Weiheng Engineering • Version 1.0 • October 30, 2026
Robot Joint Temperature-Drift Compensation and Error-Separation Plan
Temperature compensation and encoder or reducer error separation require independent reference planes and controlled thermal, direction, load, and position states. Retain raw uncompensated results, the compensation-model version, and compensated outputs separately. Correlation alone cannot uniquely assign error to an encoder, reducer, fixture, or control model.
Project and reference architecture
| Project / DUT | | Revision | | Date | |
|---|
| Command plane | | Motor/input plane | | Joint output plane | |
|---|
| Independent reference | | Load plane | | Controller / model version | |
|---|
Keep four outputs separate
| Output | Required definition | Not equivalent to | Project metric / evidence reference |
|---|
| Raw position / angle error | Command or independent reference minus measured result at named planes before compensation | Compensated accuracy | |
|---|
| Temperature drift | Change in a named metric across declared thermal state and time under controlled conditions | All long-term drift or wear | |
|---|
| Compensation effect | Raw versus compensated result using a declared model version and independent evaluation set | Proof that the physical source was removed | |
|---|
| Error-source estimate | Bounded contribution supported by independent measurement or controlled intervention | Unique root cause from correlation alone | |
|---|
Candidate source and attribution controls
| Candidate source | Evidence needed | Controlled comparison | Attribution limit | Project record |
|---|
| Motor/input encoder | Input reference, direction, speed, temperature, interpolation state | Controlled mechanics or independent input reference | Output accuracy remains outside this evidence | |
|---|
| Output encoder | Independent output reference, mounting, direction, load, temperature | Raw output encoder versus independent reference | Fixture/reference error may remain | |
|---|
| Reducer transmission | Synchronized input/output references, verified ratio, direction, load, position, temperature | Input-output relation across cycles and states | Includes coupling/fixture unless bounded | |
|---|
| Fixture / alignment | Reference stability and fixture characterization | Approved rotate, reseat, or reconfigure study | Correction uncertainty remains | |
|---|
| Controller / compensation | Command, feedback, firmware, model and state | Compensation off/on with same hardware state | Changed control behavior is part of result | |
|---|
Versioned compensation workflow
- Freeze the DUT, fixture, controller, firmware, calibration, processing, and method versions.
- Acquire raw uncompensated records across declared direction, position, load, and thermal states.
- Separate identification or fitting data from independent validation data.
- Record model inputs, structure, parameters, fitting method, and valid domain.
- Apply the versioned compensation without deleting raw measurements.
- Compare raw and compensated results using the same named metrics and conditions.
- Test repeats, reversals, thermal cycles, and out-of-domain behavior.
- Preserve changes, exceptions, invalid runs, retests, and the reviewer decision.
Evidence limit: If independent reference or controlled intervention cannot distinguish candidate sources, record not identifiable with current evidence. This plan does not claim a Weiheng accuracy, repeatability, drift, compensation performance, sampling rate, diagnostic result, customer result, or universal limit.