When the Truss Analysis App Overpredicts Failure Load
In a principles-of-engineering class, students design balsa trusses spanning 30 cm and load them at midspan until failure. Instruction covered tension, compression, and joint design; teams used a free truss-analysis app that predicted failure loads from member geometry. In testing, most trusses fail at 45–60 N, while the app predicted 150–200 N. Failures cluster at glued gusset joints and at the top chord near the load point. One team reaches 130 N after a student doubled the gusset plates but did not record the change. Students conclude that the app is wrong and ask to stop using analysis. The teacher has one week before the unit assessment and must decide what to reteach, how to explain the predicted-versus-actual gap, and how to keep the cutting, gluing, and loading procedures safe.
1. Diagnose the likely causes of the predicted-versus-actual gap and describe what to reteach and how, within the one-week constraint.
Read the worked answer
Start from the size of the gap. Predicted 150–200 N versus observed 45–60 N is a factor of three, which is too large to be measurement noise and too systematic to be bad luck. The clustering at gusset joints identifies the mechanism: the app models members as pins or rigid frames with continuous material, but the real trusses fail in glue-line shear and gusset tear-out. Balsa also varies in density, and hand-cut joints have small bond areas and poor fit. Load eccentricity at the midspan point adds bending the model may ignore.
What to reteach, in order:
- Joint efficiency - compare a glued lap joint, a gusset joint, and a pinned joint in coupon tests.
- Failure-mode identification - students inspect and sketch each fracture surface before discarding the truss.
- Model assumptions - have teams list what the app assumes and which assumptions their build violates.
Design revision: enlarge gusset area, extend gussets onto both faces, improve fit before gluing, and change one variable at a time. A retest with two or three trusses per team is feasible in a week if building is done outside class.
Safety: keep knife and saw use at supervised stations with eye protection, ventilate the glue area, and load trusses behind a shield with hands clear.
2. Students claim the app is wrong and want to abandon analysis. How should the teacher respond, and how should the undocumented gusset change be handled?
Read the worked answer
Reframe the claim. The app is not wrong; it is a model with stated assumptions, and the test measured something the model does not represent. Students should compute the ratio of observed to predicted load for every truss (roughly 0.25–0.4) and ask whether that ratio is consistent - a consistent ratio points to a systematic factor such as joint efficiency, while scatter points to construction variability. Either finding is useful engineering information.
Teach the response, not just the fact: derive an empirical joint-efficiency factor from class data and use it to revise predictions. This is how safety factors and design allowables emerge in practice, and it converts a demoralizing result into a quantitative lesson.
Handle the undocumented gusset doubling explicitly. The 130 N result is not trustworthy evidence until the team can reconstruct what it built and why. Give the team credit for the insight only after it produces a sketch, a rationale, and a prediction; then make notebook documentation a graded requirement for all teams. The lesson is reproducibility: an undocumented success cannot be learned from.
Close with a claim-evidence-reasoning write-up rather than a verdict on the software, so the assessment targets reasoning.
3. How should the unit assessment be adjusted so that it measures the intended learning, and what should the teacher avoid doing with the scores?
Read the worked answer
Change the assessment to match what was learned. The original test presumably asked for a predicted failure load and a built truss. Revise it to require:
- a prediction with assumptions stated and a joint-efficiency factor applied;
- a failure-analysis sketch identifying the mode and location;
- a redesign proposal with one named variable and an expected effect;
- an engineering notebook showing dated entries, including the undocumented change and its correction.
Weighting: process evidence should carry substantial weight - perhaps 60 percent - because the unit's learning goal is iterative analysis, not peak load. Keep a performance component so students still build and test.
Do not curve the scores to hide the prediction gap; that would erase the most instructive result. Instead, grade the explanation of the gap.
Practical constraint: with one week left, reteach joints and failure modes in two periods, run coupon tests in one, and use the remaining time for the revised write-up. A second truss test can be optional enrichment for teams that finish, so the assessment does not depend on building speed.
Equity check: teams whose trusses failed early have more failure data to analyze; make sure the rubric rewards the quality of analysis rather than the strength of the artifact.