Artificial Intelligence and Data Methods
Documents any machine-learning or data-driven method used in the project, including training data, validation, and limitations.
Section progress
0% of the workflow complete
Computational Engineering · Build, calibrate, verify and validate the numerical model that supports your design decisions.
Deliverable: AI/data-methods report with training data provenance, validation metrics, and stated domain of applicability.
Minimum tables, figures and equations for Artificial Intelligence and Data Methods
Tables — at least 6
- Table — trial sections or sizes considered, with the capacity of each and the selection decision
- Table — final selected geometry for every element: dimensions, thickness, grade, spacing, elevation
- Table — slab, beam, column and shear wall schedule with governing demand
- Table — ultimate limit state check summary: demand, capacity, ratio, pass or fail, governing clause
- Table — serviceability check summary: deflection, crack width, settlement, freeboard or velocity against its limit
- Table — factors of safety achieved against the factor required, per failure mode
Figures — at least 4
- Figure — free body diagram of each isolated element, fully labelled with loads, reactions, dimensions and axes
- Figure — shear and moment (or pressure and velocity) diagrams for each force-carrying element
- Figure — dimensioned section or plan of each designed element
- Figure — capacity versus demand plot, interaction diagram, or rating curve as applicable
Equations — at least 8
- Equation — equilibrium equations written out for each free body (sum of forces and sum of moments, or continuity and energy)
- Equation — the internal force relations V(x) and M(x), or the momentum/thrust relation, used to compute each element's demand
- Equation — the resulting demand at the critical section of each element, with numeric substitution
- Equation — the capacity expression for each element type, shown with full numeric substitution and units
- Equation — the sizing criterion that sets the final dimension (for example required area, depth or diameter)
- Equation — each limit state check written as demand over capacity with numbers substituted
- Equation — the factor of safety calculation for each failure mode checked
- Equation — punching shear, drift and deflection checks with limits
Number every table and figure (Table 4.x, Figure 4.x), caption it, and refer to it by number in your text. Number displayed equations and show the substitution with units. These counts are minimums — add whatever else your design needs.
Engineering documentation standard — required in every Chapter 4 subsection
These rules are graded on every subsection. Work that misses them is capped on technical accuracy, exhibits, codes and communication, whatever the quality of the prose.
Code and standard references
- Every requirement, factor, coefficient, limit and allowable you apply cites the governing document AND the exact section, article or sub-article number — e.g. ACI 318-19 §22.5.5.1, AISC 360-22 Chapter J, Section J3.6, AASHTO LRFD 10th Ed. Article 3.6.1.2.2, ASCE 7-22 §12.8.1, ASTM D2487, state DOT manual section, local stormwater manual chapter.
- Give the edition or year of every document the first time it appears, then use a consistent short form.
- Where a code equation is used, quote the equation number (e.g. Eq. 22.5.5.1) next to your displayed equation.
- Where you depart from a code provision, state the clause you are departing from and the engineering justification.
- List every code, standard and manual actually used in a Codes and Standards table at the start of the subsection.
Citations for statements
- Every statement of fact, value taken from elsewhere, material property, soil parameter, rainfall depth, unit cost or published method carries an in-text citation (APA) to its source.
- Field and lab data cite the report, boring log, gauge, survey file or test number and its date.
- Manufacturer data cites the product literature and revision date; software results cite the program, version and model file name.
- Uncited assertions are treated as assumptions and must appear in the assumptions table with a justification.
- Every in-text citation resolves to a full entry in the reference list.
Step-by-step calculations
- Structure every calculation the same way: (1) objective, (2) governing code clause, (3) equation in symbolic form with the equation number, (4) definition of each symbol, (5) numerical substitution, (6) result with units, (7) comparison against the limit and the pass/fail statement.
- Show the substitution line — never jump from the formula to the answer.
- Number displayed equations sequentially (Eq. 4.1, 4.2, …) and refer to them by number in the text.
- State the load or flow combination governing each calculation by name.
- Carry consistent significant figures and round only at the reported result; state the rounding convention once.
- Present repetitive element checks in a calculation table with one row per element and the same column order throughout.
Free body diagrams and figures
- Draw a separate free body diagram for each isolated element — no combined sketches standing in for several members.
- Dimension every FBD: span, depth, thickness, cover, eccentricity, embedment, slope, pipe diameter, wall height — with the dimension lines and values shown.
- Label every force, pressure, reaction and moment with its symbol, magnitude and units, and show the sign convention and coordinate axes.
- Show supports and boundary conditions explicitly (pin, roller, fixed, elastic, buoyant, hydrostatic).
- Accompany each FBD with its shear, moment, thrust, pressure or hydraulic grade diagram at the same scale reference.
- Number and caption every figure (Figure 4.x) and refer to it by number in the narrative; add a scale or north arrow to plans.
Units and notation
- Every number in text, tables, figures and equations carries its unit — no bare numbers.
- Use one unit system throughout (US customary or SI); if both appear, give the converted value in parentheses consistently.
- Check dimensional homogeneity of each equation and say so — the units of both sides must match.
- Provide a nomenclature table defining every symbol with its unit.
Checking and verification
- Every calculation is checked by an independent route — hand check against software, alternative method, order-of-magnitude estimate, or a published worked example — and the check is shown, not just claimed.
- Report demand-to-capacity ratios and factors of safety against the required values, with the source clause for each required value.
- Include a verification/checking table: item, method of check, expected, obtained, difference, accept or revise.
- Sanity-check every result (magnitude, direction, plausibility) and state the conclusion.
- Record who checked the work and on what date; flag anything still unverified as an open item.
- State limitations and the range over which the result is valid.
How to complete this section
Do this next: Read the Artificial Intelligence and Data Methods lecture and the worked example so you know what "AI/data-methods report with training data provenance, validation metrics, and stated domain of applicability." has to contain.
Not sure how to start or how much depth is expected? Read the fully written model example for this deliverable first — it shows the structure, tables and level of justification your advisor grades against.
Modeling & Simulation Center — what this workspace teaches
Build, calibrate, verify and validate the numerical model that supports your design decisions.
- Selecting analysis software for the engineering question (STAAD, SAP2000, ETABS, HEC-RAS, OpenRoads, Civil3D, ArcGIS, MATLAB, Python)
- Model geometry idealization and simplification
- Boundary conditions, supports, restraints and their effect on results
- Load application and load-case management in software
- Mesh and element selection; convergence studies
- Model calibration against measured or benchmark data
- Sensitivity analysis of governing input parameters
- Verification (solving the equations right) vs. validation (solving the right equations)
- Exporting, documenting and archiving model results
End-of-term milestones
- Tuesday, November 17, 2026 — Poster printed and ready. 36 in × 48 in poster finalized and printed one week before the November 24 showcase.
- Wednesday, November 18, 2026 — Final document package uploaded for scoring. Chapters 4–5, calculation package, drawings and appendices uploaded in the app for advisor scoring.
- Wednesday, November 18, 2026 — Poster presentation to faculty and industry. Wednesday poster session — printed 36 in × 48 in poster presented in person; industry reviewers score communication and impact.
- Wednesday, November 25, 2026 — Oral presentation and defense (scored). Scored oral presentation and defense held on Wednesday, November 25.
Artificial Intelligence and Data Methods
Documents any machine-learning or data-driven method used in the project, including training data, validation, and limitations.
Section B
Engineering story
A real project situation that frames this module
It is week 5 of implementation and the engineering computation and digital delivery team has reached artificial intelligence and data methods. Documents any machine-learning or data-driven method used in the project, including training data, validation, and limitations. The independent model checker asks one question: what establishes that supervised learning workflow?
Reporting training-set accuracy instead of held-out test performance. Because model evaluation metrics, the error does not stay local: it is carried into the design of record that drawings, quantities and cost are generated from, and every downstream product inherits it before anyone notices.
Every downstream discipline that inherits the model and the engineer who seals it carry the consequence. On this module specifically, the exposure runs through overfitting detection via learning curves and regularization, and the cost of correction rises every week the model, dataset and documented computational workflow moves closer to issue.
Decisions the engineer must make
- What record establishes supervised learning workflow, and is that record in the project data inventory?
- Does NIST AI Risk Management Framework (AI RMF 1.0, 2023), Govern, map, measure, manage, govern here — and is that the edition adopted by the jurisdiction?
- What is the acceptance criterion for model evaluation metrics, and was it written before the result was known?
- Is RMSE = √(Σ(Oi − Pi)² / n) valid over the parameter range this project actually occupies?
- If the check fails, does the team revise the model, dataset and documented computational workflow or raise a change request against the locked baseline?
Photo 1. Field review: the conversation in which a scope, a constraint or a decision is actually settled.
Capstone Studio instructional photograph
Section C
Why this matters
Professional
A licensed engineer defending artificial intelligence and data methods cites NIST AI Risk Management Framework (AI RMF 1.0, 2023), Govern, map, measure, manage, and shows the record behind each input. Your ai/data-methods report with training data provenance, validation metrics, and stated domain of applicability. is reviewed the same way — traceability is assessed before arithmetic.
Technical
Supervised learning workflow is what makes RMSE = √(Σ(Oi − Pi)² / n) usable on this project rather than a formula copied from a reference. Get it wrong and every quantity derived from it is wrong by the same factor.
Safety
The failure mode this module guards against is an unverified model output accepted as an engineering result. It reaches people through explainability requirement for engineering decisions informed by a model, which is why the safety check is recorded explicitly here rather than inferred from a passing strength or performance check.
Economic
The design of record that drawings, quantities and cost are generated from is priced from this work. Quantities, unit costs and schedule float all trace to supervised learning workflow; a late correction here is paid for as a change order, not a redline.
Environmental
Environmentally, this module fixes decisions on quantity and material that the model silently drives. Choosing conservatively without justification is not free — the excess shows up as material, energy and land that the project consumes for no measurable gain.
Community
Any computational or data-driven model output requires an independent, hand-calculable sanity check before design reliance — the same discipline applies to ML predictions. The public that depends on results no one outside the modelling team can reproduce live with that outcome long after the semester ends.
Section D
Learning objectives
By the end of this module you will be able to:
- 1.Evaluate supervised learning workflow, using this project's own conditions rather than a textbook case.
- 2.Interpret model evaluation metrics, using this project's own conditions rather than a textbook case.
- 3.Analyze overfitting detection via learning curves and regularization, using this project's own conditions rather than a textbook case.
- 4.Justify data provenance, cleaning, and bias screening before model training, using this project's own conditions rather than a textbook case.
- 5.Compute the governing quantity from RMSE = √(Σ(Oi − Pi)² / n) and Coefficient of determination: R² = 1 − Σ(Oi−Pi)² / Σ(Oi−Ō)², with a unit audit on every term.
- 6.Apply NIST AI Risk Management Framework (AI RMF 1.0, 2023), Govern, map, measure, manage, and cite the section that governs your acceptance decision.
- 7.Reproduce the worked example for a regression model predicting pavement roughness (IRI) is validated on 20 held-out sections and defend the interpretation of the result.
- 8.Produce ai/data-methods report with training data provenance, validation metrics, and stated domain of applicability. at a standard the independent model checker would accept without a second revision cycle.
Section E
Instructional content
Full lecture notes with figures and governing equations
The engineering content of artificial intelligence and data methods
Documents any machine-learning or data-driven method used in the project, including training data, validation, and limitations. That single sentence hides the substance of the module: supervised learning workflow, and model evaluation metrics. Both must be established from project evidence before anything downstream is credible.
In engineering computation and digital delivery, this work is the input to the model, dataset and documented computational workflow. Overfitting detection via learning curves and regularization — which is why this page asks you to record the source of every quantity, not just its value. The design of record that drawings, quantities and cost are generated from depends on it.
- Supervised learning workflow: train/validation/test split, cross-validation
- Model evaluation metrics: RMSE, MAE, R², classification accuracy/precision/recall
- Overfitting detection via learning curves and regularization
- Data provenance, cleaning, and bias screening before model training
- Explainability requirement for engineering decisions informed by a model
Photo 1. The engineering content of artificial intelligence and data methods in practice — Field review: the conversation in which a scope, a constraint or a decision is actually settled.
Capstone Studio instructional photograph
Governing relationships and how they are applied here
The relationships below govern artificial intelligence and data methods. RMSE = √(Σ(Oi − Pi)² / n); Coefficient of determination: R² = 1 − Σ(Oi−Pi)² / Σ(Oi−Ō)² — each is valid only inside the parameter range this project occupies, so state that range before substituting.
Model evaluation metrics sets the values you place into these expressions. Any code-prescribed factor must match NIST AI Risk Management Framework (AI RMF 1.0, 2023); a factor lifted from a different edition silently changes the answer.
RMSE = √(Σ(Oi − Pi)² / n)
- Oi = observed value
- Pi = predicted value
- n = number of samples
Coefficient of determination: R² = 1 − Σ(Oi−Pi)² / Σ(Oi−Ō)²
- Ō = mean of observed values
Photo 2. Governing relationships and how they are applied here in practice — Field review: the conversation in which a scope, a constraint or a decision is actually settled.
Capstone Studio instructional photograph
Constraints, adopted standards and the safety case for artificial intelligence and data methods
NIST AI Risk Management Framework (AI RMF 1.0, 2023), Govern, map, measure, manage, governs this module: Governs risk-based use of AI/ML in engineering decisions ISO/IEC 23053 (2022), Framework for AI systems using ML, adds the second constraint: Governs ML system lifecycle documentation
The safety case is explicit here. The failure mode is an unverified model output accepted as an engineering result; the people exposed are every downstream discipline that inherits the model and the engineer who seals it; the control that prevents it is explainability requirement for engineering decisions informed by a model together with an independent check by someone who did not perform the work.
- Controlling criterion for this module: supervised learning workflow.
- Adopted reference: NIST AI Risk Management Framework (AI RMF 1.0, 2023) — cite Govern, map, measure, manage by number.
- Failure mode guarded: an unverified model output accepted as an engineering result.
- Evidence produced: AI/data-methods report with training data provenance, validation metrics, and stated domain of applicability..
Photo 3. Constraints, adopted standards and the safety case for artificial intelligence and data methods in practice — Structural framing of a pedestrian bridge: members, connections and the load path a designer must trace.
Wikimedia Commons, CC BY-SA 4.0
Where this method stops being valid
The worked example — a regression model predicting pavement roughness (IRI) is validated on 20 held-out sections — holds only while its assumptions hold. The model explains ~85% of the variance in observed roughness; remaining scatter must be bounded and disclosed before the model informs a maintenance decision. Outside that envelope the arithmetic still returns a number, and the number is wrong in a way no unit check will catch.
For this project, the boundary you are most likely to push is explainability requirement for engineering decisions informed by a model. If you cross it, say so in writing, bound the error, and carry the limitation into your results chapter. A disclosed limitation is professional practice; a silent extrapolation is not.
Photo 4. Where this method stops being valid in practice — Field review: the conversation in which a scope, a constraint or a decision is actually settled.
Capstone Studio instructional photograph
Section F
Engineering workflow
Steps
- 1. Assemble the inputs this module needs — supervised learning workflow; model evaluation metrics — each with a unit and a source record.
- 2. Confirm NIST AI Risk Management Framework (AI RMF 1.0, 2023) is the adopted edition and locate Govern, map, measure, manage.
- 3. State the assumptions and the acceptance criterion for supervised learning workflow.
- 4. Evaluate RMSE = √(Σ(Oi − Pi)² / n) and Coefficient of determination: R² = 1 − Σ(Oi−Pi)² / Σ(Oi−Ō)² term by term, carrying one extra significant figure.
- 5. Test the result against overfitting detection via learning curves and regularization.
- 6. Audit units and run an order-of-magnitude check by hand before the number leaves your desk.
- 7. Obtain an independent check from a teammate who did not perform the work, and record their name and date.
- 8. Assemble ai/data-methods report with training data provenance, validation metrics, and stated domain of applicability. and submit it to the independent model checker for review.
Decision points
- Is every input behind supervised learning workflow traceable? If not — stop and collect the record.
- Does the result satisfy model evaluation metrics? If not — revise the work, never the criterion.
- Would the correction change the design of record that drawings, quantities and cost are generated from? If yes — raise a change-control request before proceeding.
- Have you ruled out the most common error on this module — reporting training-set accuracy instead of held-out test performance?
Quality checklist
- Documented: supervised learning workflow
- Documented: model evaluation metrics
- Documented: overfitting detection via learning curves and regularization
- NIST AI Risk Management Framework Govern, map, measure, manage cited by section number
- Units audited on every expression
- Acceptance criterion recorded before the result
- Independent check signed and dated
- AI/data-methods report with training data provenance, validation metrics, and stated domain of applicability. attached and named per the course convention
Section H
Interactive visualization
Artificial Intelligence and Data Methods — step-through
Advance one frame at a time. Each frame adds one engineering decision to the previous state.
Step 1 of 6
Assemble and clean the dataset; document provenance.
Section I
Applicable codes and standards
NIST AI Risk Management Framework
AI RMF 1.0, 2023 · Govern, map, measure, manage
Adopted design/analysis reference governing this module.
Relevance: Governs risk-based use of AI/ML in engineering decisions
Reference the section number and edition in your calculation package. Do not reproduce code text.
ISO/IEC 23053
2022 · Framework for AI systems using ML
Adopted design/analysis reference governing this module.
Relevance: Governs ML system lifecycle documentation
Reference the section number and edition in your calculation package. Do not reproduce code text.
Section J
Worked examples
Full engineering solution format
Section K
Common mistakes and how to avoid them
- Reporting training-set accuracy instead of held-out test performance.
- Using a model outside the range of its training data without disclosure.
- Treating model output as ground truth without an independent engineering check.
- Treating supervised learning workflow as a given instead of establishing it from a project record.
- Producing ai/data-methods report with training data provenance, validation metrics, and stated domain of applicability. without showing how model evaluation metrics was satisfied.
- Substituting into RMSE = √(Σ(Oi − Pi)² / n) outside the range where it is valid, and reporting the number anyway.
- Missing explainability requirement for engineering decisions informed by a model, which is exactly the path to an unverified model output accepted as an engineering result.
- Designing to the average condition when the governing condition is the controlling one.
- Freezing a design before the constructability and access review that would have changed it.
- Stopping at output and skipping verification — an unverified number is not an engineering result.
- Confusing results (what the analysis produced) with conclusions (what the engineer decided).
- Ignoring constructability: a design that cannot be built safely is not a completed design.
Section L
Industry case study
Hartford Civic Center roof collapse (1978)
Hartford, CT arena
Official findings
- Investigation found the original computer structural model omitted key secondary bending effects in the space-frame connections, and the output was trusted without independent hand-check.
Field observations
- The controlling assumption was documented nowhere in the design record.
- No independent check existed at the stage where the error entered the work.
Engineering interpretation
- Interpretation below is student analysis for instructional purposes, not an official finding.
- Map the failure to a step in your own workflow and state where your process would have caught it.
Lessons learned
- Any computational or data-driven model output requires an independent, hand-calculable sanity check before design reliance — the same discipline applies to ML predictions.
Source: Summarize the published investigation; cite it in your reference list. Do not reproduce copyrighted report text.
Section M
FE Civil exam connection
Handbook FE Reference Handbook — engineering computation and digital delivery section (record the section number from your handbook edition).
Exam topics
Handbook formulas
- RMSE
- R²
Weak results here feed your FE Civil Academy weak-area queue for targeted practice.
Question 1 of 2
Score: 0/2In artificial intelligence and data methods, which item must be established BEFORE the analysis is run?
Section N
Apply it to your project — Artificial Intelligence and Data Methods
Complete this using your own capstone project data. Every field is saved to your project record and routed to your advisor with this module's submission.
Inputs and sources
Every value needs a traceable source.
| Quantity | Value | Unit | Source / record |
|---|
Assumptions and consequences
| Assumption | Basis | Consequence if wrong |
|---|
Self-check before submission
Section O
Design challenge
Consulting challenge — Artificial Intelligence and Data Methods
Your firm has been retained to deliver the artificial intelligence and data methods scope for a municipal client on a compressed schedule. Produce the technical position your firm would defend at a public meeting.
Client request: The client wants a defensible recommendation, the basis of design, and an honest statement of what remains unresolved.
Constraints
- Adopted local code edition governs; no exceptions without written variance.
- Budget and schedule are fixed; scope changes require change control.
- Public safety and accessibility requirements are non-negotiable.
Deliverables
- One-page basis of design
- Supporting calculation extract
- Risk and limitation statement
Evaluation
- Technical correctness
- Standard compliance
- Clarity of engineering judgment
- Honest treatment of uncertainty
Section P
Documentation workspace
Write the report section for this module in the academic editor
Section Q
File uploads
Accepted: PDF, DOCX, XLSX, CSV, PNG, JPG, ZIP
No files uploaded yet.
Section R
Deliverable and advisor review
AI/data-methods report with training data provenance, validation metrics, and stated domain of applicability.
Submissions route to your assigned faculty advisor and are scored independently by faculty and administrator rubrics.
Reflection
What was the hardest engineering judgment in this module, and how did you resolve it?
Section S
ABET outcome mapping
AI/data-methods report with training data provenance, validation metrics, and stated domain of applicability. with advisor review and dual scoring.
Assessment: Faculty rubric score and administrator rubric score on this module's submission.
Rubric: Engineering design · Target: 70% of students at or above 'meets expectations'.
AI/data-methods report with training data provenance, validation metrics, and stated domain of applicability. with advisor review and dual scoring.
Assessment: Faculty rubric score and administrator rubric score on this module's submission.
Rubric: Engineering design · Target: 70% of students at or above 'meets expectations'.
Section T
References and further study
NIST AI Risk Management Framework (AI RMF 1.0, 2023)
Adopted reference — cite section numbers, do not reproduce text.
ISO/IEC 23053 (2022)
Adopted reference — cite section numbers, do not reproduce text.
Artificial Intelligence and Data Methods — instructor design procedure
Course template for the calculation package format expected in the final report appendix.
NCEES FE Reference Handbook
Locate the equations used here and note the handbook section for exam recall.
Advisor meeting agenda item
Bring the unresolved decision from this module to your next weekly advisor meeting.