Atomizer/studies/M1_Mirror/m1_mirror_flatback_lateral

M1 Mirror Flat Back - Lateral Supports Optimization

See ../README.md for project overview and optical specifications.

Study Overview

Field	Value
Study Name	m1_mirror_flatback_lateral
Algorithm	CMA-ES
Status	Ready to run
Created	2025-01-13
Trials	100 planned
Focus	Lateral support geometry only

Purpose

Optimize lateral support parameters only using the new U-joint parameterization:

• lateral_inner_u and lateral_outer_u replace the old lateral_inner_pivot and lateral_outer_pivot
• All other parameters (whiffle, ribs, thickness) are fixed at baseline values
• Mass is NOT an objective - only WFE and MFG are optimized

Design Variables (5)

Variable	Min	Max	Baseline	Units	Notes
lateral_inner_u	0.2	0.95	0.4	unitless	U-joint ratio for inner lateral (NEW)
lateral_outer_u	0.2	0.95	0.4	unitless	U-joint ratio for outer lateral (NEW)
lateral_middle_pivot	15.0	27.0	22.42	mm	Middle pivot position
lateral_inner_angle	25.0	35.0	31.96	degrees	Inner lateral angle
lateral_outer_angle	8.0	17.0	9.08	degrees	Outer lateral angle

Objectives

Objective	Weight	Target	Description
wfe_40_20	6.0	4.0 nm	Filtered RMS WFE at 40° relative to 20°
wfe_60_20	5.0	10.0 nm	Filtered RMS WFE at 60° relative to 20°
mfg_90	3.0	20.0 nm	Manufacturing deformation at 90° polishing

Weighted Sum Formula: 6*wfe_40_20 + 5*wfe_60_20 + 3*mfg_90

Note: Mass is NOT in the objective function. A hard constraint (mass ≤ 120 kg) is still enforced.

Fixed Parameters (Baked in Model)

All non-lateral parameters are fixed at the model's current values (from V10 iter123 best design):

Parameter	Value	Units
whiffle_min	70.77	mm
whiffle_outer_to_vertical	86.56	degrees
whiffle_triangle_closeness	69.02	mm
lateral_closeness	11.18	mm
rib_thickness	9.36	mm
ribs_circular_thk	7.12	mm
rib_thickness_lateral_truss	12.52	mm
mirror_face_thickness	15.00	mm
center_thickness	67.92	mm
blank_backface_angle	0.0	degrees

These are not pushed during optimization - the model already contains the correct values.

Why CMA-ES?

Factor	This Problem	Why CMA-ES
Dimensions	5 variables	CMA-ES optimal for 5-50D
Variable type	All continuous	CMA-ES designed for continuous
Landscape	Smooth (physics-based)	CMA-ES exploits gradient structure
Correlation	Lateral params likely correlated	CMA-ES learns correlations automatically
Convergence	100 trials budget	CMA-ES converges 2-3x faster than TPE

CMA-ES Settings

• sigma0: 0.3 (30% of range for initial exploration)
• restart_strategy: IPOP (restarts with larger population if stuck)
• seed: 42

Model Files

File	Purpose
ASSY_M1_assyfem1_sim1.sim	Simulation file
ASSY_M1_assyfem1.afm	Assembly FEM
M1_Blank.prt	Geometry part (102.9 kg)
M1_Blank_fem1.fem	FEM mesh
M1_Blank_fem1_i.prt	Idealized part (critical for mesh updates)
M1_Vertical_Support_Skeleton.prt	Support skeleton
M1_Vertical_Support_Skeleton_fem1.fem	Support skeleton FEM

Extraction Method

• Type: ZernikeOPDExtractor with ANNULAR aperture
• Inner radius: 135.75 mm (271.5 mm central hole excluded)
• Zernike modes: 50
• Filter orders: J1-J4 removed for WFE, J1-J3 for MFG
• Subcases: 90° (1), 20° (2), 40° (3), 60° (4)
• Reference: 20° (subcase 2)

Usage

# Single test trial
python run_optimization.py --test

# Full optimization (100 trials)
python run_optimization.py --start

# Custom trial count
python run_optimization.py --start --trials 50

# Resume interrupted run
python run_optimization.py --start --resume

Directory Structure

m1_mirror_flatback_lateral/
├── 1_setup/
│   ├── model/                    # NX model files
│   ├── optimization_config.json  # Study configuration
│   └── model_expressions.json    # Introspected expressions
├── 2_iterations/                 # FEA iteration folders
├── 3_results/                    # Results database & summaries
│   ├── study.db                  # Optuna SQLite database
│   ├── optimization.log          # Run log
│   └── optimization_summary.json # Final results
├── run_optimization.py           # Main optimization script
├── introspect_model.py          # Model introspection utility
└── README.md                     # This file

Results

Study not yet run. Results will be updated after optimization completes.

References

• Parent study: m1_mirror_cost_reduction_flat_back_V9
• Baseline design source: V10 iter123 (WS=194.46, mass=102.38kg)