Phase 3.3: Multi-objective optimization fix, updated docs & Claude skill

- Fixed drone gimbal optimization to use proper semantic directions
- Changed from ['minimize', 'minimize'] to ['minimize', 'maximize']
- Updated Claude skill (v2.0) with Phase 3.3 integration
- Added centralized extractor library documentation
- Added multi-objective optimization (Protocol 11) section
- Added NX multi-solution protocol documentation
- Added dashboard integration documentation
- Fixed Pareto front degenerate issue with proper NSGA-II configuration

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
2025-11-24 07:49:48 -05:00
parent f76bd52894
commit dd7f0c0f82
7 changed files with 1529 additions and 220 deletions

390
docs/DASHBOARD.md Normal file
View File

@@ -0,0 +1,390 @@
# Atomizer Dashboard
**Last Updated**: November 23, 2025
---
## Overview
The Atomizer Dashboard is a real-time web-based interface for monitoring and analyzing multi-objective optimization studies. Built with React, TypeScript, and Tailwind CSS, it provides comprehensive visualization and interaction capabilities for NSGA-II based structural optimization.
---
## Architecture
### Frontend Stack
- **Framework**: React 18 with TypeScript
- **Build Tool**: Vite
- **Styling**: Tailwind CSS with custom dark/light theme support
- **Charts**: Recharts for data visualization
- **State Management**: React hooks (useState, useEffect)
- **WebSocket**: Real-time optimization updates
### Backend Stack
- **Framework**: FastAPI (Python)
- **Database**: Optuna SQLite studies
- **API**: RESTful endpoints with WebSocket support
- **CORS**: Configured for local development
### Ports
- **Frontend**: `http://localhost:3003` (Vite dev server)
- **Backend**: `http://localhost:8000` (FastAPI/Uvicorn)
---
## Key Features
### 1. Multi-Objective Visualization
#### Pareto Front Plot
- 2D scatter plot showing trade-offs between objectives
- Color-coded by constraint satisfaction (green = feasible, red = infeasible)
- Interactive hover tooltips with trial details
- Automatically extracts Pareto-optimal solutions using NSGA-II
#### Parallel Coordinates Plot
**Research-Based Multi-Dimensional Visualization**
Structure: **Design Variables → Objectives → Constraints**
Features:
- **Light Theme**: White background with high-visibility dark text and colors
- **Color-Coded Axes**:
- Blue background: Design variables
- Green background: Objectives
- Yellow background: Constraints
- **Interactive Selection**:
- Hover over lines to highlight individual trials
- Click to select/deselect trials
- Multi-select with visual feedback (orange highlight)
- **Type Badges**: Labels showing DESIGN VAR, OBJECTIVE, or CONSTRAINT
- **Units Display**: Automatic unit labeling (mm, MPa, Hz, g, etc.)
- **Min/Max Labels**: Range values displayed on each axis
- **Feasibility Coloring**:
- Green: Feasible solutions
- Red: Infeasible solutions (constraint violations)
- Blue: Hover highlight
- Orange: Selected trials
**Implementation**: [ParallelCoordinatesPlot.tsx](atomizer-dashboard/frontend/src/components/ParallelCoordinatesPlot.tsx:1)
**Line colors**:
```typescript
if (isSelected) return '#FF6B00'; // Orange for selected
if (!trial.feasible) return '#DC2626'; // Red for infeasible
if (isHovered) return '#2563EB'; // Blue for hover
return '#10B981'; // Green for feasible
```
### 2. Optimizer Strategy Panel
Displays algorithm information:
- **Algorithm**: NSGA-II, TPE, or custom
- **Type**: Single-objective or Multi-objective
- **Objectives Count**: Number of optimization objectives
- **Design Variables Count**: Number of design parameters
### 3. Convergence Monitoring
- **Convergence Plot**: Best value vs. trial number
- **Real-time Updates**: WebSocket-driven live updates
- **Pruned Trials**: Visual indication of pruned trials
### 4. Parameter Space Exploration
- **2D Scatter Plot**: Design variable relationships
- **Color Mapping**: Objective values mapped to color intensity
- **Interactive Tooltips**: Trial details on hover
### 5. Trial History Table
- Comprehensive list of all trials
- Sortable columns
- Status indicators (COMPLETE, PRUNED, FAIL)
- Parameter values and objective values
- User attributes (constraints)
---
## API Endpoints
### Studies
#### GET `/api/optimization/studies`
List all available optimization studies.
**Response**:
```json
[
{
"id": "drone_gimbal_arm_optimization",
"name": "drone_gimbal_arm_optimization",
"direction": ["minimize", "maximize"],
"n_trials": 100,
"best_value": [3245.67, 165.3],
"sampler": "NSGAIISampler"
}
]
```
#### GET `/api/optimization/studies/{study_id}/trials`
Get all trials for a study.
**Response**:
```json
{
"trials": [
{
"number": 0,
"values": [3456.2, 145.6],
"params": {
"beam_half_core_thickness": 7.5,
"beam_face_thickness": 2.1,
"holes_diameter": 30.0,
"hole_count": 11
},
"state": "COMPLETE",
"user_attrs": {
"max_stress": 95.3,
"max_displacement": 1.2,
"frequency": 145.6,
"mass": 3456.2,
"constraint_satisfied": true
}
}
]
}
```
#### GET `/api/optimization/studies/{study_id}/metadata`
Get study metadata including objectives and design variables.
**Response**:
```json
{
"objectives": [
{
"name": "mass",
"type": "minimize",
"unit": "g"
},
{
"name": "frequency",
"type": "maximize",
"unit": "Hz"
}
],
"design_variables": [
{
"name": "beam_half_core_thickness",
"unit": "mm",
"min": 5.0,
"max": 10.0
}
],
"sampler": "NSGAIISampler"
}
```
#### GET `/api/optimization/studies/{study_id}/pareto-front`
Get Pareto-optimal solutions for multi-objective studies.
**Response**:
```json
{
"is_multi_objective": true,
"pareto_front": [
{
"trial_number": 0,
"values": [3245.67, 165.3],
"params": {...},
"user_attrs": {...},
"constraint_satisfied": true
}
]
}
```
### WebSocket
#### WS `/ws/optimization/{study_id}`
Real-time trial updates during optimization.
**Message Format**:
```json
{
"type": "trial_complete",
"trial": {
"number": 5,
"values": [3456.2, 145.6],
"params": {...}
}
}
```
---
## Running the Dashboard
### Backend
```bash
cd atomizer-dashboard/backend
python -m uvicorn api.main:app --reload --port 8000
```
### Frontend
```bash
cd atomizer-dashboard/frontend
npm run dev
```
Access at: `http://localhost:3003`
---
## Configuration
### Vite Proxy ([vite.config.ts](atomizer-dashboard/frontend/vite.config.ts:1))
```typescript
export default defineConfig({
plugins: [react()],
server: {
host: '0.0.0.0',
port: 3003,
proxy: {
'/api': {
target: 'http://127.0.0.1:8000',
changeOrigin: true,
secure: false,
ws: true, // WebSocket support
}
}
}
})
```
### CORS ([backend/api/main.py](atomizer-dashboard/backend/api/main.py:1))
```python
app.add_middleware(
CORSMiddleware,
allow_origins=["http://localhost:3003"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
```
---
## Component Structure
```
atomizer-dashboard/
├── frontend/
│ ├── src/
│ │ ├── components/
│ │ │ ├── ParallelCoordinatesPlot.tsx # Multi-objective visualization
│ │ │ ├── ParetoPlot.tsx # Pareto front scatter plot
│ │ │ ├── OptimizerPanel.tsx # Strategy information
│ │ │ ├── common/
│ │ │ │ └── Card.tsx # Reusable card component
│ │ │ └── dashboard/
│ │ │ ├── MetricCard.tsx # KPI display
│ │ │ └── StudyCard.tsx # Study selector
│ │ ├── pages/
│ │ │ └── Dashboard.tsx # Main dashboard page
│ │ ├── hooks/
│ │ │ └── useWebSocket.ts # WebSocket connection
│ │ ├── api/
│ │ │ └── client.ts # API client
│ │ └── types/
│ │ └── index.ts # TypeScript types
│ └── vite.config.ts
└── backend/
└── api/
├── main.py # FastAPI app
└── routes/
└── optimization.py # Optimization endpoints
```
---
## Data Flow
1. **Optimization Engine** runs trials and stores results in Optuna SQLite database
2. **Backend API** reads from database and exposes REST endpoints
3. **Frontend** fetches data via `/api/optimization/*` endpoints
4. **WebSocket** pushes real-time updates to connected clients
5. **React Components** render visualizations based on fetched data
---
## Troubleshooting
### Dashboard Page Crashes
**Issue**: `TypeError: Cannot read properties of undefined (reading 'split')`
**Fix**: Ensure all data is validated before rendering. ParallelCoordinatesPlot now includes:
```typescript
if (!paretoData || paretoData.length === 0) return <EmptyState />;
if (!objectives || !designVariables) return <EmptyState />;
```
### No Data Showing
1. Check backend is running: `curl http://localhost:8000/api/optimization/studies`
2. Verify study exists in Optuna database
3. Check browser console for API errors
4. Ensure WebSocket connection is established
### CORS Errors
- Backend must allow origin `http://localhost:3003`
- Frontend proxy must target `http://127.0.0.1:8000` (not `localhost`)
---
## Best Practices
### For Multi-Objective Studies
1. **Always use metadata endpoint** to get objective/variable definitions
2. **Extract constraints from user_attrs** for parallel coordinates
3. **Filter Pareto front** using `paretoData.pareto_front` array
4. **Validate constraint_satisfied** field before coloring
### For Real-Time Updates
1. **Use WebSocket** for live trial updates
2. **Debounce state updates** to avoid excessive re-renders
3. **Close WebSocket** connection on component unmount
### For Performance
1. **Limit displayed trials** for large studies (e.g., show last 1000)
2. **Use React.memo** for expensive components
3. **Virtualize large lists** if showing >100 trials in tables
---
## Future Enhancements
- [ ] 3D Pareto front visualization for 3+ objectives
- [ ] Advanced filtering and search in trial history
- [ ] Export results to CSV/JSON
- [ ] Custom parallel coordinates brushing/filtering
- [ ] Multi-study comparison view
- [ ] Hypervolume indicator tracking
- [ ] Interactive design variable sliders
- [ ] Constraint importance analysis
---
## References
- **Optuna Documentation**: https://optuna.readthedocs.io/
- **NSGA-II Algorithm**: Deb et al. (2002)
- **Parallel Coordinates**: Inselberg & Dimsdale (1990)
- **React Documentation**: https://react.dev/
- **FastAPI Documentation**: https://fastapi.tiangolo.com/

View File

@@ -0,0 +1,197 @@
# NX Multi-Solution Solve Protocol
## Critical Finding: SolveAllSolutions API Required for Multi-Solution Models
**Date**: November 23, 2025
**Last Updated**: November 23, 2025
**Protocol**: Multi-Solution Nastran Solve
**Affected Models**: Any NX simulation with multiple solutions (e.g., static + modal, thermal + structural)
---
## Problem Statement
When an NX simulation contains multiple solutions (e.g., Solution 1 = Static Analysis, Solution 2 = Modal Analysis), using `SolveChainOfSolutions()` with Background mode **does not wait for all solutions to complete** before returning control to Python. This causes:
1. **Missing OP2 Files**: Only the first solution's OP2 file is generated
2. **Stale Data**: Subsequent trials read old OP2 files from previous runs
3. **Identical Results**: All trials show the same values for results from missing solutions
4. **Silent Failures**: No error is raised - the solve completes but files are not written
### Example Scenario
**Drone Gimbal Arm Optimization**:
- Solution 1: Static analysis (stress, displacement)
- Solution 2: Modal analysis (frequency)
**Symptoms**:
- All 100 trials showed **identical frequency** (27.476 Hz)
- Only `beam_sim1-solution_1.op2` was created
- `beam_sim1-solution_2.op2` was never regenerated after Trial 0
- Both `.dat` files were written correctly, but solve didn't wait for completion
---
## Root Cause
```python
# WRONG APPROACH (doesn't wait for completion)
psolutions1 = []
solution_idx = 1
while True:
solution_obj_name = f"Solution[Solution {solution_idx}]"
simSolution = simSimulation1.FindObject(solution_obj_name)
if simSolution:
psolutions1.append(simSolution)
solution_idx += 1
else:
break
theCAESimSolveManager.SolveChainOfSolutions(
psolutions1,
NXOpen.CAE.SimSolution.SolveOption.Solve,
NXOpen.CAE.SimSolution.SetupCheckOption.CompleteDeepCheckAndOutputErrors,
NXOpen.CAE.SimSolution.SolveMode.Background # ❌ Returns immediately!
)
```
**Issue**: Background mode runs asynchronously and returns control to Python before all solutions finish solving.
---
## Correct Solution
### For Solving All Solutions
Use `SolveAllSolutions()` API with **Foreground mode**:
```python
# CORRECT APPROACH (waits for completion)
if solution_name:
# Solve specific solution in background mode
solution_obj_name = f"Solution[{solution_name}]"
simSolution1 = simSimulation1.FindObject(solution_obj_name)
psolutions1 = [simSolution1]
numsolutionssolved1, numsolutionsfailed1, numsolutionsskipped1 = theCAESimSolveManager.SolveChainOfSolutions(
psolutions1,
NXOpen.CAE.SimSolution.SolveOption.Solve,
NXOpen.CAE.SimSolution.SetupCheckOption.CompleteDeepCheckAndOutputErrors,
NXOpen.CAE.SimSolution.SolveMode.Background
)
else:
# Solve ALL solutions using SolveAllSolutions API (Foreground mode)
# This ensures all solutions (static + modal, etc.) complete before returning
print(f"[JOURNAL] Solving all solutions using SolveAllSolutions API (Foreground mode)...")
numsolutionssolved1, numsolutionsfailed1, numsolutionsskipped1 = theCAESimSolveManager.SolveAllSolutions(
NXOpen.CAE.SimSolution.SolveOption.Solve,
NXOpen.CAE.SimSolution.SetupCheckOption.CompleteCheckAndOutputErrors,
NXOpen.CAE.SimSolution.SolveMode.Foreground, # ✅ Blocks until complete
False
)
```
### Key Differences
| Aspect | SolveChainOfSolutions | SolveAllSolutions |
|--------|----------------------|-------------------|
| **Manual enumeration** | Required (loop through solutions) | Automatic (handles all solutions) |
| **Background mode behavior** | Returns immediately, async | N/A (Foreground recommended) |
| **Foreground mode behavior** | Blocks until complete | Blocks until complete ✅ |
| **Use case** | Specific solution selection | Solve all solutions |
---
## Implementation Location
**File**: `optimization_engine/solve_simulation.py`
**Lines**: 271-295
**When to use this protocol**:
- When `solution_name=None` is passed to `NXSolver.run_simulation()`
- Any simulation with multiple solutions that must all complete
- Multi-objective optimization requiring results from different analysis types
---
## Verification Steps
After implementing the fix, verify:
1. **Both .dat files are written** (one per solution)
```
beam_sim1-solution_1.dat # Static analysis
beam_sim1-solution_2.dat # Modal analysis
```
2. **Both .op2 files are created** with updated timestamps
```
beam_sim1-solution_1.op2 # Contains stress, displacement
beam_sim1-solution_2.op2 # Contains eigenvalues, mode shapes
```
3. **Results are unique per trial** - check that frequency values vary across trials
4. **Journal log shows**:
```
[JOURNAL] Solving all solutions using SolveAllSolutions API (Foreground mode)...
[JOURNAL] Solve completed!
[JOURNAL] Solutions solved: 2
```
---
## Related Issues Fixed
1. **All trials showing identical frequency**: Fixed by ensuring modal solution runs
2. **Only one data point in dashboard**: Fixed by all trials succeeding
3. **Parallel coordinates with NaN**: Fixed by having complete data from all solutions
---
## References
- **User's Example**: `nx_journals/user_generated_journals/journal_solve_all_solution.py` (line 27)
- **NX Open Documentation**: SimSolveManager.SolveAllSolutions() method
- **Implementation**: `optimization_engine/solve_simulation.py`
---
## Best Practices
1. **Always use Foreground mode** when solving all solutions
2. **Verify OP2 timestamp changes** to ensure fresh solves
3. **Check solve counts** in journal output to confirm both solutions ran
4. **Test with 5 trials** before running large optimizations
5. **Monitor unique frequency values** as a smoke test for multi-solution models
---
## Example Use Cases
### ✅ Correct Usage
```python
# Multi-objective optimization with static + modal
result = nx_solver.run_simulation(
sim_file=sim_file,
working_dir=model_dir,
expression_updates=design_vars,
solution_name=None # Solve ALL solutions
)
```
### ❌ Incorrect Usage (Don't Do This)
```python
# Running modal separately - inefficient and error-prone
result1 = nx_solver.run_simulation(..., solution_name="Solution 1") # Static
result2 = nx_solver.run_simulation(..., solution_name="Solution 2") # Modal
# This doubles the solve time and requires managing two result objects
```
---
**Status**: ✅ Implemented and Verified
**Impact**: Critical for all multi-solution optimization workflows