CE2134 tank simulation showing water flowing around mangrove seedlings and the coloured wakes downstream

Flow and wake simulation

AI-integrated
Interactive Platforms
for CE2134
Fluid Mechanics

the digital component of my
experiential learning

Gary Lei holding a complete mangrove seedling, including its leaves, stem and pot

Photo credit: NUS

Assistant Professor Lei Jiarui, Gary
Civil and Environmental Engineering
National University of Singapore

Laboratory forEco-hydrodynamicInteractions

Motivation & My Approach Many students can do routine calculations, but struggle to see what the equation means physically. Students often get feedback too late. Motivation & My Approach Experiential learning! Physical + Digital (2022 onwards) (2025 onwards) Many students can do routine calculations, but struggle to see what the equation means physically. Students often get feedback too late.
Experiential learning: Digital See the physics with interactive platforms Immediate feedback https://jiaruilei.github.io/WaveAttenuationNbS_2/
https://jiaruilei.github.io/WaveAttenuationNbS_2/
Loading webpage…

Three CE2134 assignments in AY2026/27 Sem 1

109 students

320+ submissions
analysed

PSet 1: AI coach vs. rule-based replies

A/B test with 109 assessable submissions

PSet 2: Confidence improvement

47 paired ratings from 106 submissions

PSet 3: Learning evidence

Webpage interactions and quiz outcomes

PSet 1 A/B test: AI vs rule-based support https://hydrostatic-pressure.onrender.com/study/a/PlVZ63vpbIY95IDhImg8hg https://hydrostatic-pressure.onrender.com/study/b/fjd479fQILU0IXeMCvQLfA PSet 1 A/B test: AI vs rule-based support https://hydrostatic-pressure.onrender.com/study/a/PlVZ63vpbIY95IDhImg8hg https://hydrostatic-pressure.onrender.com/study/b/fjd479fQILU0IXeMCvQLfA
AI coach vs. rule-based replies
A · AI coach
Loading webpage…
B · Rule-based replies
Loading webpage…
PSet 1 A/B test: AI vs rule-based support Observed outcome A: API-connected AI B: rule-based chat Off-topic/nonsense counts None identified (0/55) 12/54 (22.2%) Mean confidence change on a 1-5 scale +0.20 (n = 45) -0.03 (n = 49) Students reporting lower confidence 8/45 (17.8%) 14/49 (28.6%) API-connected AI coach showed fewer reported relevance problems. Difference in confidence change (A - B): +0.23 Confidence change did not differ significantly between A and B (Welch’s t-test, p value = 0.238). 45 and 49 paired confidence levels for A & B, respectively
PSet 2 Confidence gain with AI support https://hydrostatic-force-surfaces.onrender.com
https://hydrostatic-force-surfaces.onrender.com/
Loading webpage…
PSet 2 Confidence gain with AI support Confidence outcome Results Mean confidence on a 1-5 scale 4.06 before AI-coach usage, 4.37 after Mean confidence change on a 1-5 scale +0.31 (n = 47) Direction of confidence change 19 higher, 23 unchanged, 5 lower Confidence increases outnumbered decreases (p value = 0.0033, statistically significant). PSet 2 shows confidence gain alongside AI use for conceptual clarification 47 paired confidence levels
PSet 3 AI learning support beyond confidence https://bernoulli-pipe-hgl-egl.onrender.com/
https://bernoulli-pipe-hgl-egl.onrender.com/
Loading webpage…
PSet 3 AI learning support beyond confidence. Learning support: Evidence from the chat logs. Conceptual help: 60.7% of IDs (37/61) asked at least one concept, reasoning or equation question. Repeated engagement: 54.1% of IDs (33/61) had two or more recorded AI exchanges. For example, a user corrected velocity from 1.2 to 2.7 m/s and pressure from 125.5 to 122.575 kPa. 144 exchanges across 61 anonymous IDs.

Summary and future work

AI support shows promise for building confidence and helping students explore concepts

In the future, I will further enhance my physical+digital experiential learning

https://ce2134-interactive-lab-tank.onrender.com/
Loading webpage…