EcoSwitch Download PDF
Low-voltage prototype

Power the classroom.Not the emptiness.

EcoSwitch is a low-cost, camera-free classroom energy management system that combines PIR motion sensing with 24 GHz mmWave presence detection to control electrical loads zone by zone.

Open PDF in a new tab
B.Tech Year 1, ITPCCOE, PuneGuide: Rohit Tathe~12 weeks5-member team
Live view, linked to the simulator

A serious, low-cost, experimental smart-classroom controller.

Lights and fans in classrooms, labs and offices are often left running in empty rooms. PIR sensors are cheap but respond to movement, so they can switch off on someone reading or writing an exam. Cameras avoid that but cost more and raise privacy concerns.

EcoSwitch is a low-cost, offline controller. An ESP32-S3 reads a PIR sensor and a 24 GHz mmWave presence sensor in each of three zones, and turns a zone's loads off only after both sensors have reported that zone clear for a set time. The team is building a three-zone, low-voltage (5 V DC) prototype with an LED light indicator and a DC motor/fan per zone, then testing hybrid against PIR-only and mmWave-only sensing on false-vacancy rate, false-occupancy time, response time and energy saved.

How to read the numbers on this site

Measured

Collected from the prototype or a real room. Nothing on this page is in this category yet.

Assumption

An input chosen for the scenario model, to be verified by an audit.

Hypothesis

A claim the experiments are designed to test.

Expected

A goal or target, not a prediction or a result.

Scenario estimates Scenario modelNot yet measured

~2,000 kWh/yrLow-case waste for 60 rooms. In this case the model does not pay back.
~9,100 kWh/yrBase-case waste. Simple payback is about 3.5 years in the model.
~27,000 kWh/yrHigh-case waste. Simple payback is under 1 year in the model.
Savings depend on how much waste really occurs, which is why a measurement audit (Phase 0) comes first. These figures come from the scenario model, not from experiments.

Objectives and scope

Objectives
  • Estimate waste with a transparent, measured-where-possible model.
  • Build a three-zone prototype that controls a light indicator and a fan per zone from hybrid detection.
  • Compare PIR-only, mmWave-only and hybrid sensing.
  • Measure false vacancy, false occupancy, response time and energy saved.
  • Document the design for a larger deployment.
In scope
  • One bench-scale classroom model with three zones
  • Per zone: one LED indicator and one 5 V DC motor/fan
  • Offline local decisions on an ESP32-S3
  • Per-zone manual override
  • OLED status display and buzzer feedback
  • Scenario tests and energy calculation
Out of scope
  • Switching building mains (unless under qualified supervision)
  • AC integration
  • People counting
  • Cloud dashboards
  • Machine learning
  • Whole-institute installation

Empty rooms keep drawing power.

Buildings use a large share of electricity. Schools run on fixed timetables with gaps (between lectures, breaks, after practicals, early and late hours) where lights and fans stay on because switching off is nobody's explicit job Hypothesis: to be validated by audit

Surveys of occupancy sensing (Nguyen & Aiello 2013; Dong et al. 2019; Chen et al. 2018) cover cost, accuracy, privacy and installation trade-offs. A lighting-control meta-analysis (Williams et al. 2012) found roughly a quarter average lighting savings, mostly in commercial buildings abroad, so it cannot be applied directly to an Indian campus.

Problem: manual or timer switching wastes energy. Motion-only automation can make false "vacant" calls on still occupants, which annoys users until they bypass or remove it. A low-cost, privacy-preserving approach that stays reliable during low-motion periods is needed.

Existing solutions

PIR switchesWidely sold wall and ceiling occupancy switches.
Dual-technologyPIR plus ultrasonic or microphonic sensing.
BMSScheduling, CO₂ or camera-based occupancy.
ECBCIndia's building code has lighting-control provisions; the clauses for educational buildings still need to be read.

Why they are not more common Hypotheses to test with the college electrician

  • Tariffs and budgets make payback unattractive unless waste is high.
  • Retrofits need wiring changes and trained labour.
  • PIR-only systems can switch off on seated people, so users bypass them.
  • Split incentive: whoever pays the bill is not who chooses or uses the equipment.
  • Cameras raise privacy concerns and need more processing.

Two sensors with different blind spots.

Table 4 compares sensing technologies. Relevance ratings are the team's judgement for this application. Filter by minimum stars.

Cameras and thermal cameras lose on privacy, cost and processing. CO₂ is too slow. Wi-Fi/BLE misses people without devices. Beam-break needs alignment and counting. PIR and mmWave balance cost, privacy and capability, and their weaknesses differ, which suggests combining them.

The hybrid approach

A PIR sensor reacts to infrared change from moving warm bodies. A 24 GHz mmWave presence/motion sensor reports reflected-energy changes; exact range and detection performance depend on module configuration, placement and calibration Confirm against the S3KM1110 datasheet

Turn ONAs soon as either sensor sees presence in a zone.
Stay ONWhile either sensor still reports presence in that zone.
Turn OFFOnly after both sensors report none, continuously, for the timeout. This is designed to reduce false-vacancy events compared with PIR-only sensing.

See how each sensor sees the room

Choose what the person is doing. These are conceptual diagrams of sensing principles, not measured coverage.

PIR: passive infrared

PIR sensor looks for IR change Movement detected Little change: may read clear Field clear

Low-motion periodsA person reading or writing may produce little change in the infrared pattern.
Sunlight and warm objectsHeat sources near the lens can trigger it; to be tested.
Field leakageWide lens fields can see into a corridor or the next zone depending on placement.

mmWave: 24 GHz radar

24 GHz mmWave module transmits and listens fancurtain Possible false returns Moving target detected Low-motion presence (module-dependent) No target

24 GHz mmWave presence/motion sensor; exact range and detection performance depend on module configuration, placement and calibration.

What can go wrong with mmWave

None of these is a reason to avoid the sensor; each is a test case in the experimental matrix.

ReflectionsRadar energy bounces off walls and metal, creating returns away from the intended area.
FurnitureMetal cupboards and benches can reflect or shadow part of the zone.
FansA rotating fan is a moving target and can look like occupancy if it is not tuned out.
CurtainsMoving fabric near a window can create motion returns.
RangeDetection range is configurable and limited; each zone must be sized to the module.
PlacementMounting height and angle change what the sensor sees.
CalibrationSensitivity must be tuned per zone and verified in testing.

An empty zone does not force the entire classroom to switch off.

The classroom is split into three independent zones. Each zone has its own mmWave and PIR pair, its own timer, its own light indicator and fan, and its own manual override. The ESP32-S3 evaluates them separately.

Plan view of a stylised classroom model. Field shapes and positions are conceptual, not measured coverage. This illustration is linked to the simulator, so toggling sensors there changes the zones here.

Live zone status

One controller, three zones, one switching pattern.

Everything below describes the same prototype: an ESP32-S3, three 24 GHz mmWave sensors (S3KM1110 modules), three PIR sensors, three LED light indicators and three 5 V DC motor/fan loads switched by logic-level N-channel MOSFETs.

ESP32-S3reads 6 sensors · runs 3 zone state machinesdrives MOSFET gates, LEDs, OLED (I²C), buzzer
Zone 1
mmWave + PIRS3KM1110 module + PIR sensor
Light + FanLED indicator + 5 V DC motor/fan via MOSFET
Zone 2
mmWave + PIRS3KM1110 module + PIR sensor
Light + FanLED indicator + 5 V DC motor/fan via MOSFET
Zone 3
mmWave + PIRS3KM1110 module + PIR sensor
Light + FanLED indicator + 5 V DC motor/fan via MOSFET
0.9-inch I²C OLEDzone states and timers
Buzzeraudible feedback
Override buttonsone manual override per zone
Low-voltage supply5 V load rail · common ground

Per-zone signal chain

SensorsPIR + mmWave
ESP32-S3fusion + timers
Zone stateOCCUPIED, GRACE, VACANT…
GPIO3.3 V logic
MOSFET gateswitch only
5 V motor/fanand LED indicator

Workflow: read sensors, classify each zone as OCCUPIED, GRACE or VACANT (or MANUAL OVERRIDE / DEGRADED), switch that zone's loads, show state, and log events for energy and accuracy analysis.

Power electronics: switching a 5 V motor with a MOSFET

Low-voltage prototype · 5 V DCOne channel shown; the same circuit is repeated for each zone
+5 V supply rail GND rail (common ground) Low-voltage5 V supply +− M Motor + to +5 V Motor − to drain 1N4007flyback diodeacross motor DSG Logic-level N-channel MOSFET ESP32-S3GPIO output 3.3 V logic to gate only controller GND 10 kΩgate pulldown

Connections to build

  1. Motor positive to the 5 V supply.
  2. Motor negative to the MOSFET drain.
  3. MOSFET source to GND.
  4. 10 kΩ resistor from the gate to GND (gate pulldown).
  5. Flyback diode across the motor terminals, cathode to +5 V and anode to the drain.
  6. Controller GND and load supply GND joined: a common ground.
  7. ESP32-S3 GPIO connects to the MOSFET gate only.
Never direct. A GPIO pin cannot supply motor current. It only drives the gate; the MOSFET carries the load current from the supply.

Part selection: choose a MOSFET whose on-resistance is specified at a 3.3 V gate voltage and whose current rating comfortably exceeds the motor's start-up current. Size the 5 V supply for all three motors starting together.

Known engineering challenges

Fans and curtainsThey reflect radar. Tune sensitivity, limit range, mount away from the fan's line of sight, and test "empty room, fan running".
PlacementHeight and angle change what both sensors see, so several placements are tested.
Sunlight and heatThey can trigger PIR; the test is documented, not assumed.
Zone boundaries and corridorsDetection leaking from a corridor or neighbouring zone causes false occupancy; limit range and tune per zone.

Every component, and why it is there.

Fourteen parts make up the prototype. Hover or focus a card to lift it; filter by role.

Bill of materials

Indicative costs are not yet confirmed for the three-zone design. The original single-zone estimate in the project plan was ₹2,500–4,000; the total for three zones will be recalculated from the final parts list.

Connection map

Pin numbers are intentionally left to be filled in from the built prototype, so this page never disagrees with the firmware.

Why these technologies

Energy loss calculator.

Scenario ModelIllustrative AssumptionsNot Yet Measured

Annual waste = rooms × load (kW) × vacant-but-ON hours per day × 220 days. Pick a scenario or drag the sliders.

Each control node is assumed to draw 1.5 W continuously. Tariff ₹8/kWh, deployment ₹3,000 per room, 0.75 kg CO₂/kWh. These are assumptions to verify, not measurements, and the real standby power of the three-zone controller is listed under Prototype Measurements as not yet measured.

Table 1: Scenario assumptions

Table 2: Estimated annual loss (lights and fans)

Table 3: Recoverable savings, net of node consumption

Takeaways from the model. Savings scale with vacant-but-ON hours; in a well-run campus (Low case) it is not worth installing. Node standby use (about 788 kWh/yr under the 1.5 W assumption) is about 11% of base-case recovered energy, so low-power design matters. AC would add about 2,640 kWh and ₹21,000 per year for 10 rooms (1.5 kW, 1 h/day, 80% capture) but needs compressor-safe handling, so it is out of scope.

Prototype Measurements

Not measured yet

These cards stay empty until the experiments are run. Nothing here is estimated, simulated or copied from the scenario model. To fill them in, edit the MEASUREMENTS list at the top of the script.

Try the three-zone controller.

Switch the sensing mode, then give each zone its own inputs. Simulate a person who sits still: PIR drops out, the mmWave sensor may keep the zone occupied. Zones never affect each other.

Simulation timingAccelerated: the grace period here is 8 s so you can watch it. This is not a recommended timeout.
Real-room timingSet per room, likely several minutes, and to be decided from testing. The 20–30 s bench timeout in the project plan is also a test setting only.

Sensing mode

State machine

Automatic occupancy cycle OCCUPIEDLoads ON GRACELoads ON · timer running VACANTLoads OFF both sensors clear:start grace timer presence returns:reset timer timer expires:loads OFF PIR OR mmWave detects presence: OCCUPIED, loads ON MANUAL OVERRIDELoads as chosen · automation paused DEGRADEDOne sensor down: longer graceBoth down: loads fail ON button pressed button again sensor fault or timeout sensor recovers

Solid highlight: the zone's current state. Dashed outline: the automatic state running underneath a manual or degraded mode.

Recent transitions

    Table 7: Controller states

    Tunable parameters

    The research question.

    Core question

    Can a low-cost, camera-free hybrid PIR + 24 GHz mmWave system improve zone-level occupancy detection and reduce unnecessary classroom energy consumption?

    Where the novelty is

    Implementationa complete, low-cost build
    Zone-based controlthree independent zones
    Hybrid sensingPIR with 24 GHz mmWave
    Experimental evaluationagainst PIR-only and mmWave-only
    Classroom-specific testingfans, curtains, corridors
    Not claimed as novel: PIR sensing, mmWave presence sensing, and occupancy-based automation are all established. The contribution is how they are combined, applied per zone and evaluated in a classroom-like setting.

    Literature and research gap

    OverviewsNguyen & Aiello (2013), Dong et al. (2019), Chen et al. (2018): sensor range, accuracy limits, privacy and cost trade-offs.
    In practiceAgarwal et al. (2010): PIR and door sensors drove building control in a university, so campus-scale control is feasible.
    PIR limits and fusionLabeodan et al. (2015) show PIR struggles with low motion; Dodier et al. (2006) show fusing sensors improves detection.
    Savings potentialWilliams et al. (2012): meaningful occupancy-sensor savings, US-focused, hence scenarios plus a local audit.
    RadarManufacturer and chip-vendor documentation describe mmWave presence sensing. Add 2–3 peer-reviewed radar-occupancy papers.
    Gap: a low-cost, camera-free, offline PIR + 24 GHz mmWave controller with zone-level switching, evaluated in an Indian classroom-like setting against PIR-only and mmWave-only baselines, with ceiling-fan interference in focus.

    Comparison with existing approaches

    Experimental matrix.

    Three configurations (PIR-only, mmWave-only, hybrid) under identical conditions, with at least 20 trials per scenario per configuration (to fix with the guide). No results have been collected yet, so every cell is a placeholder.

    No data collectedHypotheses shown only on request

    Metrics by configuration

    Metric definitions

    Targets are goals, not predictions. If fan reflections force conservative mmWave settings, the report will say so.

    Expected results Hypotheses

    • Hybrid sensing is designed to lose seated-still occupants less often than PIR-only.
    • Hybrid may have more extra ON time than PIR-only if mmWave triggers on fans.
    • The prototype shows how the scenario-model estimates become a measurable test.

    Possible applications: classrooms, labs, seminar rooms, offices, libraries, hostel common areas, restrooms and corridors, each with its own timeout.

    Feasibility and 12-week plan.

    TechnicalCommon modules with plenty of documentation. The main risk is mmWave tuning across three zones, which has scheduled time.
    EconomicPrototype cost is to be confirmed from the final parts list. Deployment payback depends on real waste, so audit first.
    SafetyLow-voltage 5 V DC loads. Any 230 V work needs rated parts and a qualified electrician; students never touch live mains.
    SkillsBasic C/C++, breadboard wiring, data logging, with guidance.

    Timeline

      Team roles

      • Hardware and circuit
      • Firmware and fusion logic
      • Testing and data analysis
      • Research, energy audit and economics
      • Enclosure, integration, documentation and demo

      Safety, privacy and ethics.

      Low-voltage prototype

      The student prototype runs from a low-voltage supply (5 V DC loads). It does not switch live 230 V mains.

      Any building-mains implementation requires appropriately rated hardware, protection and qualified electrical supervision. Students do not wire, open or test anything connected to building mains.

      Risk register

      Limitations: one test room will not generalise to all rooms; no AC control; costs and tariffs are estimates; all sensor performance statements depend on module configuration, placement and calibration.

      Future improvements

      A modest, testable design for a real problem.

      EcoSwitch tackles electricity wasted on empty rooms with a three-zone, low-voltage prototype. PIR brings low cost and speed; 24 GHz mmWave sensing adds awareness during low-motion periods, with no cameras. The scenario model shows that value depends on actual waste, so the project starts with measurement and reports honestly, including what has not been measured yet.

      Request: approval to proceed with Phase 0 (campus audit) and Phase 1 (design).