From a refinery architecture perspective, this is not merely a change of technology; it is a change in system design philosophy. Measurement becomes distributed at the process, while intelligence becomes centralised. One optical platform can serve multiple
Classical vs optical analysers in process measurement
Classical approach
Optical approach
Analyser lives in the process Electronics in hazardous area One analyser = one stream Mechanical sampling systems Reactive quality monitoring
Sensor lives in the process Electronics in control room One analyser = many streams
Optical interaction, minimal sampling
Continuous optimisation input
measurement points. Response time improves, installation complexity is reduced, and expansion becomes economically feasible. This structural shift enables CDU optimisation, gasoline blending, and hydrogen purity monitoring to be treated as components of a single digital ecosystem. Measurement becomes continuous, scalable, and robust enough to support advanced optimisation strategies. More importantly, optical measurement creates the foundation for discovering and maintaining what may be described as ‘golden setpoints’. These are operating conditions where safety margins are preserved, sustainability targets are met, cost is minimised, and performance is maximised simultaneously. Unlike static operating targets defined by engineering judgement, golden setpoints are learned and continuously refined through observation of real process behaviour. In this context, optical analysers are not simply instruments. They are the sensory foundation of intelligent process optimisation (see Table 1 ). AI, chemometrics, and DRL: from measurement to golden setpoints Optical analysers create visibility, but visibility alone does not produce optimisation. The transformation occurs when measurements are translated into real-time process intelligence. This is achieved through chemometrics, AI, and deep reinforcement learning (DRL). Chemometric models convert spectral and photonic signals into physical and chemical properties that operators and control systems can act upon. In a CDU, this includes salt and water content, crude quality indicators, and boiling curve proxies. In blending, it includes octane, vapour pressure, and density. In hydrogen systems, it includes oxygen partial pressure and purity margins. Chemometrics therefore transforms raw measurement into meaningful process variables. AI then operates on top of this information layer. Table 1
It combines measured properties with process constraints, safety limits, economic objectives, and control authority to recommend or execute optimal actions. Its role is not to replace conventional control, but to guide it toward operation that is consistently closer to the true optimum. DRL represents the most advanced evolution of this concept. Unlike rule-based optimisation or static models, DRL learns directly from the behaviour of the plant. It observes the process state through real-time analyser data, applies control actions, and evaluates results against a reward function built on the four pillars: safety, sustainability, economics, and performance. Over time, the system learns which operating regions maximise value while respecting all constraints. Trained on the digital twin of the process and informed by the process analysers, the DRL-based controller implicitly leverages the advantages of predictive control alongside feedback control corrections. Predictive control allows minimal effective changes in the manipulated variables, enabled by knowledge of the future impact of the changes made to the process. However, in practice, feedforward (predictive) control alone is not sufficient to guarantee safe and stable plant operation. Feedback control, informed by the process analysers deployed on the product streams, allows responding to the influence of the unmeasured disturbances. These are factors influencing the process that cannot be practically measured. These regions define the so-called golden setpoints. They are not fixed targets determined once by engineering judgement, but dynamic operating envelopes continuously refined by experience. In a CDU, golden setpoints balance furnace severity, energy efficiency, cut quality, and corrosion risk. In blending, they minimise giveaway while maintaining specification
Refining India
19
Powered by FlippingBook