AI-Enhanced Lean-EVM Framework for Railway Infrastructure Delivery: Integrating WBS, LPS, LBMS, and Deep Learning

Abstract

Railway construction often feels like orchestrating dozens of moving pieces—earthworks, track-laying, Bridges, approach roads, signalling—scattered over many kilometres. Traditional tools such as Work Breakdown Structures and Earned Value Management give teams clear budgets and high-level timelines, but they struggle to reflect what’s happening daily. Lean approaches like the Last Planner System and Location-Based Management System bring in weekly commitments and spatial flow-line charts. Yet, they still rely on manual updates and can’t foresee emerging problems. In this paper, we weave these methods within a digital-twin environment and layer on artificial intelligence to bridge that gap. We first map the entire corridor into zones, assign each task a budget and schedule, and generate classic S-curves for cost and progress. Simulated lean check-ins capture crew commitments and roadblocks weekly, while flow-line charts visualize how far each crew has advanced along the track. Behind the scenes, a language model reads meeting notes and field logs to flag new risks, a deep-learning network learns from past cost, schedule, and flow-line data to forecast future performance and a lightweight neural detector watches for early signs of bottlenecks. Our paradigm provides a theoretical path towards location-aware, real-time decision assistance by conceiving Schedule Performance Index (SPI), Cost Performance Index (CPI), and Percent Plan Complete (PPC) as elements of a single digital twin. Through the conversion of scattered data into coherent advice, this integration promises to balance contractual rigour with flow-centric reactivity, maintaining linear infrastructure delivery in line with both financial and geographical goals.

Introduction

1. Context & Challenges

• Railway construction involves large-scale, linear, and interdependent tasks (earthworks, track laying, electrification, etc.).

• Traditional project controls like Work Breakdown Structures (WBS) and Earned Value Management (EVM) focus on cost/schedule tracking but ignore spatial and real-time site conditions.

• Lean construction methods (Last Planner System - LPS, and Location-Based Management System - LBMS) improve local coordination and flow using zone-based planning and short-term lookaheads.

• However, even lean methods can lack automation and struggle with large data volumes.

2. Proposed Framework: AI-Enhanced Lean–EVM

A novel system integrates lean practices, EVM baselines, and AI-driven prediction modules into a continuous planning loop:

1. Geo-referenced WBS defines zones and assigns cost/schedule baselines (S-curves, SPI, CPI).

2. Lean planning adds weekly lookaheads, flowlines, and PPC tracking.

3. AI modules perform:

   • Constraint detection (via LLMs) from unstructured logs.

   • Forecasting of SPI, CPI, and PPC using LSTM–Transformer models.

   • Bottleneck detection via CNNs on flowline residuals.

4. A real-time dashboard integrates insights for adaptive planning.

3. Methodology & Simulation

To validate the framework, simulated datasets were created:

• Meeting transcripts mimicking progress reports and constraints.

• Performance indicators (SPI, CPI, PPC) over 10 weeks.

• Flowline data for spatial progress across zones with delays introduced.

Simulation Modules:

1. Constraint Detection:

   • A simple logistic regression classifier correctly identifies site risks from text (100% accuracy in tests).

2. Schedule Forecasting:

   • Linear models forecast near-term SPI but struggle with complex trends.

   • More advanced models (e.g., LSTM/Transformer networks) are better suited for capturing temporal dependencies.

3. Spatial Bottleneck Detection:

   • Flowline comparisons between planned and actual progress flag delays when residuals exceed thresholds (e.g., 0.3 zone-units).

4. Decision-Support Dashboard:

   • A unified digital control interface visualizes EVM curves, lean performance (PPC), constraint alerts, and spatial flow, supporting fast, informed decisions.

4. Key Innovations

• Combining top-down (EVM) and bottom-up (Lean) systems via AI.

• Use of LLMs for extracting site constraints from unstructured notes.

• Deep learning models (LSTM-Transformers, CNNs) for accurate forecasting and anomaly detection.

• A real-time, multi-layered dashboard for dynamic project control.

5. Implications

This integrated AI-Lean–EVM approach enhances:

• Proactive risk detection

• Schedule and cost prediction

• Spatial flow monitoring

• Decision-making and plan adaptation

It offers a scalable, intelligent control architecture tailored to the complexities of railway megaprojects.

Conclusion

This paper has introduced a cohesive theoretical framework that unites Earned Value Management (EVM), Lean Construction practices (Last Planner System and Location?Based Management), and artificial intelligence into a single, real?time control paradigm for linear infrastructure projects. By conceptualizing performance indices (SPI, CPI, PPC) within a geo?referenced digital twin, the model leverages pull?planning commitments and spatial flowlines as leading indicators, while automated language models, time?series learners, and spatial anomaly detectors supply risk alerts, forecasts, and bottleneck warnings. Incorporating lean and spatial insights into the EVM baseline may improve forecasting accuracy, could detect delays faster, and align schedule and cost performance closer to more accurate levels. This integration elevates project control from retrospective variance analysis to a proactive, anticipatory method, allowing teams to mitigate local disruptions. Looking ahead, the framework offers a foundation for future work: empirically validating its benefits on live projects, enriching temporal and spatial models with LSTM/Transformer and advanced NLP techniques, and integrating real-time sensor feeds to close the virtual-physical feedback loop. By blending contractual rigor, flow?centric lean disciplines, and AI-driven insight, this integrated Lean-EVM-AI approach charts a path toward more predictable, responsive delivery of complex linear infrastructure.

References

[1] Agrawal, A. K., Zou, Y., Chen, L., Abdelmegid, M. A., & González, V. A. (2024). Moving toward lean construction through automation of planning and control in Last Planner System: A systematic literature review. Developments in the Built Environment, 18, 100419. https://doi.org/10.1016/j.dibe.2024.100419 [2] Al-Kasasbeh, M., Abudayyeh, O., & Liu, H. (2021). An integrated decision support system for building asset management based on BIM and Work Breakdown Structure. Journal of Building Engineering, 34, 101959. https://doi.org/10.1016/j.jobe.2020.101959 [3] Brioso, X., Humero, A., & Calampa, S. (2016). Comparing point-to-point precedence relations and Location-Based Management System in Last Planner System: A housing project of highly repetitive processes case study. Procedia Engineering, 164, 12–19. https://doi.org/10.1016/j.proeng.2016.11.586 [4] Chen, G., Alsharef, A., Ovid, A., Albert, A., & Jaselskis, E. (2025). Meet2Mitigate: An LLM-powered framework for real-time issue identification and mitigation from construction meeting discourse. Advanced Engineering Informatics, 64, 103068. https://doi.org/10.1016/j.aei.2024.103068 [5] Colin, J., Martens, A., Vanhoucke, M., & Wauters, M. (2015). A multivariate approach for top-down project control using earned value management. Decision Support Systems, 79, 65–76. https://doi.org/10.1016/j.dss.2015.08.002 [6] Emblemsvåg, J. (2024). Lean project planning – Bridging last planner system and earned value management. Heliyon, 10, e37810. https://doi.org/10.1016/j.heliyon.2024.e37810 [7] Gajo, P., & Barrón-Cedeño, A. (2025). Natural vs programming language in LLM knowledge graph construction. Information Processing and Management, 62, 104195. https://doi.org/10.1016/j.ipm.2025.104195 [8] Gao, S., & Low, S. P. (2014). The Last Planner System in China’s construction industry—A SWOT analysis on implementation. International Journal of Project Management, 32(7), 1260–1272. https://doi.org/10.1016/j.ijproman.2014.01.002 [9] Heigermoser, D., García de Soto, B., Abbott, E. L. S., & Chua, D. K. H. (2019). BIM-based Last Planner System tool for improving construction project management. Automation in Construction, 104, 246–254. https://doi.org/10.1016/j.autcon.2019.03.019 [10] Heo, S., & Na, S. (2025). Ready for departure: Factors to adopt large language model (LLM)-based artificial intelligence (AI) technology in the architecture, engineering and construction (AEC) industry. Results in Engineering, 25, 104325. https://doi.org/10.1016/j.rineng.2025.104325 [11] ?nan, T., Narbaev, T., & Hazir, Ö. (2022). A machine learning study to enhance project cost forecasting. IFAC PapersOnLine, 55(10), 3286–3291. https://doi.org/10.1016/j.ifacol.2022.10.127 [12] Khanh, H. D., & Kim, S. Y. (2016). A survey on production planning system in construction projects based on Last Planner System. KSCE Journal of Civil Engineering, 20(1), 1–11. https://doi.org/10.1007/s12205-015-1412-y [13] Lagos, C. I., Herrera, R. F., Mac Cawley, A. F., & Alarcón, L. F. (2024). Predicting construction schedule performance with Last Planner System and machine learning. Automation in Construction, 167, 105716. https://doi.org/10.1016/j.autcon.2024.105716 [14] Li, Q., Yang, Y., Yao, G., Wei, F., Li, R., Zhu, M., & Hou, H. (2024). Classification and application of deep learning in construction engineering and management – A systematic literature review and future innovations. Case Studies in Construction Materials, 21, e04051. https://doi.org/10.1016/j.cscm.2024.e04051 [15] Naderi, H., & Shojaei, A. (2025). Large-language model (LLM)-powered system for situated and game-based construction safety training. Expert Systems With Applications, 283, 127887. https://doi.org/10.1016/j.eswa.2025.127887 [16] Nesteby, Å. I., Aarrestad, M. E., Lohne, J., & Bohne, R. A. (2016). Integration of BREEAM-NOR in construction projects: Utilizing the Last Planner System. Energy Procedia, 96, 100–111. https://doi.org/10.1016/j.egypro.2016.09.110 [17] Robertson, J., Ferreira, C., Botha, E., & Oosthuizen, K. (2024). Game changers: A generative AI prompt protocol to enhance human-AI knowledge co-construction. Business Horizons, 67(4), 499–510. https://doi.org/10.1016/j.bushor.2024.04.008 [18] Sackey, S., Lee, D.-E., & Kim, B.-S. (2020). Duration Estimate at Completion: Improving Earned Value Management Forecasting Accuracy. KSCE Journal of Civil Engineering, 24(3), 693–702. https://doi.org/10.1007/s12205-020-0407-5 [19] Sequeira, S., & Lopes, E. (2015). Simple Method Proposal for Cost Estimation from Work Breakdown Structure. Procedia Computer Science, 64, 537–544. https://doi.org/10.1016/j.procs.2015.08.559 [20] Siami-Irdemoosa, E., Dindarloo, S. R., & Sharifzadeh, M. (2015). Work breakdown structure (WBS) development for underground construction. Automation in Construction, 58, 85–94. https://doi.org/10.1016/j.autcon.2015.07.016 [21] Sirimewan, D., Kunananthaseelan, N., Raman, S., Garcia, R., & Arashpour, M. (2024). Optimizing waste handling with interactive AI: Prompt-guided segmentation of construction and demolition waste using computer vision. Waste Management, 190, 149–160. https://doi.org/10.1016/j.wasman.2024.09.018 [22] Veganzones, D., & Séverin, E. (2025). Earnings management visualization and prediction using machine learning methods. International Journal of Accounting Information Systems, 56, 100743. https://doi.org/10.1016/j.accinf.2025.100743 [23] Zohoori, B., Verbraeck, A., Bagherpour, M., & Khakdaman, M. (2019). Monitoring production time and cost performance by combining earned value analysis and adaptive fuzzy control. Computers & Industrial Engineering, 127, 805–821. https://doi.org/10.1016/j.cie.2018.11.019 [24] Zheng, C., Wong, S., Su, X., Tang, Y., Nawaz, A., & Kassem, M. (2025). Automating construction contract review using knowledge graph-enhanced large language models. Automation in Construction, 175, 106179. https://doi.org/10.1016/j.autcon.2025.106179

Copyright

Copyright © 2025 Vishal Kumar Chaudhary, Sneha Kumari. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.