Digital twin platform bolsters resilience for pharma plants

A new platform that uses artificial intelligence (AI) and real-time plant data to enhance fault detection, system monitoring, and predictive maintenance in pharmaceutical manufacturing has been co-developed by a Singapore-based research consortium. The platform creates a digital replica of a manufacturing plant, allowing CDMOs, equipment vendors, and other manufacturers to better understand operations, manage risks, and test potential changes before implementing them.

The digital twin platform, created by the Cambridge Centre for Advanced Research and Education in Singapore (CARES) and Agency for Science, Technology and Research (A*STAR), offers a new lens to oversee pharmaceutical manufacturing. By replicating a plant’s processes and combining that with predictive analytics, the platform helps companies optimize production while preparing for both technical and external challenges.

“The digital twin hosts an idealized description of the manufacturing system, represented by calibrated first principle and hybrid models, and the digital replica of the actual plant,” says Alexei Lapkin, lead principal investigator on the project and a professor of sustainable reaction engineering at the University of Cambridge.

How the digital twin platform works

CARES led the development of the digital twin’s ontology and physical modeling, while A*STAR’s Institute for Infocomm Research (I2R) developed an AI agent for anomaly detection, which was aimed at supporting predictive maintenance needs, and created a full cloud implementation of the digital twin system on the Microsoft Azure platform.

Knowledge graph connects physical models, properties data, and data plant sensors with relationships between them mapped by ontologies to create the digital twin. The ontology is used to explicitly represent how elements of the digital twin relate to each other. And the two complementary modeling approaches, the first principles and machine learning models are also placed in the knowledge graph.

One is a calibrated replica, a plant-specific model that mirrors the layout, equipment, and operating parameters of a given facility. This digital copy is fine-tuned to reflect real-world performance and incorporates AI to support knowledge management. Complementing it is the first-principles model, a more generic, physics-based representation of plants with similar functions. This physics-based representation does not factor in exact equipment details of the plant. 

"The (first-principles model) is a really useful tool for the company to collect everything engineers know about their processes in the form of equations and their parameters," according to Lapkin. "But they also might be interested in the digital replica which represents specific instances — a particular device or particular configuration. That model works together with the knowledge-based model."

Dr Lianlian Jiang, Co-Lead Principal Investigator of the project, Unit Lead of Digital & Sustainable Manufacturing at  A*STAR I2R, said:  “The AI agent in this ontology-based digital twin can be extended beyond anomaly detection to support quality monitoring, production scheduling, and resource planning. By embedding domain knowledge into the system, the technology helps capture and transfer critical expertise while complementing staff expertise. We are glad to work with CARES and industry partners to further scale this technology to meet the needs of more complex manufacturing environments.”

Addressing the needs of pharma manufacturing

Pharma manufacturers face unpredictable stresses that can disrupt operations, from equipment issues to global supply chain challenges like tariffs or pandemics. Digital twin platforms are designed to alleviate these stresses by offering foresight and resilience, Lapkin says.

For example, predictive maintenance can identify potential failures before they escalate, thereby extending equipment life, reducing downtime, and cutting costs. 

“If you have a digital replica of your plant, you can do a lot more with this,” Lapkin contends. “You can think about how to schedule production for the next year, or how do you schedule the throughput of your plant.

Beyond cost savings, the platform also acts as a living record of operations. “Large manufacturers will have the history of all the decisions they have made because they are replicated in the digital twin,” Lapkin explains. “They will have a history of all the runs, standardized and matched to equipment to see what worked well, what didn’t, where there are problems, and bottlenecks.”

A foundation for digital factories

Many pharma companies are now looking at having their own digital factories, says Lapkin, which will require investing in technologies like the digital twin platform.

“They will not successfully operate unless you have a full digital twin,” according to Lapkin. “And, so this technology we’ve built is basically the foundation for digital factories.”

The push for digitalization is particularly urgent given how dramatically supply chains have shifted in recent years. Companies are seeking ways to reconfigure supply networks with less disruption and cost.

“This drive towards digitalization and digital factories is supporting that need, because you have the information through modelling enabled by digital twins,” says Lapkin. “You’re able to do process development, and design, and validation with the digital tools much more effectively compared to what was used to do in the past. The drivers behind this — business pressures, political realities, climate change, and economic sustainability — all highlight the need for agile digital technologies in manufacturing.”

A benefit toward sustainability

The platform also advances sustainability. By enabling more localized production and optimized supply chains, transport distances, and costs are reduced. 

“When we were starting to think about how supply chains need to be restructured for resilience, it turns out that it’s better for sustainability,” says Lapkin. “This is because supply chains become shorter, transport costs are lower, local feedstocks could be used, and local production is being used. This is way better for all aspects of sustainability. So, this is where the outputs or the benefits of technology could be multiple KPIs.”

From research to real-world impact

The system was successfully demonstrated on a real-time manufacturing testbed provided by Accelerated Materials Ltd., a Cambridge startup. Hosted on Microsoft Azure cloud, it allowed engineers to monitor plant performance and identify potential faults in real time.

A part of this technology, the model development and maintenance knowledge graph, is now being commercialized by Chemical Data Intelligence (CDI) Pte Ltd., a CARES spinoff, supported by the Pharma Innovation Programme Singapore (PIPS) Consortium. The initial product roll-out will be done with PIPS core members during 2026 with the full commercial roll out expected shortly after.