in silico
Phase 1 : In Silico Validation: Discrete Element Method (DEM) Simulation of Vibration-Induced Unjamming
- Objectif
- To computationally falsify the core mechanism (reduction of effective wall friction μ_w via vibration) before any physical experiments, using validated DEM software. Determine if the predicted μ_w reduction of ≥60% at Γ=3.0 is achievable in silico for the reference case (100 μm glass beads, D_h/d_p=5).
- Coût estimé
- €500-2000 (assuming free LIGGGHTS + existing workstation; €2k if cloud GPU needed)
- Durée estimée
- 4-6 weeks
- Critères de réussite
- Reduction in μ_w at Γ=3.0 relative to Γ=0 · ≥60% reduction (from 0.45 to ≤0.20) · (Average μ_w over 1000 time steps per simulation; compare 10 simulations per condition)
- Reduction in P_clog at Γ=3.0 · ≥70% reduction (from ≥0.80 to ≤0.25) · (Fraction of simulations with clogging events; two-proportion z-test (α=0.01))
- Critical Γ_c for non-cohesive particles · Γ_c between 1.0 and 2.0 (consistent with prediction of 1.5) · (Stepwise sweep; Γ_c defined as first Γ where μ_w < 0.20 for >50% of time)
- On continue si
- All three success criteria met (μ_w reduction ≥60%, P_clog reduction ≥70%, Γ_c in [1.0,2.0])
- On arrête si
- Any criterion fails: e.g., μ_w reduction <40% or P_clog reduction <50% or Γ_c > 3.0
- On réoriente si
- If μ_w reduction is achieved but P_clog reduction fails, consider that the mechanism might be incomplete (e.g., vibration-induced particle segregation). Pivot to test alternative mechanisms (e.g., particle size segregation) in Phase 2.
- Risques
- DEM simulations may not capture cohesive forces (van der Waals) accurately for particles <50 μmProbabilité : mediumParade : Use JKR cohesion model in LIGGGHTS; validate against known cohesion benchmarks (e.g., Rumpf model). If cohesion dominates, adjust Γ_c prediction upward.
- Computational cost: 10 simulations × 1 s real time at 1e-7 s timestep = 1e8 timesteps, requiring weeks on a single GPUProbabilité : highParade : Use coarse-graining (scale particle size by 2x, adjust density to maintain Stokes number); run on cloud GPU cluster (AWS EC2 G4dn instance, ~$1/hr).
- Liquid damping may be underestimated in DEM (Stokes drag only, no turbulence)Probabilité : mediumParade : Compare with CFD-DEM (e.g., CFDEM coupling) for one validation case; if discrepancy >20%, use CFD-DEM for all critical runs.