Skip to content
Back to skills

Abaqus Dynamic Analysis

ASecurity

Complete workflow for dynamic analysis. Use when user mentions impact, crash, drop test, transient, or time-varying response. Handles explicit and implicit dynamics.

  • 666 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added May 31, 2026
toolsbashapi

Works with

  • api

Security analysis

A100/100

Scanned May 31, 2026

npx -y skills add majiayu000/claude-skill-registry --skill abaqus-dynamic-analysis --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Abaqus Dynamic Analysis?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Abaqus Dynamic Analysis
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-abaqus-dynamic-analysis/badge)](https://www.skillsdirectory.com/skills/majiayu000-abaqus-dynamic-analysis)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: abaqus-dynamic-analysis
description: Complete workflow for dynamic analysis. Use when user mentions impact, crash, drop test, transient, or time-varying response. Handles explicit and implicit dynamics.
allowed-tools:
  - Read
  - Write
  - Edit
  - Glob
  - Grep
  - Bash(abaqus:*)
  - Skill
---

# Abaqus Dynamic Analysis Skill

This skill handles explicit and implicit dynamics for impact, crash, drop test, and transient response analysis.

## When to Use This Skill

**Route here when user mentions:**
- Impact, crash, collision, drop test
- Transient response, time-varying response
- Shock loading, blast loading, explosive loading
- High-speed events, wave propagation
- "What happens when it hits..."

**Route elsewhere:**
- Natural frequency extraction → `/abaqus-modal-analysis`
- Static/constant loads → `/abaqus-static-analysis`
- Harmonic/sinusoidal response → modal + steady-state dynamics
- Very long transients (minutes+) → consider implicit or quasi-static

## Prerequisites

Before dynamic analysis:
1. Geometry and mesh ready
2. **Material MUST have density defined** (required for mass matrix)
3. Understand event duration and loading type

## Workflow: Setting Up Dynamic Analysis

### Step 1: Gather Information from User

Ask if unclear:
- **What's the event duration?** Milliseconds, seconds, or longer?
- **Initial velocity?** For drop tests or impact
- **Is contact involved?** Parts colliding or touching
- **What output needed?** Stress, velocity, acceleration, energy?

### Step 2: Choose Explicit vs Implicit

| Factor | Explicit | Implicit |
|--------|----------|----------|
| Time scale | Short (us to ms) | Longer (ms to s) |
| Step size | Automatic (very small) | User-controlled |
| Nonlinearity | Handles well | May need iterations |
| Memory | Lower | Higher |
| Contact | Natural handling | Needs care |
| Best for | Impact, crash | Vibration, long transient |

**Decision rule:**
- Event < 10ms with impact/contact → **Explicit**
- Event > 100ms without severe nonlinearity → **Implicit**
- In between → Either can work, explicit often easier

### Step 3: Set Time Period

| Event Type | Typical Duration |
|------------|------------------|
| High-speed impact | 0.1-10 ms |
| Drop test | 1-100 ms |
| Blast loading | 1-50 ms |
| Seismic/vibration | 1-100 s |

### Step 4: Define Initial Conditions

For drop tests and impact:
- Set initial velocity on the impacting part/region
- Velocity is applied in the Initial step

### Step 5: Configure Output

Field outputs: `S` (stress), `U` (displacement), `V` (velocity), `A` (acceleration), `PEEQ` (plastic strain)

History outputs for energy balance (explicit): `ALLKE`, `ALLIE`, `ALLWK`, `ETOTAL`

### Step 6: Consider Mass Scaling (Explicit Only)

| Option | Effect | When |
|--------|--------|------|
| None | True inertia | Very short events, accuracy critical |
| At beginning | Scale once | Quasi-static explicit |
| Throughout | Continuous scaling | When inertia less important |

**Warning:** Mass scaling speeds up analysis but affects inertial response.

### Step 7: Run and Validate

Use `/abaqus-job` to submit, then check:
- Energy balance (ETOTAL approximately constant)
- Stable time increment (explicit)
- Results physically reasonable

## Key Parameters

| Parameter | Explicit | Implicit |
|-----------|----------|----------|
| Time period | Event duration | Event duration |
| Time increment | Automatic | Specify initial, min, max |
| Element library | EXPLICIT | STANDARD |
| Element type | C3D8R recommended | C3D8R or C3D8 |
| Hourglass control | ENHANCED | Default |

## Validation Checklist

- [ ] Density defined in material
- [ ] Time period appropriate for event
- [ ] Initial conditions applied (velocity, position)
- [ ] Output frequency captures behavior (100+ frames typical)
- [ ] Energy balance acceptable (ETOTAL constant for explicit)
- [ ] Results physically reasonable

## Troubleshooting

| Problem | Likely Cause | Solution |
|---------|--------------|----------|
| "Time increment too small" | Small/distorted elements | Use mass scaling or coarsen mesh |
| Energy balance error | Hourglass or instability | Check hourglass energy, add control |
| Analysis takes forever (explicit) | Long time period | Consider implicit instead |
| Convergence failure (implicit) | Severe nonlinearity | Use explicit or smaller increments |

## Related Skills

- `/abaqus-material` - Define density (required)
- `/abaqus-amplitude` - Time-varying loads
- `/abaqus-field` - Initial velocity and predefined fields
- `/abaqus-interaction` - Contact for impact problems
- `/abaqus-odb` - Results extraction

## Code Patterns

For API syntax and code examples, see:
- [API Quick Reference](references/api-quick-ref.md)
- [Common Patterns](references/common-patterns.md)
- [Troubleshooting Guide](references/troubleshooting.md)

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.