Faster to fix, faster to fly: AI for aircraft maintenance

Last week, my friends took a flight from the Dominican Republic to New York – or they tried to. The flight took off, circled for an hour, and landed again. There was an issue closing the cargo door. They waited on the ground for several hours as the maintenance technicians hurried to resolve it. People began disembarking and booking alternative flights. They waited some more. There was a missing part or a misdiagnosed defect or an unresolved chronic issue. Whatever it was, they missed their shot; the crew timed out and the flight was cancelled.

Effective maintenance work makes the difference between whether a flight takes off on time or delays cascade – leading to compensation payouts, regulatory fines, and lost revenue that costs airlines millions every year. It all rests in the hands of technicians.

With fleets, parts, and maintenance crew spread across the globe, maintenance planning is too complex to account for last minute events.

That’s where AI comes in: to transform the terabytes of MRO data into real-time recommendations for the next best action. These insights can help achieve significant savings on the bottom-line, but they also form part of a revenue-driving strategy, improving customer experience and creating the reliability airlines need to take on more service routes.

Kubrick, a Databricks consulting partner, is working with airlines to transform their maintenance footprint – manuals, log history, inventory, and more – and turn it into a blueprint for defect remediation and planning. Here, we take a deep dive into Kubrick’s Defect Resolution AI Assistant: an end-to-end AI solution that puts insight straight into the hands of technicians.

The opportunity of AI-enabled MRO

For airlines, reducing maintenance costs has the potential for critical bottom line impact; they are the third highest outlay on the balance sheet after fuel and employee salaries[1] - and are far more controllable.

The wealth of data underpinning MRO processes holds the insight to streamline processes and create cost-reducing efficiencies. Manuals, maintenance logs, work orders, inventory stocks, and in-flight data provide the blueprint for understanding the most effective fixes to keep flights on schedule. However, the manual methods of interacting with this data slow down work:

• Manual data logging processes can take up to 24 hours, preventing real-time decision-making, with added risk of inaccuracies in recording.
• Engineers manually search through extensive documentation to diagnose and resolve, often with adding complexity/cost when incorrect resolutions are made.
• Lack of connectivity between data sources prevents critical contextualization of issues to link maintenance events and their solutions.

The aviation industry is also facing a severe technician shortage, projected to exceed 70,000 roles by 2033[2]. With every skilled technician that retires, maintenance teams lose critical institutional knowledge, which is difficult to teach junior talent; searching through thousands of pages of manuals and logs doesn’t make it any easier.

Generative AI provides a unique opportunity to distill complicated information into easy-to-understand text – and allow people to query for data in their natural language. When it comes to putting data into the hands of technicians, this is a game changer.

Introducing the Defect Resolution AI Assistant

Kubrick’s Defect Resolution AI Assistant is acompound AI system that leverages the Databricks Data Intelligence Platform to combine the wealth of data to create context-aware analysis on what needs to be fixed and when, and with which parts and technicians. This architecture is interfaced with a chatbot designed specifically for technicians, providing quick and easy insight that saves them thumbing through manuals and misaligning resolutions by providing the full picture.

By connecting manuals, historic logs, inventory, work orders, flight schedules, and live flight data, the system enables:

• Intelligent troubleshooting: it queries manuals and historic defect logs to suggest likely defect based on fault codes and historic issues
• Guided repairs: it quickly retrieves instructions for defect resolution with suggestions based on historic fixes
• Maintenance prioritization: it contextualizes with flight and maintenance schedules as well as inventory so technicians know how long they have to make repairs and if they have the parts to do it.

Under the hood

At a high-level, the system is comprised of:

• Source systems: Data from the maintenance database of equipment/vehicle parts and inventory is combined with relevant live and historical data, such as defects, work orders, out-of-service events, as well as relevant regulatory/maintenance codes and manuals.
• Ingestion: Tools such as Azure Data Factory (ADF), Fivetran, etc., are employed to ingest the data
• Storage: Azure Data Lake Storage (ADLS) Gen 2 on Microsoft Azure is used for storage
• Data processing: All unstructured, semi-structured and structured source files are processed on the Databricks Data Intelligence Platform using Delta Live Tables (DLT) and streaming jobs to build bronze, silver, and gold tables in Unity Catalog. Unity Catalog ensures data governance, integrity, lineage, and high-quality monitoring through established standards for each medallion architecture level. Finally, the unstructured text of the associated repair manual is embedded into Databricks Vector Search for retrieval-augmented generation (RAG) using LLMs.
• Data visualization: The database supports multiple dashboards, which offer views for senior stakeholders on pressing maintenance issues, historical fleet health and current work orders, out-of-service events, and defects.
• Generative AI: Databricks Mosaic AI is used to build an end-to-end compound AI system. Mosaic AI Model Serving is used to host a fine-tuned Llama 3 model for text-to-SQL.

Defect Resolution AI Assistant in action: Use cases

1. Defect identification and resolution
The day-to-day tasks of technicians are transformed: from thumbing through thousands of pages of manuals and searching work orders to instantly being pointed to the right guidance. Conversing with a chatbot, the technician simply enters a fault code or defect description, and the AI returns the diagnosis, steps for remediation, and parts required, customized to the aircraft’s maintenance history.


2. Preventative maintenance
The AI system quickly becomes a long-term strategic asset for maintenance planning. By recording the common defects that result in out-of-service events, managers can factor in chronic defects into maintenance schedules and optimize parts inventory to match.


3. Technician efficacy
With the Defect Resolution AI Assistant, technicians have an ‘AI-memory’ of the defects they have tackled and the steps to remediate them. It becomes a training assistant to quickly upskill juniors while creating an accessible hub of institutional knowledge for all.

Putting trust in AI

The model’s performance is evaluated in two critical ways. First, another LLM acts as a judge for all modules that generate human-readable text. This LLM-as-a-judge model ensures that the generated responses accurately answer the question, avoid hallucinations and match the expected output format. The second evaluation method involves the TextToCypher module. Since this model generates code rather than human-readable text, it cannot be evaluated by another LLM in the same way. Instead, it uses a custom evaluation function in Databricks Managed MLflow. This function runs the generated code on Kubrick’s database to verify its functionality and then compares the results to those produced by the ground truth code. A match results in a positive evaluation, while a discrepancy results in a negative one.

The impact of the Defect Resolution AI Assistant

In operation with leading airlines, the Defect Resolution AI Assistant is helping technicians reduce troubleshooting time by up to 50% and reducing overall maintenance-related delays by 30%. In one major airline, this has so far saved $14 million in compensation payouts, fines, and wasted maintenance costs. Read the case study here.

The power of the tool is to apply it to your goals. Kubrick is working with clients to shape the solution to strategic use cases that will maximize impact, for example customizing it to specific aircraft types that are expensive or difficult to fix, or to improve the longevity of aging fleets. In one airline, the Defect Resolution AI Assistant is minimizing defects in first-class and business cabins to maximize capacity and drive revenue.

The value of the solution derives from Kubrick’s expert understanding of Databricks tools such as Delta Live Tables (DTL), streaming jobs, Unity Catalog, and Mosaic AI, making the sum of its parts all the more efficient and powerful, with security and governance in-built. Kubrick’s clients benefit from both a robust and scalable solution and their industry-specific knowledge that transform technology into impact on their MRO workflows and balance sheet.

About the author

Lewis Allsop is a Director at Kubrick specializing in Databricks solutions for Aviation. To learn more about his work and the Defect Resolution AI Assistant, get in touch: lewisallsop@kubrickgroup.com

[1] IATA

[2] McKinsey