System Architecture and Component Selection
Choosing the platform, sizing the controller and its scan time, remote I/O, network topology, redundancy and standardizing.
The overall architecture
A typical machine has a PLC, an HMI, drives, remote I/O and sensors joined by an industrial Ethernet network, with the safety devices hard-wired alongside. Draw this early, as one page: it settles the I/O count, the network protocol, the cabinet layout and who is responsible for what.
| Layer | What lives there | Typical technology |
|---|---|---|
| Enterprise and plant | MES, historian, reports, remote support | Ethernet/IP, firewalls, VPN (see HMI, Alarms, Data and Cybersecurity) |
| Supervision | HMI, SCADA, alarms, recipes | Panel HMI, PC, web |
| Control | PLC, motion controller, safety PLC | IEC 61131-3 controllers (see PLC and Logic Control) |
| Field network | Drives, remote I/O, valve islands, instruments | PROFINET, EtherNet/IP, EtherCAT, Modbus (see Industrial Communication) |
| Field devices | Sensors, actuators, motors | 24 V I/O, 4–20 mA, IO-Link (see Sensors and Field Devices) |
| Safety | Emergency stops, guards, light curtains, STO | Hard-wired relays or a safety PLC (see Machine Safety Systems) |
Choosing the platform
The controller brand and family fix the programming tool, the supported drives and the preferred network, and for the next ten years the spares, the training and the support. A fair comparison scores each option against weighted criteria, with the weights agreed before the scores.
Worked example: three platforms
| Criterion | Weight | Platform A | Platform B | Platform C |
|---|---|---|---|---|
| Fits the customer’s skills and spares | 25 % | 5 | 3 | 4 |
| Supports the drives and devices needed | 20 % | 4 | 5 | 3 |
| Network and protocol fit | 15 % | 5 | 4 | 3 |
| Performance and capacity | 15 % | 3 | 5 | 4 |
| Safety integration | 10 % | 4 | 5 | 3 |
| Total cost over 10 years | 15 % | 3 | 3 | 5 |
| Weighted score (out of 5) | 100 % | 4.10 | 4.05 | 3.70 |
- Platform A scores highest with 4.10, but only 0.05 above the runner-up. A margin that small is within the error of the scores: change a weight a little and see if the order changes. If it does, the decision is really about something not in the table (the customer’s standard, the supplier’s support).
- Weight the customer’s skills and spares heavily: a machine that the customer’s maintenance team cannot fix is a worse machine.
- Check the ten-year view: the product’s lifecycle status, firmware support and the supplier’s record on keeping old parts available.
Sizing the controller
| Item | How to size it |
|---|---|
| I/O capacity | Points from the I/O list plus 20 % spare, local and remote (see Requirements, Specification and Project Stages) |
| Program and data memory | Estimate per function, then multiply by 1.5 to 2 for growth |
| Scan time | Instruction speed, I/O update time and communications load (below) |
| Axes and motion | Number of axes, update rate, synchronization (see Motion Control and Actuators) |
| Communication connections | Ports, protocols, number of devices and connections to the HMI, SCADA and engineering tools |
| Safety | Integrated safety controller or a separate safety PLC |
| Retentive memory, clock and battery | What must survive a power loss |
Worked example: memory
About 60 bytes of memory per I/O point for tags and logic (132 points: 7.9 kB), 150 kB of program and 40 kB of data and recipes gives 198 kB. With a factor of 1.5 for growth: 297 kB. Choose the controller with at least that, and check the datasheet’s memory is the figure for the program and not the total including the firmware.
Scan time and response
A PLC runs a cycle: read the inputs, execute the program, write the outputs, handle communications. The time of one cycle is the scan time, and it sets how quickly the controller can react. For details of the scan cycle see PLC and Logic Control.
Worked example: estimating the scan
- I/O update: 132 points × 5 µs = 0.66 ms. Program: 24 000 instructions × 0.2 µs = 4.8 ms. Communications: 1.5 ms.
- Scan time = 7.0 ms. A watchdog is normally set to two to three times that: about 21 ms.
Worked example: how fast is the response?
- With a scan of 20 ms, an input filter of 8 ms and a 12 ms relay, the response is between 42 ms (the input changes just before the scan reads it) and 62 ms (it changes just after): a spread of one whole scan. The typical figure is 52 ms.
- A faster scan of 5 ms (a smaller program or a faster controller) gives 27 to 32 ms. Design for the worst case, not the average.
- The input filter matters as much as the scan: a 10 ms default filter hides any pulse shorter than that. A part passing a sensor for 20 ms (a 40 mm part at 2.0 m/s) is seen by a 20 ms scan only if it is long enough: the signal must stay on for at least one scan plus the filter, here 28 ms, longer than the 20 ms pulse. The part would be missed. Use a high-speed counter input, an interrupt, or slow the line.
- The parts are 100 mm apart center to center, so they arrive every 50 ms (20.0 per second): a counter must handle that rate.
- Use fast hardware for fast events: high-speed counter and capture inputs, a motion controller, or a drive with its own logic.
- Watch the scan time as the program grows; log the maximum, and keep it below the watchdog with margin.
- Put slow work in slow tasks (periodic tasks of 100 ms, 1 s) and keep the fast task short.
Central panel or remote I/O
On a large or spread-out machine, I/O blocks near the devices replace long multicore runs with one network cable. On a compact machine with a few dozen points, I/O in the main panel is simpler.
Worked example: 132 points on a long machine
| Item | Central panel | Remote I/O blocks |
|---|---|---|
| Cable to each point | 30 m average | 6 m average |
| Trunk (network and power) | – | 80 m, 6 blocks at $250 |
| Material cost (estimate) | $3 564 | $2 389 |
| Wiring labor (1 minute a meter) | 66 h | 15 h |
Remote I/O saves about $1 175 and 51 hours here, and makes the commissioning faster (a device is tested at its block). Against that: a network that must be reliable and diagnosed, a block that can fail and a cabinet to be mounted on the machine.
- Use IO-Link or valve islands where devices are in groups.
- Put each block’s power on its own protected branch (see Power Distribution and the Control Supply).
- Choose the enclosure for the machine environment, not the panel room (see Panel Layout, Heat and Enclosures).
Network topology and load
| Topology | Strength | Weakness |
|---|---|---|
| Star (switch in the middle) | One failed cable loses one device; easy to diagnose | More cable; the switch is a single point of failure |
| Line (daisy chain through devices) | Least cable; devices with two ports | A break or a switched-off device loses everything beyond it |
| Ring (with a redundancy protocol) | Survives one break; little extra cable | Needs managed switches or ring-capable devices and configuration |
Worked example: the load on the network, and a broken cable
- 12 devices each exchange 64 bytes of data every 4 ms, in each direction. With about 66 bytes of frame overhead each, that is 12 × (64 + 66) × 8 × 2 ÷ 0.004 s = 6.2 Mbit/s, 6.2 % of a 100 Mbit/s link. The cyclic load is small; keep it below about 30 to 50 % and leave room for engineering, diagnostics and the HMI.
- In a line of 10 devices, a broken cable after the 4th loses the other 6 devices; in a star only the one on that cable; in a ring none.
- See Industrial Communication for the protocols, the addresses and the cabling limits (100 m per copper segment).
Redundancy and availability
Duplicating a part (a controller, a power supply, a network link) makes a failure of one harmless. The availability of two parallel parts is 1 − (1 − A)², but only if the changeover works.
Worked example: a hot-standby controller
- A single controller with an availability of 99.90 % is down 8.8 hours a year.
- A perfect pair would give 99.9999 %. If the changeover itself works only 99 % of the time, the pair gives 99.9989 %, 0.10 hours a year.
- The changeover is part of the system: test it regularly. Most redundant systems fail because the standby had silently failed or the switch did not work. Common causes (a shared power feed, the same software fault) defeat redundancy.
- Redundancy costs more than twice as much as it seems: wiring, I/O, programming and testing. Use it where a stop is costly, and use a better single part, spares and fast repair where it is not (see Documentation, Handover and Lifecycle).
Standardizing
Worked example: fewer part types
A machine built with 14 different types of relay, terminal and contactor needs two spares of each: 28 lines on the spares list. Standardizing to 6 types cuts it to 12, with fewer errors in ordering and less training. Using the same few component types across a machine cuts spares, training and mistakes; the same applies to drives, sensors, cables and connectors.
- Standardize the software too: function blocks for motors, valves and alarms, the same tag naming, the same HMI faceplates.
- Use the customer’s approved parts list where there is one.
- Prefer parts with a long service life and a stated last-time-buy policy.
The environment
| Factor | What to check |
|---|---|
| Temperature | Ambient range, heat from the machine, the panel’s own heat (see Panel Layout, Heat and Enclosures) |
| Humidity and condensation | Heaters or breathers; conformal coating for boards |
| Dust, water, washdown | IP or NEMA rating, stainless enclosures, sealed connectors |
| Vibration and shock | Mounting, locking connectors, the components’ ratings |
| Altitude | Derating of drives and clearances above about 1 000 m |
| Explosive atmospheres | ATEX or Class/Division equipment and installation methods |
| Electromagnetic environment | Drives, welding, radios (see Grounding, Shielding and Noise) |
| Standards and markings | CE, UKCA, UL, local electrical codes where the machine will run |
Related
- System Integration: the overview of this guide.
- Requirements, Specification and Project Stages: the I/O list and specification this page starts from.
- Power Distribution and the Control Supply and Panel Layout, Heat and Enclosures: the electrical and thermal design of the cabinet.
- PLC and Logic Control and Industrial Communication.
FAQ
How do I choose a PLC platform?
Score the candidates against weighted criteria agreed in advance: the customer’s skills and spares first, then device support, network fit, performance, safety integration and ten-year cost. If the scores are close, decide on support and standardization.
Should I use remote I/O or put everything in the main panel?
Remote I/O near the machine saves cable and commissioning time on large or spread-out machines. For a compact machine with a few dozen points, I/O in the main panel is simpler. Compare cable cost, network reliability and how easy each choice is to maintain.
Why does my sensor miss fast parts?
The input filter and the scan time set the shortest pulse the PLC can see. A pulse shorter than the filter plus a scan can be missed. Use a high-speed counter or interrupt input, or a faster sensor path.
Is redundancy worth it?
Only where a stop is very costly. Redundancy adds cost and complexity, and fails if the changeover is not tested. A better single part, spares and fast repair are often more effective.
References
- IEC 61131-2, Programmable controllers: equipment requirements and tests; IEC 61158 and IEC 61784 (industrial networks).
- ISA-95 (enterprise–control integration) and the Purdue reference model; IEC 62443 (security architecture, zones and conduits).
- Controller manufacturers’ sizing guides and the network profile documents (PROFINET, EtherNet/IP, EtherCAT).
Last reviewed: October 2026. Scan times, loads and costs are typical and illustrative: use the datasheets and your own prices.