output of SAF, the CO₂ savings are in the order of 0.25 kg per kg of SAF. In an average hydroprocessing unit, this implies that, compared to a standard zeolite benchmark, the accessible zeolites generate CO₂ savings of about 50 metric tonnes per kg of zeolite. These benefits, combined with more than $500 of increased output per kg of zeo - lite, make accessible zeolites a technological smart play for advancing SAF technology. That is, if developers can scale and manufacture the accessible zeolite-based catalysts in a commercially relevant manner. Q How is AI being applied to enhance predictive mainte- nance for refinery and petrochemical equipment? A Mark Fronek, Manager, Digital Solutions, Becht, mark. fronek@becht.com On a scale of 0 to 7 of AI maturity (0 = no AI; 1 = descriptive analytics; 2 = diagnostic analytics; 3 = predictive analytics; 4 = prescriptive analytics; 5 = augmented intelligence, 6 = autonomous systems; and 7 = transformative AI), predic - tive maintenance is being explored at a ‘Level 3’ by nearly all mid-major and larger organisations. Many have a goal of automated work order generation at a fractional level of accuracy, which pushes into ‘Level 4’ maturity. A key difference between these efforts and long-estab - lished predictive models for rotating equipment is that the analysis looks at maintenance, inspection, and operational data. This introduces a significant challenge to success, as maintenance data sets are not easily audited for validity and have only more recently been leveraged in AI-driven systems. As a result, they have a shorter history of data clean-up and institutional emphasis on data quality. In general, inspection data is more robust than mainte - nance repair history, and AI-driven results based on this data are more likely to be accurate and require less human clean-up. However, the lengthy periods of data needing review and the likely changes in methodologies and sys - tems over those time periods create additional challenges. 1 Siters, K. Alkegen Fiber based specialty catalyst material maximises surface area and catalyst contact. PTQ Catalysis 2022 , pp.45-49. 2 Verboekend, D. Economic and environmental versatile technologies in refining. PTQ Q4 2025 , pp.45-48. A Doug Cooper, Product Management Director, Emerson’s Aspen Technology business, douglascooper@ emerson.com AI is being applied to predictive maintenance in refinery and petrochemical operations in several ways, ranging from model development and alerting strategy to diag - nostics and workflow support. The objective is to improve how early issues are detected, interpreted, and addressed before equipment problems affect production, cost, or safety. In these environments, AI helps identify early deg - radation patterns to predict likely failure before traditional alarms or maintenance intervals indicate a problem. One important application is in the development of pre - dictive models themselves. AI can be used to determine training ranges for machine learning agents, identify sensor
requirements, and establish alerting ranges that are more representative of actual operating behaviour. This is par - ticularly relevant for refinery and petrochemical equipment because operating conditions are variable and asset response is dependent on process context. In these environments, static thresholds alone are often not sufficient to detect early degradation or support reliable failure prediction. AI is also being used to improve diagnostic quality. Rather than only indicating that abnormal behaviour exists, it can help determine the likely failure mode so that more prescriptive guidance can be provided for remediation. This supports a more actionable maintenance programme, especially for assets such as pumps, compressors, heat exchangers, and distillation-related equipment where multiple process and mechanical factors can contribute to failure. Supporting technologies may include failure, anomaly, and process agents, depending on the type of asset and monitoring objective. When linked to an embedded Failure Modes and Effects Analysis (FMEA) library, the system can map likely causes, effects, and recommended corrective actions, making predictive alerts more practical for main - tenance teams. A further application is in the design and scaling of asset health strategies. AI can support the creation of asset tem - plates for refinery and petrochemical equipment; help define asset health KPIs across rule-based, condition monitoring, and first-principles approaches; and improve consistency across similar asset classes. When combined with domain expertise and physical properties data, this can provide a more complete view of asset condition than conventional monitoring methods alone. AI is also being applied to reduce inefficiency in main - tenance workflows. For example, it can help prevent alert fatigue by grouping multiple alerts associated with a single underlying issue. This allows teams to focus on the most rel - evant root cause, instead of spending time sorting through secondary symptoms. AI is also enhancing predictive main - tenance by operationalising workflows through enterprise asset management (EAM) software and computerised maintenance management system (CMMS) integration. Rather than leaving predictive insights in a dashboard, the objective is to connect maintenance intelligence directly into business processes so alerts can be reviewed, priori - tised, and translated into recommended corrective actions or work orders. This supports faster review, better prioriti - sation, and a closed-loop workflow from prediction to exe - cution. Overall, AI is making predictive maintenance more practical and scalable for industrial operators. AI helps identify early degradation patterns to predict likely failure before traditional alarms or maintenance intervals indicate a problem
16
PTQ Q3 2026
www.digitalrefining.com
Powered by FlippingBook