Computational Engineering / Digital engineering workflows

Loud and silent losses

Six things that go wrong in a handover, all applied to the same truss, all re-solved.

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.
Verify the lab exchange-capability tables against each actual exporter/importer before presenting them as real software behaviour.
Check truss load-loss and force examples against the complete model; independent arithmetic here checks scaling laws and counts only.

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

A transfer error may stop analysis, or it may leave a model that solves successfully with unintended inputs. Both need investigation. The truss examples below show how changed sections, materials, loads and connectivity affect results and which checks can reveal the changes.

By the end of this lesson you should be able to

Classify a loss as loud, silent or harmless, from what the model does
Explain why a defaulted section multiplies the deflection by exactly the area ratio
Recognise the loss that makes a structure look safer than it is
Say what a handover check should look for, given that half of these are silent

What you should already know

Module 9's stiffness method — the deflection ratios here follow directly from it
Module 12's diagnostics — the loud failures are loud because those checks fire

One truss, six transfer changes

The exercise uses a linear elastic, pin-jointed Warren truss, 24 m long and 3 m deep, with six panels and 4000 mm² member areas. It applies 60 kN at each interior bottom node.

In this exercise, changing the pin to a roller creates a mechanism, and leaving coincident nodes disconnected prevents a valid solution. Section and material defaults, a missing load and changed member names can leave the model solvable. These are outcomes of this model and its diagnostics, not guarantees for every solver or structure.

Stiffness changes

With geometry, supports and applied forces fixed, each axial member stiffness is proportional to EA/L. Scaling every area from 4000 to 1000 mm² scales the assembled stiffness by one quarter, so displacements multiply by four. Changing every modulus from 210 to 200 GPa gives a displacement ratio of 1.05. These statements assume linear force-controlled analysis without additional stiffness terms or imposed-displacement effects.

Under those assumptions, uniform stiffness scaling leaves member forces unchanged. It does not establish that a section-based resistance or stress check is unchanged: the section properties used in those checks also matter.

Load and naming changes

The exercise reports a selected deflection ratio of 0.72 when a load is omitted. Removing a load does not generally scale every response by one factor or make every result smaller; signs and load patterns matter.

Compare received loading with the intended load schedule. Also check equilibrium against the loads actually present. A model can satisfy equilibrium after a load was omitted, so its own applied loads and reactions alone will not reveal the omission.

Regenerated member names may leave the analysis numbers unchanged while breaking links to drawings and schedules. Check both numerical behaviour and traceability.

Try it

Model transfer exercises

One truss, damaged one attribute at a time, re-solved each time. The verdict is computed from what the model does.

What was lost in transfer

Both supports hold vertically; nothing holds the truss horizontally.

Does the received model solve?: no
Warnings raised by the model itself: 1
Largest displacement, against the sent model: —
Largest member force: —
Verdict: loud

The receiving model will not solve. Nobody can act on a wrong number, because there is no number.

What the receiving model says on its own

error: Nothing holds the model horizontally. It does not matter that there is no horizontal load: the matrix has zero stiffness in that direction and the solve divides by zero.

Losses that stop the model: 2
Losses that solve and are wrong: 3
Losses that change no number: 1

What this shows: A loss that stops the model is the safe one. Three of these six leave it solving and wrong.

Worked example

Why a quarter of the area is exactly four times the deflection

Given

Every member's area falls from 4000 mm² to 1000 mm² on import
Nothing else changes: geometry, material, loads and supports are all correct

Find

The exact factor on every displacement, before running anything

Practice

Members arrive with an area of 2500 mm² instead of the 4000 mm² that was sent. By what factor is every displacement in the received model multiplied?

Practice

In the six-loss table, how many of the losses leave the model solving with no warning at all?

Predict first

What concern is illustrated by the omitted gravity loading in this transfer exercise?

Worked example

Finding a silent loss with one sum

Given

A received truss model solves cleanly and reports no warnings
The sending engineer applied 60 kN at each of five interior nodes
The received model's vertical reactions total 240 kN

Find

Whether anything was lost

Practice

A model is sent with E = 210 GPa and arrives with E = 195 GPa. By what factor are all displacements multiplied?

Practice

The intended loading totals 300 kN, while a converged received model carries 240 kN according to its reactions. What percentage shortfall should be investigated?

Practice

Members arrive with an area of 6000 mm² instead of the 4000 mm² sent. By what factor are the displacements multiplied?

Check yourself

Why do member forces stay unchanged when every section area is scaled by the same factor?

Check yourself

Which of these losses is caught by the receiving software, without anyone checking anything?

Predict first

In this linear truss, all areas are uniformly reduced while geometry, material, supports and applied forces stay fixed. Which statement follows?

Summary

Two of the six losses are loud, three are silent and one is harmless
A quartered area gives exactly four times the deflection and identical member forces
A defaulted modulus gives a 5 % error, comfortably inside 'about right'
Missing loads may understate demand; compare loading and results with independent expectations
The solver checks that the model is solvable, not that it is the model you sent
Only a check against something outside the model catches a silent loss