RESEARCH AND INNOVATION PROJECT
Hyperion Production Online Software with Artificial Intelligence
HYPPOS AI
A two year project to develop and commercialise an AI soft sensor module for polypropylene production that holds its accuracy through steady state and transient operation, by learning online as the process changes.
DURATION
24 months
TIMELINE
Jul 2026 to Jun 2028
TOTAL BUDGET
€ 900.300
REQUESTED FUNDING
€ 464.572
GENERAL PROFILE
About the project
HYPPOS AI extends the HYPPOS toolkit with machine learning models that keep predicting product quality accurately while the plant moves between grades and operating regimes.
Objective
Further development and commercialisation of the HYPPOS AI module for polypropylene production under steady state and transient operations, with online learning.
Challenge
There are no AI soft sensor options customised for polypropylene production today. Producers are left maintaining quality metrics through every industry transient without a cost effective way to do it, and static models lose accuracy exactly when the process moves.
Methods
The module uses a multi expert machine learning architecture. Specialised sub models, or experts, each cover part of the operating space and are combined so that reliability and the maximum possible accuracy are preserved in complex industrial environments.
Technical KPIs
Module accuracy is measured against the dynamic industrial environment using statistical parameters.
<5%
M.A.P.E.
Mean absolute percentage error
<5%
M.A.E.
Mean absolute error
PROJECT DATA
PROJECT TITLE
Hyperion Production Online Software with Artificial Intelligence
ACRONYM
HYPPOS AI
​
STARTING DATE
01 Jul 2026
END DATE
30 Jun 2028
DURATION
24 months
TOTAL BUDGET
€ 900.300
REQUESTED FUNDING
€ 464.572
CONSORTIUM
HOST ORGANISATION
Hyperion Systems Engineering
PARTNER
The Cyprus Institute
FUNDING
PROGRAMME
RESTART 2016-2020
CO-FUNDED BY
The European Union
THROUGH
Research and Innovation Foundation
ACTIVITIES
What we are doing
Five streams of work run across the two years, from building the industrial dataset to preparing the module for market.
Building the data foundation
Collecting and structuring industrial process and laboratory data from polypropylene production, and identifying the features that carry real predictive signal.
Developing the AI module
Designing, training and benchmarking the multi expert models with online learning, then hardening them into deployable software.
Testing with industry
Running the module against live plant conditions, measuring accuracy through transients and assessing readiness for commercial release.
Preparing the route to market
Working through the go to market strategy, business model, financial planning and investment path for the finished module.
Sharing the results
Publications, a conference, open access release of models and knowledge transfer support, alongside management of the resulting IP.
OUTCOMES
Work packages and deliverables
The work is organised in five work packages, listed below with their deliverables and the month each one is due.
WP1 - Project Management
DELIVERABLES
D1.1 Intermediate Report
M12
D1.2 Final Report
WP2 - Dissemination, Exploitation and Communication Activities
M24
DELIVERABLES
D2.1 Project websites, web media presence and logo
M2
D2.2 Reporting Document on IPR and Data Management Plan
M2
D2.3 Request for the Central Knowledge Transfer Office (KTO) of the RIF
M13
D2.4 Publications
M24
D2.5 One conference
M24
D2.6 AI/ML models in open access digital platforms
WP3 - Data Collection and Management
M24
DELIVERABLES
D3.1 Report on datasets and features
M24
D3.2 Structured Database
WP4 - R&D and Innovation Activities
M24
DELIVERABLES
D4.1 HYPPOS AI Software
M24
D4.2 MVP / Beta Testing Report and Commercialization Readiness Assessment
M24
WP5 - Business Planning and Development
DELIVERABLES
D5.1 Go to Market Strategy
M24
D5.2 Business Model Canvas
M24
D5.3 Investment Attraction Plan and Exit Strategy Plan
M24
D5.4 Financial Planning Report
M24
CASE STUDY
Where the module is put to work
The plant
HELLENiQ ENERGY's polypropylene production facility in Thessaloniki, Greece, part of the only vertically integrated petrochemical complex in the country. The plant runs multiple grades on the same train, and quality is defined by parameters such as Melt Flow Rate and Xylene Solubles, confirmed by laboratory analysis on a sampling interval measured in hours.
Starting point
HYPPOS soft sensors have been in place at the site for some time and moved into active use this year. The system predicts key quality parameters from data the plant already produces, integrating with the DCS, the real-time database, LIMS and enterprise systems without any new analyser hardware. Predictions refresh every few minutes, so operators keep visibility of quality between laboratory samples.
What this project adds
The deployed system is the baseline. HYPPOS AI takes it further with multi expert models and online learning, so that accuracy is maintained not only in steady state but through the transients where a static model degrades: grade changes, feedstock variation, catalyst ageing and equipment drift. The same site serves as the pilot for the new module.
How it will be validated
Module output is compared against laboratory reference values across steady state periods and grade transitions, tracking mean absolute percentage error and mean absolute error against the 5% targets. Results will be published here as validation progresses.
20%
Less off-spec material produced
25%
​​Shorter grade transition time
€350k
​​Estimated annual margin uplift
10.5 kt
​​CO₂ reduction per year
Figures for the existing HYPPOS deployment, as presented by HELLENiQ ENERGY at SuSChemE-SUSTENS2 in Athens. They describe the baseline installation, not results of the HYPPOS AI project.
