Keeping Transit Networks Moving: How AI-Powered Predictive Maintenance Transforms Public Transport – Future Transport-News

Report on AI-Powered Predictive Maintenance for Sustainable Public Transportation Infrastructure

Introduction: Aligning Urban Transit with Sustainable Development Goals

Public transportation systems are fundamental to the development of sustainable cities and communities, directly supporting Sustainable Development Goal 11 (SDG 11). With millions of daily passenger journeys, as exemplified by Transport for London’s 3.5 billion journeys in 2024/25, the reliability of fare collection infrastructure is paramount. However, traditional reactive maintenance models for this equipment present significant challenges, leading to service disruptions, increased operational costs, and a diminished passenger experience, thereby hindering progress towards creating resilient and efficient urban environments.

This report details a collaborative research initiative between industry and the Artificial Intelligence and Data Analytics (AIDA) Lab at Imperial College London. The project explores the application of Artificial Intelligence (AI), specifically Large Language Models (LLMs), to transition from a reactive to a predictive maintenance paradigm. This innovation directly addresses SDG 9 (Industry, Innovation, and Infrastructure) by leveraging technology to build resilient, reliable, and sustainable infrastructure, ultimately reinforcing the objectives of SDG 11.

The Challenge: Inefficiencies of Reactive Maintenance

The conventional “break-fix” approach to maintaining fare collection systems is inherently inefficient and unsustainable. When a fare gate fails, a multi-step, resource-intensive process is initiated, often requiring multiple site visits by engineers. This model results in:

• Increased Downtime: Service interruptions lead to passenger delays and frustration, undermining public confidence in transit systems.
• Operational Inefficiency: Unplanned callouts and repeated site visits for diagnosis and repair increase labor and logistical costs.
• Resource Mismanagement: Inefficient inventory management and the potential for premature or unnecessary parts replacement conflict with the principles of SDG 12 (Responsible Consumption and Production).

These inefficiencies create a barrier to developing the robust and reliable public transport networks essential for sustainable urban growth.

The Solution: Predictive Maintenance as a Catalyst for Sustainable Infrastructure

Leveraging AI to Support SDG 9

Predictive maintenance represents a paradigm shift, utilizing data analytics to forecast equipment failures before they occur. This proactive strategy aligns with SDG 9 by integrating innovative technology into critical infrastructure management. The core of this project was the development of an LLM, nicknamed “PartLlama,” designed to analyze fare gate maintenance logs. These logs contain a combination of structured error codes and unstructured, free-text descriptions from technicians.

Methodology and Technological Innovation

The process involves several key stages:

1. Data Ingestion: The model processes historical maintenance data, converting textual descriptions of incidents into a format suitable for AI analysis.
2. Part Classification: Instead of general text prediction, the LLM is fine-tuned to classify each incident log and accurately predict the specific spare part required for the repair from a predefined list.
3. Efficient Fine-Tuning: To ensure cost-effectiveness and scalability, the project employed Low-Rank Adaptation (LoRA). This technique updates only a small fraction (0.0524%) of the model’s parameters, significantly reducing the computational resources required for training while preserving the model’s core language capabilities.

Results and Impact on Sustainable Development

The model demonstrated high accuracy and practical applicability, proving its potential to transform transit maintenance into a more sustainable practice.

Key Performance Metrics

• 83.5% Top-1 Accuracy: The model’s primary recommendation for a required part was correct in the vast majority of cases.
• Over 95% Top-5 Accuracy: The correct part was almost always included within the model’s top five suggestions, providing engineers with a highly reliable shortlist.

Contributions to Sustainable Development Goals

The successful implementation of this AI-driven model yields compelling benefits that directly support multiple SDGs:

1. SDG 11 (Sustainable Cities and Communities): By minimizing gate downtime and service interruptions, the system enhances the reliability and availability of public transport. This improves the passenger experience, encourages ridership, and contributes to creating more efficient, inclusive, and sustainable urban transit networks.
2. SDG 9 (Industry, Innovation, and Infrastructure): The research demonstrates a successful application of cutting-edge AI to modernize infrastructure management. It shows the potential to reduce on-site engineering visits by up to 70%, making maintenance operations more resilient and efficient.
3. SDG 8 (Decent Work and Economic Growth) & SDG 12 (Responsible Consumption and Production): The model drives significant cost savings by reducing unplanned downtime, which costs industrial manufacturers an estimated $50 billion annually. By optimizing maintenance schedules and spare-part inventories, it lowers labor and storage costs and reduces material waste, promoting sustainable economic practices and responsible resource consumption. Reports suggest predictive maintenance can reduce overall maintenance costs by 18-25%.

Future Outlook: Scaling Innovation for Global Urban Sustainability

This research serves as a proof of concept with significant potential for expansion. Future development will focus on:

• Real-Time Integration: Developing a deployment pipeline to provide instant predictions from live maintenance logs.
• Expanded Coverage: Training the model to predict a wider range of spare parts, including those used less frequently.
• Enhanced Explainability: Incorporating features that highlight the specific data points driving a prediction, increasing technician trust and confidence in the AI’s recommendations.

The long-term vision is to create a comprehensive, data-driven maintenance ecosystem that anticipates failures proactively. This approach provides a blueprint for transit agencies worldwide to enhance operational resilience and advance their sustainability objectives.

Conclusion

The integration of AI-powered predictive maintenance into public transportation systems is a critical step toward building the smart, reliable, and sustainable infrastructure mandated by the Sustainable Development Goals. By shifting from a reactive to a proactive model, transit agencies can reduce costs, minimize disruptions, and improve resource management. This technological advancement not only enhances the passenger experience but also solidifies public transport’s role as the backbone of sustainable modern cities, ensuring that urban mobility is efficient, resilient, and environmentally responsible.

Analysis of Sustainable Development Goals (SDGs) in the Article

1. Which SDGs are addressed or connected to the issues highlighted in the article?

• SDG 9: Industry, Innovation, and Infrastructure: The article’s core theme is the application of innovative technology (AI and large language models) to improve the reliability and efficiency of public transport infrastructure (fare collection gates).
• SDG 11: Sustainable Cities and Communities: By focusing on enhancing public transportation systems, the article directly addresses the goal of making cities more sustainable, inclusive, and resilient. Reliable public transport is a cornerstone of a sustainable city.
• SDG 8: Decent Work and Economic Growth: The article highlights how predictive maintenance leads to significant operational efficiencies and cost savings for transit agencies, which contributes to economic productivity.
• SDG 12: Responsible Consumption and Production: The shift from reactive or blanket preventative maintenance to a predictive model reduces waste by optimizing the use of spare parts, aligning with principles of resource efficiency.
• SDG 17: Partnerships for the Goals: The project described is a collaboration between a private company (Cubic Transportation Systems, the original publisher) and an academic institution (Imperial College London’s AIDA Lab), exemplifying a partnership to achieve sustainable development.

2. What specific targets under those SDGs can be identified based on the article’s content?

1. Target 9.1: Develop quality, reliable, sustainable and resilient infrastructure.
   • The article directly addresses this by aiming to improve the reliability of fare collection infrastructure. It discusses reducing “unplanned breakdown,” “longer downtime,” and “service interruptions” to create a more resilient system.
2. Target 9.4: Upgrade infrastructure and retrofit industries to make them sustainable, with increased resource-use efficiency and greater adoption of… sound technologies.
   • The use of AI (“PartLlama”) is a technological upgrade that increases resource-use efficiency. The article notes benefits like “optimized spare-parts inventory” and reducing “blanket part changes,” which makes the maintenance process more sustainable.
3. Target 11.2: Provide access to safe, affordable, accessible and sustainable transport systems for all.
   • By reducing gate downtime and delays (“queues build, boarding delays increase”), the AI-powered system improves the accessibility and reliability of public transport, contributing to a better and more sustainable transit experience for passengers. The article states the goal is to keep “passengers moving smoothly through stations.”
4. Target 8.2: Achieve higher levels of economic productivity through… technological upgrading and innovation.
   • The article details how this technological innovation leads to higher productivity. It cites reports that predictive maintenance can “reduce maintenance costs by 18-25%” and generate “up to 40% more [cost savings] than reactive maintenance,” which are direct measures of increased economic productivity for the transit agency.
5. Target 12.5: By 2030, substantially reduce waste generation through prevention, reduction…
   • The predictive model prevents waste. Instead of routine “blanket part changes” where functional parts might be discarded, the system ensures a part is replaced only when “the data tells you a failure is imminent.” This reduces the generation of waste from discarded spare parts.
6. Target 17.17: Encourage and promote effective public, public-private and civil society partnerships.
   • The article explicitly mentions the project was a collaboration: “Working in collaboration with independent researchers from the Artificial Intelligence and Data Analytics (AIDA) Lab at Imperial College London, our team set out to apply large language models to fare gate logs…” This is a direct example of a public-private/academic partnership for innovation.

3. Are there any indicators mentioned or implied in the article that can be used to measure progress towards the identified targets?

• Reduction in Downtime/Increase in Availability: The article implies this can be measured, stating that some organizations report “up to 50% fewer unplanned outages” and that the goal is to reduce “gate downtime.” This is a direct indicator for infrastructure reliability (Target 9.1) and transport system accessibility (Target 11.2).
• Operational and Cost Efficiency: The article provides several quantifiable indicators.
  • A potential “reduce on-site engineering visits by up to 70 percent.”
  • Cost savings are explicitly mentioned, with reports suggesting a reduction in “maintenance costs by 18-25%.”
  • These metrics measure progress towards increased economic productivity (Target 8.2) and resource efficiency (Target 9.4).
• Model Accuracy Rate: The success of the AI model is measured by its accuracy. The article states it achieved “83.5 percent Top 1 accuracy” and “Over 95 percent Top 5 accuracy.” This indicator measures the effectiveness of the technology being implemented (Target 9.4).
• Passenger Experience Metrics: While not quantified with numbers, the article implies that progress can be measured through “fewer station delays and fewer complaints from passengers” and improved “ridership metrics.” These are key indicators for the quality of a sustainable transport system (Target 11.2).
• Optimized Resource Use: An indicator for Target 12.5 would be the reduction in the number of spare parts used compared to a preventative maintenance schedule. The article mentions the benefit of an “optimized spare-parts inventory,” which could be tracked.

4. SDGs, Targets, and Indicators Summary

Source: futuretransport-news.com