Predicting Engine Failures on Grand Central’s Fleet — Before They Happen
Grand Central’s Class 180 trains carry Remote Condition Monitoring on their engines, sampling at 1Hz. That’s roughly 12 million rows of data per engine, per month.
The Challenge
Manual weekly review of one engine was feasible. Scaling that to 50 engines would produce 1.4 billion rows a year — impossible to review by hand, and it was already slowing decisions down.
Engine failures during service cause delays and cancellations. The question Chrome Angel Solutions and Amygda set out to answer: could machine learning on that condition-monitoring data predict engine health early enough to act on it?
The team entered the BridgeAI / Innovate UK AI Feasibility Competition, focused on transport innovation, to test it.
The Approach
The system ingests two streams: continuous engine telemetry (pressure, temperature, performance metrics) and operational logs tracking faults, maintenance actions and behaviours. Machine learning models convert that into an interpretable health score from 0–100 for each engine — falling scores mean rising failure risk, aggregated at fleet level with drill-down diagnostics underneath.
- Hybrid ML architecture: supervised anomaly detection plus unsupervised pattern recognition, with deep learning for high-frequency time-series telemetry.
- Large Language Models: interpret maintenance logs and extract insight from unstructured text.
- Explainable AI: SHAP values keep every prediction traceable — engineers see why a score moved, not just that it did.
- Signal analytics: rate-of-change detection and tuneable thresholds catch early degradation.
- Cloud-native MLOps: continuous retraining, controlled testing, and real-time inference at scale.
Not automating the maintenance decision. Informing it — faster, and with the reasoning attached.
The Results
“The collaboration on this project has demonstrated how effective alignment between industry experts, technology partners and subject-matter experts can drive truly innovative solutions. Shifting from reactive to predictive maintenance is essential to improving reliability, and AI will be vital in making our approach more proactive and resilient. This is just one example of an AI application in rail. The potential is still being unlocked, and collaboration is the key to unlocking it.”
What’s Next
The project is moving from proof-of-concept into operational deployment: continuing live trials to tune model thresholds and reduce false alerts, launching a follow-on BridgeAI project to build AI literacy among engineers, and testing how the health-score framework scales across fleets and systems.
About Amygda
Amygda builds predictive intelligence systems that work across any transport equipment — from aviation to railway. Explainable AI that enhances rather than replaces human expertise.
About Chrome Angel Solutions
Founded in 2006, Chrome Angel Solutions implements transformational technology for railways, maritime and other asset-intensive industries.
Talk to us about your own fleet’s data.
- What data you already have — CMMS records, fault logs, sensor feeds.
- What a health score looks like for your specific fleet or equipment.
- What implementation actually takes — typically 4–6 weeks.
FAQ
What data does the health score use?
Two streams: continuous engine telemetry (pressure, temperature, performance metrics) and operational logs tracking faults, maintenance actions and behaviours. No new sensors were installed — this runs on Grand Central’s existing Remote Condition Monitoring data.
How accurate is the prediction?
In proof-of-concept testing, the system detected roughly 84% of engine failures in advance — about 8 in 10 — giving days of warning rather than a same-day alert.
Does this replace engineers’ maintenance decisions?
No. The system was deliberately designed to inform decisions, not automate them — responsibility for maintenance and service calls stays with engineers. Every health score comes with the contributing factors behind it, not just the number.
What is the BridgeAI / Innovate UK programme?
BridgeAI is an Innovate UK programme supporting AI feasibility projects in transport and other sectors. This project was one of the funded AI Feasibility Competition proof-of-concepts, delivered in partnership with Chrome Angel Solutions, Amygda, Angel Trains, and Grand Central Railway.
Is this in production, or still a proof of concept?
Proof-of-concept as of this case study. The project is moving into live operational trials to tune model thresholds and reduce false alerts, with a follow-on BridgeAI project planned to build AI literacy among engineering teams.
Does this approach work outside rail?
Yes — the combination of high-frequency telemetry, machine learning, and explainable AI applies to any asset-intensive sector, including aerospace, automotive fleets, energy and utilities, and manufacturing equipment.
