The basic idea
Many physical problems in engineering are described by equations. How does a bracket bend under a load? How does heat spread through an engine part? How does air move around a wing? The equations that describe these situations are well known in many cases, but they can seldom be solved exactly for a real, irregular shape. Engineering simulation is the practice of finding approximate answers with a computer.
The usual strategy is to cut the problem into small pieces and solve a simple version of the problem on each piece. The pieces are connected, so what happens in one affects its neighbors. A computer can handle a very large number of such pieces by repeating simple arithmetic, which is what made these methods practical only after electronic computers appeared in the 1940s and 1950s.
Stress
Structural analysis asks how a part deforms and where it is most strongly loaded. The finite element method, which developed in aerospace and university groups from the 1950s, divides an object into small elements, such as little triangles or blocks. Each element has simple rules for how it stretches or bends. Combining these produces a large set of equations that, when solved, gives estimated displacements and stresses across the whole part.
Engineers use such results to look for regions where stress is high, which may indicate a place where a part could crack or yield. Early models were coarse because memory and speed were limited, so an analyst had to choose carefully where detail was needed and where it could be left rough.
Heat
Heat conduction can be treated in a similar manner. The object is divided into small regions, each with a temperature, and the program calculates how heat moves between neighboring regions over time. Such analysis was valuable for engines, electronics and structures exposed to extreme temperatures, including parts of spacecraft and reactors. Weather forecasting and climate modeling use related ideas on a far larger scale. Methods first developed for structures were often adapted to heat, which helped the field grow, since a well-built mesh could be reused for different kinds of analysis.
Fluid flow
Fluid flow is generally harder. The equations describing how liquids and gases move are complicated, and flows can become turbulent, with swirling motion on many scales at once. Computational fluid dynamics began to develop in the 1950s and 1960s, with national laboratories playing an important role because they had access to the most capable machines of their day. Early flow calculations were often limited to simple shapes and two dimensions.
Later, as supercomputers and then ordinary workstations improved, three-dimensional flow around aircraft, cars and buildings became practical to study. Even so, turbulence in particular is usually represented by simplified models, which means that results come with a degree of uncertainty that analysts have to understand.
What the process looked like
The typical workflow had three stages, which later software labeled preprocessing, solving and postprocessing.
- Building the model. Define the geometry, divide it into elements, assign material properties, and specify loads and constraints.
- Solving. Run the program, which assembles and solves the equations.
- Interpreting. Examine the output, which might be tables of numbers, and later plots, with color showing stress or temperature.
In early years, the first and third stages consumed most of the human effort. Errors in building the model, such as a missing connection or a wrong unit, were common, and the output did not always make such mistakes obvious.
What simulation can and cannot predict
Simulation is now routine in scientific computing and in industry, but its results should be read with care.
- A model is a simplification. Real parts have imperfections, variations in material and complicated joints that a model may leave out.
- Results depend on inputs. If loads, material properties or boundary conditions are wrong, the answer will be wrong, and it may still look convincing.
- Resolution matters. A coarse mesh can miss local effects such as stress concentrations near sharp corners, while an overly fine one wastes time. Analysts commonly check whether results change when the mesh is refined.
- Some phenomena are hard to capture. Turbulence, fracture, material fatigue and the behavior of complex assemblies often need special methods and remain active areas of study.
For these reasons, engineers generally compare simulation with physical tests, at least for the first designs in a new area. This practice, often called validation, helps establish where a model can be trusted. Simulation reduces the number of experiments needed, but it does not remove them, and an analysis that has not been checked against reality should be treated as a prediction, not a measurement.
A gradual change in practice
Early on, simulation belonged to specialists at organizations with large computers. Over the decades, as hardware became cheaper and software friendlier, more engineers could run basic analyses on their own designs, and some CAD packages began to include simple analysis tools. This broadened use, but also raised new questions about training: an easy-to-run simulation can produce an attractive but misleading picture in the hands of someone unfamiliar with its assumptions.
The history of these methods therefore includes both technical improvement and a long-running effort to teach users how to judge results, a task that remains as important as improving the software.