First, sort the failure into one of three types - the routes are completely different
When an EMC test comes back failed, resist the urge to start reworking the board. Sort the failure into one of three types first, because the root causes, localization methods, fix directions and retest scope differ completely, and mixing them simply wastes a retest slot.
- Emissions above the limit - the device disturbs others.
- Functional failure under immunity testing - the device is disturbed by others, producing abnormal motion or loss of function.
- Lock-up, reset or loss of communication under electrostatic discharge and transients - an immunity issue, but with a different mechanism and different remedies from RF field immunity.
EMC for powered wheelchairs and mobility scooters sits under ISO 7176-21, and the electrical and electromagnetic requirements EN 12184 places on powered wheelchairs point in the same direction. The composition of test items follows the relevant part; confirm against the currently valid version of the standard text.
The decision rule in one line: first establish whether the symptom is the device disturbing others or others disturbing the device, then whether the failure recovers by itself or needs manual intervention. Without answers to those two questions, every subsequent fix is guesswork.
Failure type reference table
| Symptom | Common root cause direction | Localization method | Fix priority |
|---|---|---|---|
| Conducted disturbance above the limit in the lower frequency range | Drive switching noise returning through the supply and charging leads; common-mode path untreated | Disconnect the load and measure segment by segment, swapping harnesses and filter parts one at a time | Noise at source, then common-mode filtering, then routing |
| Radiated disturbance above the limit in the mid and upper frequency range | Harness long enough to act as a radiator; poor bonding of the controller enclosure; switching edges too steep | Near-field probe scanned along the harness, seams and connectors | Routing and loop area, then bonding and shielding, then ferrites |
| Wheelchair starts, accelerates or steers by itself in an RF field | Joystick analog signal disturbed; signal leads unshielded or untwisted; reference ground contaminated | Shield segment by segment to eliminate candidates, and swap the joystick assembly for comparison | Signal integrity, then hardware filtering, then software criteria |
| Display or prompts misbehave in an RF field but the drive does not move | Communication between the display unit and the main controller is disturbed | Isolate the display unit on a separate supply and retest | Communication checksum and retransmission, then local shielding |
| Lock-up after electrostatic discharge, requiring a power cycle | No defined discharge path at enclosure seams, keypad openings or connectors | Reproduce point by point, mark the sensitive points and build a sensitivity map | Discharge path design, then port protection, then watchdog |
| Charging interrupted or controller reset under transients or surges | Protection components on the charging port missing or wrongly selected | Apply to the charging circuit alone and compare against the complete-machine result | Port protection components, then front-end supply design |
How to use the table: take the failed item, locate it on one row, then work through that row's priority order. Do not skip the earlier priorities and jump to the later ones. Bolting on ferrites and shielding foil often scrapes a pass, but the moment a supplier ships a different batch of parts in production, the machine fails again.
Immunity failures: the wheelchair moved on its own, and that is the serious category
Within immunity failures, the grading of the performance criterion matters more than pass or fail. Broadly there are three levels.
- Acceptable: no abnormality during exposure, or a brief display fluctuation with drive and braking unaffected, recovering by itself once the disturbance is removed.
- Requires assessment: function interrupted but the device stops in a safe state, with a manual reset needed before use can continue.
- Not acceptable: unintended starting, unintended acceleration, steering deviation, loss of braking, loss of the ability to stop.
The third level is not negotiable. It fails on sight, and the fix cannot rely on software masking alone. The joystick is an analog input and is an easy entry point for RF field coupling. The effective fix directions, in order, are: terminate the joystick signal cable shield reliably to the enclosure at one end, twist the signal pair, add differential filtering close to the controller, plan the signal reference ground separately from the power ground, and add plausibility criteria in software - for example, treating an input whose rate of change exceeds anything a human hand could produce as invalid.
Note that software criteria are a supplementary measure, not the primary one. If software alone is left to filter out the abnormal input, the criteria stop working as soon as the disturbance level or the coupling path changes.
Emissions above the limit: separate conducted from radiated first
The two are handled along very different lines.
Conducted emissions mostly originate in motor drive switching noise returning through the supply and charging leads. Localization path: split the machine into segments - controller plus motor, battery plus harness, charging circuit - and power each segment on its own to see which one exceeds on its own. Once the source is identified, fix it at the source first by softening the switching edges, improving the freewheeling loop layout and shrinking the area enclosed by the power loop. Only when those cannot be moved should you add common-mode chokes and filter capacitors.
Radiated emissions are mostly the harness acting as an antenna. A complete wheelchair has a long harness that takes an indirect route across structural members, which easily forms a large loop. Localization path: scan with a near-field probe along the harness, the enclosure seams and the connector positions to find where field strength concentrates. Fix in order: shorten the harness, run the outgoing and return conductors together to shrink the loop area, establish a low-impedance bond between metal structural members and the controller enclosure, add local shielding where needed, and put ferrites last.
Localization method: eliminate by segment, change one thing at a time
This is what decides whether a retest passes first time, and it is also where most of the mistakes happen.
Failure case: a single rework round adds ferrites, swaps in shielded cable and changes the software filtering logic all at once. The retest passes, but nobody knows which change did the work. When purchasing later switches to a different ferrite specification, the product fails again, and because the critical change was never identified, the whole localization exercise has to start over.
The correct approach is a change log. Change one thing at a time and record the change, the pre-scan result and the conclusion. Pre-scan work can be done under informal conditions at a fraction of the cost of a formal retest. Once the log can point to how much improvement each individual change produced, book the formal retest. The time spent here is repaid in full during production.
What has to be retested after rework: not just the failed item
This is the knock-on cost most manufacturers overlook. EMC rework usually comes with hardware and structural changes - rerouted harnesses, added shielding cans, a different controller, altered enclosure bonding. Three questions decide the scope.
| Change made | Does it touch the load path | Does it change mass in running order and center of gravity | Does it change control logic | Knock-on retest direction |
|---|---|---|---|---|
| Harness shielding material changed only | No | Essentially not | No | Generally only the EMC-related items |
| Metal shielding can or additional bonding parts added | Possibly | Yes | No | Stability and strength directions need assessment |
| Controller platform replaced | No | Yes | Yes | Drive control, braking and EMC all need redoing |
| Enclosure structure changed to add a discharge path | Yes | Yes | No | Structural strength and protection-related items need assessment |
Conclusions in the ISO 7176 series relating to structure, stability and braking are all built on a specific sample condition. Once rework changes that condition, the related conclusions have to be confirmed again. Skipping this step means shipping against a report that is internally inconsistent. The composition of items across directions on the complete machine is set out under powered wheelchair testing, and mobility scooters follow the same pattern - see mobility scooter testing.
Things you can design out before any of this happens
None of the following adds material cost, and all of them measurably reduce the chance of failing the first attempt.
- Partition the power ground and the signal ground, connect at a single point, and keep power return current out of the signal reference region.
- Shield and twist joystick and sensor signal leads over their full length, and fix the shield termination method at the design stage rather than leaving it to the production line.
- Fix harness routing during structural design, avoiding long parallel runs along metal edges and avoiding detours that create large loops.
- Provide bonding points between the controller enclosure and the frame, using serrated washers for durable low impedance rather than relying on a painted surface for contact.
- Treat the charging port as a separate electrical port and reserve footprints for protection components, even if the first build leaves them unpopulated.
- Reserve discharge paths around enclosure seams, keypad openings and connectors, instead of modifying tooling after ESD finds the problem.
Changing this at the structural design stage costs nothing, changing it once samples have arrived means rescheduling, and changing it after tooling is cut means paying for both the tooling and a fresh round of verification.
Three common misconceptions
One: assuming the complete machine is fine because the controller is compliant. A component-level conclusion does not cover the complete machine. Harness length, routing, structural materials and grounding arrangements have all changed, so the complete machine is a new electromagnetic environment.
Two: treating an immunity failure as a software problem. Software can only address the criteria layer; a coupling path problem has to be solved in hardware.
Three: piling on parts to scrape a pass. Ferrites, shielding foil and filter capacitors stacked together may pass on test day, but production consistency is not assured. The way to build margin is to solve the problem at the source and along the path, not to patch it at the end.
A note on accreditation and rework
One point worth stating: an accreditation mark only demonstrates that the laboratory has the corresponding technical competence within its accredited scope, and it is not a commitment regarding market access in the target market. When selecting a laboratory, what matters more is whether they will work through root cause with you during pre-scanning rather than simply hand over a failed result. Capability by direction is set out under standards and scope.
Stuck on EMC? Start with a pre-scan
SUNGO Mobility Testing Lab is the dedicated wheelchair and mobility aid testing lab within the same group, accredited by CNAS, CMA and IAS (USA), with laboratories in Shanghai and Hefei. We have worked on complete-machine EMC for powered wheelchairs and mobility scooters for years, and the localization paths for joystick misoperation, charging circuit transient resets and harness radiation are well established.
Send over the failed items, the complete electrical schematic, the harness routing drawing and the controller details. We will come back with a first view on root cause and a fix priority list, arrange a pre-scan where useful, and confirm the improvement before booking a formal retest. That is a full round trip cheaper than reworking and going straight back into the chamber.
Call +86 132 4819 8029, or request a quote.