PTQ Q4 2024 Issue

Work closely with its customers to capture improvement opportunities via increased operational awareness.

Optimisation of operating parameters

Data cleansing & reconciliation

Performance normalisation

Secure data transfer

Machine learning model tuning

Information & knowledge exchange

Catalyst expertise

Operational experience

Data science

Process design knowledge

Connect ‘In .

Decision support & resource planning tools

Messenger & email alerts

Data driven performance prediction

Energy eciency & emission reduction

KPI dashboard & proactive advisor

No-code data exploration

Figure 1 Flow chart of the Connect’In solution

benefits users and technology providers alike. The Group does this through its digital offer, known as Connect’In. Technology experts, technical services engineers, and data scientists work closely with end users to capture improvement opportunities via increased operational awareness. In concrete terms, Connect’In is a digital super - vision tool and interface of exchange for all aspects of cus - tomer installations. It delivers real-time performance data for these installations, enabling Axens experts to: • Identify any discrepancies between observed and expected performance This solution is designed for process engineers and oper - ations managers, who have major responsibilities in terms of the continuity of operations, product quality, and produc - tivity. The solution also simplifies the monitoring of all these parameters and gives team members the peace of mind of knowing that any performance deviation will be detected quickly and that they will receive corrective action from experts just as quickly. Enabled by real-time secure data transfer and automated KPI calculations, technical service engineers remotely moni - tor unit performances via Connect’In and supply operational guidance to increase unit profitability. Using a combina - tion of high-fidelity models and innovative ML techniques, teams convert data into timely operational recommenda - tions to help customers achieve the best performance from their range of products. • Diagnose the causes of these discrepancies • Propose corrective measures to customers. Corrective measures can also be proposed continuously through the dedicated interface. For instance, for some units, the best set of operating conditions for a given per - formance target is continuously suggested to the customer. This is made possible using a hybrid model, which com - bines Axens’ high-fidelity first principle model with ML. Collaboration is also key to achieving operational excellence in this area. Leveraging advances in web and

data-sharing technologies, customers are provided with continuous and direct access to process expertise, catalyst and process know-how, and data analytics capabilities. Benefits Digital tools such as high-speed data networks, cloud- based systems, and advanced data techniques are rapidly changing how the refining industry monitors and optimises its operating assets. Refineries that adopt real-time unit monitoring tools, integrating on-demand, advanced mod - elling of process performance, immediately gain a compet - itive advantage. By monitoring a core set of KPIs, Connect’In allows unit troubleshooting through the continuous comparison of actual and normalised performances, which are calcu - lated based on reconciled data (see Figure 1 ) and Axens’ high-fidelity models. In addition to its advanced digitalisa - tion efforts for increasing data quality and technical advice relevancies, catalyst usage and lifespans can be optimised, improving unit economics and, therefore, lowering the CO₂ emissions associated with catalyst manufacturing. The advancement of ML models enables refineries to analyse collected data and infer additional data-propos - ing soft sensors that would otherwise be costly or even impossible to collect directly. This process involves rig - orous data cleaning steps to enhance data quality and enable the creation of alternative hybrid models that com - bine knowledge-based principles with trained and adapt - ed-based principles derived from historical data. Through this approach, based on past learnings, the models lead to a series of diagnostics that pinpoint issues and recommend appropriate corrective actions. Automatic monitoring not only predicts potential issues but also identifies their root causes, thereby recommending alternative solutions aimed at reducing energy consumption. Today, this approach is readily applicable to separation columns, which can be effi - ciently optimised through algorithmic processes.

PTQ Q4 2024