Engineering simulation glossary

Computational fluid dynamics (CFD) is the use of numerical methods and computers to solve the equations that govern fluid flow and heat transfer. Engineers model a geometry, divide the fluid region into a mesh and compute fields such as velocity, pressure and temperature, to predict how a design behaves before it is built or tested.

Design of experiments (DOE) is a structured method for planning which tests or simulation runs to perform, so the results support valid, objective conclusions from as few runs as possible. It chooses the combinations of input factors to test, and is used to screen influential variables, fit response surfaces and train surrogate models.

Design space exploration is the systematic evaluation of many design alternatives across the ranges of their design variables, such as geometry, materials and operating conditions. It shows which regions of the design space perform well, which variables matter most, and which concepts deserve detailed analysis. In engineering simulation it usually combines a design of experiments, simulation runs and surrogate models.

Finite element analysis (FEA) is a numerical method for predicting how a part or structure responds to loads, heat, vibration and other physical effects. It divides the geometry into many small elements, approximates the governing equations on each one, and solves the assembled system to compute quantities such as displacement, stress and temperature.

A graph neural network (GNN) is a neural network that operates on graphs: nodes connected by edges. In engineering simulation, the mesh becomes the graph, and the GNN predicts quantities such as velocity, pressure or displacement by passing learned messages between neighboring nodes, so it works directly on irregular meshes.

A high-fidelity simulation is a numerical model that represents a system's physics in the detail needed for the accuracy a decision requires, usually on a fine mesh with few simplifying assumptions. It is the expensive reference run, such as a well-resolved CFD or FEA case, against which cheaper models are measured.

A neural operator is a neural network that learns a mapping between functions rather than between fixed-size vectors. In simulation, it learns the solution operator of a family of partial differential equations: given an input field, such as a boundary condition or a material property, it predicts the solution field, and the same trained model works across different discretizations.

A physics-informed neural network (PINN) is a neural network trained to satisfy the governing equations of a physical system, not only to fit data. The equations, such as the Navier-Stokes equations, enter the training loss as a penalty, so the network can learn a solution from little data or infer hidden quantities from measurements.

Pre-training and fine-tuning is a two-stage way to train a machine learning model. The model is first pre-trained on a large, broad dataset to learn general patterns, then fine-tuned on a smaller dataset for a specific task. In engineering simulation, fine-tuning adapts a model pre-trained on many simulations to a team's own geometries, physics and conditions.

A reduced order model (ROM) is a simplified version of a high-fidelity simulation that keeps the dominant behavior of the system while solving far fewer unknowns. Projection-based ROMs extract a small basis from snapshots of full simulations and project the governing equations onto it, so the reduced model can be evaluated cheaply across many parameter values.

Response surface methodology (RSM) is a set of statistical techniques that uses a sequence of designed experiments to fit a simple model, usually a second-order polynomial, linking several input variables to a response. Engineers use the fitted surface to see how the response changes with each input and to find settings that improve performance.

A surrogate model is a fast approximation of an expensive simulation. It is built from a limited set of solver runs and learns to map design inputs, such as geometry or operating conditions, to the outputs engineers need. Teams use it to evaluate far more design candidates than the full simulation allows, then confirm the best ones with the solver.

Topology optimization is a computational method that decides where material should go inside a given design space, for given loads, supports and constraints. It usually runs a finite element analysis at every iteration and updates a material density in each element, producing free-form layouts that engineers use as a starting point for lightweight structural design.

A transformer is a neural network architecture built on attention, a mechanism that lets each element of the input weigh its relation to every other element. In engineering simulation, transformers treat mesh points, or groups of points, as tokens and learn to predict physical fields such as pressure, velocity or stress across varied geometries.

Uncertainty quantification (UQ) is the practice of measuring how uncertain a simulation or model prediction is, and why. It traces how variations in inputs, physical parameters and the model itself change the result. In engineering simulation, UQ tells a team how far to trust a prediction before it informs a design decision.

Verification and validation (V&V) are the two checks that establish how far a simulation can be trusted. Verification asks whether the computational model solves its equations correctly. Validation asks whether those equations represent the real system, by comparing predictions with experimental data. Together they set the credibility of a simulation result.