What Are the Self-Learning Wires on an Ebike Controller? Safe Setup and Direction Reversal

Direct Answer: What They Do and Do Not Do

A setup step, not a performance feature. Treat it as commissioning work.

Self-learning saves you from matching phase and hall wires by trial and error. It does not make the bike faster, and it is not something you leave connected.

What the Controller Is Actually Learning

A brushless motor is driven by energising its windings in a repeating sequence, and the controller has to know which of its phase outputs corresponds to which motor winding, and how the motor’s hall sensors report rotor position. Get that relationship wrong and the motor judders, runs roughly, draws heavy current or refuses to turn. Traditionally you found the right combination by trying phase and hall pairings until the motor ran smoothly. A learning routine automates that: the controller drives the motor briefly, watches how it responds, and stores the arrangement that works. Our guide to hub motor internals covers how phases and hall sensors work together.

It is also worth being clear that learning mode is temporary by design. It exists to be entered, completed and left. A system that only runs while the learning wires are joined has not finished the routine, and riding it in that state is not what the mode is for.

How the Routine Generally Works

The shape is consistent across systems that support it, even though the specifics are not. Power down and confirm the wheel is free. Connect the learning pair identified in your documentation. Power on and observe the motor as it turns under its own power. Watch whether rotation is smooth rather than rough or stuttering. Confirm the wheel turns in the correct direction for forward travel. Disconnect the learning wires at the point your documentation specifies. Then power cycle and test normal throttle or pedal-assist operation with the wheel still raised.

Two general cautions that apply whatever your controller. Do not borrow a sequence written for a different make and assume it transfers — connector arrangements, entry conditions and completion behaviour all differ. And do not treat repeated attempts as a diagnostic method: if two properly executed attempts do not produce smooth rotation, something else is wrong, and continuing to cycle a motor into a fault condition risks damage. Our installation mistakes guide covers the wiring faults that most often sit underneath a failed routine.

Reading the Symptoms

Two stages, because what happens during the routine and what happens after you disconnect point at different things.

Motor judders or chugs during learning. Usually a hall sensor or phase connection problem rather than a failure of the routine itself. Stop, power down and inspect the hall and phase connectors for security, corrosion, bent pins and damaged insulation before trying again.

Motor spins smoothly in learning mode but not afterwards. Points at the routine not completing or not saving, or at the learning wires being disconnected at the wrong moment. Repeat the documented sequence exactly, paying attention to the specified disconnection point.

Motor only works while the learning wires stay connected. The routine has not completed. This is not a workaround to ride with — the system is sitting in a commissioning state rather than a running one. Repeat the routine per your documentation, and contact support if it will not complete.

Motor runs backwards after learning. Common, and usually resolved by repeating the routine as documented, since some systems establish direction during learning. Some controllers instead offer a direction setting in software. Either way, do not start by swapping phase wires on an auto-learning controller: that pre-empts the routine and can leave you with a combination the controller then learns around.

Motor does not move at all. Check power reaching the controller, connector seating and whether you have identified the correct learning pair. If the display reports a hall, motor or communication fault, record the exact code and what you were doing before resetting anything.

Self-Learning or a Hall-Sensor Fault?

These present similarly and are worth telling apart before buying parts. Hall sensors report rotor position to the controller so it knows when to switch each phase. When one fails, the controller loses part of that picture.

Diagnostic starting points, not confirmed causes. Connectors are cheaper to check than parts are to replace.

On our smart controllers the Ride Power app’s live data includes Hall state and fault codes, so you can see whether the controller is reading the sensors before you replace anything.

Check connectors and pins before replacing anything, because a poor connection produces the same symptoms as a failed sensor and costs nothing to rule out. Measurement has its place afterwards, but keep it within safe limits: low-voltage signal checks with the system de-energised are one thing, and probing battery or controller power conductors is another. Do not probe high-voltage conductors, and do not open a battery pack — internal motor or controller diagnosis belongs with a qualified technician (GOV.UK).

Self-Learning vs Manual Phase and Hall Matching

Different routes to the same result. Neither removes the need for the wheel to be raised.

Manual matching means working through phase and hall combinations until the motor runs correctly, which is legitimate on controllers without a learning function but carries more risk of running a motor in a bad combination while you search. Where a controller supports learning, use it. And note that a hall-sensor fault can prevent a learning routine from completing at all — if the controller cannot read position reliably, it cannot establish the relationship, which is why a failed routine sometimes points at the sensors rather than the procedure.

How Self-Learning Differs Across Controller Systems

Four systems, listed alphabetically. These are described rather than ranked, and the point of the section is that each documents its own approach — a routine from one does not transfer to another.

1. Flipsky FSESC 75100 Pro

An open-platform controller configured through a computer application rather than a wire pair, where motor identification is run as a detection routine in software. That gives unusually detailed control and logging, and it expects a laptop, a considered setup and a reasonable understanding of the parameters. The detection process is documented by the platform rather than by a physical learning connector, so anyone coming from a wire-pair system should expect a different workflow entirely.

2. Generic Sine-Wave Controllers with a Learning Jumper

Many budget sine-wave controllers use a two-wire learning jumper: join it, power up with the wheel raised, check the direction, then disconnect. Owners report that some reverse direction when the routine is repeated or with a throttle input during learning, and that some must be powered down to save the result — exactly why the steps for one unit should not be assumed for another. Documentation is often thin, so ask the seller for the procedure for the exact model before you start.

3. KirbEbike Smart Controllers

Our controllers support self-learning, Hall-sensor and Hall-free operation and Hall compensation. The learning lead is labelled on the harness diagram, self-learning is also listed in the Ride Power app’s settings, and the app’s live data shows Hall state and fault codes, which helps separate a learning problem from a sensor fault — see the controller and display range. Honest limits: harness layouts can change between generations, so confirm yours with support; they pair with our own motors and displays rather than universally, and the units are potted, so a fault means replacement rather than repair.

4. Sabvoton SVMC7260

A programmable FOC controller with a documented self-learning function for motor identification, configured through its app or PC tooling. The routine is part of a broader parameter set covering current limits and other behaviour, so identification is one step within a fuller commissioning process. Documentation quality varies between resellers, which matters here more than usual: an approximate procedure for a controller that spins a motor under power is not a safe substitute for the manufacturer version.

Listed alphabetically and not ranked. Procedures are not interchangeable between systems.

Common Mistakes to Avoid

If a step requires guessing which wire is which, stop and get the documentation.

Does Self-Learning Change Performance or Legality?

No to both. Self-learning establishes correct commutation so the motor runs as intended; it does not raise current limits, add torque or increase speed. If the bike feels quicker after a successful routine, that is a motor running properly rather than a motor running harder. And it has no bearing on classification: what determines whether a completed cycle is a road-legal electrically assisted pedal cycle is the motor’s continuous rated power, working pedals and where assistance cuts off — in Great Britain, 250W and 15.5mph (GOV.UK). Commissioning a controller correctly changes none of that.

Conclusion: Learn the Motor Once, Then Return to Normal Mode

Self-learning wires exist to save you from matching phase and hall combinations by hand. They put the controller into a temporary commissioning mode, it works out how to drive your motor, and then you disconnect them and ride normally. That is the whole feature — it is not a performance setting and not something to leave connected.

So treat it as workshop work rather than a tweak. Raise and secure the wheel, keep everyone clear, identify the connector from your own documentation rather than by colour or by a video for another make, and know how you will cut power before you apply it. If the motor judders, will not run, or works only while the wires are joined, stop and check connectors before repeating. And stop entirely on heat, smoke or a burning smell (GOV.UK) — that is a fault, not a step to push through.

FAQs

Sources