Particle instability / sudden particle ejection in rotary drum with internals / particle motion and heat transfer analysis

I am experiencing sudden, uncontrolled particle ejections (particles "flying away") in a CFD-DEM coupled simulation of a rotating drum with internals (segment) for heat transfer investigation (conductive and convective; low temparature area). The simulation runs stable for approximately 20-30 seconds, then particles are suddenly ejected from the packed bed. This behavior appears to correlate with the output/dump intervals.

Key observation: When I increase writeInterval (CFD) and dumptime (DEM) from 0.05s to 1.0s, the instability does NOT occur. This suggests a potential coupling or I/O synchronization issue.

System Configuration
ComponentVersionCFDEM®couplingPublic versionOpenFOAM5.xLIGGGHTS3.8.0Operating SystemUbuntu 20.04 (Docker container)SolvercfdemSolverPisoScalarParallelization12 cores (2×3×2 decomposition)
All official CFDEM tutorials (e.g., ErgunTestMPI, settlingTest) run without any issues.

Case Description
Geometry:
Cylindrical geometry with singel segment, D = 300 mm, L = 50 mm
STL meshes for walls (imported via LIGGGHTS mesh/surface)
Rotating at 2 rpm (period = 30 s) around Y-axis

Particles:
3750 spherical particles
Diameter: 3 mm
Density: 2500 kg/m³
Fill level: ~20%
Initial temperature: 293 K (20°C)

Heat Transfer:
Wall temperature: 573 K (300°C)
Particle-particle conduction via heat/gran/conduction
Convective heat transfer via LaEuScalarTemp

Mesh:
Simple blockMesh (cartesian)
Domain: 350 × 185 × 350 mm
Resolution: 39 × 21 × 39 cells (~9 mm cell size)
Ratio: ~3 cells per particle diameter

Problem Description:
1. Simulation starts normally with stable particle bed
2. Particles rotate with the rotary drum walls as expected
3. After approximately 20-30 seconds (roughly 1 rotation), particles suddenly get ejected
4. Kinetic energy increases by factor ~28× within milliseconds:
t ≈ 30.0s: KinEng = 5.3e-05 J (stable)
t ≈ 31.25s: KinEng = 1.47e-03 J (unstable!)
5. Strong pressure spikes visible in ParaView during ejection events

Visual Evidence:
The transition happens within a single output timestep:
- Timestep 617 (t = 30.85s): Particles in stable packed bed
- Timestep 618 (t = 30.90s): Multiple particles ejected high above the bed

Critical Observation:
When writeInterval and dumptime are set to 1.0s instead of 0.05s, the problem does NOT occur! This suggests the issue may be related to:
- I/O operations interfering with coupling
- Synchronization between CFD write and DEM dump
- Memory/buffer issues during frequent file output