We’re excited to announce a major upgrade to the AI Defect Classification feature on the Railnova platform. This enhancement brings a new level of intelligence and usability to the way defects are categorised in the maintenance lifecycle, powered by Large Language Models (LLMs) and multi-agent AI systems. If you want to know more about our AI performances, scroll down to the “How Railnova AI Powers Smarter Classifications” section.
Smarter AI, Smoother Workflow
Functional breakdown structures are essential for quality maintenance planning, but their complexity often creates friction for users. With this release, we’ve introduced a dramatically improved experience that helps fleet managers and maintenance personnel categorise defects faster and more accurately, without digging through endless dropdowns.
What’s New?
Real-time suggestions powered by LLM-based AI
Get the top 5 recommended categories as soon as defect details are entered. Suggestions update dynamically as more information is added, acting as both a guide and a soft check on input quality.
Context-aware interaction at key moments
Categorisation now appears as a dedicated panel during the creation of a defect report (if authorised) and again at closing, with updated AI suggestions based on Return of Experience (REX) notes.
Support for user roles and workflows
Not everyone needs to categorise defects or has the technical background to do so. The Railnova platform now allows users without technical expertise to skip initial categorisation, preserving data quality while reducing friction.
Continued support for powerful analytics
As always, full maintenance history and classification data remain available for export and integration, along with insightful Pareto graphs on immobilisations and traffic impact.
UI Enhancements
A new categorisation panel appears when creating or closing a defect. It is designed for clarity and speed and offers clear, one-click AI suggestions tailored to your input.
Availability
This feature is now available to all customers with a subscription to the AI Defect Classification feature. If you’re not yet using this module and want to explore how AI can streamline your defect workflows, contact us at sales@railnova.eu.
Why It Matters
This upgrade directly supports the ECM2 (Maintenance Development) function by improving the quality and consistency of defect classification. It makes it easier to identify defect trends, reduce downtime, and optimise intervention planning.
Stay tuned for more updates: we continue investing in AI-driven features that elevate the entire rail maintenance lifecycle.
How Railnova AI Powers Smarter Classifications
Railnova's AI Defect Classification uses a Large Language Model specialised in railway terms to analyse defect descriptions and suggest the top 5 most relevant categories. Initially, the AI achieves 50% accuracy for new categories, improving to 90% as users provide feedback.
Multilingual Support
The AI understands defects written in English, German, French, Dutch, and Italian, making it usable across Europe. It provides accurate suggestions regardless of the technician's language.
Railway Expertise
Unlike general AI models, this one is trained with real railway defect reports, manuals, and maintenance records. It understands specific railway terms and distinctions, such as "bogie noise" versus "compressor failure."
Learning from User Input
Every time a user confirms or changes a category suggestion, the AI learns and improves. This feedback loop helps the AI refine its suggestions, especially for rare or unusual defects.
Standard and Custom Categories
Railnova offers pre-set, optimised categories for immediate use. Users can also work with the Railnova team to create custom categories tailored to their specific needs while ensuring the AI continues to perform as well.
Performance and Improvement
Performance is assessed through metrics, including the frequency of correct category identification within the top five suggestions, user acceptance rates, and category coverage. Preliminary findings indicate accuracy levels of 90% for the primary suggestion in frequently encountered categories, with continuous enhancements achieved through iterative learning from user interactions.