Observe the process
Telemetry travels through an eight-phase modeled production sequence.
Eco-Twin Oracle
INDUSTRIAL ML / DIGITAL TWINModeled process prototype
An interesting prediction. A feasible decision.
THE CONTEXT
Team SpicyJalebi’s industrial process intelligence prototype.
THE KEY DECISION
Check recommendations against process phases and modeled operating limits.
THE RESULT
A visible path from replayed observations to constrained recommendations.
Eco-TwinPROCESS → RESPONSE
MECHANISM STUDY / 04SCHEMATIC
TAKE A CLOSER LOOKSELECT A STAGE ↓
SCROLL TO TRACE. SELECT TO EXPLORE.
Follow the connections as you scroll, or select a stage to see what happens there.
Map the deviation
A SOM topology and LVQ refinement locate unusual observations.
Constrain the response
Phase legality and operating rules bound the available recommendations.
A pharmaceutical-process digital-twin prototype built with Team SpicyJalebi for the AVEVA hackathon context. It joins learned patterns with modeled physical and chronological constraints.
PROJECT TYPE
Team project · process intelligence
FOCUS
AI / ML + Research + Full stack + Data / analytics
STATUS
Modeled process prototype
/01
The question.
An anomalous trace is only part of the problem. A proposed adjustment must also make sense for the process phase and the modeled operating limits.
/02
The approach.
A Kohonen self-organizing map and LVQ refinement characterize process observations. An eight-phase state machine checks transition and parameter legality. Analytics surface energy/process deviations; the FastAPI/WebSocket interface streams replayed observations into a React view.
/03
What came out of it.
The prototype makes the path from observation to constrained recommendation visible. Its traces and operating rules describe the modeled dataset and process; they do not establish factory deployment or measured real-world savings.
01
A topology for comparison
The SOM offers a map of process behavior against a reference signature, while LVQ supports anomaly categorization.
02
Chronology matters
The DFA encodes eight process phases and constrains adjustments to the modeled sequence. A rejected path stops visibly at that boundary.
03
A replayable process
Telemetry replay and WebSocket updates make the decision sequence inspectable without implying a connection to an operating plant.
Under the surface.
- Modeling
- Python · NumPy · Kohonen SOM · LVQ
- Constraints & delivery
- DFA · process analytics · FastAPI · WebSockets
- Interface
- React · TypeScript · Vite · process visualization
SCOPE & LIMITS
- A modeled digital-twin prototype, not a validated industrial control system.
- No measured factory energy savings or regulatory compliance is claimed.
- Team project; the case study describes the shared system, not sole authorship.
NEXT PROJECT / 05
Have an idea of your own? Let’s talk