Three-Phase Supply Quality Monitor with Sag, Swell and Unbalance Event Log

Three-Phase Supply Quality Monitor (SQM-3P) is an instrument our final-year EEE team built to watch a real 415 V LT feeder the way a power-quality engineer would: continuously, three-phase, and with every disturbance written down. Isolation potential transformers and clip-on current transformers bring the three phases into safe measurement range; an ESP32 samples all six channels, computes per-phase RMS voltage, frequency, fundamental phasors and the negative-sequence unbalance factor, and runs an event engine that flags sags (below 0.9 per-unit), swells (above 1.1 per-unit), interruptions and unbalance excursions — each with a timestamp, magnitude, duration and a captured waveform snippet stored on microSD.

A desktop gateway imports the logs into a SQLite events database, and Python analytics turn the events into the questions maintenance staff actually ask: at what hour do sags cluster, which phase is worst, how long was the deepest event, and did the lab compressor or the colony pump start at that moment? The measurement style follows the spirit of IEC 61000-4-30 Class S (10-cycle aggregations) and the terminology of IEEE 1159, which makes the project viva-defensible at a standard level rather than a collection of ADC readings.

The Indian context motivates it: Hyderabad colony feeders see daily pump starts, cooler motors and single-phase unbalance, and a college machines-lab feeder is a perfect test bed — a motor start gives a real, repeatable sag to validate against. The build stays under ₹7,000 by reusing instrument-transformer principles from the machines and measurements labs rather than a commercial analyser.

Colleges, hostels and small industries across Telangana run three-phase feeders whose daily disturbances — motor starts, pump switching, single-phase load unbalance — go unrecorded. When equipment misbehaves, there is no evidence: nobody can say whether a sag occurred, when, on which phase, how deep or how long. Commercial power-quality analysers answer these questions but cost more than a lakh rupees, so they never reach the people who need them; cheap protection relays trip without keeping a history. What is needed is an open, standards-aware three-phase monitor that measures per-phase RMS, frequency and voltage unbalance with proper isolation, detects sags, swells, interruptions and unbalance excursions using IEEE 1159 thresholds, stores every event with magnitude, duration, timestamp and waveform evidence in a queryable database, and produces maintenance-ready analytics — all within a student budget of about ₹7,000 and a B.Tech major-project timeline.

Abstract

This project designs, builds and validates a low-cost three-phase power-quality monitor. Six channels (three voltages via isolation potential transformers, three currents via SCT-013 clip-on transformers) are sampled by an ESP32; firmware computes 10-cycle RMS values, frequency, fundamental phasors by DFT and the negative-sequence unbalance factor; an event engine detects sags, swells, interruptions and unbalance excursions per IEEE 1159 thresholds, writing timestamped records and captured waveforms to microSD. A gateway imports events into a SQLite database and Python analytics produce dashboards, hourly sag histograms and exportable reports. The instrument is validated on a college feeder using a staged motor-start sag and a switched single-phase unbalance as reference disturbances.

Introduction

Power quality problems in India rarely look dramatic: a pump start dims the lights, an overloaded phase runs hot, a PC reboots for no visible reason. Utilities care about these events at feeder level, and the standards give precise vocabulary: IEEE 1159 classifies disturbances, IEC 61000-4-30 prescribes measurement. But a Class A analyser costs over a lakh rupees, so college labs and small industries never see their own disturbance history. Our monitor brings that capability to a student budget with standards-aware measurement choices.

Existing system and gap

• Commercial power-quality analysers cost ₹1 lakh+ and are locked instruments; small industries and college labs rarely own one.
• Cheap protection relays only indicate that something happened: no record, no duration, no three-phase view.
• Utility smart meters report energy, not disturbance events, at the consumer end.
• Published student builds usually log single-phase instantaneous voltage — no aggregation window, no unbalance, no database, hence no analysis.

Proposed system

• Safe analog front end: three 230 V-to-12 V isolation potential transformers and three SCT-013-000 current transformers with burdens, then anti-alias filtering and mid-rail biasing.
• Measurement core: ESP32 sampling six channels at about 1–2 kHz each, computing 10-cycle (200 ms) RMS aggregates, zero-cross frequency, 50 Hz DFT phasors and the negative-sequence unbalance factor.
• Event engine: sag/swell/interruption thresholds at 0.9/1.1/0.1 per-unit with hysteresis and minimum-duration rules; each event stores a one-second pre/post waveform buffer to microSD.
• Data path: microSD CSVs → desktop gateway (C) → SQLite database with a documented schema → Python dashboards, hourly histograms and CSV exports.
• Validation: multimeter cross-checks, a staged motor-start sag, a switched single-phase unbalance, and a week-long endurance run on the department feeder.

Feasibility

• Technical: the signal chain uses instrument-transformer isolation and sampling rates the ESP32 handles comfortably; the DFT and sequence computations are final-year signal-processing territory, which is exactly what makes this an advanced major project. The riskiest item — ADC non-linearity — is handled with a calibration table and stated uncertainty.
• Economic: transformers, CTs, ESP32, SD module and enclosure come to roughly ₹6,800, against ₹1 lakh-plus for a certified analyser.
• Operational: the monitor clamps onto an existing feeder with no supply break; the lab technician installs CTs with the feeder switched off; the instrument is fuse-protected and fully isolated from the mains side.

In scope

• One 415 V LT feeder in the department machines lab, monitored for a multi-week endurance run.
• Measurement of per-phase RMS voltage (10-cycle aggregates), frequency, line currents and negative-sequence voltage unbalance.
• Event detection for sags (below 0.9 pu), swells (above 1.1 pu), interruptions (below 0.1 pu) and unbalance beyond 2 per cent, with hysteresis and duration rules.
• microSD logging of aggregates and event waveform snippets; SQLite database and Python dashboards on the PC.
• Validation against multimeters, a staged motor-start sag, and a switched single-phase unbalance.

Out of scope

• Certified IEC 61000-4-30 Class A compliance (we follow Class S-style aggregation; the difference is documented).
• Sub-cycle transients and high-frequency harmonics beyond the 50th — the sampling rate does not support them.
• Harmonic spectrum analysis beyond the fundamental (THD estimation is future scope).
• Tripping or protection action; the instrument observes and records, it never switches the feeder.
• Web/cloud dashboards; analytics run on the department PC for viva-day reliability.

The project follows the V-model, matching JNTUH R22's two-stage structure: Stage-I (EE704PC) produces the verified design and plan, Stage-II (EE801PC) delivers the built, tested instrument.

Test plan: (1) unit tests — each measurement channel against calibrated meters at five points; (2) integration tests — injected signals from the lab function generator produce correct RMS, frequency and phasors; (3) functional tests — staged motor-start sag recorded with correct depth and duration on the correct phase; switched-load unbalance crossing the 2 per cent threshold and clearing on load removal; (4) endurance — a week of unattended logging with zero lost events verified by checksum counts. Every case is tabulated with expected, actual and status in the Stage-II report.

SQM-3P is a four-stage pipeline: isolated sensing, measurement computation, event capture, and analytics — with a database at the centre so every claim in the report is a query, not a screenshot.

flowchart TD
  FEEDER["3-phase LT feeder 415 V 50 Hz"] --> PT["3 isolation PTs 230 V to 12 V"]
  FEEDER --> CT["3 clip-on CTs SCT-013-000, one per line"]
  PT --> COND["Conditioning: divider + RC anti-alias + mid-rail bias"]
  CT --> BURDEN["Burden resistors + same conditioning"]
  COND --> ESP["ESP32: 6-channel ADC, 1-2 kHz per channel"]
  BURDEN --> ESP
  ESP --> CALC["Firmware: 10-cycle RMS, frequency, DFT phasors, unbalance factor"]
  CALC --> SD["microSD: aggregate CSV + event waveforms"]
  CALC -->|USB serial| GW["C import gateway on lab PC"]
  SD --> GW
  GW --> DB[("SQLite events database")]
  DB --> AN["Python analytics: dashboards, histograms, CSV exports"]
  RTC["DS3231 real-time clock"] --> ESP

Sensing. Isolation potential transformers step each phase-to-neutral down to a measurable level and give galvanic isolation — we never put a resistive divider directly across the mains. SCT-013-000 clip-on CTs sense line currents without breaking the feeder; their burdens stay permanently connected because an open-circuited CT is a safety hazard, and the report explains why.

Measurement core. The ESP32 samples all six channels on a hardware-timer schedule. Firmware maintains a rolling two-second buffer; every 200 ms (one 10-cycle window at 50 Hz, the Class S-style aggregation) it emits per-phase RMS, zero-cross frequency, and — on a slower tick — the fundamental phasors from a rectangular-window DFT, from which the negative-sequence unbalance factor is computed. The ADC's known non-linearity is handled by a piecewise calibration table measured against laboratory meters, and the residual appears in the uncertainty budget.

Event engine and storage. Thresholds compare the aggregate RMS against 0.9/1.1/0.1 per-unit with hysteresis to stop chatter; unbalance events use the 2 per cent planning level. On trigger, the engine dumps one second of pre-buffered waveform around the event to microSD with a DS3231-stamped timestamp.

flowchart LR
  A["Aggregate RMS every 200 ms"] --> B{"Below 0.9 pu?"}
  B -->|Yes| C["Sag begins: stamp and buffer"]
  C --> D{"Back above 0.9 pu + hysteresis?"}
  D -->|Yes| E["Sag ends: write magnitude, duration, waveform"]
  B -->|No| A

Analytics. The C gateway imports SD files into SQLite (tables: events, daily_summary, calibration), and Python answers maintenance questions: sags by hour, phase-wise unbalance profiles, the ten deepest events, and CSV exports for the report — each plotted figure traceable to one SQL query.

• Add a THD estimator (up to the 25th harmonic within our sampling budget) for a fuller voltage-quality picture.
• Move the database to a Raspberry Pi at the feeder so the gateway runs unattended and dashboards are live over the LAN.
• Implement IEC flicker (Pst) estimation and add flicker events alongside sags and swells.
• Correlate the event log with energy-meter readings to quantify the cost of poor quality to a small industry.
• Train a simple classifier on event waveforms to label causes (motor start, capacitor switching, load step) automatically.
• Pursue Class A alignment with a better ADC front end and the standard's full aggregation and marking-FLAG requirements.