using NLP can help a lot in such cases. NLP which is supposed to fulfill the dream of translation of languages from one to another through machine translation can also help in Intelligent document search engines for product life cycle management, tapping business policies and procedures, including huge quantity of documents, are now being used heavily by both mechanics and engineer.
Transfer of maintenance manuals from one language to another (using Machine Translation) will help cut across language barriers and engineers around the world speaking different languages can easily understand and adapt to the details of technical maintenance manuals easily.
Incident reports of pilots, controllers, Air Crew and Passengers can be collected and processed (e.g. ASRS Aviation Safety Report System of USA). Such systems can be used for detail analysis thereby avoiding accidents and repetition of such incidents.
NLP tools are widely used in Aviation (maintenance, incident reporting, crew support etc). Research and surveys are continuously being undertaken in order to improve existing systems like moving away from traditional search methodology results, such as binary answer (Is the specific key word present? Yes or No), and which then leads to false positives ( the key word is present but in the wrong context) and false negatives (the right context but without the key word), to read and analyze detail flight reports. [4] [5]

NLP TOOLS USED IN AVIATION
The following are some NLP tools that are being developed or used in the field of Civil Aviation.
The source maintenance manuals were in English. The dictionary consisted of core vocabulary and had the following data: 4054 unique entities and the English-French Bilingual dictionary consisted of 3280 unique entities. [6] The Machine Translation that was used was fully automatic. Input of text was done in a photocomposition format which was ready to be processed. Human intervention in preediting was not necessary. whenever unidentified words were encountered, these were entered after lookup in the dictionary. [6] Unfortunately the Project was terminated before reaching a stage where a proper and detail assessment could be made.

IBM WATSON
It is a NLP tool that uses NLP-MT and Cognitive computing in aviation. IBM has worked in many aspects of airline information processing. IBM with the wonderful questionanswering capacity of Watson and using Machine Translation and Natural Language Processing, helps pilots deal with real time maintenance challenges. IBM has taken help of customer service agents, flight attendants, pilots and technical staff, to provide real time service to the aviation maintenance and support .
Questions put up to Watson in human spoken language are completely intelligible to it. It can recommend suggestions in real time based on earlier occurrences. Neural Machine translation based on RNN helps WATSON in understanding scenarios and consider previous occurrences. Machines such as Watson are real time systems used not only in LAW, Medicine but also in Aviation [7] Let us take a scenario: if an aircraft suddenly develops a problem with a hydraulic system, The pilot can simply describe / speak the problem, that is in natural human language. The work of Watson is to interpret the problem from a logical viewpoint and then to consider all the concerned technical data pertaining to a solution. It then makes a series of recommendations to the pilot on how to modify operation of the aircraft to mitigate the problem, on how to troubleshoot it, or an alternate or nearest airport to divert to. Using Neural structure Watson will consider all possible options and recommend the optimal one. [7] Such type of AI systems using NLP and MT are going to form the backbone of incident management and reporting system of the aviation world. Courses are available at (https://www.watsonacademy.info/course/index.php?categoryid=35) Which introduces state-of-the-art natural language processing methods , with a focus on the technicalities related to IBM Deep Question Answering (Watson) system.

AMRIT
1Ansah is a company located in Sydney which is trying to help mechanics easily use technical documents while working on helicopters at Airbus Group Australia Pacific using a tool called AMRIT.
AMRIT uses NLP to "READ" 1.2million documents or digital manuals of the following types : To deal with huge amount of manuals and queries, help of multiple servers are taken so that parallel processing can be done to achieve the final product. Hardware configuration is of a Linux server farm consisting of seven servers, which strictly follows Airbus specified infrastructure and configuration. [8][9] AMRIT in its next stage is being prepared to take in manuals and documents for fixed-wing aircraft and other rotor based models. Then training and testing are to be done in order to field it for real time testing. 1Ansah's support product of aircraft maintenance and others like it have prospects that are limitless and after Airbus prospective future clients of AMRIT includes Engineering branch of Etihad Airways, Singapore Airlines Engineering Co. ,MRO of Emirates and Lufthansa MRO.

BLUE
Boeing's NLP system, BLUE (Boeing Language Understanding Engine), consists of the following: 1.Parser , 2.logical form (LF) generator, 3. initial logic generator, and 4. processing modules. [10]  initial logic generator performs a straightforward transformation of the LF to first-order logic syntax 4 Processing processing modules then perform word modules sense disambiguation, semantic role labeling, co-reference resolution, and some limited metonymic and other transformations BLUE, is able to generate representational structures for many texts, using numerous linguistic aspects while also missing a variety of others. Each sentence of every paragraph are considered individually, and each sentence is processed . BLUE has a pipelined architecture with multiple transformation steps through which it processes each sentence. Though BLUE produces a output that is good enough for controlled architecture language , such as technical manuals , maintenance manuals etc., but in some cases human assistance may be required to find out the real meaning of the sentence. [10] 2.5 "PLUS" by SAFETY DATA and "Partial Automation of Aircraft Parts" by ATOS

PLUS
Safety Data uses 'PLUS', which uses NLP to read and analyze flight reports. Traditional search methodology generally results in a binary answer (Is the specific key word present? Yes or No), and therefore leads to false positives (the key word is present but in the wrong context) and false negatives (the right context but without the key word). NLP which is a part of AI, Linguistics and Computer Science allows them to weigh a text (here flight reports) depending on its likelihood of addressing the desired topic .

[4] [5]
The PLUS Process can be summarized as follows: PLUS's mediation layer is completely programmable in order to filter and standardize incoming data, which can adapt to any static or dynamic data source.

INVESTIGATE
Queries coupled with logic help in searching for similar occurrences.

NEW KNOWLEDGE
Human assistance and the ability of PLUS to learn allows it to add new knowledge and data related to the concerned work.

GENERATION of WATCH REPORTS
Incoming data is always watched and watch-reports are created in real time.
It also helps in making suggestions and decisions in risk handling and prioritization

: Automation of Aircraft Parts by AToS
AToS, is an international information technology services company Serving a global client base, it delivers IT service, based on Machine Learning has undertaken a project which aims to partly automatize the design of aircraft parts and use NLP applications in the process.
These techniques could bring about substantial improvements in the aviation industry, from enhancing safety to minimizing environmental impact. One of the disadvantages of Neural and Statistical Machine Translation and other NLP systems are that many of them suffer from data deficit. They unconditionally depend on access to large and sufficient data to perform complex analyses and get meaningful results. In such cases the quality of the input determines the quality of the output. Fortunately now open source projects and Data sources have been thriving; from 3D printing to the peer-to-peer economy, many of the most revolutionary inventions of the last decade have been the result of open-source collaborations. Indeed, the crowdsourcing of data collection in aviation has begun, with Automatic Data Surveillance -Broadcast data exchange websites and apps, such as the Flightradar24.com, radarbox24.com, planefinder.com, Open Sky network. The European Union encourages open data exchange and associated projects and has involved sixteen airlines and Air Navigation Service providers who share data with the aim to improve aviation safety.

ASRS and ECCAIRS:
In 2012 the probability of dying on a single flight on one of the top 39 airlines was one in twenty million. air travel safety is continuously improved. In 2012 ICAO (2013) reported 2012 as have had the least rate of incidents and accidents (3.2 accidents for one million arrivals and departures) [11] In most of the cases, even when something serious occurs, example engine problems or Gear and hydraulics(front and rear) problems, the accident is avoided and the aircraft is able to land safely. Equipments such as ILS, ADS-B, Primary and secondary radar, training in Simulators and safety procedures imposed all throughout has contributed to safety. Air Transportation being a complex procedure requires efforts in all level to make it safe. Procedures incident reporting improves safety of flying.
Now moving on to reporting systems, although several systems exist at different levels for companies, government agencies and NGOs, two of the most widely used are ASRS (a North-American database of incident reports), and ECCAIRS (European Coordination Centre for Accident and Incident Reporting Systems) a software system proposed by Europe for managing incident reports at different levels. [11]

ASRS
The oldest (established in 1976) and most well known incident reporting system is known as ASRS which stands for Aviation Safety Report , which is managed by NASA. It is responsible for collecting reports of aviation events in the United States . [11][12] Its information sources contains data for maintenance, policy development, training. ASRS (https://asrs.arc.nasa.gov/) has processed over a million incident reports. ASRS receives reports from pilots, air traffic controllers, cabin crew, dispatchers, maintenance technicians, ground personnel and others involved in aviation operations. reporting to ASRS has been remarkable all throughout, with 400 reports per month. In recent years, report intake has grown exponentially which averages to 1,769 reports per week and more than 7,664 reports per month.
[12] It has 4 types of forms to report into : ASRS products and services are as follows:

ECCAIRS :
ECCAIRS (European Coordination Centre for Accident and Incident Reporting Systems) standardize accident and incident data collection and exchange within the European Union. ECCAIRS was developed by the European Commission's Joint Research Center. Its aim is to assist entities in Europe to collect, share and analyze safety data and try to improve safety in transportation. Its information are free to be used by researchers and others . it is a software platform using the following taxonomy of ICAOs Accident/incident Data Reporting and covers collection, and knowledge extraction of incident reports. The European Coordination Centre for Accident and Incident Reporting Systems and the Safety Recommendation Information System are based on the (ECF). ECF is a software developed by the Joint Research Centre of European commission responsible to report and share accident and incident reports. It is used by national agencies of several countries, including the French DGAC. The reports collected by DGAC are similar to those of ASRS, with the notable distinction of being written in several languages (French and English). The Report goes through the various types of translation methods like Direct, Indirect (Interlingua, transfer), controlled language, sublanguage systems etc. and translations demands such as Dissemination, Assimilation, and other methods. Ultimately the report narrows down to Stand Alone machine and Server based Machine Translation systems and does a comparative study of the MT systems/tools available in between different languages and the cost of each of them. The report also goes through the disadvantage of both the Server based and Stand alone MT S/W's. some of the Stand alone MT S/W's which could be used for ICAO languages as of 2004 are listed in Table 6. we see that there is a coverage gap for six IACO languages (English, French, Chinese, Arabic, Spanish, Russian)  In this paper we have come across some NLP tools that is helping various Aircraft manufacturers produce crafts faster and better and also some NLP tools that help us in storingincident reports and help us to have a safer future for aircrafts by helping to identify the problems and how to avoid the incidents and in turn accidents. Let's compare the tools covered in this paper in the table below: languages are being employed. NLP based Speech to Speech, Speech to Text translation systems based on Machine Translation specifically designed for aviation domain are a must.