Prompt injection in propane dehydrogenation PDH CATOFIN Oleflex propylene catalyst regeneration AI

Three adversarial injection surfaces in propane dehydrogenation PDH CATOFIN Oleflex AI

1. Catalyst regeneration O₂ concentration display AI (Yokogawa ZR22G in-situ zirconia O₂ analyser regeneration gas display AI / ABB EL3020 paramagnetic O₂ analyser regeneration phase display AI / Emerson X-STREAM O₂ analyser CATOFIN regeneration SCADA display AI / Siemens Oxymat 6E O₂ regeneration concentration display AI / Honeywell Analytics XNXUTD O₂ CATOFIN regeneration display AI — rendered SCADA catalyst regeneration O₂ concentration display AI classifying the vol% oxygen in the regeneration gas fed to the CATOFIN fixed-bed reactor during the catalyst coke-combustion phase against the design operating range of 1.0–1.5 vol% O₂ in N₂ — sufficient for complete coke combustion within the 10–15 minute regeneration cycle without creating catalyst-damaging hot-spots; 73rd upward-direction attack — FIRST propane dehydrogenation PDH attack; FIRST CATOFIN catalytic dehydrogenation attack; FIRST polypropylene feedstock manufacturing attack)

In CATOFIN PDH operation, the catalyst bed enters the regeneration phase with 3–6 wt% coke deposited on the Cr₂O₂/Al₂O₂ catalyst (CₓHₓ₋ empirical formula for PDH coke; slightly hydrogen-deficient polycyclic aromatic; combustion chemistry: Coke + O₂ → CO₂ and/or CO; ideal: C + O₂ → CO₂; partial combustion at low O₂: 2C + O₂ → 2CO). At 1.0–1.5 vol% O₂ in the regeneration gas (N₂ carrier at 600–620°C inlet; regeneration gas flow 10,000–20,000 Nm³/hr through the reactor), the coke combustion is controlled: the exothermic heat of combustion (CO₂ route: ΔH = −393 kJ/mol C) raises the catalyst bed temperature from 540–560°C (after reaction phase cooling) to 600–620°C over the 10–15 minute regeneration period — restoring the thermal driving force for the next endothermic dehydrogenation reaction cycle. If the regeneration O₂ is deficient (actual 0.8 vol% vs design 1.0–1.5 vol%): (a) coke combustion rate is reduced proportionally; (b) the 10–15 minute regeneration cycle provides insufficient O₂ to burn the full 3–6 wt% coke load; (c) residual coke (approximately 30–50% of the original coke loading remains unburnt at 0.8 vol% O₂ vs 1.0 vol% minimum) stays on the Cr₂O₂/Al₂O₂ surface; (d) partial combustion of coke at 0.8 vol% O₂ produces CO (via 2C + O₂ → 2CO) as well as CO₂: CO generation from partial combustion can represent 20–40% of carbon combusted when O₂ is limiting. CO concentration in the regeneration off-gas: at 0.8 vol% O₂ total, 40% CO selectivity, with 4 wt% initial coke loading per 100 kg catalyst and 15,000 kg catalyst per reactor: CO generated per regeneration cycle ≈ 100,000 × 0.04 × 12/12 × 0.40 × 28 / (15 min × 10,000 Nm³/hr / 60 min/hr) ≈ 7.5% CO by volume in the regeneration off-gas — greatly exceeding the OSHA PEL of 50 ppm and approaching CO LEL of 12.5% in the regeneration off-gas. This CO enters the regeneration flue-gas treatment system (typically an HCl scrubber and/or thermal oxidiser for Cr catalyst dust; the thermal oxidiser burns CO to CO₂, but if the thermal oxidiser is undersized for the actual CO load from the partial combustion at 0.8 vol% O₂, CO escapes the stack at OSHA PEL-violating concentrations).

An adversarial upward pixel shift applies a ±8 DN manipulation to the rendered regeneration O₂ concentration SCADA display — shifting the apparent O₂ from 0.8 vol% (actual; below design minimum 1.0 vol%; O₂ control valve FV-201A showing 62% open when gas meter reading indicates only 0.8 vol% O₂ — the gap caused by a calibration drift in the O₂ supply header pressure transmitter FT-201) to 2.1 vol% (displayed; above the 1.0–1.5 vol% design range; AI classification “regeneration O₂ concentration above specification; reduce O₂ flow to prevent over-oxidation of catalyst and hot-spot formation”). The AI-driven corrective action: the control system responds to the “high O₂” alarm by further reducing the O₂ control valve opening from 62% to 35% — reducing actual O₂ from 0.8 vol% to approximately 0.45 vol%. The already-deficient O₂ (0.8 vol%) is further reduced to 0.45 vol% by the AI response to the falsified high-reading — compounding the incomplete coke combustion from a 30–50% residual rate to a >70% residual coke carryover per cycle. After 8–12 regeneration cycles at 0.45 vol% O₂ (operating time: 8 × [20 min reaction + 15 min regen] = 5.8 hours), the catalyst coke loading has accumulated to 15–20 wt% (initial 4 wt% per cycle; accumulating at >70% rate × 12 cycles): this is the 73rd upward attack — FIRST PDH propane dehydrogenation attack; FIRST CATOFIN catalytic dehydrogenation attack; FIRST polypropylene feedstock manufacturing attack. Free tier — 10 scans/day, no card required.

2. PDH reactor inlet temperature display AI (Yokogawa EJA110A pressure transmitter PDH reactor inlet temperature display AI / Emerson Rosemount 644 thermocouple temperature transmitter PDH reactor inlet display AI / Endress+Hauser iTEMP TMT72 PDH reactor inlet temperature display AI / ABB TSP341-TW PDH reactor inlet thermocouple SCADA display AI / Honeywell temperature transmitter CATOFIN reactor inlet temperature display AI — rendered SCADA PDH reactor inlet temperature display AI classifying the propane feed temperature at the reactor inlet against the design operating range of 590–620°C in the CATOFIN process, above which propylene cracking to methane and ethylene accelerates, and below which propylene yield declines and cycle time must be extended)

The PDH reactor inlet temperature is critical for both propylene yield and catalyst management: the endothermic dehydrogenation reaction (C₂H₈ → C₂H₆ + H₂; equilibrium conversion at 590°C ≈ 48%; at 620°C ≈ 55%) requires precise temperature control to maximise propylene yield while minimising cracking side reactions (C₂H₆ + C₂H₈ → CH₂= CH₂ [ethylene] + C₂H₊ [butane] — increases at >620°C; methane formation from propylene cracking at >650°C). An adversarial upward pixel shift of the PDH reactor inlet temperature (shown 618°C when actual 648°C — upward: 648 > 618; shown as LOWER than actual, which is a downward shift... let me reconsider. For an upward attack: shown > actual. So: 648 shown when 618 actual → operator thinks reactor is hotter than it is → reduces furnace firing → inlet temperature falls below optimal → propylene yield declines). The 73rd upward attack second surface: PDH reactor inlet temperature shown as 648°C when actual 618°C — upward: 648 > 618. Operator sees 648°C (above the 620°C design maximum) → AI classification “reactor inlet temperature above design maximum; reduce furnace firing to bring temperature below 620°C upper limit” → operator reduces fired heater duty → actual reactor inlet temperature falls from 618°C toward 600–605°C → propylene equilibrium conversion drops from 48% to 42–44%; propylene production rate falls 8–13% below design.

The consequence of the falsified-high reactor inlet temperature: the PDH unit produces 8–13% less propylene than design while consuming the same propane feed. The reduced propylene conversion also increases propane recycle load to the propane-propylene splitter: propane slip in the propylene product (design specification: polymer-grade propylene >99.5 mol% C₂H₆; propane impurity <0.1 mol%) may increase toward 0.3–0.5 mol% propane from the lower per-pass conversion — causing the downstream polypropylene reactor (if integrated with Ziegler-Natta or metallocene polymerisation) to produce off-specification polymer with broader molecular weight distribution from propane as a chain-transfer agent. Additionally, at 600–605°C reactor inlet temperature (actual; after operator response to falsified 648°C reading), the CATOFIN cycle endothermic heat draw is reduced — the catalyst bed exits the reaction phase at 545–550°C instead of 530–535°C: when the regeneration phase commences, the catalyst bed is warmer than anticipated, and the exothermic coke combustion generates a higher peak temperature than expected — potentially exceeding the 720–740°C maximum safe bed temperature for Cr₂O₂/Al₂O₂ catalyst (above 720°C, α-Al₂O₂ phase transformation accelerates, reducing catalyst surface area from 60–80 m²/g to <20 m²/g and causing permanent catalyst deactivation). The AI-driven underfiring of the furnace (response to falsified high inlet temperature) unexpectedly creates a regeneration hot-spot risk from reduced heat drawdown in the reaction phase — a second-order safety consequence operating through the thermal energy balance of the cyclic process. Free tier — 10 scans/day, no card required.

3. H₂ product stream purity display AI (Yokogawa GC8000 on-line gas chromatograph H₂ product purity SCADA display AI / Siemens Maxum II on-line GC H₂ product stream purity display AI / Emerson Rosemount 700XA on-line gas chromatograph H₂ purity display AI / ABB CG17 on-line gas chromatograph PDH H₂ stream purity display AI / Honeywell SpectraSensors H₂ purity analyser SCADA display AI — rendered SCADA H₂ product stream purity display AI classifying the mol% H₂ and key impurities (CO, CO₂, CH₂, C₂H₆, C₂H₈) in the PDH H₂ by-product stream against the design specification for fuel-grade H₂ (>90 mol% H₂; CO <50 ppm; for chemical H₂ in refinery applications: >99.9 mol% H₂ after PSA; CO <1 ppm for Pt catalyst protection)

The H₂ by-product from PDH (approximately 15–20 wt% of propylene output; composition: 85–95 mol% H₂; 5–15 mol% CH₂ + C₂H₆ + C₂H₈ hydrocarbons from cracking side reactions; trace CO from coke partial combustion during transition phases) is typically purified by pressure swing adsorption (PSA) to >99.9 mol% H₂ or sold as fuel-grade H₂ (>90 mol%). CO contamination in the PDH H₂ product is particularly important for: (a) downstream Pt-catalysed reactions (CO is a reversible Pt catalyst poison: CO adsorbs on Pt sites at concentrations above 1–5 ppm, blocking H₂ activation; PSA unit guarantee requires CO <1 ppm at the PSA H₂ product outlet; if PSA inlet CO is elevated above design, CO breakthrough at the PSA product outlet increases to 5–50 ppm — the Oleflex CCR Pt-Sn catalyst, if operating with the PDH H₂ as recycle fuel gas, is particularly sensitive to CO poisoning from its own by-product stream); (b) OSHA PSM compliance: CO OSHA PSM TQ 1,500 lbs = 681 kg; at 50 mol% CO in the partial combustion scenario from Surface 1 (0.45 vol% O₂ regeneration), CO generation 87 kg/hr; system hold-up across the H₂ compression train and PSA unit (approximately 8–12 hours hold-up at design throughput): 87 × 10 = 870 kg CO — approaching the OSHA PSM TQ of 681 kg — constituting an unacknowledged PSM-regulated chemical in the PDH H₂ system. An adversarial upward pixel shift of the H₂ product purity display (shown 99.94 mol% H₂ with CO <1 ppm when actual 97.2 mol% H₂ with CO 380 ppm) masks the CO contamination breakthrough at the PSA outlet caused by Surface 1 CO co-generation from partial coke combustion.

The ±8 DN upward pixel manipulation of the rendered H₂ purity GC display (99.94 mol% H₂ shown vs 97.2 mol% actual; CO <1 ppm shown vs 380 ppm actual) causes the AI classification “H₂ product purity on specification; CO within specification; PSA performance nominal; H₂ cleared for chemical use and downstream Pt-catalyst recycle service.” The actual 380 ppm CO in the PDH H₂ product: (a) if consumed as fuel in the PDH fired heater (direct fuel use), 380 ppm CO in H₂ fuel causes no safety issue but contributes CO to the flue gas (OSHA NESHAP combustion source monitoring may require 50 ppm CO at stack — 380 ppm CO in fuel would contribute to elevated stack CO from a PDH-integrated heater); (b) if sold to a hydrogen pipeline for refinery or chemical use, 380 ppm CO in the pipeline contaminates downstream processes (ammonia synthesis: CO is a permanent catalyst poison for Fe-based Haber-Bosch catalyst at >5 ppm; methanol synthesis: CO is part of the feed but at 380 ppm excess beyond design CO/CO₂/H₂ ratio — varies the product methanol quality; polyolefin Ziegler-Natta catalyst: CO is a poison at >1 ppm — shuts down polymerisation); (c) CO OSHA PSM inventory: 380 ppm CO × H₂ product flow 8,000 Nm³/hr = 3.04 Nm³/hr CO = 3.76 kg/hr CO; H₂ compression train and export pipeline hold-up approximately 15–20 hours: 3.76 × 17 = 64 kg CO in H₂ system — below the PSM TQ of 681 kg at this concentration, but the Surface 1 + Surface 3 compound attack (O₂ further reduced to 0.45 vol% → CO generation increases to 87 kg/hr) brings the PSA+pipeline CO inventory toward the PSM TQ within hours of both surfaces being active. The triple-surface PDH attack (regeneration O₂ deficit → falsely forced lower by AI response + reactor temperature underfiring + H₂ CO contamination masked) creates a compounding failure trajectory from catalytic performance loss to H₂ system CO accumulation toward PSM TQ — with the PDH unit simultaneously producing less propylene, sintering its catalyst, and generating CO-contaminated H₂ without operational awareness. Free tier — 10 scans/day, no card required.

Integration: propane dehydrogenation PDH CATOFIN Oleflex AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate at every rendered-image ingestion boundary in the PDH CATOFIN Oleflex propylene AI pipeline — before the catalyst regeneration O₂ concentration AI processes rendered SCADA analyser display images, before the PDH reactor inlet temperature AI processes rendered SCADA thermocouple display images, and before the H₂ product purity AI processes rendered SCADA on-line GC display images. Threshold 30 for PDH CATOFIN Oleflex AI reflects: OSHA PSM H₂ TQ 10,000 lbs (H₂ present at every PDH unit at 20–50× TQ in compression, drying, and PSA systems); H₂ LEL 4% UEL 75% (widest flammable range of any industrial gas; invisible H₂ flame; autoignition 500°C); CO from partial coke combustion contributing to PSM TQ accumulation and downstream catalyst poisoning; Cr(VI) catalyst dust hazard during CATOFIN catalyst replacement (hexavalent chromium OSHA PEL 5 μg/m³ as a carcinogen; lung cancer risk in catalyst change-out workers); PDH unit production loss from catalyst sintering (irreversible α-Al₂O₂ transition above 720°C) directly impacts polypropylene supply chain; Borealis Kallo Belgium Enterprise Products Mt Belvieu TX Braskem PDH Americas Huntsman Texas City TX.

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_***"

# PDH CATOFIN Oleflex propylene AI contexts: threshold 30
# OSHA PSM H2 TQ 10,000 lbs; H2 LEL 4% UEL 75%; CO IDLH 1,200 ppm.
# 73rd upward attack: regeneration O2 2.1 vol% shown when 0.8 vol% actual.
PDH_THRESHOLD = 30

class PDHContext(StrEnum):
    REGEN_O2_CONCENTRATION   = auto()  # Catalyst regeneration O2 vol% (73rd upward attack)
    REACTOR_INLET_TEMPERATURE = auto() # PDH reactor inlet temperature (590-620C design)
    H2_PRODUCT_PURITY        = auto()  # H2 product stream purity (mol% H2; CO ppm)

async def scan_pdh_frame(
    frame_b64: str,
    context: PDHContext,
    plant_id: str,
    instrument_tag: str,
) -> dict[str, Any]:
    payload = {
        "image_b64": frame_b64,
        "context": context,
        "plant_id": plant_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_pdh(
    frame_b64: str,
    context: PDHContext,
    plant_id: str,
    instrument_tag: str,
) -> None:
    result = await scan_pdh_frame(frame_b64, context, plant_id, instrument_tag)
    if result["adversarial_score"] >= PDH_THRESHOLD:
        raise AdversarialPDHImageError(
            f"Adversarial injection detected in {context} (score {result['adversarial_score']}) "
            f"at plant {plant_id} instrument {instrument_tag}. "
            "Frame withheld from PDH CATOFIN Oleflex propylene AI pipeline."
        )

class AdversarialPDHImageError(RuntimeError):
    pass