PTQ Q2 2026 Issue

Next-gen digital twins: automating model lifecycle management

Optimising simulation frameworks for enhanced reliability and predictive accuracy

Soni Malik and Michelle Wicmandy KBC (A Yokogawa Company)

R efineries depend on process simulation and linear programming (LP) models to support daily oper - ations, long-term planning, and emissions-reduc - tion strategies. As assets become more integrated and operating conditions shift more frequently, keeping these digital twins accurate has become increasingly difficult. Even well-built models can drift from reality as equipment performance changes, feed variability increases, and data quality fluctuates. This degradation often happens gradually, without clear warning, until planning assumptions no longer match actual operating constraints. The consequences can include reduced optimisation, higher energy use, and diminished confidence in simulation-based decisions. But what happens when these simulation models or digital twins fall out of sync with reality? When the digital reflection of the plant no longer mirrors its physical twin, how quickly does performance drift, and what does that mean for profitability and sustainability? Despite growing reliance on digital twins, model calibra - tion and validation remain largely manual in many refiner - ies. Engineers must reconcile data from multiple sources, evaluate deviations, and adjust model parameters to reflect current plant conditions. This work requires time, domain expertise, and coordination across teams. Unfortunately, human resources are becoming increasingly scarce. At the same time, market pressures are intensifying. According to Wood Mackenzie, more than 20% of global refining capacity is at risk of closure by 2035, based on an analysis of 420 sites.1 Operational performance will deter - mine which facilities remain competitive. To remain com - petitive, refiners are accelerating investment in automation and digital technologies. The global digital transformation market in oil and gas is projected to grow by $56.4 billion between 2025 and 2029. This equates to a compound annual growth rate (CAGR) of about 14.5%.2 This surge reflects a broader shift. Digitalisation is no longer experi - mental but a core element of operational strategy. Divergence of model sophistication and maintenance scalability As refinery units or assets become increasingly integrated, dynamic, and data-intensive, model sophistication is out - smarting the maintenance of these simulation models affecting scalability. Manual recalibration protocol simply

cannot keep pace with the frequency of changes in mod - ern refining. If the models appear to be outdated and do not align with actual predictions, the lag creates a credibil - ity gap, hence diminishing optimisation potential. Without the ability to automate model or digital twin maintenance, these deviations compound. In practice, model degradation rarely presents sudden failure. Instead, small deviations accumulate over time as equipment fouling, catalyst ageing, feedstock variability, and instrumentation drift alter plant behaviour. Optimisation decisions based on outdated assumptions gradually move operating targets away from true process constraints. Energy integration is becoming less efficient. Hydrogen management margins are tightening, and planning models are beginning to reflect assumed rather than achievable performance. Since these deviations occur incrementally, they can be accepted as normal operational To overcome the limitations of manual interventions or sudden recalibrations, a new generation of automated maintenance applications integrates process simulation, machine learning, and cloud computing variability rather than recognised as model inaccuracy. Over time, however, this silent divergence reduces optimisation effectiveness and erodes confidence in simulation-driven decision-making. Maintaining digital twin accuracy is much like monitor - ing a refinery’s heartbeat. Without regular diagnostics, its rhythm falters. Without timely care, the system weakens. Daily monitoring and quantifying the extent of deviation acts as preventive maintenance, keeping the model’s pulse steady and its predictions trustworthy. Streamlining model maintenance The integrity of a digital twin is defined by its ability to mirror physical reality through precise predictions. To overcome the limitations of manual interventions or sudden recalibrations,

PTQ Q2 2026