Structures, water, computers, languages and people (not necessarily in this order)

The 7 main causes of numerical rubbish

Finite element analysis (FEA) is an everyday engineering tool but it is very easy to make mistakes. I expose the seven most common traps I have found in engineering consultancy and suggest ways of working aroud them:

1- The virtual reality trap. Analyst can become so involved in FEA that the link between the real structure and the model may be forgotten.

2- The time trap. It is usual that the modelling takes so much time and effort that little is left for verification and validation of results, alternative designs, writing reports and documentation, backups and reviews by other analysts.

3- The "we-use-FEA-on-everything" proud. FEA is an expensive method and it s easy to forget that in large structures FEA tend to offer a deceiveing level of detail. Other methods, such as calculating stresses from equations in Roark's formulas for stress and strain or a simple model of linear elastic bars can be faster, less expensive and more reliable.

4- Averaging stresses. Stress averaging provides visually pleasant results but does little to improve accuracy and masks the highests stresses providing misleading results. In fact, comparing averaged to nonaveraged stresses gives a intuitive measure of the quality of the numerical solution.

5- The CAD trap. The geometry should be defined explicitly and clearly: it is absolutely impossible to calculate anything without a previous design. The use of preprocessors such as Patran or Catia gives the engineer the abily to generate complex geometries but this does not necessarily mean that they will be more accurate that simple AutoCAD generated models. On the other hand, oversimplification can leave out critical load paths. The geometrical model should represent a realistic compromise between the accuracy of the source of geometry and the correctness of the model.

6- Coarse mesh + low order elements + few modes of vibration. This is probably the worst possible combination when shear analysis is critical (f.ex. in walls under seismic load) so my advice is use a fine mesh, use at least second order elements if possible and ensure that the masses are at their centre of gravity, that shear-locking has been avoided and that the participation factor in all directions is close to 100%.

7- The management trap. Economic competition among engineering firms implies low time with which calculations are executed. However, the attitude of the manager must ensure through policy and example that engineering staff can allocate time to check analyses that are sufficiently accurate and correct for design of new structures and investigation of existing ones instead of simply putting pressure on the analysts.

I, therefore, recommend to the computer analyst that she or he adopt procedures to ensure meaningful and accurate analysis results and to the management of structural design firms that they enforce validation and verification procedures as a matter of quality control.

6 comments:

  1. I got involved in another trap. Let me call it "non linear time costs trap", meant to be how modelling and processing times and difficult level grow rather exponentially as you try to increase model accuracy.

    Very good text, Xose Manuel, I´d wish you to translate it into Galician some day.

    ReplyDelete
  2. I do not think it is a question of complexity versus simplicity of the models. Everyone wants the pretty FEA plots but no one wants to allocate the required time to learn how to create models properly or to compare them to approximate analytical results. I truly believe that most management people think FEA replaces the need for fundamental analytical skills or they expect the engineer to perform the analytical verification tasks, the gross numbers, on her/his personal time and without peer review. In my experience, there is so much pressure to just get the job out of the building. People are forced to work fast, not smart. This is the main reason of lack of credibility of the Spanish simulation industry.

    ReplyDelete
  3. Prof Alex Scordelis often admonished his students (notably Mark Ketchum, Reinhard Ludke, Bruce Maison, John Eidinger): You should CHECK Sum Verticals = total weight!!!

    Course 220A, UC Berkeley, 1975.

    Still good advice.

    ReplyDelete
  4. In FEA models, simple change in end constrain give large differnec in the results.
    Engineer should have an idea of what type of stresse will indused and he should always use his experience and knoleddge when analysing the resluts. Otherwise results may mislead the enginner.

    ReplyDelete
  5. I got involved in another trap. Let me call it "non linear time costs trap", meant to be how modelling and processing times and difficult level grow rather exponentially as you try to increase model accuracy.

    Very good text, Xose Manuel, I´d wish you to translate it into Galician some day.

    Posted by F. Míguez to not only bridges at Tuesday, January 19, 2010

    ReplyDelete