A research agenda.
My research lives at the intersection of optimisation, diagnostics, and decision-making for complex engineering assets. The PhD established the methods; the ongoing work extends them into domains and ventures where the cost of a wrong signal is highest.
The central question
How do we make better decisions when signals are incomplete, noisy, and expensive — and how do we know which signals are worth paying for in the first place?
Three threads, one method.
All of my research uses multi-objective optimisation as its language. The threads differ in what they optimise over — sensors, decisions, or whole asset architectures.
Sensor optimisation
What to measure on a complex asset, where to measure it, and how to score what each sensor contributes — formalised in MOSOF and the Normalised Diagnostic Contribution Index (NDCI).
Diagnostic decisions
Once the signals are in, how do we turn them into actions a stakeholder can defend? Frameworks that surface uncertainty and trade-offs honestly rather than hiding them in a black-box label.
Stakeholder-aware design
An OEM, an operator, and an MRO weight the same problem differently. Methods that make those weightings explicit and let the right answer for each be different — without re-running the analysis from scratch.
Multi-objective by default.
Real engineering decisions never have a single objective. Performance fights weight, reliability fights cost, accuracy fights latency. Forcing a single scalar out of that pretends a trade-off doesn't exist; surfacing the Pareto front shows the trade-off honestly and lets a human pick.
The methods are search-based — NSGA-II-family genetic algorithms over carefully encoded candidate spaces — and the validation is empirical, not just analytical: every framework gets tested on a real reference system before it ships.
- 01Encode — what are we choosing between, and what objectives does the choice trade off?
- 02Search — multi-objective evolutionary algorithms over the candidate space.
- 03Score — diagnostic contribution, not just dominance, when the candidates are sensors.
- 04Surface — Pareto front plus a knee-point recommendation, with explicit trade-offs for stakeholders.
From paper to product.
Research that doesn't get used isn't research that matters. The work flows in two directions — back into peer-reviewed publications, and forward into ventures and advisory engagements where the methods meet a real customer.
Working on a related question?
If your team is grappling with sensor selection, diagnostic-system design, or decisions under expensive uncertainty — in academia or industry — I'd be glad to discuss whether there's overlap.
hello@buraksuslu.com →