Scientific AI for industrial decisions

Your plant data, turned into decisions. Made to measure.

artXdata builds scientific AI for industrial systems — combining process knowledge, advanced modelling and explainable decision intelligence.

Built around your process. Trained on your data. Designed to be understood.

Explore use cases Discuss a project
data · ΔP bar · NTU · µS/cmart · custom model

Companies, institutes and academies we have worked with

Water & industry
Veolia-VWT (DB&T)OTV-VeoliaSIDEM-VeoliaInSight-Veolia
Data & AI companies
Aquila Data EnablerArondor
AI risk & governance
VeoliaIPSENFERMA
Institutions & academia
ILOIATURPolytechnique Montréal

Why artXdata

Why art × data

Data

The evidence.

Signals, measurements, documents and operational history.

×

The intelligence.

Science, process knowledge and modelling transform observations into understanding.

Art

The craft.

Understanding becomes a solution designed for the people who will use it.

Our proposition

We don't make black boxes.
We make industrial intelligence you can see through.

01 · Evidence

Not a concept. Already operating.

Scale of our Water-Tech AI deployments

  • 23plants
  • 88years of operational data
  • 145trains
  • 90M+data points
  • 9Water-Tech AI innovations
  • 19AI use cases delivered
  • 15+years of quantitative modelling & AI

Selected engagements

Four engagements show the range, from real-time plant optimization to enterprise AI governance.

Water · Desalination

AI suite for water treatment plants

Veolia Water Technologies · 2021 – present

Challenge

Desalination and treatment plants run under tight quality limits, with chemical, energy and membrane costs that react to changing raw water.

Approach

  • Custom AI modules for membranes, clarification, dosing and energy (see Innovations)
  • Test cases validated with plant operations

Result

  • ML pipelines run in UAT and production on real-time plant data
  • Optimization algorithms developed with the Veolia Water Technologies digital team

Governance · Water utility

AI product risk management

Veolia · AI Act readiness

Challenge

A growing portfolio of AI products needed consistent risk classification before the EU AI Act applies.

Approach

  • Governance framework aligned with EU AI Act and ISO/IEC 42001
  • Impact assessments per ISO/IEC 42005

Result

  • Risk identification, classification and mitigation across the AI lifecycle
  • A dedicated AI risk management team set up and guided

Enterprise risk · SaaS

Automated enterprise risk management

Arondor · FERMA expert team · 2019 – 2021

Challenge

Risk registers were built and valued by hand, from long audit reports, and updated too rarely.

Approach

  • APIs scoring supplier, shortage, antitrust, fraud and cyber risk
  • Operational risk valuation from audit reports

Result

  • Risk products delivered as SaaS and APIs
  • Methodology contributed to the FERMA scientific expert team

Pharma · Automotive

Supply chain risk & fraud analysis

IPSEN · FORVIA Faurecia

Challenge

AI used in supply chain and fraud controls had to be assessed and governed like any other critical system.

Approach

  • AI risk assessment methodology and governance strategy
  • Compliance support under ISO 27001 and 27005

Result

  • Automated supply chain risk assessment
  • Governed fraud analysis with documented AI risk controls

02 · Innovations

Nine water-tech AI innovations, invented by artXdata

Three families of AI modules cover the core units of a treatment or desalination plant: reverse-osmosis membranes, clarification and energy. Each module turns plant data into an early warning, a forecast or a recommendation that operators can act on.

23

Smart Membrane and Smart Clarification run on 23 water treatment and desalination plants worldwide.Proven in daily operation across different feed waters, climates and plant designs.

Smart Membrane

Reverse osmosis & nanofiltration · from normalization to replacement

01Inventor

Smart Membrane AI Normalization

An AI-based normalization framework that removes the effect of changing operating conditions from membrane KPIs, so performance degradation is assessed consistently over time.

Output · normalized differential pressure, specific flux, feed conductivity, feed pressure, permeate flow

02Inventor

Smart Membrane AI-Based Biofouling, Scaling & Colloidal Fouling Detection and Forecasting

Detects, distinguishes and forecasts biofouling, scaling and colloidal fouling early enough to support proactive membrane operation and maintenance.

Output · real-time fouling indicator forecasts

03Inventor

Smart Membrane Fouling Diagnosis, Early Warning & Performance Intelligence

Combines membrane condition indicators, trends and process relationships to diagnose degradation mechanisms and deliver early warning and actionable performance intelligence.

Output · fouling type, alerts, performance insight

04Inventor

Smart Membrane CIP Recommendation & Forecasting AI

Forecasts membrane cleaning needs and recommends the timing of clean-in-place before fouling causes significant performance or production losses.

Output · CIP date, CIP components

05Inventor

Smart Membrane MR Recommendation & Remaining-Life Forecasting AI

Estimates remaining membrane operating life and supports membrane replacement recommendations from degradation, performance and maintenance-history signals.

Output · membrane replacement forecast, membrane ageing

06Inventor

Smart Membrane Multi-Train Fouling & CIP Intelligence

Integrates fouling and cleaning information across multiple RO trains to reveal train-level deterioration, plant-level risk and the priority for intervention.

Output · train ranking, plant-level risk

Smart Clarification

Coagulation, flocculation & settling

07Inventor

Smart Clarification NTU Forecasting AI

Forecasts outlet turbidity so operators can anticipate clarification performance and adjust operating conditions before water-quality limits are approached.

Output · HRT, 30-min NTU or UV forecast

08Inventor

Smart Clarification OPEX Simulation & Optimization AI

Simulates operating scenarios and optimizes clarification OPEX while maintaining the required treatment performance and process constraints.

Output · optimal scenario, OPEX saving

Energy

Pumps & high-pressure trains

09Inventor

Energy Module AI

Forecasts and optimizes energy-related process performance to reduce pump energy consumption and cost while preserving treatment objectives and operational constraints.

Output · pump set-points, kWh/m³ saving

NormalizeClean signalKPIs freed from temperature, pressure and feed effects (01)
DiagnoseMechanism & warningBiofouling, scaling or colloidal fouling, flagged early (02 · 03)
RecommendCIP & replacementWhen to clean, when to replace (04 · 05)
PrioritizeAcross trainsWhere to intervene first, plant-wide (06)

03 · Industries

Built around the problems of each industry

Start from your sector and its constraints. Each industry maps to the problems it faces, the AI that answers them and the use cases already delivered.

Core expertise WATER · WASTEWATER · DESALINATION

Water & Wastewater

Challenges

  • Chemical over-dosing and cost volatility
  • Membrane fouling, scaling and unplanned cleaning
  • Raw water salinity, UV and turbidity swings
  • Clarifier outlet turbidity close to limits

What we deploy

  • Smart Membrane suite: normalization, fouling, CIP, remaining life
  • Smart Clarification: NTU forecasting, OPEX optimization
  • Reinforcement learning for dosing
ENERGY · UTILITIES

Energy & Utilities

Challenges

  • Energy is the largest variable cost of treatment
  • Pumps and high-pressure trains run off-optimum
  • Decarbonisation targets need measurable savings

What we deploy

  • Plant energy optimization
  • Real-time forecasting of demand and load
  • Anomaly detection on assets

Our methods also travel.

The same quantitative foundations have been applied to supply chains, financial services, enterprise risk, documents, recommendation systems and AI governance.

  • Supply chain & qualitySupply chain risk automation, fraud analysis and quality-loss recommendation engines for chemicals, pharma and manufacturing.
  • Financial servicesFraud detection in SWIFT, SEPA and card flows; credit and counterparty risk.
  • Enterprise riskAutomated risk registers and scoring of supplier, fraud and cyber risk.
  • DocumentsLLM and NLP document intelligence across the whole document lifecycle.
  • Recommendation systemsHedonic and ML recommenders for pricing, product ranges and maintenance.
  • AI governanceGovernance frameworks (ISO/IEC 42001) and impact assessments (ISO/IEC 42005) aligned with the EU AI Act.

04 · Selected use cases

Use cases delivered in production

Every card below comes from a real project. Filter by industry to see the problem, the method and the data it runs on.

Water · Desalination

Chemical consumption optimization

Recommends coagulant and antiscalant doses in real time to cut chemical use while keeping treated water on specification.

Method
Reinforcement learning, statistical learning, micro-econometric custom models
Data
Real-time time series
Water · Raw water quality

Salinity, UV & turbidity forecasting

Forecasts salinity, UV transmittance and turbidity (NTU) so operators can anticipate changes in feed water.

Method
ML, neural network and statistical models
Data
Time series & panel data, real time
Water · Membranes

Membrane performance normalization

Computes normalized permeate flow, salt passage and ΔP (ASTM D4516 practice), so membrane decline is visible independently of temperature, pressure and feed salinity.

Method
ML models, behavioural regression models
Data
Real-time time series
Water · Maintenance

Cleaning detection & recommendation

Detects past clean-in-place (CIP) events in the historian and recommends the next CIP date from the normalized performance trend.

Method
Recommendation engines, ML models
Data
Cross section & time series, real time
Water · Membranes

Biofouling, scaling & colloidal fouling

Separates the three fouling mechanisms on RO trains using normalized ΔP, salt passage, SDI and scaling indices, and forecasts each one before it costs production.

Method
ML models, custom mathematical models
Data
Real-time time series
Energy · Treatment plants

AI for plant energy optimization

Models the energy consumption of treatment plants and finds operating points that use less energy for the same output.

Method
ML, neural network and statistical models
Data
Real-time time series
Energy · Assets

Anomaly detection

Flags abnormal behaviour in operational and business data so teams investigate the few cases that matter.

Method
ML and neural network models
Data
Cross section
Pharma · Supply chain

Supply chain risk automation

Automates supply chain risk assessment with an AI risk methodology and responsible-AI compliance support.

Method
AI risk assessment, governance strategy
Frame
ISO 27001 / 27005
Automotive · Controls

Fraud analysis

AI-based fraud analysis with a risk management framework and governance processes around the models.

Method
AI system assessment, risk frameworks
Frame
ISO 27001 / 27005
Industry · Quality

Quality-loss recommendation engine

Links quality losses to their likely causes and recommends corrective actions.

Method
Custom hedonic recommendation system
Data
Cross section, real time
Enterprise risk · API

Supplier & operational risk API

Scores supplier failure, stock shortage, antitrust, corruption and fraud, and cyber-security risks through one API.

Method
ML, NLP, risk valuation models
Data
Cross section & text
Enterprise risk · ERM

Automated risk register & valuation

Builds the risk register, values operational risks from audit reports and recommends risk scenarios.

Method
NLP, ERM optimization, scenario recommendation
Data
Audit reports, text
Water utility · AI products

AI product risk management

Identifies, classifies and mitigates the risks of AI products across their lifecycle, with AI Act evidence.

Method
AI governance framework, impact assessment
Frame
EU AI Act, ISO 42001, 42005, 27001
Banking · Payments

Fraud detection in SWIFT, SEPA & cards

Multidisciplinary econometric and ML specifications to detect fraud in international transfers (ISO 20022) and card payments.

Method
NLP, neural networks, micro-econometrics, ML
Data
Text, cross section, time series
Banking · Compliance

Abnormal behaviour in financial e-mail

Surfaces communication patterns that may signal antitrust or conduct issues.

Method
NLP, ML models
Data
Text & cross section, real time
Banking · Credit

Credit risk & price prediction

Predicts credit risk and prices, and analyses customer behavioural tendencies.

Method
ML, neural networks, econometrics
Data
Cross section
Services · Customer

Client sentiment satisfaction service

Measures customer satisfaction in real time from free text with an LLM and NLP service (SENT-AI).

Method
LLM, NLP
Data
Text, real time
Documents · Lifecycle

Document indexing & compliance

Indexes, classifies and scores documents and contracts, including table extraction and compliance scores.

Method
NLP entity recognition, classification
Data
Text
Retail · Luxury

Product recommendation & pricing

Recommends products and optimizes prices for luxury and tourism offers.

Method
Recommender systems, micro-econometrics
Data
Cross section

05 · How we work

From a first question to governed AI in production

Each service is a clear entry point with a defined scope and deliverables. Start where you are: with a question, a pilot, a model to deploy or an AI portfolio to govern.

What we build

Predict · Diagnose · Optimize · Recommend · Govern

Forecasting, predictive maintenance, optimization, recommendation systems, GenAI and governed AI.

Predict

Forecasts of water quality, fouling, scaling, demand and equipment behaviour, early enough to act.

Diagnose

Early warning and root-cause diagnosis: which mechanism is degrading performance, and why.

Optimize

Set-points, chemical doses, cleaning schedules and energy use, optimized under real operating constraints.

Recommend

Recommendation engines and GenAI assistants grounded in your documents, procedures and plant history.

Govern

Risk classification, impact assessment and monitoring aligned with the EU AI Act and ISO/IEC 42001.

How we work

Five steps, each with a defined scope

Start with the step that matches where you are.

01Diagnose

AI Opportu­nity Diagnostic

Find where AI pays off before investing in it.

ForPlant and business leaders with data but no clear AI priorities
You receive
  • Data audit and maturity assessment
  • Ranked use-case portfolio with value cases
  • Roadmap agreed with plant and business owners
Typical duration2–4 weeks
02Prove

Proof of Value

One use case, modelled on your data and tested against today's baseline.

ForTeams with a priority use case to validate
You receive
  • A working model on your historical data
  • Validation report run with your operators
  • Measured gain and a clear go/no-go decision
Typical duration8–12 weeks
03Deploy

Deployment & Industrial­ization

From a validated model to a tool operators use every shift.

ForSites ready to put a validated model into operation
You receive
  • Integration with your plant data sources
  • MLOps pipeline and model monitoring
  • Documentation, UAT and handover to your teams
Typical durationScoped per site
04Govern

AI Governance & EU AI Act Readiness

Know which AI you run, its risk level and the evidence auditors will ask for.

ForOrganizations that use or buy AI, industrial or not
You receive
  • AI inventory and risk classification
  • Impact assessments aligned with ISO/IEC 42005
  • ISO/IEC 42001 readiness and audit evidence
Typical durationScoped per portfolio
05Improve

Run & Improve

Keep models accurate as the plant, the water and the seasons change.

ForClients with models in production
You receive
  • Drift monitoring and retraining
  • Periodic performance reviews
  • On-call expert support
Typical durationOngoing

Specialist offers

Available on their own or alongside any step above.

Ready-made modules

Smart Water AI Modules

AI modules for membranes, clarification and energy, configured to your plant's data instead of built from scratch.

Smart MembraneSmart ClarificationEnergy Module
Teams & methods

Training & Scientific Advisory

AI literacy workshops for operators and managers, and expert review of the modelling choices your own teams make.

AI literacy · EU AI ActMethodology reviews
Every engagement

Fixed scope

Scope, deliverables and timeline are agreed before work starts.

Every engagement

Your data stays yours

Your data remains your property and is used only for your project.

Every engagement

White-box delivery

Every model comes with documentation your team can read, question and maintain.

Not sure where to start? Most clients begin with the AI Opportunity Diagnostic.

Discuss a projectSee use cases

06 · Technology & science

The methods behind the models

We apply scientific methods to industrial problems. Each method is chosen for the physics of the process, the data available and the decision it must support.

White-box by design. A painting rewards attention: step closer and the brushstrokes become visible. We design our models with the same principle. Forecasts and recommendations are accompanied by the variables, operating conditions, assumptions and evidence that shaped them — so engineers can inspect, verify and challenge the result.

The data is the pigment. Understanding is the art.

Methods and where they apply

Reinforcement learning
Sequential control of chemical dosing, cleaning intervals and energy set-points, trained offline on plant history and bounded by process limits.
Time-series forecasting
ML, neural and state-space models on real-time and panel data, with prediction intervals for salinity, turbidity (NTU), UV transmittance and ΔP.
Econometric & behavioural models
Micro-econometric identification separates operator actions from process drift, so a model explains what changed and why.
Physics-informed & custom mathematical models
Mass-balance and membrane transport equations combined with learned terms for fouling, scaling and colloidal deposition.
Recommendation engines
Collaborative filtering, matrix factorisation, hedonic and transformer-RL recommenders for maintenance actions and offers.
NLP, LLMs & agents
Entity recognition, topic models and agentic LLMs grounded in procedures, reports and contracts.

Standards we work to

EU AI ActISO/IEC 42001ISO/IEC 42005ISO/IEC 23894ISO/IEC 27001ISO/IEC 27005ASTM D4516

Governance is built into delivery: each model ships with its risk classification, technical documentation and monitoring plan. RO performance data is normalized following the ASTM D4516 practice before any model sees it.

Scientific approach, industrial constraints

HypothesisStart from the process

Every model begins with the physics and chemistry of the unit operation: what drives permeability loss, which variables the operator actually controls, which are disturbances.

IdentificationSeparate cause from correlation

Econometric identification and controlled back-testing show the effect of a dose or set-point change, net of temperature, feed quality and load.

UncertaintyQuantify, then decide

Forecasts carry confidence intervals and recommendations carry a risk bound, so operators know when to trust the model and when to override it.

ValidationProve it on the plant

Test cases are run with plant operations in UAT before production, and model drift is monitored against the same KPIs afterwards.

07 · Expertise

Scientific depth, industrial delivery

artXdata brings together four competencies that are rarely found in one team.

Science

Quantitative modelling

Econometrics, reinforcement learning, forecasting and custom mathematical models, designed for the process at hand and validated against plant data.

Engineering

Industrial AI in production

Real-time pipelines, MLOps and technical documentation, from UAT to production, built with plant operations, data engineers and product teams.

Trust

AI risk & governance

Risk classification, impact assessment and lifecycle monitoring aligned with the EU AI Act, ISO/IEC 42001 and 42005, ISO/IEC 27001 and 27005.

Craft

Data visualization & data art

Results made readable for operators and managers, and custom artworks generated from a client's own data.

Different domains. The same fundamental questions.

artXdata’s modelling philosophy grew out of academic research in applied microeconomics, behavioural modelling, asymmetric information and decision-making under uncertainty. Research on time use, life satisfaction and hedonic valuation applied the same methods to quantities that are hard to observe or to price.

From human behaviour to industrial processes, many modelling problems share the same underlying structure: How do we infer what cannot be directly observed? What should the reference point be? What explains a decision? And how do we value something for which no market price exists?

These questions now shape the way we build industrial AI: combining economic reasoning, process knowledge and machine learning to turn imperfect observations into interpretable, actionable decisions.

Signature

Every dataset has a shape.

Beyond the model, selected artXdata engagements can transform a client’s own operational history into a unique generative artwork.

The same data that informed the decision becomes its visual signature.

Example: 365 daily salinity readings as rays · dots mark clean-in-place events (simulated data)

08 · Contact

Tell us about your plant, your data or your AI portfolio

A first call is enough to see whether a use case is worth modelling. We work in French, English and Turkish.

E-mail
okay.gunes@artxdata.com
Web
artxdata.com
Languages
Français · English · Türkçe