What role are AI systems expected to play when optimising plant-wide operations?Jan-2024
Michael Keaton, R2E Group, firstname.lastname@example.org
Our colleagues have provided very good answers. AI will play an increasing role in our business with its ability to reason over large datasets and find patterns and high probability answers. One of the major keys to successful AI implementation is the data fed to it. Unfortunately, data in the refining and petrochemical industry tends to be inadequate, which limits the scope of applications. For one, data typically resides in multiple systems, so AI must integrate with multiple systems or data must be pulled together in one spot (e.g. big data cloud platforms). Another is the lack of one source of the truth. That doesn't mean data resides in only one spot, but that there is only one path to the source data. Another is data quality: is the value good, is it right, is it timely, and so on.
Most refiners and petrochemical companies have poor data quality and don't have GOOD data governance. Data governance and data quality management need to be beefed up for broad AI success. Also, engineers that know how the facility operates and how to maintain it must be involved in data processes. Don't leave it to IT and Data Scientists, not to take anything away from them, but that engineering expertise is needed for success.
Lisa Krumpholz, CSO, Navigance GmbH, Lisa.Krumpholz@clariant.com
The major challenges of optimising plant-wide operations are the vast amount of data available and the high complexity of interconnected unit operations. Traditional first-principles models and tools for optimisation have disadvantages in coping with these challenges as they usually require high effort and thus cost to develop, maintain, and adapt to changes in the operation.
In contrast, AI systems with machine learning-based models at their core can be designed with reasonable effort for complex systems. They offer the opportunity to learn automatically from continuous data streams in the plant and adapt quickly to changing conditions.
Thus, AI systems will see rapid adoption in the coming years to replace, complement or enhance existing optimisation approaches. Like the developments in autonomous driving, AI systems are expected first to be adopted as assisting systems to support and enable better human decision-making for plant-wide optimisation.
Ezequiel Vicent, Oli Systems, Ezequiel.Vicent@olisystems.com
AI will play a major role in the optimisation of plant-wide operations both during steady-state times and during shutdowns and start-ups. There are many examples of how AI is being used today to optimise a plant, but the decision to go from an open system to a closed system is still a few years away< and the technology has not yet caught up.
A prime example of AI being used in plant optimisation is in the area of energy and emissions management. There are energy optimisers that use first principles to look at the current energy status of the unit and are able to optimise fuel consumption and steam production while accounting for combustion emissions to minimise the amount of energy needed for the steam demand.
They will account for steam Cogen units and heat integration. However, to predict future demand, AI models ‘learn’ where the peaks and valleys come in and are able to predict the input changes before they happen. This helps the energy optimiser capture changes more quickly and have additional energy savings.
Another area where AI, or in this case a Machine Learning (ML) model, can make a big impact is in dynamic processes, like the start-up or shutdown of a unit. Consider a unit where a process upset occurs upstream, and a column needs to be quickly shut down with a precise sequence of events. The outcome largely relies on the operators’ experience. In such a case, a ML model can be ‘taught’ that exact sequence under varying process and environmental conditions.
Various dynamic simulations can be created to show the different types of upsets that can trigger a shutdown, and the shutdown sequence can be included. The ML model, once tested against multiple simulations, can now be added as a closed-loop system and allowed to ‘operate’ the shutdown or start-up of the column to avoid damage to the unit or unwanted chemical releases.
However, for more complex systems, AI still needs to evolve as a technology. Several refiners we have worked with have started on the path of AI implementation but have stopped short of full ‘autonomous plant’ systems. We have heard that the complexity of the processes at the refinery and the constant variation of feedstock and pricing have made it difficult to gain value from full AI implementation.
Isabelle Conso, AXENS, Isabelle.Conso@axens.net
AI is expected to play a significant role in optimising plant-wide operations. Here are some key roles and benefits that AI systems can provide in this context: Safety and compliance: AI can monitor safety conditions in the plant and detect anomalies or potential hazards. It can also assist in compliance with regulatory requirements by ensuring that processes and products meet the necessary standards.
Process optimisation: AI can continuously analyse vast amounts of data from various sensors to uncover patterns in view of production process optimisation. It can also produce soft sensors generated through surrogate models that will provide insights for adjustments of parameters such as temperature, pressure, and flow rates to maximise efficiency and product quality.
Predictive maintenance: AI can monitor equipment and machinery in real-time, analysing data from sensors to predict when maintenance is needed. This can help prevent unplanned downtime and reduce maintenance costs.
Production scheduling: AI can create optimised production schedules that balance production efficiency with demand fluctuations and resource constraints.