Improvement based on number of attempts
As each formulation change requires a new round of experiments, development time and material costs become unpredictable.
/ INDUSTRIAL SOLUTION
22
AI Recipe System (gPROMS)
Develop formulations with a validated model of the product and its manufacturing process.
With Siemens gPROMS FormulatedProducts, we model formulation and production together, grounding AI-assisted recipe decisions in a mechanistic model.
Explore the solution
Develop the formulation not by number of trials, but by a validated model that describes the product and the process together.
Results that do not hold during scale-up
A recipe that works in the laboratory may not produce the same result on a production scale due to differences in mixing, temperature, and residence time.
The risk of a data model without physics
Models trained with historical data alone do not produce reliable predictions outside the data range and cannot explain the decision reasoning.
Design the product and process
within one model.
Siemens gPROMS FormulatedProducts combines formulation behavior and manufacturing processes in the same mechanistic model family. Data-driven methods are added to this model as a hybrid layer, not replaced.
Model the formulation
Define ingredients, solubility and product properties with physics-based models.
→Link the production steps
The crystallization, drying, granulation and tableting steps are solved in the end-to-end flow.
→Define the design space
Critical parameters, risk and operational range are scanned on the model to prepare for scaling.
→Deploy the recipe to the plant
The verified recipe is transferred to the ISA-88-based batch system and control layer.
/ gPROMS FORMULATEDPRODUCTS SCOPEBase recipe decisions
Base recipe decisions
on physics.
The platform supports one workflow from development to production for pharmaceutical and formulated products, using mechanistic and hybrid model libraries.
Formulation design
Compare candidate formulations for solubility, stability and manufacturability before physical trials.
Review the product →Crystallization, drying, granulation
In solid form, particle size, moisture and flow behavior are calculated together throughout the process steps.
Review the product →
Tableting and oral absorption
Compression behaviour and absorption models link product performance to formulation decisions.
Review the product →
Bioreactor and separation
Digital twin models are built for the bioreactor, membrane filtration and chromatography steps.
Explore the platform →/ COMBINING AI AND MODELSThe learned layer,
The learned layer,
Within the limits of physics.
Data-driven methods are used to accelerate the mechanistic model and fill gaps; each suggestion is presented with the validity range of the model.
Mechanistic and data-driven integration
The learning component works together with the physical model; the prediction is based on an explanatory basis.
Parameter scanning and risk
Critical process parameters and acceptable operational range are systematically determined.
From lab to production.
Using the same model at different equipment scales, the transfer risk is pre-evaluated.
Critical quality attributes
Variables determining product quality are monitored together on the formulation and process side.
Deviation analysis
Production deviations are compared with model scenarios to narrow down possible causes.
Batch migration
The verified recipe parameters are transferred to the ISA-88 recipe structure and control layer.
/ ENGINEERING INPUTSDefine the decision problem
Define the decision problem
before building the model.
- 01Product and quality goal
Critical quality characteristics, specification range, and acceptance criteria are clarified.
- 02Process and equipment
Existing equipment scales, cycle times and constraints are modeled.
- 03Data scope
The adequacy of laboratory, pilot and production data and sources of uncertainty are evaluated.
- 04Verification plan
It is written down with which experiments the model will be tested and in which decisions it will be used.
From the decision problem to a verified recipe, Six controlled steps.
- 01Select usage scenario
It determines which of the development time, yield, quality, or scale-up risks is targeted.
- 02Collect data and information.
The formulation history, experimental results and process records are gathered.
- 03Build the model
Product and process steps are modeled using the gPROMS FormulatedProducts libraries.
- 04Verify by experiment.
The target experiment set is planned according to the model's prediction and the results are compared.
- 05Scale up
Production-scale equipment and cycle conditions are tested on the model.
- 06Transfer to production
Recipe parameters are moved to the batch system; deviation and performance tracking is set up.
Where formulation determines product performance:
production areas.
Pharmaceuticals and life sciences
Solid form development, scale up and verifiable process design.
Specialty chemicals
Optimization of yield and product characteristics in reaction and crystallization steps.
Cosmetics and cleaning products
Prediction of emulsion, viscosity and stability behavior at the formulation stage.
Food and beverage
Correlation of product quality with process conditions in mixing, thermal processing and drying steps.
Let’s clarify your requirements
with the right engineering team.
Tell us briefly about your operation so we can establish the right next step together.
Discuss your requirements Your request is routed directly to the relevant engineering team.