Smart RO Controller: System Architecture

Two-tier architecture, not one

A single Raspberry Pi doing everything — HMI, sensor polling, relay switching — is the tempting design and the wrong one for an unattended appliance sitting next to a 3-phase pump. Split the job instead.

They talk over a simple UART link (JSON packets, ~1 Hz). The Pi never touches the pump directly. That single decision is what makes "unattended operation" — a claim already on your site for the SM-379 — actually true rather than aspirational, and it's the difference customers eventually notice as reliability.

Electronics & control schematic

Four bands, top to bottom: the wetted/process path, the field sensors reading it, the control electronics, and the electrical power section. Boxes are grouped by function, not by literal wire routing — your panel wiring diagram will fan out from here.

Note on "backwash": spiral-wound RO membranes themselves generally aren't backwash-tolerant — the automatic backwash cycle applies to the sediment/multimedia pre-filter stage ahead of the membrane, with effluent routed to the same waste line. If your SM-379 uses a backwashable UF or capillary stage instead, the same V1–V3 solenoid + timer/ΔP logic applies unchanged — worth confirming with your mechanical drawing so the trigger (timer vs pressure-drop) is set correctly.

Sensor & actuator map

One row per feature you listed. "Interface" is what actually lands on the ESP32 — RS-485 for the two parameters customers read most closely (TDS, pH), simple pulse/analog for the rest.

Reject/waste flow isn't in your list of two flow points — it's cheap to derive as (LPH in − LPH out) for display, or add a third paddlewheel sensor on the waste line if you want a true, independently-measured recovery % on the HMI. Recovery % is a strong "premium" number to show a customer that a base fabricated unit can't.

Control logic

This is the loop that runs on the ESP32 — the Pi only observes and displays it. Pump start is gated on level interlocks; backwash interrupts filtration on its own schedule; any fault de-energizes the pump and returns to idle.

flowchart TD
    A[Power ON] --> B[ESP32 self-test: sensors, relays, comms]
    B --> C{Fault found?}
    C -- Yes --> D[HMI shows fault code · pump locked out]
    C -- No --> E[Idle — waiting for START]
    E --> F{START pressed?}
    F -- No --> E
    F -- Yes --> G{Antiscalant OK AND feed-tank level OK?}
    G -- No --> H[HMI alert: low level · pump does not start]
    H --> E
    G -- Yes --> I[Energize pump contactor]
    I --> J{Run-feedback received within 3s?}
    J -- No --> K[Trip: pump fault · de-energize · alarm]
    K --> E
    J -- Yes --> L[Filtering — stream TDS, pH, flow, pressure to HMI]
    L --> M{Backwash due? timer or pre-filter ΔP}
    M -- Yes --> N[Pause filtration · run backwash sequence]
    N --> L
    M -- No --> O{Alarm? high TDS / high membrane ΔP / low antiscalant / STOP}
    O -- No --> L
    O -- Yes --> P[De-energize pump · close valves · log event]
    P --> E
  

Automatic backwash sequence

Two valid triggers — pick one, or offer both as a setting:

1. Pause normal filtration — de-energize (or bypass) the feed path to the membrane.
2. Open drain-path solenoid(s), reverse flow direction through the pre-filter media.
3. Run for a fixed duration (typically 1–3 min, tune on commissioning).
4. Close backwash valves, re-open normal flow path.
5. Resume filtration; log the event with timestamp and trigger reason for the service history screen.

Bill of materials — two tiers

Figures are indicative single-unit market prices in India (Sept 2026) gathered for planning, not a quote — get formal pricing from your electrical BOM supplier at your volumes before costing the product. Essential = lowest defensible cost for the full feature set. Pro = what actually justifies a premium price tag next to a "just automatic start/stop" competitor.

Rough incremental electronics BOM: Essential ≈ ₹20,000–24,000 per unit · Pro ≈ ₹58,000–78,000 per unit, before your usual margin — heavily dependent on pump HP (drives contactor size) and whether you pick the all-in-one industrial touch panel.

Safety & panel-build notes

• STOP must not depend on software. Wire the STOP button in series with the contactor coil circuit (a hardwired interlock), not only into an ESP32 GPIO. If the ESP32 or Pi hangs, the customer can still cut the pump.
• Galvanic isolation. Everything touching 3-phase (contactor coil sensing, pump feedback) should cross to the low-voltage side through opto-isolators or isolated relay modules — never a direct tap.
• Phase-failure / single-phasing protection on the pump circuit is cheap (~₹500–1,500) and prevents the single most common cause of burnt 3-phase pump windings in the field — worth it even in the Essential tier.
• RS-485, not long analog runs, near the pump. A 3-phase contactor switching is an EMI source; differential RS-485 signalling shrugs it off far better than a single-ended 4-20mA loop run alongside power cable.
• DIN-rail everything. Relays, terminals, SMPS, MCB — standard panel-builder practice, and it reads as "industrial," not "hobby project," to anyone who opens the door. That perception is part of what you're charging for.

HMI screen set

One kiosk app, full-screen, no OS chrome exposed. Suggested screens:

• Home / Dashboard — TDS, pH, LPH in/out, recovery %, pump status, antiscalant %, membrane health, total run hours. This is the screen a customer looks at 95% of the time.
• Settings — TDS alarm setpoint, backwash interval or ΔP threshold. PIN-gate this so end customers can't drift it away from spec; your technician's PIN is different from theirs.
• Alarms / history — timestamped fault and backwash log, exportable for AMC visits.
• Service reminder — membrane run-hours, pH-probe calibration due date. A small feature that reads as "the machine looks after itself."

Where the margin actually comes from

Don't cost-cut the two parameters a customer reads with their own eyes — TDS and pH. Putting both on real RS-485 transmitters instead of hobby analog boards is a small BOM delta (a few thousand rupees) against a control system that already runs into tens of thousands; it's the cheapest credibility you'll buy. Flow and level, by contrast, are where the Essential tier is genuinely fine — the customer sees a number on screen either way and won't know or care which sensor produced it.

The bigger lever isn't the BOM at all — it's reuse. Build one ESP32 firmware image and one Pi HMI app, and ship it across the whole 1000/2000/3000/4000+ LPH range with only calibration constants (flow K-factor, pressure range, pump HP) changing per model. That's one engineering effort amortised over your entire product line, which is what actually moves margin at scale — not which brand of pressure transducer you pick.

And the electronics you're building here is also a second revenue line on its own: remote monitoring / AMC-as-a-subscription, sold against the same Wi-Fi/4G hook already sitting on the Pi. That's recurring revenue on top of the one-time unit sale, and it's a natural extension of "Automatic Fault Diagnostics" and "Nationwide Service Support," which are already on your site.

Open items to confirm against your mechanical design

Mechanical schematic — 1000 LPH vertical-membrane skid

The process/piping layout of the SM-379 itself, separate from the electronics diagram above: what's physically on the skid, how it's plumbed, and where each of the seven monitored points sits.

Instrument tags follow the same names used in §03's sensor map (FI = flow, PT = pressure, LS = level) so the two diagrams read as one system: this one shows where each device physically sits on the skid, §02 shows how it's wired.