Predictive & Prescriptive Maintenance

The evolution of industrial systems requires moving beyond merely predicting failures to actively orchestrating interventions. Our research bridges the gap between predictive analytics (estimating when a failure might occur) and prescriptive analytics (deciding exactly what to do about it under real-world constraints).

In a highly complex operational environment, knowing that a machine will fail is insufficient if the maintenance plan cannot accommodate the intervention. Our lab develops end-to-end decision support tools that connect these two layers.

Predictive Analytics: Anticipating Degradation and Anomalies

We design and deploy advanced machine learning models to monitor operational environments, process complex signals, and detect faults before they disrupt critical systems.

• System Log Anomaly Detection: We develop tools for continuous anomaly detection in massive system logs to support operational monitoring at the Paranal Observatory.
• Degradation and Remaining Useful Life: We implement frameworks to identify slow degradation faults and estimate remaining useful life within harsh environments, allowing for early intervention at the ALMA Radiotelescope.

Prescriptive Analytics: From Data to Decisions

To translate predictions into actionable strategies, we build combinatorial and stochastic optimization models that generate robust maintenance schedules.

• Optimization Under Uncertainty: We incorporate post-prognostic information and uncertainty metrics directly into our models to build true prescriptive maintenance systems that dictate the optimal time for intervention.
• Large-Scale Maintenance Planning: We tackle resource-aware scheduling limitations to optimize long-term maintenance planning for fleets. Our models explicitly address strict safety requirements, regulatory constraints, and fleet availability targets, as demonstrated in our collaborative work with LATAM Airlines.

Funding and Support

This continuous line of research is supported by the following grants:

• FIE-2016-V022
• 16IFI6626 CORFO
• FONDECYT 1231245
• FONDEF IT2410031