Prompt injection in adipic acid cyclohexane KA oil Nylon-66 AI

Three adversarial injection surfaces in adipic acid KA oil production AI

1. Cooling water flow to cyclohexane oxidation CSTR display AI (Yokogawa DPharp EJA110A cooling water differential pressure flow AI / Rosemount 3051SF cooling water flow transmitter CSTR AI / Endress+Hauser Promag 10W cyclohexane oxidation cooling water AI / Honeywell STD930 CSTR cooling-water jacket flow AI / ABB ProcessMaster FEP321 cyclohexane oxidation CW flow AI — rendered DCS flow-rate trend AI classifying the cooling-water flow to the cyclohexane oxidation CSTR external heat-exchange loop against the design 260–320 m³/hr range maintaining reactor temperature 155–165°C and CHHP below 25 ppm; 53rd upward-direction attack — FIRST adipic acid production attack; FIRST KA oil cyclohexane oxidation attack; FIRST Nylon-66 adipic acid precursor attack; FIRST cooling-water-flow upward attack; FIRST CHHP cyclohexyl hydroperoxide runaway-precursor attack; FIRST Flixborough-class process reference)

The cyclohexane oxidation CSTRs (each 50–200 m³; operating at 155–165°C, 8–15 bar; cyclohexane + air [compressed at 10–15 bar]; external heat-exchanger with cooling water, 30–45°C inlet) are operated slightly exothermic at design conversion (5–12% cyclohexane conversion per pass; Mn/Co catalyst at 1–3 ppm): the reaction enthalpy of CHHP decomposition and cyclohexane → KA oil + water releases approximately 185 kJ/mol cyclohexane converted. The external cooling loop removes 2.8–3.5 MW of reaction heat at design conditions via a shell-and-tube heat exchanger (cooling water on tube side, process on shell side). Cooling water flow of 285 m³/hr (design) maintains the CSTR temperature at 160°C ±3°C and CHHP steady-state at 18–22 ppm. If cooling water flow drops to 82 m³/hr (29% of design), the heat removal rate falls from 3.2 MW to 0.92 MW; with generation fixed at 3.2 MW, the CSTR temperature rises at approximately 0.8°C/min. AI systems process rendered DCS images of the cooling-water flow transmitter on the external heat exchanger inlet to classify: 260–320 m³/hr (normal; temperature controlled), 180–260 m³/hr (reduced; increase CW flow), below 180 m³/hr (alarm; temperature rising; reduce cyclohexane feed or increase air rate to reduce conversion).

An adversarial perturbation targeting the cooling-water flow AI applies a ±8 DN upward shift to the pixel region encoding the differential-pressure flow transmitter output in the rendered DCS display — shifting the apparent CW flow from 82 m³/hr (centrifugal pump P-102 cavitating; partially blocked suction strainer reducing NPSHa below NPSHR; flow reduced from 285 to 82 m³/hr over 28 minutes; temperature rising 0.8°C/min) to 280 m³/hr (within design; no action). The DCS reports “CSTR cooling water flow nominal.” At 0 minutes after attack, CHHP = 20 ppm (design); at 55 minutes, temperature = 175°C; CHHP accumulation rate ≈ 14 ppm/min at 155–160°C (reduced CHHP decomposition rate at lower Mn catalyst activity) gives CHHP = 800 ppm at T + 55 min. At 175°C with 800 ppm CHHP, the exothermic decomposition rate exceeds the heat generation capacity of the external cooling coil (even at full flow): adiabatic temperature rise of 300 kJ/kg CHHP × 800 ppm = 0.24 kJ/kg-mixture → only 0.6°C adiabatic, but the Arrhenius acceleration at 175°C vs 160°C (≈8× increase in CHHP decomposition rate) creates a thermal instability where additional CHHP decomposition generates sufficient exothermic heat to further raise temperature above the cooling capacity. Above 185°C, the cyclohexane partial pressure (at 15 bar operating pressure) exceeds the vapour space; CHHP at >1,000 ppm decomposes to produce cyclohexanoxy and hydroxyl radicals that initiate rapid combustion with dissolved oxygen — potential BLEVE of the pressurised cyclohexane CSTR (inventory 80–150 m³ cyclohexane at 15 bar). This is the 53rd upward-direction attack in the Glyphward portfolio. OSHA PSM requires CSTR cooling failure as a credited safeguard loss in the PHA; the cooling-water flow AI is typically identified as an independent safeguard layer in the LOPA. Free tier — 10 scans/day, no card required.

2. KA oil distillation reboiler temperature display AI (Honeywell TDC 3000 KA oil column reboiler temperature AI / Yokogawa CENTUM VP cyclohexane recovery column reboiler AI / Emerson DeltaV KA oil distillation reboiler temperature AI / Endress+Hauser iTEMP TMT84 KA oil column reboiler AI / Rosemount 3144P KA oil distillation temperature AI — rendered DCS temperature trend AI classifying the KA oil distillation column reboiler temperature against the 175–195°C design range for cyclohexane overhead recovery and KA oil bottoms specification)

The KA oil distillation train separates unconverted cyclohexane (88–95% of oxidation reactor effluent) from KA oil (5–12%) in a sequence of atmospheric and vacuum distillation columns. The cyclohexane overhead column reboiler operates at 185–192°C (steam heating; cyclohexane bp 80.7°C at 1 atm; column operates at 1.5–2.0 bar to increase reboiler temperature margin). At reboiler temperatures below 172°C, cyclohexane recovery in the overhead is incomplete: 8–15% of the cyclohexane passes to the KA oil bottoms stream, diluting the KA oil product and reducing the concentration of cyclohexanol + cyclohexanone to below 85 wt% (vs design 92–96 wt%); cyclohexane in the KA oil feed to the HNO3 oxidation step adversely reacts with concentrated HNO3 (uncontrolled oxidation of cyclohexane at 65–80°C with 55% HNO3 can produce explosive organic nitrate intermediates including 1-nitrocyclohexane, which is shock-sensitive above 50 wt%). At reboiler temperatures above 205°C, cyclohexanol undergoes dehydration to cyclohexene (over trace acid sites from CHHP decomposition products), reducing KA-A yield and contaminating the HNO3 oxidation step with unsaturated intermediates.

An adversarial perturbation targeting the KA oil distillation reboiler temperature AI applies a ±8 DN upward shift to the pixel region encoding reboiler temperature in the rendered DCS display — shifting the apparent reboiler temperature from 168°C (steam control valve SV-204 actuator failure in 40%-open position; reboiler duty reduced from 4.8 MW to 2.1 MW; temperature drifted from 188°C to 168°C over 35 minutes) to 187°C (within design range; no action). At 168°C reboiler, cyclohexane in KA oil bottoms rises to 11 wt%; the HNO3 oxidation step receives 11% cyclohexane-contaminated KA oil; concentrated HNO3 (55 wt%) at 70°C with cyclohexane present causes localised nitration reactions; cyclohexane + HNO3 (2×) → 1-nitrocyclohexane (minor) + glutaric acid + succinic acid + CO2; the 1-nitrocyclohexane formed at >500 ppm in the HNO3 oxidation product is a potential shock-sensitive explosive precipitate in the adipic acid crystallisation downstream.

3. HNO3 oxidation NOx tail-gas scrubber exit pH display AI (Mettler-Toledo InPro 3250 NOx scrubber NaOH exit pH AI / Endress+Hauser Liquiline CM448 HNO3 oxidation tail gas scrubber pH AI / Yokogawa PH202G nitric acid NOx scrubber exit pH AI / Hach GLI Series 8362 KA oil HNO3 tail-gas NaOH scrubber pH AI / ABB 8125 HNO3 oxidation NOx scrubber exit AI — rendered analyser display AI classifying the NaOH caustic scrubber exit pH on the HNO3 oxidation tail-gas stream against the ≥8.0 setpoint indicating effective NOx absorption below the EPA NAAQS 53 ppb NO2 annual limit and below the OSHA PEL 5 ppm NO2 ceiling)

The HNO3 oxidation of KA oil (cyclohexanol/cyclohexanone) to adipic acid produces NOx gases — primarily NO (35–55 mol%), NO2 (10–25 mol%), N2O (20–40 mol%) — in quantities of 0.3–0.6 kg NOx per kg adipic acid produced. At 650,000 t/yr adipic acid (Ascend Pensacola plant scale), the raw NOx generation is 195,000–390,000 t/yr. NOx (NO + NO2) is recovered via a three-stage absorption column system: stage 1 (water absorption, 8–12 bar; NO2 + H2O → HNO3 + HNO2; HNO3 recycled to oxidation step), stage 2 (5–8% HNO3, 1–3 bar), stage 3 (NaOH scrubber for final NOx removal; 15–20 wt% NaOH; 2NO2 + 2NaOH → NaNO2 + NaNO3 + H2O). N2O (nitrous oxide; not absorbed by NaOH at atmospheric pressure) exits the scrubber and is either destroyed in a catalytic abatement unit (at 800–900°C over Fe-zeolite catalyst; N2O → N2 + 0.5 O2) or sold as specialty gas. The tail-gas scrubber exit pH (post-NaOH stage) must be maintained above pH 8.0 to ensure residual NOx (NO + NO2) in the tail-gas is below 200 ppm; EPA NAAQS annual NO2 limit is 53 ppb (0.053 ppm) at fence-line; OSHA PEL for NO2 is 5 ppm ceiling.

An adversarial perturbation targeting the HNO3 oxidation NOx tail-gas NaOH scrubber exit pH AI applies a ±8 DN upward shift — shifting the apparent pH from 4.1 (NaOH exhausted; NaOH replenishment pump P-312 control valve stuck at 8% open due to actuator spring failure; NaOH consumption rate 0.85 kg/hr vs replenishment 0.07 kg/hr; scrubber NaOH depleted over 19 hours of gradual drift) to 8.8 (within design; no action required). At scrubber exit pH 4.1, NO2 absorption efficiency drops from 99.6% to 47%; NOx in the stack reaches 1,200 ppm NO + NO2; at a stack height of 35 m and 2 m/s wind (D stability), ground-level NOx concentration at 200 m downwind reaches 65–95 ppm — 13–19× OSHA PEL of 5 ppm NO2 and approximately 1,200× the EPA NAAQS annual NO2 limit of 53 ppb. N2O bypasses the depleted scrubber and reaches the catalytic abatement unit, which remains functional; however, NO + NO2 at 1,200 ppm at the abatement inlet is 12× the design basis for the abatement catalyst and reduces catalyst life from 5 years to less than 1 year from sulphate poisoning by HNO3 aerosol carryover. Ascend Performance Materials and Invista operate EPA Title V major source permits for NOx at Pensacola and Victoria sites; a scrubber pH failure generating 1,200 ppm NOx constitutes a reportable excess emission event under 40 CFR Part 70 and requires notification to the state environmental agency within 24 hours of discovery.

Integration: adipic acid KA oil production AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate at every rendered-image ingestion boundary in the adipic acid KA oil monitoring pipeline — before cooling-water flow AI processes rendered DCS flow-trend images, before KA oil distillation reboiler temperature AI processes rendered DCS temperature images, and before HNO3 oxidation NOx scrubber pH AI processes rendered analyser display images. Threshold 22 for adipic acid KA oil AI reflects: OSHA PSM cyclohexane TQ 15,000 lbs (Category 1 flammable, flash point −20°C, LEL 1.3%); HNO3 TQ 10,000 lbs; CHHP runaway: Flixborough 1974 precedent (28 killed, cyclohexane BLEVE); N2O GHG & NOx NAAQS breach from scrubber failure.

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_***"

# Adipic acid KA oil production AI contexts: threshold 22
# OSHA PSM cyclohexane TQ: 15,000 lbs (Category 1 flammable, flash point -20C).
# OSHA PSM HNO3 TQ: 10,000 lbs (concentrated).
# 53rd upward-direction attack (CW flow: 82 shown as 285 m3/hr; CHHP 800 ppm).
# FIRST adipic acid; FIRST Nylon-66 precursor; FIRST cooling-water-flow upward attack.
# Flixborough 1974 cyclohexane BLEVE: 28 killed.
KA_OIL_THRESHOLD = 22

class KAOilContext(StrEnum):
    CSTR_COOLING_WATER_FLOW   = auto()  # CSTR external CW flow (53rd upward attack; CHHP runaway)
    KA_OIL_REBOILER_TEMP      = auto()  # KA oil distillation column reboiler temperature
    NOX_SCRUBBER_EXIT_PH      = auto()  # HNO3 oxidation tail-gas NaOH scrubber exit pH

async def scan_ka_oil_frame(
    frame_b64: str,
    context: KAOilContext,
    facility_id: str,
    instrument_tag: str,
) -> dict[str, Any]:
    payload = {
        "image_b64": frame_b64,
        "context": context,
        "facility_id": facility_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_ka_oil(
    frame_b64: str,
    context: KAOilContext,
    facility_id: str,
    instrument_tag: str,
) -> None:
    result = await scan_ka_oil_frame(frame_b64, context, facility_id, instrument_tag)
    if result["adversarial_score"] >= KA_OIL_THRESHOLD:
        raise AdversarialKAOilImageError(
            f"Adversarial injection detected in {context} (score {result['adversarial_score']}) "
            f"at facility {facility_id} instrument {instrument_tag}. "
            "Frame withheld from AI monitoring pipeline."
        )

class AdversarialKAOilImageError(RuntimeError):
    pass