Intelligent quality control for sustainable manufacturing
Data-driven surrogate models trained on stochastic FE simulations. Real-time quality prediction, process adjustment, and anomaly detection for automotive sheet metal forming.
Featured projects.
Explore our ongoing research initiatives in sustainable steel manufacturing.
EU Horizon Europe 2024–2028. Developing AI-driven tools to optimize the use of recycled steel in mass manufacturing.
Vinnova FFI 2020–2023. Advanced material models and AI-based process adjustment to predict and prevent forming failures in DP-steels and aluminium.
Platform capabilities.
From material certificate to real-time quality decision on the shop floor.
ANN trained on stochastic AutoForm-Sigma FE simulations. Replaces expensive simulations with instant predictions of failure and springback.
GREEN / YELLOW / RED status classification for immediate shop-floor decisions. Automatic cushion force optimisation to achieve GREEN.
Understand why a prediction was made. SHAP feature importance shows which material properties drive the quality outcome.
PCA + Local Outlier Factor flags anomalous material batches before they reach the press. Unsupervised, no labelled data needed.
Latest papers.
The three most recent publications from CiSMA & PREDICT — see the research page for the full archive.
Numerical data driven operation support for manufacturing of automotive body components
Prediction of sheet metal part production robustness using advanced tribological models, thermo-mechanical modelling and stochastic FE-simulations
On the use of Process Work as an Indicator for Process Disturbance in industrial Sheet Metal Forming
How it works.
A two-stage methodology: observe the physical process, then simulate and predict digitally.
Observe
Capture real-world forming data from the production line
Record forming process parameters, press forces, and production conditions from each stamping cycle.
Digitise manufacturing tools and press setup for FE-model geometry update and boundary conditions.
Calibrate the finite-element model against measured data to ensure simulation accuracy and reliability.
Assess forming quality using failure index, springback, and thinning against GREEN/YELLOW/RED thresholds.
Measure r-value, tensile strength (RM), and yield stress (RP) from incoming coil material certificates.
Simulate
Build surrogate models from FE data for real-time quality prediction
Stochastic FE simulations in AutoForm-Sigma map the full solution space of material and process parameter variations.
ANN surrogate models are trained on multimodal production data augmented with FE simulations. Domain adaptation bridges the gap between virtual and real-world conditions.
Input material certificate data at the press line. Get instant quality classification with SHAP explanations and optimal cushion force.
Transfer learning updates models with new production data. Version control enables rollback and continuous improvement.
Built for every role.
From the press line to the research lab — tailored tools for each stakeholder.
"I know in 30 seconds whether this coil belongs on the press today."
Enter r-value, RM and RP from the certificate. Get a GREEN / YELLOW / RED verdict before the steel reaches the press. Anomalous batches are flagged automatically.
"No guesswork. The cushion-force number lands on my screen with the coil."
Receive optimised cushion-force settings for each batch. Act on clear traffic-light signals. Log issues against a coil for later review by engineering.
"SHAP tells me which property of which coil cost us the part."
Deep-dive into SHAP attributions, browse the simulation dataset, manage surrogate-model versions. Compare runs, export findings, push updated models to the line.
"FE datasets, trained models and twelve papers — all in one place."
Access FE simulation datasets, trained ANN models, and the consortium's publication record. API access for downstream research. Citation-ready.
Consortium partners.
18 research institutions and industry partners across Europe.
