IPIS — Integrated Process Intelligence System

A hybrid digital-twin-backed framework integrating soft sensors, predictive maintenance, and real-time optimization for chemical process manufacturing.

---

What this is

IPIS is an open-source process intelligence framework that integrates three capability modules into a closed-loop decision system, unified by a fourth integration layer:

1. Soft Sensor — real-time prediction of hard-to-measure quality variables
2. Predictive Maintenance — anomaly detection and remaining useful life estimation
3. Real-Time Optimization (RTO) — constrained setpoint recommendations

All three modules sit on top of a digital twin layer (first-principles physics models) that provides baselines, constraints, and surrogate training data.

A fourth layer (Module 4) composes the three into a single closed-loop coverage certificate: a distribution-free, per-cycle guarantee on the joint safety event that the product stays in specification and the equipment survives to its next maintenance window. This is the integration that turns three calibrated components into one certified system.

Why it exists

Three documented gaps in industrial AI for process manufacturing:

• Cross-process generalization — published models don't transfer between plants with different topologies
• Grade-transition robustness — models degrade during operating regime shifts
• Proof-of-concept to production gap — 87% of industrial AI projects fail at deployment

IPIS addresses all three with one architecture, validated across heterogeneous benchmark datasets.

Architecture (high level)

        DIGITAL TWIN LAYER (DWSIM + GEKKO + CoolProp)
                            │
        ┌───────────────────┼───────────────────┐
        ↓                   ↓                   ↓
   Soft Sensor    Predictive Maintenance      RTO
   (Module 1)        (Module 2)         (Module 3)
        │                   │                   │
        └───────────────────┴───────────────────┘
                            ↓
                OPERATIONAL STATE BUS
                (MQTT + InfluxDB)
                            ↓
              OPERATOR DASHBOARD + API
              (Streamlit + FastAPI)

Project status

Module	Status	Notes
Module 1 — Soft Sensor	✅ Complete	Under review at Journal of Process Control (JPROCONT-D-26-00618; transfer from CACE)
Module 2 — Predictive Maintenance	✅ Complete	Feature-complete (2A–2D); SCC paper under review (JRESS-D-26-04700)
Module 3 — Real-Time Optimization	✅ Complete	Submitted, IEEE TCST (26-0876)
Module 4 — Integration (composed certificate)	✅ Under review	Closed-loop coverage certificate validated on the debutanizer twin; under review at Computers & Chemical Engineering (CACE-D-26-01079), reframed and ported after IECR declined on scope (paper4/)
Module 5 — Dynamic / horizon realization	🚧 In progress	Dynamic plant, closed-loop orchestrator, and adaptive-conformal horizon coverage validated (62 tests; three results frozen); Paper 5 drafting (paper5/)

Publications

Listed by module. Submission directories paperN/ are numbered by authoring order, which differs from the module numbering for Modules 2 and 3; the module mapping below is authoritative.

• Module 1 — Soft Sensor: When does a calibrated soft sensor keep its promise? A negative-control study of validity without accuracy under drift and delayed labels. Under review, Journal of Process Control (JPROCONT-D-26-00618) (CACE-D-26-00944). Source: paper/.
• Module 2 — Predictive Maintenance (Similarity-Calibrated Conformal): Similarity-Calibrated Conformal prediction: data-free coverage guarantees for remaining-useful-life intervals under operating-regime transfer. Under review, Reliability Engineering & System Safety (JRESS-D-26-04700). Source: paper3/.
• Module 3 — Real-Time Optimization: Safe real-time optimization under unmeasured disturbances: a finite-sample, distribution-free constraint-satisfaction guarantee. Submitted, IEEE Trans. Control Systems Technology (26-0876). Source: paper2/tcst/.
• Module 4 — Integration (composed coverage certificate): A composed coverage certificate for closed-loop process operation: certified joint product-quality and equipment-survival guarantees under feedback. Under review, Computers & Chemical Engineering (CACE-D-26-01079; reframed and ported after the IECR submission was declined on scope). Source: paper4/.
• Module 5 — Dynamic / horizon realization (this paper): Horizon-wide safety guarantees for closed-loop process operation via adaptive conformal calibration. In preparation, Computers & Chemical Engineering (paper5/).

Project lifecycle and roadmap

The through-line. IPIS is building toward a single idea: certified process intelligence — a plant in which every data-driven decision carries a distribution-free safety guarantee. Each module is not only a method but a capability the field did not previously have, and the modules are designed to compose. Read in sequence they move from trustworthy components, to a certified closed loop, to a certified plant.

What each module opens.

Module	The contribution	The door it opens
M1 — Soft Sensor	Calibrated uncertainty that survives delayed labels, drift, and regime shift, with a hostile negative control isolating where the performance comes from	Inferential sensing you can trust in production, not only on a benchmark
M2 — Predictive Maintenance (SCC)	An a-priori, physics-derived coverage certificate for remaining-life bounds under operating-regime transfer	Prognostics whose guarantee holds when the regime changes
M3 — Real-Time Optimization	Naming and fixing the conformal selection effect: why marginally valid back-offs over-violate, and how conditional calibration restores safety	Economic optimization that stays provably safe under uncertainty
M4 — Integration (composed certificate)	The first derived, enforceable coverage certificate for the joint closed-loop safety event (in-spec product and surviving equipment), with the feedback objection resolved by causal timing	Certified closed-loop autonomy: a plant that optimizes itself against a provable safety floor
M5 — Dynamic / horizon (next)	The certificate as a runtime monitor over a real-time horizon, via adaptive conformal inference	Certified autonomy in real time, off the quasi-static twin

The lifecycle, in phases.

• Phase A — the three pillars (DONE). Calibrated sensing (M1), transferable prognostics (M2), safe optimization (M3). Three standalone papers, each a documented gap closed.
• Phase B — composition (DONE). The closed-loop coverage certificate that turns the three guarantees into one joint guarantee (M4). The headline result of the project so far: on a calibrated debutanizer twin, a certified safety floor of 0.75 is met at 0.988 coverage under the budget and collapses to 0.000 without it.
• Phase C — realization (NEXT, Module 5). Move to a dynamic, physically realistic loop and lift the per-cycle certificate to a horizon guarantee. The bridge to plantwide.
• Phase D — scale (mid-term). Plantwide / multi-unit operation with one coverage budget allocated across units, a certificate for an integrated plant rather than a single column. The highest-leverage step. In parallel, tighten the composition with a dependence-aware joint-failure model.
• Phase E — generalize, harden, validate (long-term). New dimensionless groups per unit type (turning the universality claim into a theorem); active fault management (the certificate as a fault-tolerant-control monitor); and validation on an operating LPG / petrochemical column with real DCS data, the credibility capstone.

The endgame. A unified theory and an open-source reference implementation of certified process intelligence for manufacturing: a framework in which soft sensing, prognostics, and optimization are not three tools bolted together but one system carrying a single, enforceable, distribution-free safety guarantee from the sensor, to the setpoint, to the plant. The intended capstone is a book that documents the whole arc as a coherent sub-discipline of process systems engineering, each module a chapter and a door opened toward autonomous, certified manufacturing.

Quick start

# Clone
git clone https://github.com/beebzy-droid/IPIS.git
cd IPIS

# Create environment (Python 3.11+)
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate

# Install in editable mode with dev dependencies
pip install -e ".[dev]"

# Verify installation
pytest tests/unit -v

# Download datasets
python scripts/download_datasets.py --all

Documentation

• Project structure map
• Continuity / handoff doc
• System architecture
• Architecture Decision Records (ADRs)
• Module 1 specification — Soft Sensor
• Module 2 specification — Predictive Maintenance
• Module 3 specification — Real-Time Optimization
• Module 4 theory note — Composed coverage certificate (integration)

Tech stack

• ML: PyTorch, scikit-learn, XGBoost, River (online learning), MAPIE (conformal)
• Physics: DWSIM, CoolProp, GEKKO
• Infrastructure: OPC-UA (asyncua), MQTT (Mosquitto), InfluxDB, FastAPI, Streamlit
• MLOps: MLflow, Hydra, Docker, GitHub Actions, pytest

Citation

If you use this work in research, please cite:

@software{busico_ipis_2026,
  author = {Busico, Bien},
  title = {IPIS: Integrated Process Intelligence System},
  year = {2026},
  url = {https://github.com/beebzy-droid/IPIS}
}

License

MIT — see LICENSE.

Author

Bien Busico — Process Engineer | Chemical Engineering × AI/ML × Industry 4.0