in silico
Phase 1 : In Silico Validation
- Objectif
- To rapidly and inexpensively test the core assumptions of the hierarchical control model, specifically the constitutive model linking SMA/healing agent actuation to macroscopic stiffness, and the stability of the control algorithms under simulated sensor noise and failure.
- Coût estimé
- €1,500-2,000 (primarily software license allocation and cloud compute credits)
- Durée estimée
- 6-8 weeks
- Critères de réussite
- Constitutive Model Accuracy · Predicted vs. “ground truth” (from high-fidelity COMSOL or literature) stiffness change from SMA actuation must be within ±7%. · (Run a standalone simulation of a beam segment with SMA activation, compare the simulated stiffness change to values calculated from classical laminate plate theory or published experimental data.)
- Control Loop Stability · The hierarchical controller must maintain simulated beam stiffness within ±15% of target for 95% of a 500-cycle simulation with a single 0.3mm crack. · (Analyze the stiffness retention output from virtual experiment (a). Calculate the percentage of time steps where stiffness is within the ±15% bound.)
- Robustness to Sensor Failure · With 30% sensor failure, the hierarchical controller’s mean absolute regulation error must be at least 40% lower than the centralized controller’s error. · (Compare the mean absolute error (MAE) outputs from virtual experiment (c) between the two control architectures.)
- On continue si
- All three success criteria are met. The model demonstrates a plausible causal chain from damage detection to stiffness recovery.
- On arrête si
- The constitutive model is inaccurate (>±15% error) or the control system is fundamentally unstable (diverges) under nominal conditions. This indicates a flaw in the core mechanistic assumption.
- On réoriente si
- The hierarchical controller is stable but performs only marginally better than the baseline. Pivot by simplifying the architecture (e.g., remove the high-level strategy layer, use a direct damage-to-actuation map) for Phase 2.
- Risques
- Simulation model is over-simplified and misses key non-linearities (e.g., SMA hysteresis, crack-healing interface degradation).Probabilité : mediumParade : Validate sub-models (SMA, healing) against dedicated experimental data from literature. Use a hybrid modeling approach (data-driven + physics-based) for uncertain components.
- Simulation run-times are prohibitively long for 500-cycle tests.Probabilité : highParade : Use model order reduction techniques for the beam dynamics. Implement the simulation on a cloud computing platform (AWS ParallelCluster) and run parameter sweeps in parallel.