Computational Engineering / Diagnosing computational models

The Model Doctor

The method, applied: name the discriminating check before you name the cause.

Lesson review: expert verification remains open

This lesson has received a full-context editorial review and a check of its technical explanations. Selected arithmetic has been checked separately. This is not independent engineering sign-off.

Checks needed for project use

A computational engineering reviewer must check algorithms, convergence claims, statistical assumptions and numerical examples beyond the cases explicitly recorded.
Model Doctor lab case banks are separate from these lesson blocks and need their own full technical audit.
Confirm mesh-quality limits and punching checks for each adopted element and design standard.

These are educational models and exercises. Check their assumptions before using the results. The full technical content and every diagram have not yet been independently verified.

Concrete and steel calculations retain their stated first-generation Eurocode teaching basis. Use the standards and amendments specified for your project; do not mix generations without an appropriate design basis.

Read the lesson register and outstanding review notes

Why this matters

Everything in this module so far has been a fault and its symptoms. This lesson is about the procedure — and the single most important thing about it is an ordering that feels wrong the first time.

Name the check before you name the cause. An engineer who reaches for a diagnosis first will find evidence for it, because a model has thousands of numbers and some of them will fit any story. Committing to a discriminating check first is what stops that.

By the end of this lesson you should be able to

Apply the seven-part method to an unfamiliar model
Choose a check that discriminates between candidate causes rather than confirming one
Distinguish local from global symptoms as a first cut
Say what a correction must be verified against

Diagnostic sequence

Record symptoms, choose a discriminating check, propose a cause, test it, make a justified correction and verify the result. Separate observations from interpretations so that a plausible early explanation can still be rejected.

Local and global patterns help prioritise checks. They do not divide causes cleanly: a unit error can affect one assigned section, and a local connection fault can redistribute forces throughout a frame.

Compare quantities with independent estimates, check intended loads against entered loads, assess complete equilibrium and inspect connectivity. Test sensitivity to uncertain constraints, stiffness and mesh choices. A balanced model can still have a wrong load path; one mesh refinement cannot prove convergence or diagnose a singularity.

After a correction, confirm both that the discrepancy is resolved and that the intended behaviour remains. Replacing a stiff eccentric link with a constraint, for example, must preserve the eccentric load transfer as well as improve numerical conditioning.

Try it

Model Doctor

80 original, fictional cases across 12 fault categories. Read the symptoms, commit to a diagnosis, then open the method.

The frame that will not solve

A single-bay, single-storey portal frame modelled with beam elements. Both column bases are given vertical restraint only. Gravity load is applied to the rafter.

Symptoms

The solver stops with a message about a singular matrix or a zero pivot
No results are produced at all
Adding more load makes no difference to the message

What is your diagnosis?

Nothing is revealed until you commit. That is the discipline the module teaches: an engineer who reads the answer first will find evidence for it in any model.

About these cases

Every case is invented. None is a retelling of a real project, and none is taken from any source.
Where a real failure is relevant it is named once, in Module 13, with a pointer to the published investigation — not reconstructed.
80 cases are written against a target of 80. The categories still short are listed in the coverage matrix.

What this shows: Name the discriminating check before you name the cause — an engineer who reaches for the diagnosis first will find evidence for it.

Worked example

Diagnosing an unfamiliar model

Given

A multi-storey frame where beam deflections under gravity are almost zero at every level
Column axial loads are erratic — some far above their tributary area, others near zero
Total reactions equal the applied load exactly
No warnings

Find

The cause, following the method

Check yourself

A single bay of a floor deflects three times more than its neighbours under identical load, and the columns under it carry almost nothing. Total reactions are correct. What is the strongest first check?

Practice

A model has 96 elements. A count finds 187 moment releases. Roughly how many releases would you expect if they had been applied to every element end? Give the number.

Summary

Name the discriminating check before naming the cause
Local or global is the first cut, and it eliminates half the candidates
A check whose outcomes point at the same cause is not a check
Confirm the diagnosis rather than assuming it
Verify that the correction fixed the fault and kept the feature