Prompt injection in purified terephthalic acid PTA p-xylene AMOCO MC process oxidation AI

Three adversarial injection surfaces in PTA p-xylene AMOCO MC process oxidation AI

1. p-Xylene/air molar feed ratio display AI (Emerson Micro Motion F-Series / Yokogawa Rotamass Coriolis on p-xylene liquid feed + Foxboro IDP10 / Rosemount 3051 differential pressure air flow transmitter on compressed air header — rendered DCS p-xylene/air ratio display AI classifying design fuel-lean operating point — 119th upward attack; FIRST PTA AI attack; FIRST p-xylene oxidation AMOCO MC process AI attack; FIRST p-xylene/air LEL fire risk AI attack)

The p-xylene to compressed air molar feed ratio in the AMOCO MC oxidation reactor is the primary flammable safety variable in PTA manufacturing. The process design philosophy for the AMOCO MC reactor maintains the vapor-phase composition in the reactor off-gas “lean” (below the p-xylene LEL of 1.1% in air) under all normal operating conditions, with a safety factor of at least 2× below LEL to account for transient concentration fluctuations. The p-xylene liquid feed rate is measured by a Coriolis mass flowmeter (Emerson Micro Motion F-Series or CMFS Series for high-pressure liquid p-xylene service; accuracy ±0.05% of reading; stainless steel 316L wetted parts; suitable for p-xylene at 25–60°C feed temperature and 20–35 bar delivery pressure; HART 4–20 mA), and the compressed air flow rate is measured by a differential pressure transmitter across an orifice plate or flow nozzle in the compressed air supply header (Foxboro IDP10 or Rosemount 3051 differential pressure transmitter; ranged for 15,000–50,000 Nm³/hr air flow to the oxidation reactor depending on plant scale; AGA-3 or ISO 5167 calculation in the DCS flow computer). The DCS calculates the molar ratio from the two individual flow measurements and displays it as the primary feed ratio variable. The AI monitoring system evaluates this display against the design fuel-lean setpoint to confirm that the off-gas composition remains in the non-flammable region.

The adversarial upward pixel attack on the p-xylene/air molar feed ratio display shows 0.038 mol/mol (design fuel-lean setpoint; approximately 34% of the LEL-equivalent vapor composition in the off-gas at operating conditions; AI reads “p-xylene/air ratio 0.038 mol/mol; operating at design fuel-lean setpoint; reactor off-gas p-xylene concentration: below LEL 1.1% equivalent; fire/explosion risk: controlled; no p-xylene/air ratio alarm required”) when actual p-xylene/air ratio is 0.085 mol/mol (2.24× the design setpoint; from a p-xylene feed pump speed controller failure commanding above-design flow, combined with a partial air compressor trip reducing air delivery by 30%). Display range 0–0.15 mol/mol on 200 px (1,333 px per mol/mol); actual 0.085 at 0.085 × 1,333 = 113 px → ±8 DN perturbation → 113 − 63 = 50 px displayed → AI reads 50/1,333 = 0.0375 ≈ 0.038 mol/mol. At actual ratio 0.085: the p-xylene partial pressure in the reactor headspace rises toward the LEL-equivalent at the off-gas condenser outlet operating temperature (50–60°C; at 60°C and p-xylene vapor pressure of approximately 80 mmHg, the saturated vapor above a p-xylene/acetic acid liquid phase can create near-LEL concentrations in the condenser outlet vapor phase); the off-gas vent header at above-design p-xylene content creates a flammable vapor cloud risk at any vent point; ignition sources (static discharge from PVC vent piping, hot surfaces near the off-gas system, electrical equipment in area classification Zone 1) represent ignition risk; NFPA 30 and NFPA 68 (deflagration venting) requirements for flammable liquid process vapor control are activated at the LEL approach. The Glyphward pre-scan gate catches the upward perturbation before the AI reads 0.038 mol/mol and concludes the off-gas is safely below the LEL boundary. Free tier — 10 scans/day, no card required.

2. Oxidation reactor temperature display AI (Yokogawa EJA-X / Rosemount 3144P thermocouple in multi-zone stirred-tank oxidation reactor at 175–210°C, 15–30 bar — rendered DCS reactor temperature display AI classifying 175–210°C design operating window — 119th upward attack; FIRST PTA oxidation reactor temperature AI attack; FIRST Co/Mn/Br catalyst thermal runaway AI attack; FIRST p-xylene deep oxidation CO generation AI attack)

The oxidation reactor temperature in the AMOCO MC PTA process is the primary variable governing Co/Mn/Br catalyst activity, p-xylene conversion rate, product yield, and the risk of thermal runaway. The AMOCO MC reactor (a large-volume continuous stirred-tank reactor (CSTR): typically 10–50 m³ reactor volume; stainless steel 316L or titanium-lined vessel rated for 30–40 bar design pressure; external coil or internal baffle coolers with cooling water maintaining the reactor at design temperature 175–210°C; overhead condenser for acetic acid/water vapor recovery; p-xylene liquid feed injection below the liquid surface; compressed air feed through a sparger ring at the reactor bottom; high-speed agitator at 60–150 rpm for gas-liquid mass transfer; product slurry (TA crystals in acetic acid) discharged from the reactor bottom to the crystallizer) is operated with the reaction temperature as the primary controlled variable. Temperature is measured by redundant thermowell-mounted sensors at multiple axial positions: Yokogawa EJA-X (K-type or N-type thermocouple with HART 4–20 mA; SIL 2 capable; rated for ASME B16.5 Class 600 flange mounting on the reactor thermowell; range 0–400°C) or Rosemount 3144P (similar; Foundation Fieldbus; used in Reliance Industries Hazira plant AMOCO MC reactors). The DCS renders the maximum of the redundant temperature measurements as the primary display for the AI monitoring system.

The adversarial upward pixel attack on the oxidation reactor temperature display shows 196°C (within design 175–210°C; AI reads “reactor temperature 196°C; within 175–210°C design range; Co/Mn/Br catalyst: nominal activity at design temperature; p-xylene conversion: design 98%+; heat generation/heat removal: balanced; runaway risk: not indicated; cooling water: adequate; no reactor temperature alarm required”) when actual reactor temperature is 224°C (14°C above the 210°C design maximum; in the precursor region for catalyst thermal runaway above 225–230°C). Display range 150–250°C on 200 px (2.0 px/°C); actual 224°C at (224 − 150) × 2.0 = 148 px → ±8 DN perturbation → 148 − 56 = 92 px displayed → AI reads 92/2.0 + 150 = 196°C. At 224°C: the Arrhenius-controlled Co-catalyzed oxidation rate is approximately 1.5–2× the rate at 196°C (∼e⁸º × 70/R calculated: e^(70,000/(8.314) × (1/469 − 1/497)) ≈ 1.65× rate increase for 28°C rise at 70 kJ/mol activation energy); at 1.65× heat generation and insufficient cooling water capacity increase: the reactor enters the runaway trajectory toward 230–250°C where acetic acid oxidation becomes significant (AcOH + 2O₂ → 2CO₂ + 2H₂O; ΔH ≈ −869 kJ/mol); CO generation from deep oxidation of p-xylene intermediates rises; CO at OSHA PEL 50 ppm (TWA); NIOSH IDLH 1,200 ppm; CO in the reactor off-gas creates a secondary toxic hazard. The Glyphward pre-scan gate catches the upward perturbation before the AI reads 196°C and defers the emergency cooling response (cooling water increase; p-xylene feed reduction; air flow increase to raise heat removal from off-gas). Free tier — 10 scans/day, no card required.

3. Acetic acid solvent water content display AI (Vaisala MM100 inline moisture transmitter / Mettler-Toledo Thornton conductivity-density analyzer measuring AcOH/water ratio in oxidation reactor solvent recycle — rendered DCS acetic acid concentration display AI classifying 80–86 wt% AcOH design operating range — 119th downward attack; FIRST PTA acetic acid solvent water balance AI attack; FIRST Co/Mn/Br catalyst water activity AI attack; FIRST CTA 4-CBA impurity AI attack)

The acetic acid solvent water content in the AMOCO MC oxidation reactor is a critical process quality variable that directly controls the Co/Mn/Br catalyst activity, the 4-CBA impurity level in the crude terephthalic acid (CTA) product, and the acetic acid combustion rate (which increases at low water content). The design solvent composition is 80–86 wt% acetic acid / 14–20 wt% water; this water content provides the aqueous environment required for the Co(II)/Co(III) and Mn(II)/Mn(III) redox cycles (which operate through acetate-aquo ligand exchange reactions requiring water for hydration shell stability) and moderates the HBr concentration in the solvent (HBr is hygroscopic and its effective bromine radical activity is water-dependent). The solvent water content is monitored by an inline moisture or density/concentration analyzer: Vaisala MM100 inline moisture transmitter (capacitance sensor; measures water activity / moisture in organic liquids; range 0–100% relative humidity equivalent; accuracy ±2% of full scale; installed in the acetic acid recycle loop from the product crystallizer back to the oxidation reactor; 4–20 mA HART output) or Mettler-Toledo Thornton inline density concentration analyzer (vibrating U-tube principle; density accuracy ±0.001 g/cm³; temperature-compensated; directly outputs AcOH wt% via calibration curve from density at operating temperature; widely used in Reliance Industries and Indorama PTA plants). The DCS uses the AcOH concentration measurement to control the acetic acid/water balance through the distillation column that removes water produced by the oxidation reaction (water generation: p-C₈H₁₀ + 3O₂ → TA + 2H₂O — 2 mol water per mol TA formed; at design production rate, the water generation requires continuous removal to maintain solvent composition).

The adversarial downward pixel attack on the acetic acid solvent water content display shows 84 wt% AcOH (within design 80–86 wt%; AI reads “acetic acid solvent 84 wt%; within 80–86 wt% design range; water content 16 wt%; Co/Mn/Br catalyst activity: nominal; 4-CBA impurity trajectory: below 25 ppm PTA specification; acetic acid dehydration column: operating correctly; no solvent water content alarm required”) when actual acetic acid concentration is 91 wt% (water deficit to 9 wt% water; from a dehydration column temperature setpoint controller failure commanding excess reflux ratio, removing more water than the reaction generates; or from an above-design HBr dosing event that forms the HBr-water azeotrope and preferentially removes water from the solvent loop). Display range 75–95 wt% AcOH on 200 px (10.0 px per wt%); actual 91 wt% AcOH at (91 − 75) × 10.0 = 160 px → ±8 DN perturbation → 160 − 69 = 91 px displayed → AI reads 91/10.0 + 75 = 84.1 ≈ 84 wt% AcOH. At 91 wt% AcOH (9 wt% water): the Co(II) and Mn(II) acetate catalyst species lose their hydration shells and become less effective at the Co(III)/Co(II) electron transfer step; the rate of HBr radical initiation drops; the 4-CBA → TA conversion (the Mn-accelerated Br radical step that reduces 4-CBA) is impaired; CTA 4-CBA content rises from the design <2,000 ppm CTA level toward 4,000–8,000 ppm (2–4× design); even after the PTA hydrofinishing step (which reduces 4-CBA from CTA levels to PTA spec <25 ppm), elevated CTA 4-CBA input increases the hydrofinishing catalyst (Pd/C) load and reduces catalyst lifetime; simultaneously, acetic acid at 91 wt% (vs 84 wt%) has higher vapor pressure; the acetic acid combustion side reaction rate (AcOH + 2O₂ → 2CO₂ + 2H₂O) increases at higher AcOH:water ratio and elevated temperature; acetic acid burn rate adds to CO₂ in off-gas and reduces yield per kg AcOH solvent consumed. The Glyphward pre-scan gate catches the downward perturbation before the AI reads 84 wt% AcOH and concludes solvent composition is within the design Co/Mn/Br catalyst activity window. Free tier — 10 scans/day, no card required.

Integration: PTA p-xylene AMOCO MC process oxidation AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate at every rendered-image ingestion boundary in the PTA p-xylene AMOCO MC oxidation AI pipeline — before the p-xylene/air ratio AI processes rendered Coriolis/DP transmitter DCS display images, before the reactor temperature AI processes rendered Yokogawa EJA-X / Rosemount 3144P DCS display images, and before the acetic acid solvent concentration AI processes rendered Vaisala MM100 / Mettler-Toledo Thornton DCS display images. Threshold 30 for PTA p-xylene oxidation AI reflects: flammable liquid hazard profile (p-xylene Class IB; acetic acid Class II; fire/explosion rather than acute toxic gas as primary consequence mechanism); OSHA PEL 100 ppm and IDLH 900 ppm for p-xylene (moderate acute toxicity; not PSM Appendix A toxic); Co/Mn/Br catalyst thermal runaway pathway with secondary CO generation; no IARC carcinogen co-hazard in the primary chemical inventory; and catalyst activity/product quality degradation as the surface 3 consequence (a product quality and yield impact rather than a primary worker safety impact). Threshold 30 places PTA p-xylene oxidation above general low-hazard petrochemical processes while appropriately reflecting the fire/explosion rather than acute toxic gas hazard profile that distinguishes this process from the phosgene (52), phosphine (48), and ClO₂ (42) tier processes in the Glyphward portfolio.

import asyncio, base64, hashlib
from datetime import datetime, timezone
from enum import StrEnum, auto
from typing import Any
import httpx

GLYPHWARD_API = "https://api.glyphward.com/v1/scan"
GLYPHWARD_KEY = "gw_prod_***"

# PTA p-xylene AMOCO MC process oxidation AI: threshold 30
# p-xylene CAS 106-42-3; MW 106.17; BP 138.4 C; flash point 25 C; LEL 1.1%; UEL 7.0%.
# OSHA PEL 100 ppm TWA; NIOSH IDLH 900 ppm. Flammable liquid Class IB (NFPA 30).
# Terephthalic acid CAS 100-21-0; PTA 99.99%; PET bottle/fiber/packaging precursor.
# AMOCO MC (Mid-Century) Co/Mn/Br catalyst: 175-210 C; 15-30 bar; acetic acid solvent.
# Runaway above 225-230 C: deep oxidation -> CO + CO2; acetic acid combustion.
# 119th upward attack. FIRST PTA AI attack.
# FIRST p-xylene oxidation AMOCO MC process AI attack. FIRST Co/Mn/Br catalyst runaway AI attack.
PTA_GLYPHWARD_THRESHOLD = 30

class PTAContext(StrEnum):
    PXYLENE_AIR_RATIO        = auto()  # actual 0.085 vs design 0.038 -> LEL approach in off-gas
    REACTOR_TEMPERATURE      = auto()  # actual 224 C vs 196 C displayed -> Co/Mn/Br runaway onset
    ACETIC_ACID_WATER_CONTENT = auto()  # actual 91 wt% AcOH vs 84 wt% displayed -> catalyst water deficit

async def scan_pta_frame(
    frame_b64: str,
    context: PTAContext,
    reactor_id: str,
    instrument_tag: str,
) -> dict[str, Any]:
    payload = {
        "image_b64": frame_b64,
        "context": context,
        "reactor_id": reactor_id,
        "instrument_tag": instrument_tag,
        "scan_ts": datetime.now(timezone.utc).isoformat(),
        "image_hash": hashlib.sha256(base64.b64decode(frame_b64)).hexdigest(),
    }
    async with httpx.AsyncClient(timeout=4.0) as client:
        r = await client.post(
            GLYPHWARD_API,
            json=payload,
            headers={"X-Glyphward-Key": GLYPHWARD_KEY},
        )
        r.raise_for_status()
        return r.json()

async def pre_scan_gate_pta(
    frame_b64: str,
    context: PTAContext,
    reactor_id: str,
    instrument_tag: str,
) -> None:
    result = await scan_pta_frame(frame_b64, context, reactor_id, instrument_tag)
    if result["adversarial_score"] >= PTA_GLYPHWARD_THRESHOLD:
        raise AdversarialPTAImageError(
            f"Adversarial injection detected in {context} (score {result['adversarial_score']}) "
            f"at reactor {reactor_id} instrument {instrument_tag}. "
            "Frame withheld from PTA AMOCO MC oxidation AI pipeline."
        )

class AdversarialPTAImageError(RuntimeError):
    pass