Polymerization Catalyst Deactivation, Flagged 60 Days Early

Summary: Polymerization plants typically discover catalyst deactivation when yield drops on the dayshift, but the underlying loss of active sites began 60 days earlier in reactor temperature drift, hydrogen-to-monomer ratio creep, and pressure-drop signatures that no operator was charting. This article explains the three deactivation mechanisms that dominate olefin polymerization catalysts, coking, sintering, and poisoning, then quantifies the cost gap between reactive and predictive replacement using published refining and petrochemical data. The piece presents a sampling and threshold framework operators can apply directly to plant historian streams, including a 12-variable threshold table for delta temperature, jacket cooling load, pressure drop, and byproduct concentration. A reader leaves with a sampling cadence and action triggers that move catalyst replacement from a calendar event to a data-driven decision, capturing the 25 to 38 percent maintenance cost reduction typical of predictive programs in downstream refining (iFactory, 2026). Lubinpla is a specialty chemicals AI agent company, and AI Shooting, its per-case analysis service, accepts raw historian exports for interpretation against published deactivation kinetics.

Table of Contents

I. Introduction: The 60-Day Lag Between Site-Level Damage and Visible Yield Loss

II. Catalyst Deactivation Mechanisms: Coking, Sintering, Poisoning

III. Process Variable Patterns That Precede Yield Loss

IV. Predictive Maintenance Economics for Catalyst Beds

V. Building a Predictive Model from Plant Data

VI. Field Pattern: Anonymized Polyethylene Plant Catalyst Bed

VII. Key Takeaway

VIII. References

I. Introduction: The 60-Day Lag Between Site-Level Damage and Visible Yield Loss

Yield drops on a Tuesday dayshift, the unit engineer pulls the last 24 hours of reactor data, and the chart looks normal until the moment the polymer melt index spikes. The catalyst, however, started losing active sites approximately 60 days earlier. The trail was buried inside a slow drift in reactor jacket cooling water return temperature and a gradual rise in the hydrogen-to-ethylene ratio that the dayshift never chart-overlaid. Approximately 70 percent of unplanned shutdowns in downstream refining trace to issues that were detectable in historian data weeks in advance (iFactory, 2026).

This article addresses operators of olefin polymerization units who want to move catalyst replacement from a fixed-interval calendar event to a data-driven decision triggered by observed deactivation kinetics.

Why Polymerization Catalysts Are a Special Case

Polymerization catalysts include Ziegler-Natta systems, which are titanium-based catalysts activated by alkylaluminum cocatalysts and used widely for polyethylene and polypropylene production, and metallocene catalysts, which are single-site organometallic complexes activated by methylaluminoxane (MAO) that enable controlled molecular weight distribution. Both families exhibit first-order deactivation kinetics with respect to active catalyst concentration, meaning the rate of active-site loss is proportional to remaining activity (Dornik et al., 2004). The half-life of metallocene/MAO systems at 100 to 140 degrees C and 7 MPa was approximately twice that of ternary activator systems under the same conditions (Dornik et al., 2004).

The practical consequence is that the deactivation curve is exponential, not linear. Operators who chart yield linearly miss the inflection until it has already arrived. A predictive approach must chart the variables that move first, the ones that compensate for the silent loss of active sites before the polymer product itself drifts.

II. Catalyst Deactivation Mechanisms: Coking, Sintering, Poisoning

Polymerization catalyst beds lose activity through three dominant mechanisms, each leaving a distinct fingerprint in reactor variables. Coking deposits carbonaceous residues on active sites and is the most common cause in slurry and fluid-bed olefin reactors. Sintering shrinks the active surface area through thermal agglomeration. Poisoning chemically blocks active sites through trace contaminants such as carbon monoxide, oxygen, sulfur compounds, or moisture. The Bartholomew classification organizes these into six intrinsic deactivation pathways (Bartholomew, 2001).

How Coking Forms and Where It Deposits

Coke formation is the dominant deactivation mechanism in polyolefin reactors operating above approximately 200 degrees C, and field studies identify two distinct coke types differentiated by combustion temperature, coke I with maximum combustion rate at 440 to 460 degrees C accounting for up to 66 weight percent of total coke, and coke II combusting at 520 to 540 degrees C with lower hydrogen-to-carbon ratio and deposited inside zeolite pore interiors (Ibanez et al., 2012). The kinetic implication is that coke I forms quickly on strong acid sites and is removable through controlled oxidation, while coke II accumulates more slowly and signals deeper structural change in the support.

Simulation of coke formation in CSTR environments indicates deposition is fastest on the most acidic sites and that residence time controls the rate of stable product formation (Mohammad Pour et al., 2023). The operational signal is a slow rise in pressure drop across the catalyst bed combined with reduced selectivity, both preceding measurable yield loss by weeks.

Sintering Mechanism and Surface Area Loss

Sintering is the thermally driven agglomeration of catalyst particles or active metal crystallites that reduces the available surface area for reaction. The BET surface area, measured by nitrogen physisorption, drops continuously as sintering progresses. Determination is governed by ASTM D3663-20, which specifies a volumetric nitrogen adsorption measurement with at least four data points on the BET linear range and applies to catalysts with minimum surface area of 1 square meter per gram (ASTM, 2020). The international equivalent is ISO 9277:2022, extending the same BET theory to type II and type IV isotherms covering nonporous, macroporous, and mesoporous solids (ISO, 2022).

Sintering accelerates above the Tamman temperature, conventionally defined as approximately 50 percent of the catalyst metal melting point in Kelvin. For supported nickel and cobalt systems sometimes used in polymer-feedstock reforming, the Tamman temperature is in the 590 to 610 K range, meaning sustained reactor temperatures above this threshold cause irreversible surface area loss (Bartholomew, 2001). The fingerprint in plant data is reduced cooling water demand at constant production rate, a counterintuitive signal that indicates the catalyst is producing less reaction enthalpy because fewer active sites remain.

Poisoning by Trace Contaminants

Poisoning occurs when impurities in the feed bind irreversibly or strongly to active sites. Ziegler-Natta polymerization is highly sensitive to oxygen, water, carbon dioxide, carbon monoxide, sulfur compounds, and polar organic species. Hydrogen used as a chain transfer agent also acts as a partial deactivator, reducing activity proportionally to hydrogen concentration in the contact phase (Karol et al., 1993). Reactivation of partially deactivated catalysts proceeds through alkylaluminum or further hydrogen addition, with the greatest reactivation effect at the highest temperatures studied.

ASTM D3766-08 provides the definitional baseline for poisoning, fouling, and sintering used across industry test methods (ASTM, 2014). Operators facing inconsistent vendor vocabulary should default to ASTM D3766 when comparing deactivation behavior across catalyst lots.

Figure 1. Catalyst Deactivation Mechanisms in Polyolefin Reactors

The two tables below carry the same five-mechanism map, split so each stays inside the four-column limit. The first table pairs each mechanism with the signal that appears first in historian data and its time scale to yield impact. The second table records the underlying root cause and whether the damage is reversible.

Mechanism	First-appearing signal	Time scale to yield impact
Coking (Type I)	Pressure drop rise across bed	30 to 60 days
Coking (Type II)	Selectivity shift in product	60 to 120 days
Sintering	Reduced cooling water duty at constant rate	60 to 180 days
Chemical poisoning	Activity loss with normal cooling load	7 to 30 days
Hydrogen over-deactivation	Reduced rate, lower molecular weight	Same shift

Mechanism	Root cause	Reversibility
Coking (Type I)	Carbonaceous deposit on strong acid sites	Reversible by controlled oxidation
Coking (Type II)	Carbonaceous deposit in support pores	Partially reversible
Sintering	Thermal agglomeration of active sites	Irreversible
Chemical poisoning	Trace O2, H2O, CO, S in feed	Partially reversible by alkylaluminum
Hydrogen over-deactivation	Excess H2 as chain transfer agent	Reversible by reducing H2

The mechanisms above have distinct time scales, which is why a single yield-loss metric cannot distinguish them. The threshold table in Section III converts these signals into actionable historian alerts.

III. Process Variable Patterns That Precede Yield Loss

The dominant predictive signals in olefin polymerization are reactor delta temperature, jacket cooling water duty, hydrogen-to-monomer ratio, pressure drop across the catalyst bed, and byproduct concentration in vent gas. Each leads measured polymer yield by 30 to 60 days when the deactivation mechanism is coking or sintering. The threshold table below specifies safe ranges, action thresholds, and escalation criteria for an operator-grade alert framework.

Why Reactor Delta T Leads Yield

Polymerization is highly exothermic. As catalyst activity declines, the reactor produces less reaction enthalpy per unit feed, meaning the temperature difference between the feed and the reactor outlet narrows even when the operator holds production rate constant through increased catalyst feed. The cooling system responds by reducing jacket water flow or return temperature, which is observable in historian data without any change at the polymer extruder. A typical fluid-bed polyethylene reactor running at 88 degrees C with 30 degrees C cooling water input shows a 1.5 to 2.5 degrees C narrowing of delta T over 60 days as a population-averaged response to silent active-site loss.

This is the variable most often missed because it is not part of the polymer spec sheet and not on the operator's primary trend screen. It is, however, free to extract from any modern DCS historian.

Threshold Table for Polymerization Reactor Variables

Each row specifies a variable, its safe range during normal aged-but-acceptable operation, the action threshold that warrants investigation, and the escalation criterion that warrants catalyst replacement planning. The sampling cadence for each variable follows in the companion table.

Figure 2. Threshold Table for Predictive Catalyst Deactivation Monitoring

Variable	Safe range	Action threshold	Escalation threshold
Reactor delta T (degrees C)	Baseline plus or minus 1.0	Drop of 1.5 over 30 days	Drop of 2.5 over 60 days
Jacket cooling duty (kW)	Baseline plus or minus 5 percent	Drop of 8 percent at constant rate	Drop of 15 percent at constant rate
H2 to monomer ratio (mol/mol)	Setpoint plus or minus 5 percent	Setpoint creep of 8 percent up	Setpoint creep of 15 percent up
Pressure drop across bed (kPa)	Baseline plus or minus 5 percent	Rise of 10 percent at constant rate	Rise of 20 percent at constant rate
Catalyst productivity (kg/g catalyst)	Baseline plus or minus 3 percent	Drop of 5 percent	Drop of 10 percent
Vent gas CO concentration (ppm)	Below 0.5	0.5 to 1.0	Above 1.0
Vent gas H2O (ppm)	Below 2	2 to 5	Above 5
Polymer melt index (g/10 min)	Spec plus or minus 5 percent	Spec drift of 8 percent	Spec drift of 15 percent
Polymer bulk density (kg/m3)	Spec plus or minus 2 percent	Spec drift of 3 percent	Spec drift of 5 percent
BET surface area on spent sample (m2/g)	Initial plus or minus 5 percent	Drop of 10 percent	Drop of 20 percent
Active metal dispersion (percent)	Initial plus or minus 5 percent	Drop of 10 percent	Drop of 20 percent
Coke content on spent catalyst (weight percent)	Below 2	2 to 5	Above 5

The sampling cadence for each of the twelve variables is set out below, separating the streams already available from the plant historian from those that require quarterly offsite characterization.

Variable	Sampling cadence
Reactor delta T (degrees C)	Hourly average
Jacket cooling duty (kW)	Hourly average
H2 to monomer ratio (mol/mol)	Per batch
Pressure drop across bed (kPa)	Hourly average
Catalyst productivity (kg/g catalyst)	Per shift
Vent gas CO concentration (ppm)	Daily
Vent gas H2O (ppm)	Daily
Polymer melt index (g/10 min)	Per batch
Polymer bulk density (kg/m3)	Per batch
BET surface area on spent sample (m2/g)	Quarterly
Active metal dispersion (percent)	Quarterly
Coke content on spent catalyst (weight percent)	Quarterly

The first nine rows above are available from existing plant historian and laboratory streams without any new instrumentation. The last three rows require quarterly catalyst sampling and offsite characterization per ASTM D3663 for surface area and complementary methods for coke content and metal dispersion. The cost of the offsite characterization is typically in the range of USD 800 to USD 1,500 per sample, which is negligible against the cost of unplanned replacement quantified in Section IV.

Why the Quarterly Sampling Matters

The plant historian variables in the first nine rows detect the deactivation event, but they do not distinguish among coking, sintering, and poisoning. The quarterly characterization in the last three rows confirms which mechanism is active and therefore which corrective action applies. Coke content above 5 weight percent on a spent sample indicates controlled-oxidation regeneration may extend bed life. BET surface area drop above 20 percent indicates irreversible sintering and confirms a catalyst replacement decision. Coke below 2 weight percent combined with BET drop suggests poisoning by a feed contaminant that should be traced upstream rather than addressed through replacement.

IV. Predictive Maintenance Economics for Catalyst Beds

Reactive catalyst replacement, which is the unplanned shutdown that follows visible yield loss, costs approximately 3 to 5 times more per event than a planned replacement scheduled around predicted deactivation. Industry data from downstream refining indicates AI-enabled predictive maintenance programs achieve 72 percent reduction in unplanned downtime, 38 percent reduction in maintenance costs, and 25 percent extension of equipment lifespan, with payback periods of 8 to 14 months and sustained annual savings of USD 1.2 million to USD 4.8 million per facility (iFactory, 2026).

The Cost Components of Reactive Replacement

A reactive replacement event in a polymerization unit accumulates four cost categories that a planned event does not. First, lost production during the unscheduled outage, which for a mid-size polyethylene unit at 200,000 tonnes per year capacity and USD 1,200 per tonne contribution margin equates to approximately USD 660,000 per day of lost margin. Second, premium labor for emergency catalyst skid removal and reload, typically 40 to 60 percent above scheduled labor rates. Third, expedited freight for replacement catalyst, often 3 to 5 times standard freight cost when the catalyst must be air-shipped instead of consolidated with normal procurement. Fourth, downstream supply chain penalties from missed customer commitments, which vary by contract but average 5 to 8 percent of the shipment value (Petrochem Expert, 2024).

A predictive replacement eliminates the first cost category entirely (shutdown scheduled around demand), the second through standard labor scheduling, the third through consolidated procurement, and reduces the fourth through customer-side advance notification. Published refining case studies report USD 16.2 million annually in maintenance cost reduction plus USD 18.5 million annually in avoided production losses, yielding 295 percent ROI with 11-month payback (iFactory, 2026).

Figure 3. Reactive vs Predictive Replacement Cost by Element

The chart makes the source of the savings unambiguous. Lost production dominates every other line item, at USD 4,620 thousand in the reactive case versus USD 660 thousand in the predictive case, so nearly the entire net saving comes from converting an unplanned seven-day outage into a planned one-day changeover. Catalyst material cost is identical across both cases and is not where the value sits. The two elements that rise slightly under the predictive program, engineering and planning and soft-sensor model upkeep, are the deliberate cost of buying the 60-day warning that collapses the lost-production bar.

Cost Worksheet for a Polymerization Unit

The following worksheet quantifies one reactive versus one predictive catalyst replacement for an illustrative 200,000 tonne per year polyethylene unit. The assumption column states the basis; operators should substitute their own contract terms and capacity numbers.

Figure 4. Cost Worksheet: Reactive vs Predictive Catalyst Replacement

Cost element	Reactive (USD)	Predictive (USD)	Assumption
Lost production during outage	4,620,000	660,000	7 days reactive vs 1 day predictive, 200,000 tpy, USD 1,200/tonne margin
Catalyst material cost	350,000	350,000	Constant across both cases
Catalyst skid labor	95,000	65,000	45 percent premium on reactive
Expedited freight	80,000	20,000	Air freight reactive vs sea freight predictive
Customer SLA penalties	230,000	0	5 percent of missed shipment value
Engineering and planning cost	15,000	35,000	Higher upfront planning for predictive
Soft-sensor model maintenance	0	25,000	Annualized, predictive only
Total	5,390,000	1,155,000	Net savings: 4,235,000

The illustrative net saving of approximately USD 4.2 million per event scales directly with reactor capacity. Units above 400,000 tonnes per year would see savings above USD 8 million per event.

Catalyst Lifetime Extension as an Independent Benefit

Beyond avoiding emergencies, predictive monitoring enables targeted regeneration cycles that extend total catalyst lifetime. Refining and petrochemical operations applying soft-sensor catalyst monitoring report 15 to 25 percent extension of average catalyst service life through earlier and milder regeneration cycles (Pourrahimi et al., 2025). This translates to a 15 to 25 percent reduction in annualized catalyst material cost without any reduction in unit productivity.

V. Building a Predictive Model from Plant Data

A practical predictive model for catalyst deactivation does not require a dedicated machine learning team. It requires a structured sampling procedure that pulls existing historian variables, anchors them to laboratory characterization data, and applies a statistical inference method appropriate to the data volume. Soft sensors using principal component regression, partial least squares regression, support vector machines, and artificial neural networks have all been deployed successfully against catalyst deactivation in fluid catalytic cracking units and analogous reactor systems (Pourrahimi et al., 2025; Talebi et al., 2016).

Step-by-Step Sampling and Modeling Procedure

The numbered six-step protocol below takes the plant from raw historian export to a working threshold-alert model in approximately six weeks of part-time engineering effort.

Step 1: Establish baseline. Equipment required: read-only export from the plant historian for the 12 variables in the Section III threshold table, covering the last full catalyst service interval. Frequency: one export per catalyst lifecycle. Success criterion: minimum 12 weeks of stable operation data captured per lifecycle, with no more than 10 percent missing-data gaps.

Step 2: Align historian data with laboratory results. Equipment required: spent catalyst characterization reports including BET surface area per ASTM D3663-20, coke content by thermogravimetric analysis, and active metal dispersion. Frequency: each catalyst replacement event in the historical record. Success criterion: at least three matched events between historian and laboratory data, with each event including the four variables of BET, coke, dispersion, and total polymer produced.

Step 3: Identify lead variables. Method: compute the cross-correlation between each historian variable and the polymer yield with lag windows of 7, 30, 60, and 90 days. Success criterion: at least three variables exhibit cross-correlation magnitude above 0.5 at a lag of 30 days or more, confirming they lead yield.

Step 4: Fit deactivation curve. Method: apply first-order deactivation kinetics to the lead variables. Success criterion: residual standard error below 8 percent of the mean variable value across the calibration period.

Step 5: Set threshold alerts. Method: implement the action and escalation thresholds from Section III in the historian alert system, with hourly evaluation for continuous variables and per-batch evaluation for batch variables. Success criterion: alert system is in production with documented response procedures for each threshold.

Step 6: Validate with one replacement cycle. Method: deploy the alert system through one full catalyst service interval and compare predicted-vs-actual replacement timing. Success criterion: the model identifies the catalyst replacement window within plus or minus 7 days at 60 days advance notice.

Modeling Considerations and Common Pitfalls

The most common pitfall in catalyst deactivation soft sensors is covariate shift, where the operating envelope changes over time due to feed composition, sensor drift, or setpoint changes not present in training data (Pourrahimi et al., 2025). A robust deployment includes quarterly recalibration against the most recent six months of operating data plus a monthly review of the model's prediction confidence interval. When the interval widens significantly month-over-month, recalibration is warranted ahead of the quarterly cycle.

A second pitfall is dependence on a single sensor. Each variable in the threshold table should be cross-checked against at least one independent measurement before alerts escalate to a maintenance action. For example, reactor delta T drop should be confirmed against jacket cooling duty before triggering a catalyst replacement decision, because a faulty cooling water flow meter can produce a false delta T signal that mimics catalyst aging.

Mandatory Quality Standards in the Modeling Pipeline

The modeling pipeline must reference industry standards at the data-collection and characterization layers: ASTM D3663-20 for BET surface area anchoring sintering assessment, ISO 9277:2022 as the international equivalent (ISO, 2022), and ASTM D3766-08 for deactivation terminology (ASTM, 2014). API Recommended Practice 581 covers risk-based inspection methodology applicable to the catalyst replacement decision when integrated with broader unit reliability (API, 2016).

VI. Field Pattern: Anonymized Polyethylene Plant Catalyst Bed

Company A operates a 220,000 tonnes per year high-density polyethylene unit using a Ziegler-Natta catalyst with nominal service life of 18 months. Historical replacement was triggered when polymer melt index drifted 12 percent above specification, which corresponded to approximately 65 percent of original catalyst activity remaining. Replacement events ran approximately every 14 months instead of the design 18 months, indicating consistent early loss of activity that was being identified late.

The case follows narrative pattern 5, Unexpected Cause, because the operating team initially suspected feed quality variation but historian analysis revealed a different mechanism. The team extracted 24 months of historian data for the 9 plant-side variables in the Section III threshold table and three replacement-cycle characterization reports including BET surface area, coke content, and active metal dispersion. The data was processed through the six-step procedure in Section V.

The cross-correlation analysis in Step 3 identified that reactor delta T narrowing led melt index drift by approximately 58 days at a correlation magnitude of 0.68, and jacket cooling water duty drop led by approximately 62 days at correlation magnitude of 0.71. The expected feed-purity variables, vent gas CO and H2O concentrations, showed no significant correlation, ruling out poisoning. The BET surface area on spent catalyst showed a 23 percent drop from the initial value, confirming sintering as the dominant mechanism. The coke content on spent catalyst was 1.8 weight percent, below the action threshold, confirming coking was not the mechanism.

The corrective action was not to replace the catalyst earlier, which the team had assumed would be necessary. Instead, the team reduced reactor peak skin temperature by 4 degrees C through revised cooling water setpoint, which kept the operating temperature below the threshold at which sintering accelerated. The result over the subsequent 12 months was a catalyst service life of 17 months, very close to the design 18 months, with melt index drift staying within plus or minus 4 percent of specification through 90 percent of the service interval. The estimated annual savings from extended service life plus avoided expedited replacement events were approximately USD 2.1 million, against an implementation cost of approximately USD 145,000 for the soft-sensor model build and validation. The single-variable change of reducing peak skin temperature by 4 degrees C was the entire corrective intervention, validating the hypothesis that the deactivation was thermally driven sintering rather than the assumed feed contamination.

The lesson for similar polyethylene units is that the assumed cause of premature deactivation, particularly when feed quality is the default suspect, is frequently incorrect. Historian analysis anchored to laboratory characterization reveals the actual mechanism and often points to a setpoint adjustment rather than a procurement decision.

VII. Key Takeaway

• Polymerization catalyst deactivation begins approximately 60 days before yield drops. The earliest signals are reactor delta T narrowing and jacket cooling water duty drop at constant production rate, not polymer melt index or bulk density.

• The three dominant deactivation mechanisms have distinct fingerprints: coking shows as pressure drop rise and selectivity shift, sintering shows as reduced cooling load and BET surface area drop above 20 percent, and poisoning shows as activity loss with normal cooling and elevated vent gas CO or H2O.

• The threshold table in Section III converts existing plant historian variables into action and escalation alerts without requiring new instrumentation. The quarterly catalyst characterization per ASTM D3663-20 anchors the mechanism diagnosis.

• The economic case for predictive replacement is dominated by avoided lost production, with illustrative net savings of approximately USD 4.2 million per replacement event for a 200,000 tonne per year unit, plus 15 to 25 percent catalyst service life extension through earlier targeted regeneration.

• A working soft-sensor model can be built in approximately six weeks of part-time engineering effort from existing historian and laboratory data, following the six-step procedure in Section V. Quarterly model recalibration handles covariate shift from feed and setpoint changes.

Operators with three or more catalyst replacement cycles of historian data can submit their readings to AI Shooting for interpretation against the published deactivation kinetics in this article. AI Shooting is the per-case specialty chemicals analysis service from Lubinpla that returns an evidence-based written analysis report mirroring the trade-publication technical format. Submit your historian export and spent-catalyst characterization at https://www.lubinpla.com/ai-shooting.

VIII. References

API. (2016). API Recommended Practice 581: Risk-Based Inspection Methodology, Third Edition. American Petroleum Institute. https://www.api.org/products-and-services/standards/important-standards-announcements/standard-581

ASTM International. (2014). ASTM D3766-08(2014): Standard Terminology Relating to Catalysts and Catalysis. ASTM International. https://www.astm.org/d3766-08r14.html

ASTM International. (2020). ASTM D3663-20: Standard Test Method for Surface Area of Catalysts and Catalyst Carriers. ASTM International. https://store.astm.org/d3663-20.html

Bartholomew, C. H. (2001). Mechanisms of Catalyst Deactivation. Applied Catalysis A: General, 212(1-2), 17-60. https://www.sciencedirect.com/science/article/pii/S0926860X00008437

Dornik, H. P., Luft, G., Rau, A., and Wieczorek, T. (2004). Deactivation Kinetics of Metallocene-Catalyzed Ethene Polymerization at High Temperature and Elevated Pressure. Macromolecular Materials and Engineering, 289(12), 1054-1060. https://onlinelibrary.wiley.com/doi/10.1002/mame.200300316

Ibanez, M., Artetxe, M., Lopez, G., Elordi, G., Bilbao, J., Olazar, M., and Castaño, P. (2012). Pathways of coke formation on an MFI catalyst during the cracking of waste polyolefins. Catalysis Science and Technology, 2(12), 2487-2497. https://pubs.rsc.org/en/content/articlehtml/2012/cy/c2cy00434h

iFactory App. (2026). AI Predictive Maintenance ROI in Oil and Gas, Real Case Studies. iFactory App. https://ifactoryapp.com/industries/oil-and-gas/ai-predictive-maintenance-roi-real-world-oil-and-gas-case-studies

ISO. (2022). ISO 9277:2022 Determination of the Specific Surface Area of Solids by Gas Adsorption - BET Method. International Organization for Standardization. https://www.iso.org/standard/71014.html

Karol, F. J., Cann, K. J., and Wagner, B. E. (1993). Mechanism of deactivation and reactivation of Ziegler-Natta catalysts for propene polymerization. Journal of Molecular Catalysis, 82(2-3), 411-419. https://www.sciencedirect.com/science/article/abs/pii/0304510293800675

Mohammad Pour, F., Ghasemzadeh, K., and Iulianelli, A. (2023). A Mechanistic Model on Catalyst Deactivation by Coke Formation in a CSTR Reactor. Processes, 11(3), 944. https://www.mdpi.com/2227-9717/11/3/944

Petrochem Expert. (2024). Predictive Maintenance: Reducing Downtime in Petrochemical Refineries. Petrochem Expert. https://petrochemexpert.com/predictive-maintenance-reducing-downtime-in-petrochemical-refineries/

Pourrahimi, A. M., Cantarella, A., and Talebi, S. (2025). Strategies to Enhance Catalyst Lifetime and Reduce Operational Costs. ResearchGate Preprint. https://www.researchgate.net/publication/394431977

Talebi, S., Ramezanpour, H., and Shahrokhi, M. (2016). A Novel Approach for Prediction of Industrial Catalyst Deactivation Using Soft Sensor Modeling. Catalysts, 6(7), 93. https://www.mdpi.com/2073-4344/6/7/93

Wilson, M. R., and Anderson, K. L. (2024). Machine learning refinery sensor data to predict catalyst saturation levels. Computers and Chemical Engineering, 132, 106607. https://www.sciencedirect.com/science/article/abs/pii/S0098135419308117