IAE Academy

FOR CORE AND IT JOBS · CHAPTER 6 OF 7

SCADA, DCS and HMI

This chapter explains how a SCADA system polls its tags, what an operator sees on an HMI, and how a DCS differs from PLC plus SCADA. An engineer needs it to size a poll cycle, set alarms that do not chatter and judge what a hot standby pair buys.

Updated 30 Oct 2026

What you will be able to do

You will name the parts of a SCADA system and work out how old a polled value can be. You will read an HMI's mimic, trends and alarms and set an alarm that does not chatter. You will compare a DCS with PLC plus SCADA and judge a hot standby pair.

Before you start

  1. Chapter 5, Analog control and PID in a PLC

  2. Percentages and unit conversion (seconds and milliseconds, bits and bytes)

  3. Solving linear inequalities

  4. Sums written with the Σ sign

  5. Probability of independent events: the chance that both happen is the product

SCADA: parts, tags and polling

SCADA (supervisory control and data acquisition) lets people watch and adjust a plant from one place. PLCs or RTUs (remote terminal units, small outstations for distant sites) still run the fast control, such as the loop of chapter 5. SCADA collects their data, shows and stores it, and sends back commands such as a new setpoint.

  1. Field devices: sensors and actuators, wired as 4-20 mA, 0-10 V or on/off signals.

  2. PLC or RTU (the outstation): reads inputs, runs its program, keeps values in memory. Addresses are in IEC style, such as %MW10 for a memory word; some PLC families write them differently.

  3. Network: a shared serial line, Ethernet, radio or cellular, to the master station.

  4. Master station (the SCADA server): polls the outstations, holds the tag database and raises alarms.

  5. HMI: the operator's screens, fed by the master station.

  6. Historian: stores time-stamped tag values for later review.

A tag is a named data point with everything the system needs to use it. The master station polls: it sends a request to each outstation in turn and waits for the reply. Under plain polling, which this chapter uses, an outstation speaks only when asked. Each reply carries all of that outstation's polled tags, so a tag is read once per poll cycle, and its poll interval is met only when the cycle is no longer.

Tag field

Value for the tank level

Name

TK101_LEVEL

Source

PLC 3, memory word %MW10

Type and scaling

Analog, 0 to 100 %

Alarm limits

Low 20 %, high 80 %

Poll interval (target)

1 s

Logged by historian

Every 5 s

TIME TO SEND ONE MESSAGE

Holds for asynchronous serial, each byte framed as 10 bits (start, 8 data, stop, no parity). L is the message length in bytes, R the bit rate in bit/s, t_msg in s.

POLL CYCLE TIME

Holds when the master polls one outstation at a time on a shared line. n is the number of outstations; t_req,i and t_resp,i are outstation i's request and reply times (formula above) and t_gap,i its turnaround and wait, all in s.

WORST-CASE AGE OF A POLLED VALUE

Upper bound for plain polling, every cycle T_cycle long, outstations alike (t_resp is one reply's time), screen refresh left out. T_scan is the PLC scan time: read inputs, run program, write outputs. A change just after an input read is in PLC memory by the end of the next scan: up to 2 T_scan. The next reply starts up to T_cycle later. All in s.

WORKED EXAMPLE

A line at R = 10 000 bit/s, 10 bits per byte, links 8 outstations. Each poll: 10-byte request, 40-byte reply, 50 ms gap. PLC scan 20 ms. Find T_cycle and A_max, and how many outstations keep A_max within 1 s.

Solution
  1. Request: 10 × 10 / 10 000 = 0.010 s = 10 ms (100 bits).

  2. Reply: 10 × 40 / 10 000 = 0.040 s = 40 ms, by the same formula.

  3. One outstation: 10 + 40 + 50 = 100 ms. The eight are alike, so T_cycle = 8 × 100 = 800 ms.

  4. A_max = 2 × 20 + 800 + 40 = 880 ms: two scans, the wait for the next reply and the reply add.

  5. Each outstation adds 100 ms, so 100n + 2 T_scan + t_resp = 100n + 80 ≤ 1000 ms gives n ≤ 9.2. Outstations are whole: n = 9 gives 980 ms, n = 10 gives 1080 ms.

Answer: T_cycle = 800 ms, A_max = 880 ms; at most 9 outstations.

WATCH OUT FOR

A fast PLC scan does not make a tag fresh. Test: compare T_cycle with 2 T_scan in A_max; the larger term is the one to shorten.

The HMI (human-machine interface) is what the operator sees; three kinds of screen do most of the work.

Screen

What it shows

What the operator uses it for

Mimic

Plant picture with live values, such as a tank level and a pump coloured running or stopped

Seeing plant state, sending commands

Trend

Values against time

Seeing direction and speed of change

Alarm list

Active alarms with time, tag, priority and state

Choosing what to deal with first

  1. The value crosses its limit: the alarm is time stamped, shown unacknowledged, and may sound a horn.

  2. The operator acknowledges: the horn stops and the system records that someone saw it.

  3. The operator acts on the process.

  4. The value returns: the alarm clears and, once also acknowledged, leaves the active list.

Priority, set at configuration from the consequence of ignoring the alarm and the time left to respond, orders the list: a critical alarm shows first and loudest. A noisy signal near a limit can raise and clear an alarm again and again (chatter). The cure is a deadband below the high limit : on at process value , off only at , last state kept in between. Take a steady true value with noise within . Chatter needs and , so , possible only if .

DEADBAND THAT STOPS CHATTER

Holds for a steady true value with noise within plus or minus e. d is the deadband and e the noise amplitude, both in the unit of PV.

Older trends come from the historian, which logs each tag at its own interval , often slower than polling.

HISTORIAN STORAGE PER DAY

Holds when every tag is logged at one constant interval, uncompressed. N is the number of tags, T_log the interval in s, b the bytes per sample, S in bytes per day; 86 400 s is one day.

WATCH OUT FOR

Acknowledging an alarm fixes nothing. Test: if the process value is still beyond the limit, the alarm stays active.

DCS and redundancy

A distributed control system (DCS) spreads control over many controllers in one plant, made, configured and displayed as one system. Each tag is defined once, in one engineering database, and used by controller, display and alarm alike. PLC plus SCADA is built from separate parts joined by a network.

PLC plus SCADA

DCS

Plant type

Machines, batches, spread-out sites such as pipelines

Continuous process on one site, such as a refinery

Control style

Each PLC runs its own program; SCADA sets setpoints

Function blocks (ready-made elements such as a PID block) wired across controllers

Redundancy

Added per project, such as a server pair

Usually standard: paired controllers, networks and power

In hot standby, a second unit is powered and kept up to date with the primary's data over a link, so it can take over at once. The primary sends a heartbeat every ; if the standby misses in a row, it takes over without operator action, within plus its switchover time. A cold standby is off and must be started and loaded first.

AVAILABILITY OF A REDUNDANT PAIR

Holds when both units have availability a (the fraction of time one works), fail independently and switch over perfectly. The pair is down only when both are, with probability (1 - a)^2.

WORKED EXAMPLE

A SCADA server has a = 0.99. An identical hot standby is added (independent failures, perfect switchover). Find A_pair and the yearly downtime of one server and of the pair (8760 h a year).

Solution
  1. One server is down 1 - 0.99 = 0.01 of the time: 0.01 × 8760 = 87.6 h a year.

  2. Both must be down, and failures are independent: 0.01 × 0.01 = 0.0001, so A_pair = 0.9999.

  3. Pair: 0.0001 × 8760 = 0.876 h, about 53 min, the same fraction of the year.

Answer: A_pair = 0.9999; about 0.88 h of downtime a year against 87.6 h for one server.

WATCH OUT FOR

A pair helps only against independent failures. Test: list what the two units share, such as one power supply. Each shared item is a single point of failure that stops both.

Check yourself

CHECK YOURSELF 1

In a SCADA system, which part stores time-stamped tag values so that an operator can look at a trend from last week?

  1. (A)

    The historian

  2. (B)

    The RTU

  3. (C)

    The mimic

  4. (D)

    The final control element

Answer and solution

Answer: (A)

  1. The historian logs tag values over time.

  2. An RTU collects field data, a mimic is a live picture, a final control element acts on the process.

CHECK YOURSELF 2

A master station polls 6 outstations one at a time on a serial line at 10 000 bit/s, with 10 bits per byte. Each poll has an 8-byte request, a 22-byte reply and a 40 ms gap. Find the poll cycle time T_cycle in ms.

Answer and solution

Answer: (accepted range )

  1. Request: 8 × 10 / 10 000 = 0.008 s = 8 ms.

  2. Reply: 22 × 10 / 10 000 = 0.022 s = 22 ms.

  3. One outstation: 8 + 22 + 40 = 70 ms.

  4. T_cycle = 6 × 70 = 420 ms.

CHECK YOURSELF 3One or more options are correct

Which statements about alarm handling on an HMI are correct?

  1. (A)

    Acknowledging an alarm stops the horn and records that the operator saw it.

  2. (B)

    Acknowledging an alarm brings the process value back inside its limit.

  3. (C)

    An alarm whose condition still exists stays in the active list after it is acknowledged.

  4. (D)

    Priority lets the system show a critical alarm before a minor one.

Answer and solution

Answer: (A), (C), (D)

  1. A is correct: acknowledgement silences the horn and records it.

  2. B is wrong: acknowledging changes nothing in the process.

  3. C is correct: it leaves the list only once cleared and acknowledged.

  4. D is correct: priority orders and highlights the list.

CHECK YOURSELF 4

A historian logs 2000 tags every 5 s, storing 8 bytes per sample, with no compression. Find the storage used in one day, in MB (1 MB = 10^6 bytes).

Answer and solution

Answer: (accepted range )

  1. Samples per tag per day: 86 400 / 5 = 17 280.

  2. Samples in all: 2000 × 17 280 = 34 560 000.

  3. Bytes: 34 560 000 × 8 = 276 480 000 bytes = 276.48 MB.

Used in these GATE topics