Selection guide
How to choose between Modbus, Profibus and CANopen
Short answer
Match the protocol to the PLC master that already exists, not to the datasheet that impressed you. Then fit the cable run to the bit rate, because CAN-based trunks shorten as speed rises, and mix protocols one segment at a time through gateways or the PLC itself. A fourteen-device example shows the arithmetic.
On this page
- Where each protocol sits in the stack
- Step 1: the master already in the cabinet decides
- Step 2: fit the cable run to the bit rate
- Choosing the trunk for a 14-device line with a Profibus DP master already fitted
- Mixing protocols one segment at a time
- If you are building the device, not the plant
- Questions asked
You are not choosing in the abstract: the protocol is fixed by the master your PLC already has, the cable run it must survive, and the variants the field devices ship in. Work them in that order. A Profibus port already in the rack decides more than any feature comparison; a CAN trunk's maximum length falls as its bit rate rises; and mixed plants are normal as long as each segment runs one protocol. The decision order, the numbers that drive it, and a worked selection on a fourteen-device line follow.
Where each protocol sits in the stack #
Modbus defines only an application-level register model: numbered registers and coils that ride whatever pipe you point them down, a serial pair as Modbus RTU or TCP port 502 as Modbus TCP. CANopen and DeviceNet both sit on the CAN data-link layer, but they are different application layers with different identifier plans, so they do not interoperate. PROFINET is an Ethernet-based relative of Profibus, not Profibus on a faster cable.
| Protocol | Physical and link layer | Access model | Typical domain |
|---|---|---|---|
| Modbus RTU | Serial, multidrop twisted pair | Master polls slave addresses | Meters, drives, simple serial devices |
| Modbus TCP | Ethernet, TCP port 502 | Client/server | SCADA links, gateways, devices with an Ethernet port |
| Profibus DP | RS-485 shielded twisted pair | Master/slave | Remote I/O and drives on PLCs with a DP master |
| CANopen | CAN, up to 1 Mbit/s | Multi-master capable, object dictionary | Embedded machines, motion, mobile equipment |
| DeviceNet | CAN, 125/250/500 kbit/s | Master/slave plus peer messaging | Sensors and actuators |
For every device on the bill of materials, write down which of these columns its datasheet actually lists. A device with no stated protocol cannot be given one by wiring.
Step 1: the master already in the cabinet decides #
A fieldbus master is two things: a hardware interface, whether an onboard port, a scanner or a communication module, and the configuration tool that loads the device description files. The master decides the shortlist.
List every controller in the project, the ports and modules fitted to each, and the protocols each supports as master, not as slave or server. Then pick the protocol that covers the largest number of devices with a master you already own. Everything else becomes either a variant swap, meaning the same sensor ordered in the protocol your master speaks, or a gateway. Controllers whose serial side is occupied or absent often take Modbus RTU through an added module or gateway, as covered in connecting ControlLogix to RS232 and RS485 serial devices.
Step 2: fit the cable run to the bit rate #
CAN-based trunks, CANopen and DeviceNet, are terminated at both ends of the trunk, and their maximum length falls as the bit rate rises: the arbitration that makes CAN robust requires the signal to cross the whole bus within one bit time, so a shorter bit time means a shorter bus. DeviceNet is specified at 125, 250 and 500 kbit/s; CANopen runs up to 1 Mbit/s, where the trunk is at its shortest. Set the lowest bit rate that still meets your update-time budget and step up only when that budget forces it.
The signature of an over-long trunk is near nodes answering while the far end times out. Physically, differential signaling on a shielded twisted pair is what carries these buses past drive outputs and contactors, and CAN was designed for interference-prone vehicle wiring, so keep the pair shielded and routed away from power cables rather than betting on the protocol to absorb the noise.
Choosing the trunk for a 14-device line with a Profibus DP master already fitted #
Inputs, each with its source:
- Bill of materials for the line, an illustrative case in the shape of the question: 14 field devices, valves, remote I/O blocks and two drives.
- 9 of the 14 ship a Profibus DP variant.
- 12 of the 14 ship Modbus RTU, including all 5 that lack a Profibus variant.
- 3 of the 14 ship CANopen.
- The rack today: one Profibus DP master port free; no Modbus or CAN master hardware fitted.
The rule in words: choose the protocol whose master you already own that covers the most devices, group the leftovers by the protocol they share and put each group of two or more behind one gateway, and swap a lone straggler for a matching variant.
The arithmetic:
- Profibus: 14 − 9 = 5 devices outside the master's protocol; master cost 0, because the port is fitted.
- Modbus RTU: 14 − 12 = 2 devices outside; master cost one serial module. The overlap checks out: 12 − 5 = 7 devices carry both protocols.
- CANopen: 14 − 3 = 11 devices outside; master cost one module or gateway.
Profibus DP wins on both counts: the fewest devices outside and the only master already in the rack. The 5 leftover devices all speak Modbus RTU, so they share one RS-485 segment at slave addresses 1 through 5 behind a single Modbus RTU-to-Profibus gateway.
Decision: run the Profibus DP trunk on the fitted port and one gateway for the five. The threshold the arithmetic expresses is two or more leftovers sharing one protocol earn a gateway; a single leftover gets swapped for a variant, because a gateway costs more than one sensor in the master's protocol. The translator can be a boxed unit, or using a PLC as a Modbus RTU to Profibus gateway on capacity you already have.
Mixing protocols one segment at a time #
Mixed plants are normal; the rule is one protocol per segment, joined where translation happens, at a controller, a gateway or a software runtime. Follow one request hop by hop: a SCADA client sends Modbus TCP to the gateway, the gateway acts as Profibus master on the second segment and polls the slaves, and the reply returns along the same path. The gateway adds its own scan time to the round trip, so the end-to-end update time is the sum of the segment scans plus the copy delay.
On the CAN side, motion is the common case: running CiA 402 CANopen drives from an S7-1500 through a gateway keeps the drive profile on its native bus and hands the PLC cyclic data. Ethernet protocols mix the same way, and connecting PROFINET IO to a CompactLogix through a gateway is the same pattern one layer up. When the translator is software rather than a boxed gateway, a protocol-conversion or OPC UA runtime, it runs on an industrial mini PC mounted next to the controller.
Before you pull a node to prove the diagnostics, drive the machine to a safe state: a node dropping off the bus can leave its outputs holding the last commanded values.
- Every node carries the same address plan, bit rate, parity and framing, with no duplicate addresses.
- A scan from the master brings every configured node online with no bus fault, and the error counters stay flat across a production cycle.
- One value changed at the field end arrives at the supervisory end within the update time you calculated.
If you are building the device, not the plant #
When the question is which protocol to build into a sensor, relay or valve controller, the customers' controller lists decide. Ship Modbus first: it runs on any serial port and carries no conformance-testing requirement, which is why it appears on so many small devices. Add Profibus or DeviceNet when the customer base demands them, because those networks expect conformance-tested, certified stacks. CAN is the choice for a new, self-contained network or automotive-type equipment, and lighting has its own protocols in DALI and DMX. The committed default for an industrial device: Modbus RTU as the base variant, one certified protocol added per customer demand.
Questions asked
Why does a Modbus RTU segment time out when the wiring tests good?
Mismatched serial settings come first: baud rate, parity and stop bits must be identical on the master and every slave, and each slave needs a unique address. Reversed A/B polarity or a missing terminator at either end produces the same timeouts. To check termination without dismantling the trunk, disconnect the master and measure across the two signal conductors: two fitted terminators in parallel read about half the value of one.
Can a CANopen device sit on a DeviceNet trunk?
No. Both use the CAN physical and data-link layers, but their application layers, identifier plans and device profiles are different, so a DeviceNet master cannot read a CANopen node even at a matching bit rate. Give the node a CANopen master, or put a gateway between the two networks and map the data across it.
Is PROFINET just Profibus on a faster cable?
No. PROFINET is an Ethernet-based protocol in its own right; Profibus DP is an RS-485 fieldbus. A Profibus device does not become a PROFINET device by recabling, and the two are joined only through a gateway or a controller that speaks both.
Which bit rate should a long CAN trunk run?
The lowest one that still meets your update-time budget. Maximum CAN bus length falls as the bit rate rises, so a long trunk only ever argues for a lower rate: with DeviceNet's 125, 250 and 500 kbit/s settings, start at 125 kbit/s and step up only when the update budget forces it.