AIMM Index — Ispat Godavari, Raipur

Command Centre — Ispat Godavari Ltd., Raipur

Three AIMM compliance twins monitoring the full iron ore beneficiation-to-sponge-iron production chain

PRODUCTION CHAIN — AIMM TWIN ARCHITECTURE

UPSTREAM INPUT

Iron Ore Fines

Fe₂O₃ feed, 11.97% Ilmenite

→

TWIN 1 — HTRS

67%

Electrostatic separation

L3 — Managed

→

TWIN 2 — RED

74%

Magnetic separation

L4 — Optimised

→

TWIN 3 — KILN

71%

Rotary kiln DRI

L3 — Managed

→

OUTPUT

Sponge Iron

92% metallization

AI$M = 100 × √(ACI × SIR) — computed on every telemetry event, propagated upstream-to-downstream

⚡

HTRS AI$M

67%

L3 — Managed

🧲

RED AI$M

74%

L4 — Optimised

🔥

Kiln AI$M

71%

L3 — Managed

⚠

Open Alerts

3

2 warning · 1 info

📊

Telemetry Events

30

Real CCD data

ACTIVE AGENT FINDINGS

HTRS_RECOVERY_DEFICIT

HTRS achieving 52.3% recovery vs RED's 64.24% — 11.9% gap. Upstream ore quality limited by electrostatic constraint.

→ Consider switching primary separation to RED for higher-Fe ore batches.

HTRS_SVR_DRIFT_DETECTED

SVR model deployed for recovery prediction — LOOCV R²=0.049 (vs in-sample 0.9999). Catastrophic generalisation failure.

→ RSM model (R²=0.9887) should be used instead. This is what AIMM catches.

KILN_CONTINUITY_LOW

Rotary kiln continuity score 0.61 — below the 0.65 manufacturing floor. Unplanned downtime risk.

→ Cross-check kiln maintenance log vs planned production schedule.

SPONGE IRON KILN — LIVE METRICS

Metallization Rate

92.4%

Target: ≥90%

Total Fe Content

91.8%

Premium grade

Kiln Temperature

978°C

Range: 950-1050°C

Coal Rate

1.72

T coal / T DRI

AI$M TRAJECTORY — ALL THREE TWINS

● HTRS Separator ● RED Separator ● Sponge Iron Kiln

TWIN 1 — HTRS SEPARATOR · L3-MANAGED

High Tension Roll Separator

Electrostatic separation of iron ore fines — Ispat Godavari upstream beneficiation

67%

AI$M Score

AI$M = 100 × √(ACI × SIR) = 100 × √(0.5187 × 0.8691) = 67.1% → L3 (Managed)

ACI BREAKDOWN (Accuracy/Consistency Index)

True Positives (high-quality ore correctly passed)

482

True Negatives (subgrade ore correctly rejected)

438

False Positives (subgrade passed as high-quality)

→ reduces kiln efficiency

68

False Negatives (high-quality rejected)

→ yield loss

52

Drift (D) — process parameter drift from baseline

0.07 (7%)

ACI = 0.5187 — affected by FP rate. SVR model (deployed before AIMM) had LOOCV R²=0.049 — this confusion matrix shows why.

SIR BREAKDOWN (System Integrity & Resilience)

Encryption Score

Sensor data security

0.88

Integration Score

SCADA + MES connectivity

0.84

Modularity Score

Separator control architecture

0.78

Continuity Score (Uptime)

HTRS operational availability

0.91

SIR = 0.8691 — strong uptime (91%) but modularity could be improved for better fault isolation.

HTRS EXPERIMENTAL DATA — 30-RUN CCD (REAL, from Ilmenite Recovery Study)

${htrs_table_rows}

Run	Feed Rate (tph)	Feed Roll (rpm)	Sep. Roll (rpm)	Temp (°C)	Splitter (°)	Voltage (kV)	Recovery (obs.%)	AIMM Status

HTRS RECOVERY DISTRIBUTION

DOMAIN AGENT FINDINGS — HTRS

FAB_YIELD_VARIANCE

Yield variance (FP+FN)/total = 11.7% — above 5% fab tolerance.

→ Investigate false-pass rate; high FP degrades kiln feed quality.

HTRS_SVR_MODEL_DRIFT

SVR LOOCV R²=0.049 vs in-sample 0.9999. Model is overfitted — production predictions unreliable.

→ Deploy RSM model (R²=0.9887) instead. Switch model, then recalibrate twin.

HTRS_OPTIMAL_CONDITIONS

Best recovery (52.3%): 0.20 tph · 25 rpm · 165 rpm · 110°C · 60° · 24 kV. Current run deviates from optimal.

→ Adjust voltage to 24kV and splitter to 60° to approach optimal operating point.

TWIN 2 — RED SEPARATOR · L4-OPTIMISED

Rare-Earth Drum Magnetic Separator

Magnetic separation — higher recovery, complementary to HTRS

74%

AI$M Score

AI$M = 100 × √(ACI × SIR) = 100 × √(0.6312 × 0.8748) = 74.3% → L4 (Optimised)

CRITICAL OPERATING VALUES (Confirmed)

Feed Rate

0.22 tph

Drum Speed

75 rpm

Feed Vibration

70 rpm

Temperature

65°C

Splitter Position

60°

RECOVERY PERFORMANCE

64.24%

Ilmenite Recovery (observed)

91.51%

Product Grade

RSM predicted: 65.02% · Confirmatory trial: 64.32% — model validated

RED vs HTRS COMPARISON

Recovery

RED: 64.24% vs HTRS: 52.3%

Grade

RED: 91.51% vs HTRS: 90.4%

AI$M Score

RED: 74% vs HTRS: 67%

AIMM Level

L4 vs L3

Recommended

RED Primary

DOMAIN AGENT FINDINGS — RED (Semiconductor/Manufacturing Agent)

RED_RECOVERY_SUPERIOR

RED achieves 64.24% recovery vs 52.3% HTRS — 11.9% advantage. AI$M reflects this: L4 vs L3.

→ Recommend RED as primary separator for Fe recovery maximization.

CALIBRATION_COMPLETE

RED twin is calibrated. Baseline AI$M = 74.3%. Drift monitoring active.

→ Maintain current operating conditions within confirmed critical values.

AIMM L4 CRITERIA MET

Continuous monitoring

✓

Baseline calibrated

✓

Model validated (RSM)

✓

Critical conditions confirmed

✓

Agent drift monitoring

✓

TWIN 3 — ROTARY KILN · L3-MANAGED

Sponge Iron (DRI) Rotary Kiln

Direct Reduced Iron production — quality depends on upstream HTRS/RED ore beneficiation

71%

AI$M Score

AI$M = 100 × √(ACI × SIR) = 100 × √(0.5680 × 0.8900) = 71.1% → L3 (Managed)

↑ Upstream dependency: The Kiln twin AI$M is partially bounded by HTRS (67%) upstream. Improving HTRS recovery from 52.3% to RED-equivalent 64.24% would elevate kiln ACI and lift Kiln AI$M from 71% toward 76%.

METALLIZATION

92.4%

Target ≥90% ✓

TOTAL Fe

91.8%

Premium grade

KILN TEMP

978°C

950-1050°C range

KILN UPTIME

61%

Below 65% floor ⚠

KILN PROCESS PARAMETERS

Coal consumption

1.72 T/T DRI

Kiln speed

0.48 RPM

Air injection rate

4,200 Nm³/hr

Discharge temperature

1,010°C

Carbon in product

0.18%

Sulphur content

0.02%

Production rate

142 TPD

AGENT FINDINGS — KILN

KILN_CONTINUITY_LOW

Continuity score 0.61 — below 0.65 manufacturing floor. Uptime not meeting baseline.

→ Correlate with maintenance log; likely scheduled downtime — update SIR inputs after next maintenance window.

KILN_UPSTREAM_CONSTRAINT

Kiln ACI bounded by HTRS feed quality. 11.7% FP rate from HTRS enters kiln as subgrade ore.

→ Switching to RED as primary separator would improve feed quality and lift kiln AI$M ~5 points.

Growth Comparison — Three AIMM Twins

Geometric trajectories of all three twins plotted on the same L0–L6 scale. Real CCD data used for HTRS/RED history.

AI$M TRAJECTORIES — HTRS vs RED vs KILN (30-POINT HISTORY)● Live

● HTRS Twin (AI$M 67%) ● RED Twin (AI$M 74%) ● Kiln Twin (AI$M 71%)

Data points derived from the 30-run CCD matrix (Ilmenite Recovery study, Andhra Pradesh plant). Each run maps to a telemetry event; AI$M recomputed after each one. RED consistently outperforms HTRS — the AIMM growth chart makes this visible from run 1.

HTRS GROWTH

Start AI$M

28%

Peak AI$M

69%

Current AI$M

67%

Level

L3

Trend

↗ Improving

RED GROWTH

Start AI$M

31%

Peak AI$M

76%

Current AI$M

74%

Level

L4

Trend

↗ Optimised

KILN GROWTH

Start AI$M

32%

Peak AI$M

73%

Current AI$M

71%

Level

L3→L4

Trend

↗ Progressing

Submit Telemetry — Live Demo

Submit a real telemetry value and watch AI$M recompute in real time. This simulates the POST /api/twin/telemetry endpoint.

SUBMIT TELEMETRY EVENT

Select Twin

Metric Type

Value

RESULT

Submit a telemetry event to see the live AI$M recomputation.

TELEMETRY HISTORY (SESSION)

No events this session yet.

Compliance Annexure

Generated live from twin state — in the live platform this downloads as a .docx. Preview shown here for the NRDC demo.

Click a button above to preview the live Compliance Annexure for that twin.

Domain Agent — All Findings

Manufacturing agent active for HTRS/RED (Semiconductor/Fab mapping). Manufacturing agent for Kiln. All findings derived from real HTRS/RED data.

Active Agent: Manufacturing / Semiconductor Fab Selected for domain: "Sponge Iron / Mineral Processing"

HTRS_SVR_MODEL_DRIFT — CRITICAL HTRS Twin

SVR model deployed for Ilmenite recovery prediction: in-sample R²=0.9999 (appears perfect), LOOCV R²=0.049 (catastrophic failure). This model is driving production decisions with effectively random predictions outside training data.

→ Immediate action: Replace SVR with RSM model (R²=0.9887, LOOCV validated). This is exactly the gap AIMM was built to catch — audits never see this because the in-sample report shows 0.9999.

FAB_YIELD_VARIANCE — WARNING HTRS Twin

HTRS yield variance (FP+FN)/total = 11.7%, above the 5% manufacturing tolerance. 68 false-positives pass subgrade ore as high-grade, reducing kiln feed quality. 52 false-negatives represent recoverable yield loss.

→ Adjust voltage and splitter position toward optimal (24kV, 60°). This reduces FP rate. Cross-check with RED recovery: switching primary separation to RED reduces overall FP exposure.

KILN_CONTINUITY_LOW — WARNING Kiln Twin

Rotary kiln continuity score (SIR Cr) = 0.61, below the 0.65 manufacturing floor for process continuity. This represents unplanned downtime risk and is constraining the Kiln AI$M below L4.

→ Correlate with maintenance log. If this reflects planned outages, update SIR inputs post-maintenance to re-establish baseline. If unplanned, initiate root-cause investigation on kiln mechanical condition.

RED_SUPERIOR_RECOVERY — INFO RED Twin

RED separator consistently achieves 64.24% Ilmenite recovery vs HTRS 52.3% — a 11.9% advantage on the same feed. RED product grade (91.51%) also marginally exceeds HTRS (90.4%).

→ For Fe-priority production runs, consider RED as primary separator. Combined HTRS→RED circuit could yield further gains (electrostatic selectivity + magnetic recovery).

KILN_UPSTREAM_QUALITY_DEPENDENCY — INFO Kiln Twin

Kiln ACI score (0.568) is partially bounded by HTRS feed quality. The 11.7% false-positive rate from HTRS passes subgrade ore through to the kiln, elevating kiln FP inputs and depressing ACI. Improving HTRS → improves Kiln AI$M automatically.

→ This is the cascade effect of linked twins. Fixing HTRS (Switch to RED primary + RSM model) propagates improvement upstream-to-downstream without any separate kiln intervention.

About This Demo

WHAT IS LIVE vs SIMULATED IN THIS DEMO

HTRS/RED recovery data

30-run CCD matrix, RSM/ANOVA results, SHAP analysis

REAL PAPER DATA

AI$M formula computation

Running in your browser right now from the same JS as the live engine

REAL ENGINE

Domain agent findings

Manufacturing/Fab agent rules applied to real HTRS/RED metrics

REAL AGENT

Ispat Godavari plant identity

Named as demonstration example — survey pending NDA

ILLUSTRATIVE

Kiln metrics (temp, metallization, coal rate)

Typical values for an Indian RKSI plant of this capacity

REPRESENTATIVE

Telemetry submit demo

Client-side simulation — in live platform calls POST /api/twin/telemetry

SIMULATED

TRL EVIDENCE THIS DEMO DEMONSTRATES

AI$M formula running live, in a browser, on real industrial data

TRL 4

Domain agent (manufacturing) activating correct findings from real CCD data

TRL 4

Three linked twins with cascade dependency modelled

TRL 4

Annexure generation (live, from twin state)

TRL 4

Named industrial partner (Ispat Godavari) — survey in planning

TRL 4→5

CONTACT & LIVE PLATFORM

Platform: aimmindex.com

API: api.aimmindex.com/health

Contact: Dr. Partha Sarathy Iyengar · partha@is360technologies.com

IS360 Technologies & Services Pvt. Ltd. · Chennai, India

Patent: 202641007225 filed 24/01/2026

AIMM Level Compliance Annexure