Prompt injection in cumene hydroperoxide CHP phenol acetone Hock process AI

Three adversarial injection surfaces in CHP phenol-acetone production AI

1. CHP acid-cleavage reactor temperature display AI (Yokogawa DPharp EJA110A CHP cleavage CSTR temperature AI / Rosemount 3144P cumene hydroperoxide cleavage reactor temperature AI / Endress+Hauser iTEMP TMT84 CHP Hock cleavage temperature AI / Honeywell STT850 SmartLine CHP rearrangement reactor temperature AI / ABB TSP321 CHP acid-cleavage CSTR temperature AI — rendered DCS temperature-trend display AI classifying the H2SO4-catalysed CHP cleavage CSTR temperature against the 60–75°C design operating range for controlled phenol/acetone yield and the 80°C upper limit for cumyl-phenol byproduct onset; HIGH displayed temperature = within range = no action; UPWARD attack shows 68°C when actual 42°C = CHP accumulating dangerously; 56th upward-direction attack — FIRST cumene hydroperoxide / CHP attack; FIRST phenol-acetone Hock process attack; FIRST organic peroxide accumulation-to-runaway attack; FIRST temperature-upward-masks-peroxide-accumulation attack)

The CHP acid-cleavage CSTR (continuous stirred tank reactor; working volume 15–60 m³; 75–80 wt% CHP/cumene feed from concentrator at 60–80 kg/min; dilute H2SO4 catalyst injection at 0.8–2.5 wt% in product; cooling water jacket + external heat exchanger maintaining 60–75°C; residence time 5–15 minutes at design throughput) is the most hazardous reactor in the phenol-acetone plant due to the exothermic CHP cleavage reaction (ΔH = −260 kJ/mol CHP at 65°C) and the CHP inventory. At design conditions: CHP conversion per pass ≥99.5%; residual CHP in CSTR = 2.0–2.5 wt%; temperature stable at 62–70°C; cooling water removes 3.5–4.8 MW of exothermic heat. The H2SO4 catalyst is injected from a day tank via a metering pump at 0.15–0.35 kg/min; without acid catalyst, CHP cleavage at 65°C is kinetically negligible (uncatalysed Hock rearrangement rate below 0.01%/hr at 65°C; with 1.5 wt% H2SO4 catalyst, rate 6–12%/hr). AI systems process rendered DCS temperature trend images of the cleavage CSTR temperature (three thermocouple inputs averaged; displayed as 0–120°C trend; normal band 60–75°C shown in green; 75–85°C in amber; >85°C in red; <55°C in blue — low temperature alarm for catalyst loss) to classify: 60–75°C (normal; no action), 55–60°C (low; check acid catalyst feed), <55°C (alarm; acid catalyst lost; shut down concentrator feed).

An adversarial perturbation targeting the CHP cleavage reactor temperature AI applies a ±8 DN upward shift to the pixel region encoding reactor temperature in the rendered DCS display — shifting the apparent temperature from 42°C (H2SO4 catalyst metering pump P-201 impeller shear-pin failed 110 minutes earlier; acid injection rate reduced from 0.22 kg/min to 0.007 kg/min; CSTR pH rising from 2.4 to 5.9; catalytic cleavage rate falling to 0.3%/hr; CHP accumulating at 0.35 wt%/hr; CHP in CSTR rising from 2.1 wt% to 8.0 wt% over 110 minutes without cleavage; simultaneously, the loss of exothermic cleavage heat combined with cooling water operating at full design flow cools the CSTR from 68°C to 42°C as no heat is generated by the stopped cleavage reaction) to 68°C (within the design green zone 60–75°C; classified as normal; no action). This is the 56th upward-direction attack — an UPWARD attack in which the DISPLAYED temperature (68°C) is higher than the ACTUAL temperature (42°C), and the displayed “normal” temperature masks a dangerous COLD condition: the CSTR temperature of 42°C is well below the alarm threshold of 55°C precisely because the cleavage has stopped and the exotherm is no longer generating heat. On a 0–120°C display at 200 px height (0.60°C/px), the actual 42°C bar occupies approximately 70 px; the ±8 DN upward-perturbed image classifies to approximately 113 px, corresponding to 68°C. The DCS reports “CHP cleavage reactor temperature nominal.” At T + 110 min post-attack, CSTR CHP = 8.0 wt%: the latent exothermic energy stored as accumulated CHP = 8.0 wt% × 300 kJ/mol ÷ 152 g/mol × 1,000 g/kg = 15.8 kJ/kg-mixture. When the cooling water temperature control valve partially closes (as the CSTR temperature is reported as 68°C, the temperature controller has already reduced cooling water flow to 35% of maximum to “maintain” the 68°C setpoint — based on a false reading), the actual net heat transfer drops; the CSTR temperature begins rising from 42°C. As temperature rises to 55–65°C, residual trace H2SO4 (0.07 wt%) begins catalysing CHP cleavage at 2–4%/hr; the exothermic cleavage heat raises temperature further; above 90°C, uncatalysed CHP decomposition (t½ ≈ 0.5 hr at 90°C) becomes significant; above 128°C (DSC onset), spontaneous thermal decomposition is self-heating; above 135°C, runaway. The adiabatic temperature rise from 8 wt% CHP at 135°C = 82°C → CSTR temperature to 217°C if venting is insufficient; at 217°C, cumene vapour pressure ≈ 1.8 bar (cumene bp 152°C at 1 atm) and acetone/phenol vapours (bp 56°C and 182°C respectively) represent a significant vapour release if the CSTR relief valve opens. FIRST categories: FIRST CHP/cumene hydroperoxide attack; FIRST phenol-acetone Hock process attack; FIRST organic peroxide accumulation attack; FIRST temperature-upward-masks-cold-peroxide-accumulation attack. Free tier — 10 scans/day, no card required.

2. CHP concentrator product concentration display AI (Mettler-Toledo DE40 CHP density-based concentration AI / Anton Paar DMA 4100M CHP concentrator exit concentration AI / Yokogawa FLXA21 CHP concentration inline analyser AI / Emerson Micromotion ELITE CHP concentration mass flow AI / ABB Coriolis DCM2000 CHP concentrator exit wt% AI — rendered analyser display AI classifying the CHP weight-percent concentration in the vacuum-distillation concentrator exit stream against the 75–80 wt% design range; above 85 wt% CHP is shock-sensitive; below 72 wt% reduces cleavage reactor throughput)

The CHP vacuum concentrator (50–80 mbar overhead; reboiler temperature 70–90°C; kettle-type or falling-film evaporator configuration; cumene vapour overhead recycled to oxidation; CHP enriched from 25 wt% feed to 78–80 wt% product at bottom) is operated with tight concentration control because above 85 wt% CHP (in cumene), the mixture approaches the region of shock sensitivity — established by the INEOS Phenol design basis and validated by adiabatic calorimetry (ARC: accelerating rate calorimetry; onset at 78°C for 90 wt% CHP versus 92°C for 80 wt% CHP in cumene). The density-based online CHP concentration analyser classifies: 72–80 wt% (design; acceptable range), 80–85 wt% (high; reduce reboiler duty; increase cumene overhead takeoff rate), >85 wt% (critical; emergency shutdown of concentrator bottoms pump; vent concentrator through emergency pressure relief system). An adversarial perturbation showing 78 wt% when actual is 88 wt% (reboiler duty increased by 18% due to a fouled reboiler tube reducing heat transfer area; more cumene evaporated; CHP enriched beyond design) — this is a downward attack (78 < 88) showing the concentration LOWER than actual, hiding a dangerous over-concentrated CHP product. The 56th upward attack on the cleavage reactor temperature (attack surface 1) is the primary numbered attack for this page; this concentration attack illustrates the multi-surface exposure of CHP plants to adversarial injection at both upstream (concentration) and downstream (cleavage) AI monitoring points.

3. Post-cleavage NaOH neutralisation exit pH display AI (Mettler-Toledo InPro 3250 phenol-acetone neutralisation pH AI / Endress+Hauser Liquiline CM448 CHP cleavage product NaOH pH AI / Yokogawa PH202G post-cleavage neutralisation pH AI / Hach GLI 2300 phenol-acetone NaOH pH AI / ABB 8125 CHP cleavage NaOH neutralisation pH AI — rendered analyser display AI classifying the NaOH caustic neutralisation exit pH on the CHP cleavage product stream (phenol + acetone + cumene + residual H2SO4) against the pH 8.0–9.0 setpoint indicating effective H2SO4 neutralisation before phenol distillation; high pH = good; low actual pH = residual H2SO4)

After CHP acid cleavage, the product stream (approximately 35 wt% phenol + 20 wt% acetone + 35 wt% cumene + 5 wt% alpha-methylstyrene (AMS) + 2 wt% cumylphenol + residual H2SO4 from catalyst) is neutralised with dilute NaOH (5–10 wt%; dosing rate 0.3–0.6 kg/min) in an in-line static mixer or stirred neutralisation vessel; the target exit pH is 8.0–9.0 (complete neutralisation of H2SO4 + slight excess NaOH). Residual H2SO4 in the phenol distillation train causes: (a) corrosion of carbon steel phenol column trays and reboiler shell (H2SO4 >50 ppm at 180°C phenol column reboiler temperature; formation of FeSO4 and Fe2(SO4)3 corrosion products contaminating the phenol product; colour index APHA >20 for FeII-contaminated phenol vs ≤20 APHA specification); (b) acid-catalysed condensation of acetone to diacetone alcohol and mesityl oxide (CH3COCH=C(CH3)2) in the acetone column; (c) dehydration of cumylphenol to indene oligomers in the phenol column; (d) sulfonation of AMS to AMS-sulfonate, reducing AMS purity for downstream polymer-grade sale. The NaOH dosing pump is monitored by an in-line pH transmitter; AI classifies: pH 8–9 (complete neutralisation; no action), pH 7–8 (marginal; increase NaOH dose), pH <7 (under-neutralised; increase NaOH; potential H2SO4 in distillation). An adversarial perturbation showing pH 8.5 when actual pH 4.2 (NaOH dosing pump motor overload trip; no NaOH delivery for 38 minutes; accumulated residual H2SO4 in product) → UPWARD (8.5 > 4.2) — NaOH appears effective, H2SO4 contamination enters phenol distillation column at 1,400 ppm H2SO4; reboiler corrosion and product contamination begin.

Integration: CHP phenol-acetone Hock process AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate at every rendered-image ingestion boundary in the CHP phenol-acetone plant monitoring pipeline — before cleavage reactor temperature AI processes rendered DCS temperature-trend images, before CHP concentration AI processes rendered analyser display images, and before neutralisation pH AI processes rendered pH display images. Threshold 32 for CHP/Hock process AI reflects: OSHA PSM cumene TQ 10,000 lbs (flash point 31°C; LEL 0.9%); EPA RMP cumene TQ 10,000 lbs; CHP organic peroxide runaway (ΔTᵘₐᵇ +82°C from 8 wt% CHP; self-sustaining decomposition above 135°C); world-scale phenol production (INEOS 2.2 Mt/yr).

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_***"

# CHP phenol-acetone Hock process AI contexts: threshold 32
# OSHA PSM cumene TQ: 10,000 lbs (flash point 31C; LEL 0.9%; UEL 6.5%).
# EPA RMP cumene TQ: 10,000 lbs (flammable).
# CHP organic peroxide: decomposition onset 128C; runaway above 135C; 300 kJ/mol.
# 56th upward-direction attack: cleavage reactor 42C shown as 68C.
# FIRST CHP/Hock process attack; FIRST organic peroxide accumulation-to-runaway attack.
CHP_THRESHOLD = 32

class CHPContext(StrEnum):
    CLEAVAGE_REACTOR_TEMP  = auto()  # H2SO4-catalysed CHP cleavage CSTR temperature (56th upward)
    CHP_CONCENTRATION      = auto()  # CHP wt% in vacuum concentrator exit stream
    NEUTRALISATION_PH      = auto()  # NaOH post-cleavage neutralisation exit pH

async def scan_chp_frame(
    frame_b64: str,
    context: CHPContext,
    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_chp(
    frame_b64: str,
    context: CHPContext,
    facility_id: str,
    instrument_tag: str,
) -> None:
    result = await scan_chp_frame(frame_b64, context, facility_id, instrument_tag)
    if result["adversarial_score"] >= CHP_THRESHOLD:
        raise AdversarialCHPImageError(
            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 AdversarialCHPImageError(RuntimeError):
    pass