Ship-state forecasting and a hardware cost removed
Led sensor-fusion forecasting that achieved parametric-roll detection and replaced RTK hardware with an algorithm, saving roughly $10K per unit sold.
- hardware cost removed from every unit sold
- ~$10K
- university STEM program built on the platform
- 3-year
- Organization
- ORBIS, Inc. · Trim Management platform
- My role
- Director of Artificial Intelligence; technical lead
- Period · Focus
- 2021 – 2026 · Platforms, Strategy
In brief
Trim Management predicts a vessel's state at sea from fused shipboard sensors. I led the time-series forecasting and signal-processing work, which achieved detection of parametric roll, a dangerous resonance condition, and I initiated the full-stack digital-twin interface on top of it. Experiments I directed removed the platform's need for RTK hardware, saving roughly $10K on every unit sold, and a university partnership I set up adapted the platform into a three-year STEM program worth $80K in initial revenue.
Context and challenge
The platform needed reliable ship-state prediction under shipboard conditions, including warning of parametric roll, while its per-unit hardware cost included an RTK positioning dependency.
My mandate and scope
- Led sensor-fusion and time-series forecasting for ship-state prediction.
- Initiated and led full-stack development of the digital-twin interface.
- Directed the experiments that eliminated the RTK hardware requirement.
- Established the SIU partnership and wrote the plan and statement of work for a three-year STEM engineering program built on an adapted version of the platform.
Key decisions and why
- 1
Replace a hardware dependency with an algorithm
The RTK requirement added cost to every unit. Directing experiments to remove it turned a model improvement into a recurring per-unit saving of roughly $10K.
- 2
Run inference at the edge
Shipboard connectivity is intermittent and parametric-roll warning is only useful if it arrives in time, so inference ran on the vessel.
- 3
Open a second channel for the platform
Adapting the platform for a university STEM program created new revenue from an existing asset.
What changed
- Ship-state prediction and parametric-roll detection became platform capabilities, presented through a digital-twin interface.
- Unit economics improved by removing a hardware component.
Outcome and evidence
- Parametric-roll detection achieved from fused shipboard sensor data.
- RTK hardware requirement eliminated: approximately $10K saved per unit sold.
- Three-year SIU STEM program on an adapted platform: $80K in initial revenue.
What to expect from me
I look for the place where a model changes the economics of the product as well as its accuracy, and I will push a platform into an adjacent channel when the asset supports it.
Technical perspective
- Sensor fusion and time-series forecasting over ship-state data; signal processing for resonance detection.
- Edge deployment for shipboard inference under intermittent connectivity, where latency is operationally significant.
- Models trained on ship-state sensor data; full-stack digital-twin interface.
Artifacts
Confidentiality: Customer, vessel, and performance data are not published.