A Short and Incomplete History of Pre- and Post-Processors

As seen in our previous example, applied finite element theory requires more than just the ability to find a solution to a posed problem, one must also pose the problem!  Even after finding a solution, significant work usually awaits after results are found, and software for this purpose has evolved in tandem with FEA solvers.
The Dark Ages
Before the introduction of software designed to aid in the generation of finite element models, modeling was accomplished in a manner similar to other prehistoric engineering work: with reams of paper.  The more limited computers available in the 60s and 70s could find solutions to models of limited size.  Therefore the manpower required to take copies of blueprints, dimension the locations of grids, and connect the related elements, was more manageable.  The limited model size also made more practical reordering the grids to allow for faster solutions to structural FEAs.  As is common in engineering, analysts became very talented at work that would later be done faster and usually better by automatic algorithms.
Several phrases still in common use when referring to finite element models, particularly Nastran, come from this era.  One in particular is the term 'card', which describes the lines of text used to describe finite element entities, which were literally stored on punch cards.  Clever analysts would have large libraries of punch cards which could be readily used to describe the properties of elements and materials.

Proprietary codes and the dawn of Patran
As the 70s gave way to the 80s, computers fast enough for graphical display of geometry allowed for the development of programs which would allow for easier generation of models.  A number of large engineering companies developed their own proprietary pre-processors, while MSC corporation, the original vendor of Nastran, developed the matching pre-processor Patran.  Patran was initially given to Nastran customers at no additional cost, and rapidly became the industry standard pre-processor.  This led competing FEA codes, such as ABAQUS, to partner with MSC to have support for their solvers added to Patran.  Patran would grow to have very broad functionality, allowing for importing of geometry from CAD design systems, renumbering of grids to reduce solution times, and graphical display of element properties to allow rapid validation of models.

Hypermesh, ABAQUS CAE, and ANSYS
Beginning around the end of the 80s additional vendors began competing in the marketplace.
Hypermesh was authored with the specific goal of being solver agnostic, allowing analysts to familiarize themselves with a single solver and quickly adapt models from one solver to another.  Later iterations of Hypermesh allowed for greater modifications of existing models independent of CAD generated geometry, and allow for pre-processing specific to the optimization capabilities of its own solver, Optistruct, descended from Nastran.
ABAQUS CAE took a different track from Hypermesh, with a focus on a single solver and the ability to make full use of the unique features of that solver.  A rarity in pre-processors, CAE allowed CAD style dimensioning of 2-D sections as the origins of 3-D models.  In a preview of a growing trend in Pre-processors, emphasis was made on generating geometry then generating mesh associated with that geometry,   rather than operating on the mesh directly, as was the style of Hypermesh or Patran.
ANSYS took a similar approach to ABAQUS, creating their own pre-processor, with the additional step of hiding the input into their solver from the user.   As solvers predated pre-processors, the typical pre-processor workflow was a pre-processor that generated an ASCII text file that could be inspected and edited by the user, and also imported into a competing pre-processor.  By hiding this abstraction from the user, ANSYS created a more complete solution that was also less amenable to users mixing and matching pieces of their workflow from other vendors.

FEMAP, Solidworks simulation, and CATIA Analysis
As desktop computers became fast enough to solve large and not particularly efficient FEA models, more vendors became interested in broadening the FEA user base from dedicated analysts to typical drafters and designers.
FEMAP was authored as a typical pre-processor of the previous era, handling multiple solvers and generating an input deck, but with a focus on affordability and intuitive ease of use rather than sophisticated capabilities.
Solidworks Simulation and CATIA Analysis, both now a part of the Dassault systems product lineup, each add FEA capabilities to advanced CAD systems.  Solidworks incorporates the COSMOS solver, whereas CATIA incorporates the ABAQUS solver.  Both take care to make reasonable choices of FEA parameters for the type of analysis chosen by the user, such as element type and internal solver, and are integrated into the larger parametric workflow of the designer.

Optimization and 
The ideal future workflow for engineering is an intuitive GUI which allows a minimally trained analyst to:

1. Conceptualize a design
2. Evaluate its structural adequacy for its function
3. Optimize its design for both cost and function
4. Generate design detail, including variability tolerance, sufficient to produce the design
5. Save the results of the analysts work into a product data management system

 With Hypermesh and Solidworks/CATIA, there are two reasonable workflows that are likely to continue to be used:

• Design first: For smaller design teams working on designs with a reasonable number of pieces with shorter design cycles, such as a consumer product rather than a jet fighter, the Solidworks/CATIA workflow tends to work best.  Software licensing costs are usually more reasonable, training investments more manageable, and models sizes easier to solve by more typical computing hardware. 
• Analysis first: For larger design teams, typically with dedicated analysts teamed with dedicated designers, an analysis first methodology tends to work well.  At several steps and levels of detail, an analyst will take an initial concept of a design and, with the use of pre-processors such as Patran and Hypermesh, mature it into a more reasonable design using either hand iterations or optimization. This will yield dimensions of a design that the analyst will then pass onto the designers who will realize these dimensions into producible designs.

Closing Remarks
As computing hardware and FEA software continue to evolve, the trend will likely be towards the more intuitive design first approach becoming more common.  This trend will take some time.  An FEA model that hasn't been carefully designed to have only as much detail as necessary can take a very long time to solve, given that solution times typically grow with the square of the model detail, which for a cube of material grows with the cube of detail in any direction!  On the other hand, Moore's Law only generates a double of transistor count (not speed!!) about every 18 months.  Both design methodologies are likely to remain in force for some time to come, and their continued influence on each other should yield dividends in efficiency and intuitiveness that will benefit both.