AWS Partner Network (APN) Blog
Transforming Aviation Maintenance with the Infosys Generative AI Solution Built on Amazon Bedrock
By Dr. Ravi Kumar G. V. V. – Infosys Limited
By Devaraja Holla V., Venugopal R., Senthil S. M., Krishnasagar M. K. – Infosys Limited
By Kiran Killedar – AWS
 |
| Infosys |
 |
Commercial aircraft maintenance, repair, and overhaul (MRO) is a global business involving original equipment manufacturers (OEMs), passenger and cargo airline operators, and third-party providers. There are over 25,000 aircraft operating worldwide, each requiring unique scheduled and unscheduled maintenance, as well as regulatory compliance to ensure airworthiness for continued operation.
According to a 2022 IATA report, the global commercial MRO spend in 2022 was approximately $78 billion USD, with a compound annual growth rate (CAGR) of 4.4%. To address cost pressures and a shortage of skilled engineering labor, leveraging emerging technologies like generative artificial intelligence (AI) is crucial.
In this post, you will learn how an Infosys generative AI-based solution built on Amazon Bedrock, a fully managed service that makes it easy to build and scale generative AI applications with foundation models, enhanced productivity and leveled the playing field for newcomers in MRO. The solution aims to reduce the expenditure on document search, analysis, interpretation, and management, which currently amounts to approximately $5 billion USD.
The customer wanted assistance in crafting compelling documents, summarizing complex documents, building conversational AI agents, and generating striking, customized visuals. AWS stepped in to assist customers contending with the need to build and scale generative AI applications with Amazon Bedrock, which arms developers and businesses with secure, seamless, and scalable access to cutting-edge models from a range of leading companies.
Infosys is an AWS Premier Tier Services Partner and Managed Services Provider (MSP) that enables clients to outperform competition and stay ahead of the innovation curve.
Challenge
The aircraft maintenance engineering process involves referring to a wide range of documents, such as aircraft configuration details, operations manuals, maintenance procedures, service bulletins, warranty claims, and airworthiness directives.
The aviation industry is heavily regulated, with civil aviation authorities like Federal Aviation Administration (FAA) and European Union Aviation Safety Agency (EASA) issue airworthiness directives on a regular basis to address safety deficiencies in specific aircraft models, engines, avionics, and systems.
Compliance with these directives is essential, as non-compliance could lead to grounding the aircraft. Finding and interpreting relevant engineering documents and airworthiness directives for specific aircraft models requires careful reading and analysis. They contain unstructured text with problem descriptions, resolutions, and references to other documents.
Accessing these documents involves using filters and keyword searches, making the process time-consuming. The current search, access, and analysis process can take several days, adding cost and potentially increasing errors, and new users may find it challenging to use the system effectively.
Solution Overview
The Infosys generative AI solution aims to address these challenges by combining generative AI, knowledge engineering techniques, and an Infosys-patented framework for supporting aircraft repair processes. It utilizes large language models (LLMs) and knowledge graphs with domain-specific ontologies, and the Amazon Bedrock platform.
The solution employs vector storage and knowledge graphs to organize the data and capture relationships within. It also enhances the availability and accessibility of relevant content for aviation maintenance and repair. It aims to provide accurate answers to maintenance engineers’ queries from the aircraft document repository, reducing engineering effort and enabling faster decisions to ensure the aircraft’s airworthiness.
By understanding the intent and semantics of questions, the solution uses language models to offer precise responses. It then evaluates answers objectively for relevancy and authenticity, which are critical for aviation safety and regulatory compliance.
Solution Features
The solution offers semantic search powered by an LLM, which uses query planning and prompting techniques with the help of Amazon Bedrock Titan and Claude models. It also includes custom-trained models for improved accuracy by fine-tuning on custom datasets. Users benefit from a chatbot-like conversational experience with context management.
The solution is highly adaptable to different domains, making it suitable for various use cases with unique engineering document sets. Additionally, it supports various data formats, including unstructured text and graph data, and can connect to multiple data sources through pluggable data adapters.
Figure 1 – Infosys generative AI solution features.
Solution Architecture
The solution’s architecture utilizes AWS components and services for indexing and retrieval flow.
Figure 2 – Solution architecture.
The indexing flow works on documents like airworthiness directives (ADs), pre-processing and transforming them into vectors and triplets (hierarchical) representation. The document text and vectors are stored in a vector database (Amazon Aurora PostgreSQL), while triplets and metadata are stored in a knowledge graph (Amazon Neptune). This is generally a one-time activity unless the source documents change.
Figure 3 – Indexing flow.
The retrieval flow involves finding potential documents from the repository (Vector + knowledge graph) and providing answers to user input questions. The system synthesizes the final answer based on the shortlisted document set, along with references to these shortlisted documents themselves.
Figure 4 – Retrieval flow.
Quality of the answers generated by LLMs depends on the documents passed to it in the prompt bundle. Lots of care must be taken in designing the indexing and retrieval part of the pipeline.
The Infosys generative AI solution uses prompting techniques extensively to build and fine tune the model, as LLMs can sometimes generate inaccurate information when prompted. To minimize this issue, it instructs the LLMs to generate answers only from the relevant document set, thus reducing the likelihood of inaccurate responses.
As users are free to ask any open-ended questions, the system should be flexible enough to understand the question and create a good enough plan to answer it. An LLM-powered agents’ framework is leveraged for advanced query planning and orchestration of the step-by-step plan. The generated answers should be reliable and accurate. An evaluation approach measuring the answers on faithfulness, relevancy, and harmlessness is implemented in this solution.
The solution is deployed on AWS, including the frontend, data storage, and backend using AWS Lambda. The initial version uses publicly available ADs from the FAA and EASA, ensuring no privacy concerns. Amazon Bedrock provides secure methods to store and customize models for enterprise data, addressing data privacy as a critical feature.
Conclusion
Aircraft maintenance costs (10-15%) are the second highest after fuel costs for aircraft operators. The Infosys generative AI solution cuts down maintenance costs and speeds up decision-making by swiftly locating the right documents.
In this post, we explored how the Infosys generative AI solution utilizes a vector and graph database to ensure accurate and contextualized answers with high precision. Its architecture is decoupled and has been tested with Amazon Bedrock Titan and Claude models, making it easy to deploy with minimal changes. It offers a seamless and hassle-free conversational AI experience to access relevant content from a vast repository of engineering documents.
In addition to aviation, the solution is versatile and applicable across various industries like manufacturing, automotive, and financial services for regulatory, compliance, troubleshooting, maintenance, and more. For further information or customized solutions, please reach out to Infosys and AWS.
.
.
Infosys – AWS Partner Spotlight
Infosys is an AWS Premier Tier Services Partner and MSP that offers integrated consulting and technology solutions leveraging the flexibility and economics of cloud where IT and business services are delivered on demand.