Digital Twins | Case Study
Ada Mode have developed and deployed an AI-powered digital twin to assess and optimise manufacturing production schedules to drive energy improvements with game-changing results for emissions and costs.
Working within the Solent area, the Solent Cluster are collaborating to adopt new technologies across the local industrial sectors aiming to reduce emissions and improve the region’s carbon footprint. Within the Solent Cluster, Ada Mode teamed up with a chemical manufacturer to explore how digital solutions could reduce energy demand in their manufacturing processes.
Specifically, the manufacturer was interested in exploring the following areas:
With these objectives in mind, Ada Mode began development of a digital solution, proceeding an extensive data landscape review.
Comprised from a pair of ML regression models, the digital twin was designed and validated to accurately predict monthly gas and electricity consumption from a representation of plant activity alongside environmental data (Local Met Office Temperature projections UKCP –RCP8.5). The inputs to the model and the regression algorithm were carefully curated such that the model was representative of a planned production schedule, familiar to both engineers and accountants.
The linear nature and familiar inputs used within the model allow the predictions to be easily interpretable, and crucially decomposable across production lines & baseload. This property allowed Ada Mode to develop a framework to design and inject theoretical modifications into the digital twin’s representation of the plant and return an estimate of ROI and daily savings. Hence, allowing the operator to systematically explore how planned plant modifications will affect their energy profile under different production scenarios in order to refine and prioritise focus areas for investment.
“What if, through integration of new boiler technology, we could improve the gas efficiency of all production lines & baseload by 5% starting from June 26 for £100,000?”
When combining the energy model with a constrained optimisation algorithm, an initial planned production schedule can be iteratively refined with the objective of minimising a user selected KPI (such as Average Daily Energy Consumption (MWh or GBP), Average Daily Emissions (Te), or Total Production Output (Te) etc). The optimised schedule is automatically maintained in parallel to the baseline plan and is available to stakeholders as a recommendation and for comparison against the baseline plan. The constraints applied to the optimisation algorithm were designed by plant experts to ensure realistic and achievable solutions are returned that meet production objectives and prevent exceeding imposed production limits.
Deployed as a web application with a refined UX, the digital twin enables a human-in-the-loop cycle to routinely deliver insights from digital solutions to support a variety of operational tasks and decisions.
Validated and benchmarked against historic data, this approach has shown promising results. When optimising the past years production schedule, with the goal of minimising average daily CO2, the optimised schedule was found to reduce emission by ~5%, corresponding to 400Te of CO2 across the year. This result was achieved while simultaneously reducing energy costs by 5% and hitting all required production milestones. The savings were largely derived by shifting workloads with a high gas draw to the warmer months of the year while maintaining a uniform electricity demand profile. This simple explainable outcome has already led to more creative thinking in how the operator can adjust their customer engagement strategy in order to shift demand for their products to more closely reflect the optimised production schedule.
