Portfolio
Applied research prototypes and engineering work. Items are experimental or under active development unless a publication or demo link is listed.
Dempster–Shafer Theory for Industrial Diagnostics
Role: Research prototype / published research support
Dempster–Shafer Theory · Belief Functions · Uncertainty Modeling · Industrial Diagnostics · Continuous Casting (CCM)
- Context
- Industrial diagnostics under uncertainty, incomplete observations, and conflicting evidence — focused on continuous casting machines in metallurgy.
- Problem
- Classical probabilistic models may be brittle when evidence is sparse, imprecise, or conflicting across temperature, vibration, and acoustic sensors.
- Approach
- Use Dempster–Shafer belief functions to represent uncertainty and combine evidence from diagnostic sources; adaptive combination rules (Dempster / Yager) depending on conflict coefficient.
- Evidence
-
- Outcome
- Experimental validation on CCM diagnostic scenarios reported in the published work: improved diagnostic accuracy and reduced false confidence under high-conflict evidence (see project page for details).
- Status
- Published research support with working diagnostic prototype demo.
Industrial MRO Optimization with Hierarchical MARL
Role: Simulation-based research prototype
Reinforcement Learning · MARL · HRL · CMDP · HITL · Industrial Maintenance · Simulation
- Context
- Industrial maintenance planning under limited maintenance resources and equipment degradation.
- Problem
- Reactive or fixed-schedule maintenance policies may lead to avoidable downtime or inefficient resource use under stochastic failures and partial observability.
- Approach
- Simulation-based decision-support prototype using hierarchical RL / MARL, HITL operator feedback, constrained decision-making (CMDP / action masking), and latency-aware intervention logic at the strategic level.
- Evidence
-
- Outcome
- Simulation experiments compare learned maintenance policies against baseline strategies such as reactive or schedule-based maintenance. Reported trends include a shift toward more preventive planning under resource limits (see working paper draft; no production deployment).
- Status
- Active research prototype — under development.
RL Simulation Environments
Role: Experimental environment
Python · Gymnasium · PyGame · Simulation · RL Environments · Experiment Logging
- Context
- Reproducible testbeds for hierarchical and multi-agent RL beyond toy benchmarks.
- Problem
- Maintenance and resource-allocation experiments need configurable simulation environments with logging hooks.
- Approach
- Custom Gymnasium-style environments modeling equipment fleet dynamics, maintenance actions, and resource constraints.
- Evidence
-
- Repository: not publicly linked yet
- Status
- Experimental environment — iterative development.
Data / ML Engineering Sandbox
Role: Engineering sandbox (under development)
Python · SQL · FastAPI · Docker · Experiment Tracking
- Context
- Small research-oriented prototypes: data-processing scripts, API experiments, and reproducible experiment pipelines.
- Problem
- Bridging research experiments with lightweight data workflows and serving stubs.
- Approach
- Python / SQL-based data and ML prototypes; containerized API experiments; tracked experiment runs.
- Evidence
-
- Repository: not publicly linked yet
- Status
- Under development — no public artifact yet.