BorealSense: A Father-Son Engineering Project

Design Document — Learning, Awards, and University Pathway Edition

Project: Low-cost distributed early wildfire detection for Northern communities Team: Son (project owner/researcher) + Jerry (mentor/infrastructure) Horizon: Multi-year arc (Grade 9/10 → Grade 12), adaptable to current grade Related family context: builds on the BOREAL-UAS wildfire exploration, sits alongside (not inside) GLOM and the FRC curriculum

1. Strategy First: What Actually Wins Awards and Impresses Universities

Before designing the project, design for the judges and admissions readers. This section is deliberately blunt, because getting it wrong wastes years.

1.1 The ownership rule (the one that matters most)

Science fair judges and admissions readers are extremely good at detecting parent-driven projects — and Canada-Wide Science Fair (CWSF) explicitly requires disclosure of all assistance received. A project where Dad designed the PCB and the student presents it will lose to a cruder project the student genuinely owns, in both fairs and interviews. This is a feature of the plan, not a constraint:

Division of labor (write this down as a family contract):

The mentor role you should model is a university supervisor, not a co-author. Judges love well-mentored students — mentorship is disclosed, celebrated, and normal. What kills projects is undisclosed authorship.

1.2 What CWSF-level judges reward (in order)

1. A real question with a measurable answer — engineering fairs are won by science about the engineering (characterized performance, ROC curves, detection envelopes), not by “I built a thing that works.”
2. Iteration visible in the logbook — v1 failed → why → v2. A dated, handwritten/committed logbook is both required practice and the single best credibility artifact.
3. Rigor: controls, repeated trials, error bars, honest limitations.
4. Impact narrative: who needs this, what does it cost, who did you talk to.
5. The interview: a student who can reason live about why choices were made beats a more polished poster every time.

1.3 The honest view on university admissions (Canada + US)

• Canadian engineering (Waterloo, U of T EngSci, McMaster Health Sci-style programs): grades are the gate — low-to-mid 90s average is the entry ticket and no project substitutes for it. The differentiators are the supplemental applications (Waterloo’s AIF, EngSci profile): a sustained, authentic project with competition results and community impact is exactly what those forms are built to surface. Rule: the project must never cost grades. Time-box it.
• Major scholarships (Schulich Leaders ~$100–120k, Loran): these explicitly seek STEM entrepreneurship/innovation + leadership + community impact — a multi-year project with a community deployment story is close to a template match. School nomination required; cultivate that relationship early.
• US/UK top schools (if in scope): narrative + national/international distinction matter far more; ISEF (reachable from CWSF via Team Canada-ISEF) is the recognized currency.
• Shad Canada (summer program, Grade 10/11): widely respected signal, application strengthened by exactly this kind of project; also a network.
• No guarantees, ever. The plan below is designed so that even with zero awards, the outcome is a kid who can design experiments, build hardware, analyze data, and talk about it — which is the actual point, and also the thing that makes awards likely as a side effect.

1.4 Why a new project rather than folding into GLOM or FRC

• GLOM is the family’s shared long-arc project; FRC is team-based (great for leadership narrative, weak for individual research distinction). Fairs and supplementals need something with individual ownership and a research question. The three complement each other: FRC = teamwork/leadership, GLOM = family engineering culture, BorealSense = his research identity.

2. Project Selection

2.1 Selection criteria

A candidate project must be: (a) ownable end-to-end by a high-schooler, (b) a genuine research question, not a build, (c) cheap to iterate (< $100/node class hardware), (d) safe and legal for a minor, (e) tied to a problem Canadians care about, (f) rich enough for a 2–3 year arc with annual “chapters,” (g) adjacent to Dad’s expertise for mentorship without Dad’s fingerprints on the artifact.

2.2 Candidates considered

2.3 The research question (the heart of everything)

“How early, how far away, and how reliably can a network of sub-$50 environmental sensor nodes detect a smoldering fire, compared to the detection methods communities actually have?”

Sub-questions that give each year its chapter:

• Y1: Which cheap sensing modalities (PM2.5, VOC/CO via MOS gas sensors, ΔT/RH) respond to smoldering combustion, at what distance, wind, and time-to-detect? (Controlled experiments, single node.)
• Y2: Does a network + ML classifier beat single-node thresholds on false-positive rate? (Campfire ≠ wildfire ≠ BBQ ≠ diesel generator — the discrimination problem is the scientifically interesting part.)
• Y3: Field pilot with a real community (cottage association, camp, or a FireSmart-engaged community) — deployment data, reliability engineering, cost/coverage model.

This is honest science: the answer might be “PM sensors are useless beyond 30 m in wind” — and that’s still a gold-medal-caliber result if measured well.

3. Technical Design (scoped for student ownership)

3.1 The node (Year 1 target)

┌──────────────── BorealSense Node v1 (~$45 BOM) ────────────────┐
│  ESP32-S3 (or ESP32-C3) module        Solar: 2 W panel +       │
│   ├─ PMS5003 / PMSA003 (PM1/2.5/10)   TP4056-class charger +   │
│   ├─ MOS gas sensor (CO/VOC class,    18650 Li-ion + fuel      │
│   │   e.g. MiCS/SGP sensor)           gauge                    │
│   ├─ BME280 (T/RH/P)                                           │
│   └─ LoRa radio (SX1276/RFM95, 915 MHz ISM) → mesh/star to     │
│      gateway (Raspberry Pi + LoRa hat + cell/Starlink uplink)  │
│  Weatherproof enclosure (3D-printed + conformal coat), mast    │
│  mount, duty-cycled sampling for solar power budget            │
└────────────────────────────────────────────────────────────────┘

Deliberate choices for the student:

• ESP32 + Arduino/MicroPython first, not STM32/Rust. He must own the code; the on-ramp matters more than elegance. (If he later ports a node to embedded Rust himself, that’s a bonus chapter, not a requirement.)
• LoRa 915 MHz ISM: license-free, kilometers of range line-of-sight, and the power/range/duty-cycle trade-offs are themselves teachable physics.
• Through-hole/module-based v1, custom PCB in v2 (he learns KiCad from you — one evening a week — and the v2 board is his layout, reviewed not redrawn).
• Every node logs raw data to SD + transmits — raw data is the research asset.

3.2 The experiments (this is where medals come from)

Safety/legality first: no wildland ignition, ever. Smoldering sources = charcoal/smudge pots/approved fire-pit fires within municipal bylaws, propane fire table for a “clean combustion” control, always with adult supervision, water on hand, and burn-permit rules respected. This constraint is itself teachable: designing valid proxies is real science.

Experimental program (Y1):

1. Response characterization: fixed smoldering source (mass-controlled charcoal bed), nodes at 5/10/20/40/80 m in a line; measure signal vs distance × wind (log anemometer data); repeated trials; time-to-threshold.
2. Confounder library: BBQ, campfire, vehicle exhaust, humid morning, pollen day — recorded with identical protocol → labeled dataset.
3. Detection statistics: ROC curves per modality and fused; report detection distance envelope with confidence intervals.
4. Compare against baseline: “when could a human smell/see it?” (structured observer protocol — fun, and judges remember it).

Y2 ML chapter: classical ML first (logistic regression / random forest on engineered features — explainable in an interview), only then a small neural model on-device (TinyML / ESP32) as a comparison. The question is false-positive discrimination, measured, not “I used AI.”

Y3 field chapter: 8–12 nodes, one real site, one season. Uptime, battery performance through weather, false alarms in the wild, community feedback. Reliability engineering data is rare at fairs and reads as maturity.

3.3 What Dad builds (openly, as infrastructure)

• The lab itself: bench, tools, safety training, LCSC ordering pipeline.
• A data backhaul VPS + Grafana instance (“the family server”) — disclosed as provided infrastructure, like a school lab.
• Teaching modules on demand: KiCad evenings, “how PPK GPS works,” statistics for ROC curves — taught, with him doing the exercises.

4. Skills Curriculum Embedded in the Project

Meta-skills deliberately trained: estimating before measuring, reading datasheets, asking for help precisely, and presenting failures without shame.

5. Award & Milestone Calendar (Ontario pathway)

Verify current-year dates each fall — fair registration windows are rigid.

Backup framing: if a fair year misses, the same artifacts feed hackathons, FireSmart community recognition, and supplemental essays — nothing is wasted.

6. Governance: Keeping It His, Keeping It Fun

1. Weekly cadence, time-boxed: one 2–3 h lab session + one short data/ planning session. Grades outrank the project always (see §1.3).
2. He writes the plan: each session starts from his written goal; you ask questions, you don’t set agendas (hardest rule for an engineer-dad).
3. Motivation check every term: if the spark isn’t there, pivot the question or pause — a resentful project produces neither learning nor awards, and admissions readers can smell joylessness too.
4. Disclosure discipline: assistance log maintained alongside the logbook — converts directly into the fair’s required declarations.
5. Let v1 be his level of ugly. The gap between v1 and v3 is the story.

7. Budget

Cheap enough that he can also pitch part of it — a $200 sponsorship ask to a local business or cottage association is itself a leadership artifact.

8. Risks

9. First Four Sessions (start this month)

1. Session 1 — The pitch flows backward: show him the wildfire problem (news clips, FIRMS map of Ontario), then ask what he thinks could be done cheaply. Steer only by questions. If he lands somewhere else entirely — his project now.
2. Session 2 — First smoke: breadboard ESP32 + PM sensor, hold it near a blown-out candle, watch numbers move. Hook set.
3. Session 3 — First question: he writes one page: question, hypothesis, what Y1 success looks like. Logbook officially opens.
4. Session 4 — First experiment design: distances, trials, controls, on paper, before any more hardware. The habit that wins medals starts here.