Digital twins for heat treatment, optimization, and control of industrial furnaces – TwinHeat

In many industries, and in particularly in glass industry, furnaces are widely adopted technology involving fossil combustibles. Due to the global climate change, breakthroughs are expected to reduce the CO2 footprint. This is becoming all the more important for glass industry, where furnaces are used to melt a glass forming liquid at temperature up to 1600°C, a process today considered to be one of the most highly energy-intensive and CO2-emitting globally.

Numerical simulation is a quite standard tool for modeling furnaces. However, it remains complex and requires significant computational time to accurately design a new furnace, which limits the number of cases tested and of variants studied. This is further complicated by the ever-increasing number of operating cases and the need for flexibility in new industrial plants.

A precise numerical framework to design new furnaces and to optimize the control of existing furnaces is then a subject of major importance. It allows to improve the design of new heating systems aimed at decreasing energy consumption or find alternative sources (decarbonized energy: electricity or hydrogen), to reduce the time and cost for controlling furnaces (by reducing time experimentation), and therefore to continually develop safe and reliable products that meet the customer specifications .

Despite the evident industrial interest for accurate modeling and control of heating processes, there is no global study neither global answer addressing this problem in an industrial context. To optimize the operating conditions of existing furnaces as well as to study the optimal combinations of heating parameters, to minimize energy consumption, to adapt to energy transition, while respecting product quality constraints, an innovative numerical framework integrating the latest advances in artificial intelligence (AI) with traditional computational fluid dynamics (CFD) can therefore become a decisive cornerstone towards a more climate-neutral, sustainable, yet productive and competitive glass industry.

Two simultaneous actions are needed to achieve this breakthrough: first the development of a ground-breaking numerical strategy attended by machine learning for accurate prediction of the fluid dynamics of furnace under well-defined and controlled conditions. Second, the development of an embedded Deep Reinforcement Learning strategy for furnace control system that can be used as a basis for an automatic control and optimization system, notably through intelligent predictive models.

Four major national industrial groups facing such challenges will partner to support the TwinHeat chair, to create the digital twin for their own heating furnace systems, to leverage its understanding, its thermal treatment, its control and finally its optimization: Saint-Gobain, Verallia, Fives, and Pochet. Following the recommendation of the chair panel in 2023, a new partner will strengthen this consortium’s capability to disseminate such digital twin technology and tap into its full potential at industrial level: SCC Sciences Computers Consultants, 30 years of experience in developing and marketing numerical simulation software dedicated to industrial material transformation processes.

The TwinHeat chair contributes then to a long-term vision of high-fidelity computational tools as a basis for reliable simulations of furnace processes allowing the partners to remove several major technical barriers, to faster aid to decision for delivering high quality materials, minimizing energy consumption, preventing material reject and thus optimizing heat processes.