Resources

Enhancing Airworthiness Data Integrity using Explainable AI and Wrangling

KeepFlying
Blog
June 29, 2022
Enhancing Airworthiness Data Integrity using Explainable AI and Wrangling

The starting point of building and harnessing a data ecosystem is to define business problems and the values that can be generated by the underlying data.

Within an Airline or an MRO, this often involves individual functions (Fleet Management& Technical Records, CAMO, Part 145, Supply Chain) consolidating data sources, identifying gaps and filling in the blanks using tools and technologies to derive potential use cases that can add value to their functions.

Airworthiness Data Integrity

The historic data that often travels along with an Aircraft or Engine during a trade or a lease, or one that an MRO maintains against Hangar and Shop Visits are often spread across different sources. Less than 60% of the data is managed efficiently within a Maintenance / MIS system in most organizations.

While the paperless journey has been gaining adoption, most projects of this nature require a tremendous volume of data to be tapped through scanned PDFs and other electronic sources that lie outside a traditional MIS.

Significant progress has been made using OCR technologies to extract data out of scanned PDFs and link, say for example, Dirty Finger Prints (DFPs) to the source Work Package in an M&E/MRO/MIS. While this eases the process of data management and indexing, it is often insufficient when churning the data into useful business value.

The underlying objective, and challenge, is to achieve a level of data integrity where the paper and digital records match.

This usually entails departments filtering and enhancing the integrity of data sources on their own through defining rules that are subsequently fed into a consolidated data set. This may involve the team supporting Technical Records, for instance, writing code and rules on top of their airworthiness data to filter out, say, applicable Service Bulletins against a particular MSN when extracted from a scanned PDF or highlighting any inconsistencies in the calculation of Remaining Cycles against an Engine LLP based on the thrust rating(s) in which an Engine has historically operated.

Significant progress has been made using OCR technologies to extract data out of scanned PDFs and link, say for example, Dirty Finger Prints (DFPs) to the source Work Package in an M&E/MRO/MIS. While this eases the process of data management and indexing, it is often insufficient when churning the data into useful business value.

Similarly, an Part145 that has managed several hundreds and thousands of Hangar / Shop Visits can extract valuable data points from their own history of managing different Aircraft and Engine types of varying work scope levels. Rules can be written depending on the type of Aircraft / Engine and how the work scope levels varied by Age, Utilization, Operational Environment, EGTM and other technical and operational parameters. Text semantics can be introduced to identify patterns that help estimate, say for example, prediction of Non-Routines based on the routine card being worked on.

While each department focuses on value and the right set of tools and technologies that can support building a validated, enhanced data set that allows for further analysis, they still end up create data silos and can be expensive to manage, given each function requires data points at different frequencies depending on the mission critical impact it has.

So how can Airlines, MROs make the best use of technologies and scale that can be replicated across functions organically that allow the data streams to interact with one and other? How can such a layer help increase the accuracy of the data being processed?

Data Wrangling. And Explainable AI.

Beyond rules, machine learning models that are configured at an Aircraft and Engine Type level can create an organic framework that allows users to understand and interpret predictions made by the underlying models whilst being integrated across data streams flowing in from across functions.

By training these models with a wide variety of data samples (typically spanning across hundreds and thousands of standard documents and forms – from Form 1 /8130s to LLP status reports to Shop Visit findings etc.,), data integrity is managed as a factor of the underlying Aircraft / Engine profile.

The video below shows how this is achieved and managed.

Understanding how machine learning models arrive at the decisions they make is very crucial when managing airworthiness and maintenance data. Data Wrangling in conjunction with Explainable AI allows functions to understand this impact.

Contact us to know how this can be setup in your environment against your Aircraft, Engines, Components and MRO facilities.