Computational modeling for manufacturing is most useful when it does more than produce a plausible simulation. Industrial teams need models that connect physical behavior to measurements, make uncertainty visible and provide results that can inform engineering decisions. Work summarized by the National Institute of Standards and Technology’s Applied and Computational Mathematics Division (ACMD) for October 2018 through December 2019 offers examples of how applied mathematics can support that goal.
The projects span materials, measurement systems, biological technologies and high-performance computing. For automation engineers, system integrators and operations leaders, their common thread is the relationship between models and trustworthy data: simulation can help interpret complex systems, but its value depends on sound mathematical methods and measurement foundations.
Computational modeling for manufacturing starts with measurement
In measurement science, mathematical models help describe the physical process being measured, guide measurement-system design and turn raw observations into useful information. Algorithms may also be needed to estimate system parameters or reconstruct behavior that cannot be observed directly. These needs are familiar in industrial settings, where sensors, equipment and processes produce data that must be interpreted before it can support control or process decisions.
ACMD’s work on ill-posed, time-reversed partial differential equations illustrates a particularly challenging case. Such problems arise in inverse measurement tasks: the desired result may require reconstructing an earlier or hidden state from later observations. Straightforward step-by-step numerical methods can be unstable, causing errors to grow rapidly. Researchers developed an approach to stabilize explicit difference schemes for certain multidimensional, nonlinear dissipative equations, enabling useful reconstructions over short periods.
The industrial takeaway is not that every process model can be run backward. Rather, inverse problems require methods designed around their mathematical difficulties. When a measurement or inspection task depends on reconstructing an unobserved condition, stability and error behavior deserve attention alongside the model’s physical assumptions.
Materials modeling needs credible links across scales
Materials development often relies on simulation to explore behavior that is difficult or costly to measure directly. Molecular dynamics, for example, can represent many atoms over short time spans, but connecting that level of detail to larger-scale material properties is challenging. Coarse-grained models reduce complexity, yet approaches without quantitative error estimates or clear links to more detailed models can be difficult to assess.
ACMD researchers described a hierarchy of methods intended to connect models of differing complexity to all-atom representations. One approach reformulates interaction potentials for rigid-body molecules using a generalized multipole expansion. The broader lesson for industrial R&D is to treat simplification as a modeling decision that needs justification. Reduced models can be valuable, but engineers need to understand what has been omitted and how that affects the result.
Uncertainty quantification is part of that picture. It helps assess how assumptions, measurements and computational choices influence outputs. For teams using simulation to compare candidate materials or guide experiments, documenting uncertainty makes results easier to interpret and reuse.
High-performance computing can support measurement standards
Some simulations require computing resources beyond a desktop workstation, particularly when a model must capture complex material behavior at useful fidelity. In 2019, ACMD-led simulations helped determine certified property values for NIST Standard Reference Material 2497, a reference concrete used for rheological measurements. The work used large-scale parallel computing on an external high-performance computing facility.
The example shows how simulation can contribute to measurement infrastructure, not only product design. The cement paste and mortar reference materials were cited in a new ASTM test method for measuring rheological properties of cementitious materials. ACMD’s simulation code was also being used to study dense suspensions flowing in pipes.
For plants and equipment developers, the practical point is that computational tools can help characterize material behavior relevant to process equipment and test methods. But the results become more useful when tied to reference materials, defined measurements and methods that other groups can apply consistently.
Better data analysis strengthens complex measurement systems
Automation systems increasingly depend on data from many sensors and interacting components. This makes analysis methods important as well as data collection. ACMD work on microfluidics, for example, addressed the challenge of measuring very small volumetric flow rates. Researchers developed flowmeters able to measure rates down to approximately 10 nL/min with relative uncertainty of 5% or less. A second device could identify zero flow within 0.1 nL/min, supporting flowmeter calibration and flow-stability characterization.
These results were enabled by mathematical scaling relationships and symmetry-breaking arguments. They demonstrate how physical understanding and mathematical analysis can help design measurement approaches for regimes where direct measurement is difficult. Similar thinking can help engineers evaluate whether a sensor or test method is appropriate for a process’s operating range, rather than relying on nominal readings alone.
Data analysis also features in the division’s work on biological measurements. For two-dimensional NMR spectroscopy used to characterize monoclonal antibodies, researchers applied principal component analysis and a clustering method to spectra collected under different conditions. The analysis identified sample type and temperature, supporting the development of a more consistent measurement approach. While the application differs from conventional factory automation, the general challenge—extracting reliable information from complex, condition-dependent data—is widely relevant.
Practical implications for engineering teams
NIST’s activities point to several considerations for organizations adopting simulation and advanced analytics:
- Connect models to measurements. Define what the model represents and how its outputs relate to observable process data.
- Assess uncertainty and stability. A numerical result is not automatically a reliable result, especially for inverse problems or simplified models.
- Match computing methods to the problem. High-fidelity simulations may require parallel computing, while other tasks may call for focused mathematical analysis or data-reduction techniques.
- Build comparability into the workflow. Reference materials, calibration and repeatable measurement methods help different teams interpret results consistently.
Conclusion
Applied mathematics supports industrial decision-making when it links physical models, computation and measurement into a coherent process. NIST’s FY 2019 work—from reference-concrete simulations to microfluidic flow measurement and materials modeling—shows why model quality, data interpretation and uncertainty matter together. For industrial automation teams, the implication is clear: treat computational tools as part of the measurement system, and evaluate not only what a model predicts, but also how well its assumptions and outputs are grounded in evidence.