Subprojects

M1 Sense

Systematic Over-Instrumentation

M1 decides which sensors and actuators to add while a process is immature, and which to remove once it matures.

Illustration. The green field shows where measurements would reveal most about wrinkle formation. Candidate sensors on the sheet and grippers are ranked and selected. As the process matures, two are removed.

Work packages

Results of the first funding phase, by work package.

  1. M1.1 Optimal sensor and actuator placement and parameterisation

    Bayesian optimisation, developed with M3, identifies gripper configurations that improve product quality in a high-fidelity forming simulation from T2, and places an additional gripper in a targeted way. A second method places additional or mobile sensors for fault diagnosis under model mismatch. Validation on the real system is planned.

    Three simulated formed parts seen from above, with curvature in colour and the gripper positions marked around each part.
    Simulated forming results, with curvature in colour. (a) Configuration of eight grippers optimised by Bayesian optimisation. (b) Improved result with an additional, locally optimised gripper. (c) A less effective placement of the additional gripper. Figure: After Döhner et al. 2026.

    Publications

  2. M1.2 Discovery of latent variables

    Causal root-cause analysis combines anomaly detection with expert-defined causal graphs and process-step information, and a counterfactual extension explains why one root cause is more likely than another. Further methods learn causal structure from wavelet-based soft interventions, evaluate learned causal graphs with a normalised causal edit distance, and learn dynamic structural causal mechanisms from time series.

    Architecture diagram of the dynamic structural causal mechanisms method, from observational time series through a GRU encoder and graph-constrained neural mechanisms to counterfactual inference and optimisation.
    Overview of the dynamic structural causal mechanisms (DSCM) method. Observational time series are encoded by a GRU and environment embeddings, then combined with graph-constrained neural mechanisms to learn dynamic causal relationships. Figure: KI-FOR 5339.

    Publications

  3. M1.3 Sensitivity and influence analysis

    A method combines causal discovery, causal inference and temporal neural networks. It selects the variables with a causal influence on the outcome and estimates their effect through the conditional average treatment effect, as a basis for a minimal relevant sensor set.

    Publications

  4. M1.4 Application and validation

    Patterns printed on the laminate before forming make defects such as folds visible afterwards, and methods for automated fold detection are being developed. Transferring the Bayesian optimisation of the gripper configuration to the real forming process is planned, and the influence of low-frequency tool vibrations on forming is being measured on a dedicated test rig.

About the subproject

Early on, generous instrumentation is the fastest way to understand a new process. Later, many sensors are no longer needed, and the process should run with fewer and cheaper ones.

Principal investigators
Researchers
  • Hannah Decker
  • Frank Döhner
  • Zahra Nasrollah
  • Shahenda Youssef
  • Georg Zeeb
Former members
  • Dr. Josephine Rehak
Host institute
Institute of Vehicle System Technology (FAST), KIT
People in M1

Publications 9