refineries and other energy plants have to go beyond this and start looking at how to reduce margin leakage that occurs between planning and scheduling actual operations. The key here is an innovative modelling and optimisation approach that combines fundamental planning models with dynamic APC models. This unique approach uses a model that is consistent in material and quality balances while incor - porating dynamic models from the APC layer. This results in the ability to have consistent models, economics, and objec - tives between offline planning and online optimisation. The approach enables online optimisation of broad enve - lopes covering multiple process units within a refinery. Typical optimisation envelopes for refining include middle distillates, naphtha residue processing, and hydrogen and utility sys - tems. And as companies drive towards sustainability excel - lence, setting the objectives of the optimisation towards zero carbon, lower water use, and minimising waste is crucial. Machine learning and optimisation Optimising production calls for organisations to sift through a staggering amount of data and choices. Advances in machine learning provide a powerful tool for meeting this challenge. Artificial intelligence (AI) algorithms chew through huge reams of data, looking for patterns and associations. Industrial AI can identify the links between thousands of production variables and suggest ways to adjust them, leading to much more informed production opti - misation. AI algorithms can be holistic in their approach, over - coming the coordination problem of making all the required changes at once to find a new, more efficient equilibrium. Making the right predictions One application that makes use of machine learning and AI to help optimise plant production is prescriptive mainte - nance, which combines predictive and diagnostic capabili - ties to transform asset performance management and help achieve the best possible results from refinery and olefins plant deployments. The goal here is to minimise plant downtime by predic - tion and prevention of future failure events. All the produc - tion optimisation in the world, using the optimisation models previously described, goes to nought if the plant goes down. Prescriptive maintenance technology works on the prem - ise of why might a plant simply predict production issues when it can prescribe fixes for them and act on the prescrip - tions? It takes the already well-established predictive main - tenance paradigm onto a whole new level. Using machine learning and AI to anticipate asset main - tenance requirements, predictive maintenance can help avoid impromptu corrective maintenance by allowing for plant maintenance to be effectively scheduled prior to equipment failure. Prescriptive maintenance, on the other hand, not only looks for failure signatures but also provides information about how to delay or eliminate equipment failure. These algorithms can comb historical data for examples of a wide variety of operating conditions, extract patterns, and extrapolate data to provide hypothetical operating environ - ments. The cascade of effects and consequences from small
adjustments to an industrial process can be simulated by the prescriptive maintenance model, allowing for what might otherwise be expensive and risky experimentation to be safely performed in a computer simulation. In order for prescriptive maintenance to be effective in driving up production optimisation, a machine learning model will be trained on historical sensor and service data. The more high-quality information available, the more accu - rate the AI model will be in identifying more signs of main - tenance needs and failure signatures while producing fewer false positives. Higher-level information about an organisation can be provided to the machine learning algorithm when training a prescriptive maintenance algorithm. This allows the soft - ware to look at strategic considerations, such as the cost of repairs and manufacturing downtime. The machine learning model training occurs on specialised hardware, either stored locally or in the cloud. The model is code that can be deployed on-site or in the cloud, so some way of accessing and running the model is required. This can be directly integrated with many asset management software suites, simplifying the process of integrating the recommendations of the prescriptive maintenance model and ultimately further helping to optimise production at the plant or refinery. Upskilling is vital but can be difficult While technology is key, refineries and olefins plants must not overlook the human element of production optimisation. Plant digitalisation technology providing new opportunities for production optimisation has matured to the point of reli - ability in the last decade. But the personnel best positioned to make decisions that will take advantage of these new tools may have developed their skills before the rise of these new technologies. These workers must be prepared to fully implement the requisite changes for the firm to practice production optimi - sation or risk making costly inefficient decisions. An abun - dance of caution is an excellent strategy when managing expensive and dangerous equipment and processes but can hinder optimisation. Moving forward Ultimately, for any refinery or olefins plant, production opti - misation is critically important in driving profitability and sustainability. However, achieving it is not easy and relies on a commitment by the business concerned to continuous evolution and, in particular, migrating to digital technologies. Refineries and olefins plant operators need visibility and auditability of their sustainability performance. To achieve that, many are embracing the latest advanced and AI-driven technologies, such as vertically integrated production optimi - sation, digital twin, and prescriptive maintenance. Moreover, they need to deliver change management, ensuring they are training and upskilling employees and bringing them on the journey to production optimisation. For those operators who do all this well, it is a vision of the future that is not far away.
Ron Beck is Senior Director, Marketing at AspenTech.
93
PTQ Q3 2022
www.digitalrefining.com
Powered by FlippingBook