🔥 UHTM Dataset — IMI Project

Ultra-High Temperature Materials · Physics-Informed Feature Engineering

A collaboratively engineered 200-sample × 48-feature dataset of refractory ceramics and composites,
built by a 4-member team using physics-informed feature engineering across 8 material property domains.

📋 Website : https://imi-project.vercel.app/

🧪 Project Overview

This project constructs a structured, ML-ready dataset for Ultra-High Temperature Materials (UHTMs) — refractory ceramics used in hypersonic re-entry vehicles, thermal protection systems (TPS), and aerospace leading-edge components.

The dataset covers 10 material families (carbides, borides, nitrides) and their composites/doped variants across 8 physical property domains, with 3 supervised regression targets. All features are derived from 5 literature-anchored material properties using validated physical relations plus realistic Gaussian noise (np.random.seed(42)).

Key design principles:

Each of the 4 team members owns a distinct physical domain → clean separation of concerns
All features trace back to 5 shared anchors (Tₘ, ρ, vₑ, ΔEN, Pₛ) → internally consistent dataset
Reproducible: fixed seed, shared base file, merge validation enforced programmatically

📊 Dataset Summary

Property	Value
Total Samples	200 (100 experimental + 100 synthetic)
Feature Columns	48 (F01–F48)
Target Columns	3 (T1–T3)
Metadata Columns	5 (Sample_ID, Material_System, Source_Type, Synthesis_Method, Crystal_Structure)
Total Columns	56
Material Families	10 base + composites + doped variants
Random Seed	42 (all scripts)
Final Output	`UHTM_final_200x48.xlsx` + `UHTM_final_200x48.csv`

🗂 Repository Structure

IMI-Project-main/
│
├── Aadi-Dev/                          # Member 1 — Thermodynamic + Electronic
│   ├── dataset.py                     # ★ Generates F01–F12
│   ├── Aadi.xlsx                      # Output: 200 × 17 (meta + 12 features)
│   └── UHTM_base_200.xlsx             # Base reference copy
│
├── Krish/                             # Member 2 — Mechanical + Thermal + Infrastructure
│   ├── Intro.py                       # Branch onboarding note
│   └── LAB EVALUATION/
│       ├── Krishh.py                  # ★ Generates F13–F24
│       ├── Krishh.xlsx                # Output: 200 × 17
│       ├── UHTM_base_200.xlsx
│       └── Merge/
│           ├── Base.py                # ★★ Generates UHTM_base_200.xlsx (run first)
│           ├── mergeAll.py            # ★★ Final merge of all 4 member files
│           ├── AadiDev.xlsx           # Member 1 snapshot for merge
│           ├── Krishh.xlsx            # Member 2 snapshot for merge
│           ├── Niranjan.xlsx          # Member 4 snapshot for merge
│           ├── Salan.xlsx             # Member 3 snapshot for merge
│           └── UHTM_final_200x48.xlsx # ★★ FINAL MERGED DATASET
│
├── Niranjan/                          # Member 4 — Phase + ML Descriptors + Targets
│   ├── Niranjan.py                    # ★ Generates F37–F48 + T1, T2, T3
│   ├── Niranjan.xlsx                  # Output: 200 × 20
│   └── UHTM_base_200.xlsx
│
├── Salan/                             # Member 3 — Oxidation + Microstructural
│   ├── lab evaluation/
│   │   ├── Salan.py                   # ★ Generates F25–F36
│   │   ├── Salan_member3.xlsx         # Output: 200 × 17
│   │   └── UHTM_base_200.xlsx
│   └── Backup_datasets/
│       ├── UHTM_Complete.csv
│       ├── UHTM_Complete.xlsx
│       └── completefile.py
│
├── UHTM_final_200x48.csv              # ★★ ML-ready CSV (root copy)
└── UHTM_final_200x48.xlsx             # ★★ Final annotated Excel (root copy)

🔄 Data Pipeline

┌─────────────────────────────────────────────────────────────────────────┐
│                                                                         │
│   STEP 1: Base.py                                                       │
│   ─────────────                                                         │
│   Krish generates UHTM_base_200.xlsx                                    │
│   200 rows × 10 cols (5 meta + 5 hidden anchors: Tₘ, ρ, vₑ, ΔEN, Pₛ)  │
│                          │                                              │
│          ┌───────────────┼────────────────┐                             │
│          ▼               ▼                ▼              ▼              │
│   STEP 2a: Aadi   2b: Krish       2c: Salan      2d: Niranjan          │
│   dataset.py      Krishh.py       Salan.py        Niranjan.py          │
│   F01–F12         F13–F24         F25–F36         F37–F48 + T1–T3      │
│   Aadi.xlsx       Krishh.xlsx     Salan_m3.xlsx   Niranjan.xlsx        │
│          │               │                │              │              │
│          └───────────────┴────────────────┴──────────────┘             │
│                                   │                                     │
│                                   ▼                                     │
│   STEP 3: mergeAll.py                                                   │
│   ────────────────────                                                  │
│   Validates 200 rows + Sample_ID match across all 4 files               │
│   Horizontally joins all feature groups                                 │
│                                   │                                     │
│                                   ▼                                     │
│   STEP 4: UHTM_final_200x48.xlsx / .csv                                 │
│   200 rows × 56 cols (5 meta + 48 features + 3 targets)                 │
│   + Summary Stats sheet + Feature Legend sheet                          │
│                                                                         │
└─────────────────────────────────────────────────────────────────────────┘

👥 Team Contributions

🔵 Aadi — Member 1: Thermodynamic & Electronic

Script: Aadi-Dev/dataset.py | Output: Aadi.xlsx | Features: F01–F12

Aadi is responsible for the foundational material properties spanning thermodynamic stability and quantum electronic structure — the two groups that most directly determine a material's suitability as a UHTM candidate.

Group A — Thermodynamic / Structural (F01–F06)

Feature	Name	Formula / Physics	Unit
F01	Melting Point	Literature anchor `Tₘ` + 2% noise	K
F02	Debye Temperature	`θ_D ≈ 300 + 0.15·Tₘ` (Debye-Grüneisen scaling)	K
F03	Cohesive Energy	`E_coh = vₑ·0.82 + ΔEN·1.1` (metallic + ionic/covalent)	eV/atom
F04	Formation Enthalpy	`ΔHf = -(ΔEN·45 + vₑ·8)` (exothermic = stable)	kJ/mol
F05	Lattice Parameter a	`a ≈ 3.2 + ρ^(-0.3)·0.5` (inverse power law with density)	Å
F06	Grüneisen Parameter	`γ ≈ 0.4 + ΔEN·0.3 + vₑ·0.05` (anharmonicity index)	—

Group B — Electronic Structure (F07–F12)

Feature	Name	Formula / Physics	Unit
F07	Band Gap	`max(0, 0.5 - vₑ·0.05)` — metallic carbides/nitrides ≈ 0	eV
F08	DOS at Fermi Level	`N(Ef) ≈ vₑ·0.55 + 1.2` (more valence e⁻ → higher DOS)	states/eV
F09	Bader Charge Transfer	`Δq ≈ ΔEN·0.6` (Pauling-type charge transfer)	e⁻
F10	Fermi Velocity	`v_F = 1×10⁶ · (vₑ/8.5)` (free-electron model)	m/s
F11	Valence Electron Density	`n = ρ·vₑ·1.8×10²²` (electrons per unit volume)	×10²²/cm³
F12	Work Function	`φ ≈ 3.5 + ΔEN·0.8 - vₑ·0.05` (surface escape energy)	eV

Physical significance: F01 is the primary UHTM selection criterion. F07 distinguishes metallic from semiconducting behaviour at high T. F11 governs metallic bonding strength and shear modulus. F12 controls thermionic emission.

🟣 Krish — Member 2: Mechanical & Thermal Transport

Scripts: Krishh.py, Base.py, mergeAll.py | Output: Krishh.xlsx | Features: F13–F24

Krish owns both mechanical integrity and thermal transport features — the two most critical property groups for structural aerospace applications. Krish also authored the base dataset generator (Base.py) and the final merge script (mergeAll.py), serving as the project's data infrastructure lead.

Group C — Mechanical Properties (F13–F18)

Feature	Name	Formula / Physics	Unit
F13	Young's Modulus	`E ≈ 200 + Tₘ·0.04 + vₑ·15` (Gilman-Cohen stiffness relation)	GPa
F14	Vickers Hardness	`H_v ≈ 15 + ΔEN·8 + vₑ·1.2` (bond character + electron count)	GPa
F15	Fracture Toughness K_Ic	`K_Ic ≈ 2.5 + 1.5/ΔEN` (ionic bonds = more brittle)	MPa√m
F16	Compressive Strength	`σ_c ≈ 0.6·E` (empirical ratio for dense ceramics)	GPa
F17	Poisson's Ratio	`ν ≈ 0.18 + ΔEN·0.02` (covalent ≈ 0.18; ionic → higher)	—
F18	Flexural Strength	`σ_f = 300 + E·0.8 + H·12 - porosity·15`	MPa

Group D — Thermal Transport (F19–F24)

Feature	Name	Formula / Physics	Unit
F19	Thermal Conductivity	`κ = 20 + vₑ·4 - ΔEN·3` (electronic + ionic scattering)	W/m·K
F20	Coeff. Thermal Expansion	`CTE = 6.5 + ΔEN·0.8 - vₑ·0.3`	×10⁻⁶/K
F21	Specific Heat Capacity	`Cp ≈ 180 + 800/ρ` (inverse density, Dulong-Petit)	J/kg·K
F22	Thermal Diffusivity	`α = κ / (ρ·Cp)` (exact thermodynamic definition)	m²/s
F23	Max Service Temperature	`T_max ≈ 0.75·Tₘ` (standard refractory engineering rule)	K
F24	Thermal Shock Resistance	`R = σ_f·κ / (E·CTE)` (Hasselman R-parameter)	W/m

Infrastructure contributions: Base.py defines all 10 material anchors from literature (Materials Project, JARVIS-DFT, Fahrenholtz & Hilmas 2012). mergeAll.py validates row counts and Sample_ID alignment before joining, preventing silent merge errors.

🔴 Salan — Member 3: Oxidation Stability & Microstructural

Script: Salan/lab evaluation/Salan.py | Output: Salan_member3.xlsx | Features: F25–F36

Salan handles the chemical stability and process-microstructure features — the groups that determine how a UHTM behaves over time in oxidising environments and how its properties are influenced by the synthesis route.

Group E — Oxidation / Chemical Stability (F25–F30)

Feature	Name	Formula / Physics	Unit
F25	Oxidation Onset Temperature	`T_ox = 800 + ΔEN·150 + Tₘ·0.05`	K
F26	Parabolic Rate Constant k_p	`k_p = 1×10⁻¹² · exp(ΔEN·0.8)` (Arrhenius)	kg²/m⁴s
F27	Oxidation Activation Energy	`Ea = 120 + ΔEN·30` (diffusion barrier)	kJ/mol
F28	Gravimetric Parabolic Rate	`k_p_grav = 1×10⁻¹⁰ · ΔEN` (TGA standard unit)	g²/cm⁴s
F29	Oxide Layer Stability Index	`OLS = ΔEN·0.7 + vₑ·0.15` (protective scale adherence)	—
F30	Oxygen Diffusivity in Oxide	`D_O = 1×10⁻¹⁴ · exp(-ΔEN·1.2)`	m²/s

Group F — Microstructural Features (F31–F36)

Feature	Name	Formula / Physics	Unit
F31	Average Grain Size	`d = 2 + (Pₛ^-0.3)·10` (Hall-Petch: smaller grains → harder)	μm
F32	Relative Density	`ρ_rel = min(99.9, 92 + Pₛ·0.18)`	%
F33	Porosity	`P = max(0.1, 100 - ρ_rel)` (complement of relative density)	%
F34	Crystallite Size (XRD)	`D_xrd = grain_size_μm · 1000 · 0.08` (Scherrer ~8% of grain)	nm
F35	Dislocation Density	`ρ_disl = 1×10¹²/d^1.5` (inverse power law with grain size)	×10¹²/m²
F36	Grain Boundary Energy	`γ_gb = 0.3 + ΔEN·0.25 + vₑ·0.02`	J/m²

Physical significance: F25 is the primary chemical stability criterion. F26 determines oxide scale growth rate — smaller k_p = better protection. F31–F33 directly link sintering pressure (Pₛ anchor) to microstructure, closing the process–property loop.

🟢 Niranjan — Member 4: Phase/Composite · ML Descriptors · Targets

Script: Niranjan/Niranjan.py | Output: Niranjan.xlsx | Features: F37–F48 + Targets: T1–T3

Niranjan handles the highest-level features: composite system descriptors, dimensionless ML merit indices designed for Pareto optimisation, and all three supervised learning targets. This is the most interdependent feature block — many features in Groups G and H re-derive quantities from Groups A–F using the same seed to ensure cross-member consistency.

Group G — Phase / Composite Features (F37–F42)

Feature	Name	Formula / Physics	Unit
F37	Phase Stability Index	`PSI = Tₘ/4000 + E_coh/12` (normalised thermodynamic stability)	—
F38	Secondary Phase Vol. Fraction	0% monolithic; 5–20% composites (index-dependent)	%
F39	Interfacial Energy	`γ_int = 0.5 + ΔEN·0.3` (ionic mismatch → delamination risk)	J/m²
F40	CTE Mismatch Index	`ΔCTE =	CTE - 5.0
F41	Solid Solution Distortion δ	`δ = ΔEN·0.12 + (vₑ%3)·0.05` (Hume-Rothery lattice distortion)	—
F42	Wettability Index	`W = 1 - f_ion = vₑ·0.15/(ΔEN + vₑ·0.15)` (covalent fraction)	—

Group H — ML Descriptor Features (F43–F48)

Feature	Name	Formula / Physics	Unit
F43	Thermal Merit Index	`TMI = κ/(CTE·ρ)` (Ashby figure of merit for TPS panels)	W/kg
F44	Toughness-Stiffness Index	`TSI = K_Ic·√E` (combined crack resistance and stiffness)	GPa·√GPa
F45	Oxidation Merit Score	`OMS = T_ox/(k_p_norm + 1)` (Bayesian optimisation reward signal)	—
F46	Bond Ionicity Fraction	`f_ion = ΔEN/(ΔEN + vₑ·0.15)` (Pauling ionicity)	—
F47	Structural Stability Index	`SSI = (Tₘ/4000)·(E_coh/10)·(1 - P/100)`	—
F48	Creep Resistance Parameter	`CR = (Tₘ/3000)·(E/400)·(1/d)^0.3` (grain size + stiffness)	—

🎯 Target Variables

All three targets are defined and computed by Niranjan (Member 4).

Target	Name	Type	Description
T1	Flexural Strength	Continuous (MPa)	Multi-factor structural target. Driven by Young's Modulus, Vickers Hardness, and Porosity. Primary design metric for load-bearing structures.
T2	Oxidation Resistance Score	Continuous (0–10)	Composite weighted score: onset temperature (×3) + rate constant (×2) + stability index (×2) + CTE match (×1).
T3	Thermal Shock Cycles	Integer (cycles)	Predicted cycles-to-failure under rapid thermal cycling. Driven by K_Ic, CTE, and composite mismatch index F40.

🗃 Feature Groups at a Glance

Group	Features	Member	Domain	Colour (in Excel)
A — Thermodynamic	F01–F06	Aadi	Phase stability, bonding energetics	Teal
B — Electronic	F07–F12	Aadi	DFT / quantum properties	Blue
C — Mechanical	F13–F18	Krish	Structural integrity	Purple
D — Thermal Transport	F19–F24	Krish	Heat transport & diffusion	Orange
E — Oxidation	F25–F30	Salan	Chemical stability	Red
F — Microstructural	F31–F36	Salan	Process–structure–property	Green
G — Phase/Composite	F37–F42	Niranjan	Multi-phase composite systems	Indigo
H — ML Descriptors	F43–F48	Niranjan	Pareto / reward signals for inverse design	Teal-dark
Targets	T1–T3	Niranjan	Supervised regression targets	Dark Red

🧱 Material Classes

Base anchor values sourced from: Materials Project (mp-*), JARVIS-DFT, Fahrenholtz & Hilmas (2012), Cedillos-Barraza et al. (2016), Opeka et al. J. Eur. Ceram. Soc.

Material	Crystal	T_m (K)	ρ (g/cm³)	v_e	ΔEN
HfC	FCC	3900	12.20	8	1.3
ZrC	FCC	3420	6.73	8	1.3
TaC	FCC	3880	14.30	9	1.1
HfB₂	HEX	3380	10.50	6	0.9
ZrB₂	HEX	3245	6.09	6	0.9
TiC	FCC	3160	4.93	8	1.5
NbC	FCC	3600	7.79	9	1.2
HfN	FCC	3385	13.80	9	1.6
ZrN	FCC	2980	7.09	9	1.6
TaN	HEX	3090	16.30	10	1.4

Composite variants (experimental, index 50–99): HfC-SiC, ZrB₂-SiC, HfB₂-SiC, TaC-HfC, ZrC-TiC, HfC-TaC, ZrB₂-ZrC, HfB₂-MoSi₂, TiC-TiB₂, NbC-HfC

Doped variants (synthetic, index 150–199): HfC:Y, ZrC:La, TaC:W, HfB₂:Al, ZrB₂:Y, TiC:Nb, NbC:Ta, HfN:Zr, ZrN:Hf, TaN:Nb

⚙️ Setup & Usage

Prerequisites

pip install pandas numpy openpyxl

Step 1 — Generate the base file (run once)

# From Krish/LAB EVALUATION/Merge/
python Base.py
# Output: UHTM_base_200.xlsx
# Contains: 200 rows × 10 cols (5 metadata + 5 physical anchors)

Step 2 — Each member generates their features independently

# Member 1 — Aadi
cd Aadi-Dev/
python dataset.py
# Output: Aadi.xlsx (200 × 17)

# Member 2 — Krish
cd "Krish/LAB EVALUATION/"
python Krishh.py
# Output: Krishh.xlsx (200 × 17)

# Member 3 — Salan
cd "Salan/lab evaluation/"
python Salan.py
# Output: Salan_member3.xlsx (200 × 17)

# Member 4 — Niranjan
cd Niranjan/
python Niranjan.py
# Output: Niranjan.xlsx (200 × 20: 12 features + 3 targets + 5 meta)

Step 3 — Merge all outputs

# Copy all member xlsx files into Krish/LAB EVALUATION/Merge/
# Rename if needed: AadiDev.xlsx, Krishh.xlsx, Salan.xlsx, Niranjan.xlsx

cd "Krish/LAB EVALUATION/Merge/"
python mergeAll.py
# Validates: 200 rows + Sample_ID match across all 4 files
# Output: UHTM_final_200x48.xlsx
#   Sheet 1 — UHTM_Full_Dataset  (200 × 56, colour-coded by group)
#   Sheet 2 — Summary Stats       (describe() for all F and T columns)
#   Sheet 3 — Feature Legend       (group → member → domain mapping)

Reproducibility

All scripts use np.random.seed(42) at the top level. As long as the base file is generated first and member scripts are run with the same seed, all outputs are deterministic.

📦 Output Files

File	Location	Description
`UHTM_base_200.xlsx`	`Krish/LAB EVALUATION/Merge/`	Base dataset with material anchors. Input to all 4 member scripts.
`Aadi.xlsx`	`Aadi-Dev/`	Member 1 output: F01–F12
`Krishh.xlsx`	`Krish/LAB EVALUATION/`	Member 2 output: F13–F24
`Salan_member3.xlsx`	`Salan/lab evaluation/`	Member 3 output: F25–F36
`Niranjan.xlsx`	`Niranjan/`	Member 4 output: F37–F48 + T1–T3
`UHTM_final_200x48.xlsx`	Root + `Merge/`	★ Final merged dataset with summary and legend sheets
`UHTM_final_200x48.csv`	Root	★ ML-ready flat CSV export

👤 Authors

Member	Features	Scripts
Aadi	F01–F12 (Thermodynamic + Electronic)	`Aadi-Dev/dataset.py`
Krish	F13–F24 (Mechanical + Thermal) + Base + Merge	`Krishh.py`, `Base.py`, `mergeAll.py`
Salan	F25–F36 (Oxidation + Microstructural)	`Salan/lab evaluation/Salan.py`
Niranjan	F37–F48 (Phase + ML Descriptors) + T1–T3	`Niranjan/Niranjan.py`

IMI Project · 2026 · Ultra-High Temperature Materials Dataset

Name		Name	Last commit message	Last commit date
Latest commit History 63 Commits
Aadi-Dev		Aadi-Dev
DatasetCuration		DatasetCuration
Krish		Krish
Literature		Literature
Niranjan		Niranjan
Salan		Salan
.gitignore		.gitignore
LAB Evaluation Explaination.pdf		LAB Evaluation Explaination.pdf
README.md		README.md
UHTM_final_200x48.csv		UHTM_final_200x48.csv
UHTM_final_200x48.xlsx		UHTM_final_200x48.xlsx
index.html		index.html

Folders and files

Latest commit

History

Repository files navigation