Why Your CAN Bus Works in the Lab and Fails in the Machine

Author: Srihari Maddula  •  Founder & Technical Lead, Eurth Techtronics Pvt Ltd

Category: Communication Protocol  •  Estimated Reading Time: 18–20 minutes

Published: July 2026

The Field Failure That the Bench Cannot Reproduce

The system worked perfectly in the lab. Every node communicating, every message received, error counters at zero. The engineer was confident. The system was deployed into a running machine — a conveyor line, an industrial press, a pump station — and within hours, CAN error frames appeared. Within days, nodes were going bus-off. Within a week, the customer was calling.

Back on the bench, everything works again. The cable is the same. The termination is the same. The firmware is unchanged. The error counters are zero. The engineer is baffled.

This is the most common story in industrial fieldbus commissioning. The lab environment is clean: short cables, controlled temperature, no high-current switching loads, no ground potential difference between nodes. The industrial environment is none of these things. Understanding why a working bench implementation fails in the machine requires understanding what the machine does to the physical layer that the lab does not.

Ground Potential Difference: The Invisible Killer

CAN is a differential protocol. CANH and CANL carry a differential voltage that represents the bus state — dominant (approximately 2V differential) or recessive (0V differential). The protocol is designed to reject common-mode noise because both lines are affected equally and the receiver looks only at the difference.

What the protocol is not designed to handle is a large common-mode offset — a ground potential difference between two nodes that shifts the absolute voltage of CANH and CANL at one end of the bus relative to the other. This happens in industrial environments because each machine cabinet has its own chassis ground, connected to earth at the building's earthing point. The cable path from cabinet to cabinet passes through a plant floor where large motors and variable frequency drives inject current into the earth system, creating voltage differences between earthing points of 1 to 10 volts or more.

CAN transceivers have a specified common-mode operating range — typically -2V to +7V for standard transceivers, or -7V to +12V for isolated or automotive-grade parts. When the ground potential difference pushes the common-mode voltage outside this range, the transceiver either misreads the bus state or protects itself by entering a fault state. At the physical layer, this appears as intermittent error frames correlated with motor start events — which is exactly what the engineer sees but cannot explain.

The fix is galvanic isolation at every node — an isolated CAN transceiver (ISO1050, ADM3053, or similar) that breaks the ground path between the CAN bus and the node's local ground. With isolation, the common-mode offset appears across the isolation barrier, not across the transceiver input. The CAN bus itself floats relative to both ends' local grounds, and the differential signal is unaffected.

Stub Length: The Topology Error That Costs Baud Rate

CAN is designed for a linear bus topology — a single trunk cable with all nodes tapped onto it. The standard allows stub connections from the trunk to each node, but the stub length is strictly limited by the baud rate. At 1 Mbit/s, stubs must be shorter than approximately 30 cm. At 500 kbit/s, the limit relaxes to about 1 metre. Longer stubs create signal reflections that appear at the trunk as noise during the bit time, causing bit errors.

In practice, industrial installations often use star or tree topologies because they are convenient for the cable routing. Each node gets a stub of whatever length is convenient. At low baud rates on short stubs, this works. As the system is upgraded — more nodes added, baud rate increased to handle more traffic, cable runs extended — the stub violations that were always present start causing failures.

If a star topology is required by the installation, use a CAN repeater or a CANopen hub that provides signal regeneration and topology isolation. If a tree topology is unavoidable, calculate the stub length limit at your target baud rate and enforce it in the installation specification. Measure the actual stub lengths after installation before commissioning.

Termination: Two Resistors That Everyone Gets Wrong

A CAN bus requires exactly two 120-ohm termination resistors — one at each physical end of the bus trunk. This matches the characteristic impedance of a standard CAN cable and absorbs signal reflections that would otherwise bounce back from the cable ends and corrupt bus signals.

The most common termination mistakes: more than two termination resistors (adding nodes with built-in termination enabled at non-end positions — parallel resistance reduces total termination, increasing reflection energy); missing termination (no termination at one end — the open end reflects 100% of the signal energy); and wrong resistance value (100 or 150 ohm substituted for 120 ohm, or incorrectly calculated split termination).

Verify termination in the installed bus by measuring the resistance between CANH and CANL with all nodes powered off. The correct reading is 60 ohms (two 120-ohm resistors in parallel). Significantly higher means missing or high-value termination. Significantly lower means too many termination resistors.

Modbus RTU on RS-485: The Baud Rate Margin Trap

Modbus RTU over RS-485 is simpler than CAN in many respects — half-duplex, master-slave, no distributed error handling. Its failure modes in the field are different but equally non-obvious.

RS-485 is specified to operate at distances up to 1,200 metres and speeds up to 10 Mbit/s — but not simultaneously. The maximum reliable speed at 1,200 metres is approximately 100 kbit/s, and that assumes good cable (120-ohm characteristic impedance), correct termination, and no significant common-mode noise. Real Modbus installations often use 9,600 baud or 19,200 baud precisely because those speeds leave enough margin to tolerate cable imperfections and noise.

The baud rate margin trap: the system is commissioned at 38,400 baud on a 600-metre cable run. It works reliably at 25 degrees Celsius. At 55 degrees — summer ambient in an outdoor equipment enclosure — the cable's characteristic impedance changes, the transceiver timing margins shrink, and sporadic CRC errors appear. The system becomes unreliable. The fix is reducing the baud rate, which nobody wants to do because it affects polling throughput.

The engineering answer is to characterise your RS-485 link at the extremes of your operating temperature range during design validation, not during field commissioning. If the margin is thin at 38,400 baud and 55 degrees, find out during development — not during a customer site visit.

EMI from Variable Frequency Drives: The Silent Jammer

Variable frequency drives (VFDs) — used to control motor speed in pumps, conveyors, and HVAC systems — generate conducted and radiated EMI at frequencies up to hundreds of kilohertz. This noise couples into nearby signal cables through capacitive and inductive coupling. On an unshielded or improperly grounded CAN or RS-485 cable running parallel to a VFD power cable, the coupled noise can be comparable in magnitude to the differential signal.

The EMI correlation is the diagnostic clue: errors appear when the VFD is running and disappear when it is stopped. This narrows the root cause to EMI coupling rather than bus configuration errors. The fix requires a combination of: shielded cable (drain wire connected to chassis ground at one end only to avoid ground loops); cable routing separated from high-current power cables by at least 20 cm; ferrite cores at cable entry points; and in severe cases, isolated transceivers.

THE RULE: Test your industrial fieldbus implementation in the actual electrical environment — with motors running, VFDs operating, and all inductive loads energised.

The lab tells you the protocol works. The machine tells you whether the physical layer survives.

© 2026 Eurth Techtronics Pvt Ltd  |  eurthtech.com  |  All rights reserved.