TY  - CONF
TI  - An AI Copilot for Make-Ready Planning in the Last Planner System
C1  - Auckland, New Zealand
C3  - Proceedings of the 32nd Annual Conference of the International Group for Lean Construction (IGLC 32)
SP  - 365
EP  - 376
PY  - 2024
DO  - 10.24928/2024/0155
AU  - Lagos, Camilo I.
AU  - Herrera, Rodrigo F.
AU  - Cawley, Alejandro Mac
AU  - Alarcón, Luis F.
AD  - PhD Candidate, School of Engineering, Pontificia Universidad Católica de Chile, CONXAI Technologies GmbH, colagos@uc.cl, orcid.org/0000-0002-0648-0039
AD  - Assistant Professor, School of Civil Engineering, Pontificia Universidad Católica de Valparaíso, rodrigo.herrera@pucv.cl orcid.org/0000-0001-5186-3154
AD  - Associate Professor, Dept. of Industrial Engineering, Pontifica Universidad Católica de Chile, amac@ing.puc.cl, orcid.org/0000-0002-4848-4732
AD  - Professor, Dept. of Construction Engineering and Management, Pontifica Universidad Católica de Chile, lalarcon@ing.puc.cl, orcid.org/0000-0002-9277-2272
AB  - Many challenges in partial Last Planner System implementations can be attributed to the underutilization of Make-Ready Planning, although other factors also play a role. Failing to identify constraints in time to prevent Reasons for Noncompliance (RNCs) decreases short and long-term performance. Reducing the complexity of identifying, registering, and managing constraints systematically was found as a critical improvement opportunity. This research proposes the use of an artificial intelligence (AI) recommender system to facilitate constraint identification and RNC prevention. The system employs Large Language Model (LLM) embeddings to represent new task descriptions and find the most similar previously seen tasks. Subsequently, it fetches the set of constraints and RNCs belonging to these past tasks, represented in the embedded system, and uses it to produce three prioritized recommendations. Finally, the selected recommendations are categorized using Machine Learning Classification. The model was able to provide three sound recommendations for 69% of tasks and yielded a 60% relative improvement compared to a rule-based frequent pattern probabilistic system. The results pose three benefits for LPS practitioners: Reducing the effort needed to identify and register constraints, alerting probable RNCs needing to be prevented, and enriching data registration, allowing it to be used in future knowledge management.
PB  - 
T2  - Proceedings of the 32nd Annual Conference of the International Group for Lean Construction (IGLC 32)
DA  - 2024/07/01
CY  - Auckland, New Zealand
L1  - http://iglc.net/Papers/Details/2290/pdf
L2  - http://iglc.net/Papers/Details/2290
N1  - Export Date: 25 April 2025
DB  - IGLC.net
DP  - IGLC
LA  - English
ER  -