The buyer task and decision boundary
Fault investigation should distinguish an observed alarm from an established failed component. Preserve the first deviation, the task and the installed revisions before selecting a replacement. A successful restart after several simultaneous changes only describes the new combination. The dossier should expose that limitation and identify the next discriminating observation, rather than retrospectively turning the first suspicion into a proven cause.
Technical context: Preserve the symptom before choosing a part
A useful maintenance dossier contains the original alarm text, controller time, task stage, tool state and the changes already attempted. Distinguish operator recollection from an exported device log. Keep both, but label their origins. If the fault is cleared before collection, say so instead of reconstructing a precise timeline from memory. The purpose is to help a service party reproduce the decision path. It is not to make the first suspected component look certain. Photographs should identify a label or connection, not substitute for an electrical or mechanical inspection.
A second condition to preserve: Build a compatibility map with separate decisions
Give mechanical mounting, power and pinout, feedback, control support, firmware and parameters their own rows. Each row should identify an original document, revision, confirming party and unresolved condition. A statement that two parts fit the same bracket does not answer whether the controller supports the feedback. Likewise, a matching protocol name does not establish the same object definitions or operating mode. Do not collapse the map into one green compatibility tick. A documented unknown is a more useful result than a substitute recommended without the evidence needed to inspect it.
Preserve the first deviation
For a drive, reducer or harness symptom, distinguish the reported state from the assembly eventually implicated. Record the original observation before clearing a state, editing a table or changing a component. Separate a direct export, an operator description and an analyst inference. All may be useful, but they do not have the same evidential role. The first visible alarm or implausible figure is a starting point, not automatically the cause. Keep enough preceding and following context that a reviewer can distinguish an initiating discrepancy from its later effects.
Build competing explanations with discriminating checks
For a drive, reducer or harness symptom, distinguish the reported state from the assembly eventually implicated. List the plausible branches and the evidence that would distinguish them. A check should answer a question, not simply add activity to a checklist. Change one authorised condition at a time where the task allows it; where several changes cannot be separated, state the limit on attribution. A successful outcome from a new combination does not retrospectively prove which previous element was responsible. Retain failed and interrupted attempts instead of only the final success.
Keep identity, time and transformation together
For a drive, reducer or harness symptom, distinguish the reported state from the assembly eventually implicated. Each observation should point to the relevant configuration, original source and time basis. Preserve the raw value or state and separately record any conversion, filtering or interpretation. If a clock, unit or version is unknown, describe how that limits the conclusion. Avoid filling a gap from a similar machine or a neighbouring row. A precise-looking reconstruction without evidence is less useful than a clearly bounded sequence of known observations.
Close the issue at the level observed
For a drive, reducer or harness symptom, distinguish the reported state from the assembly eventually implicated. The closure should distinguish the symptom, the explanation, the change made and the evidence collected afterwards. State the task or data range and the observation window. A condition not covered remains open, even when no discrepancy appeared in a shorter check. Assign the next action and the material needed to resolve it. This makes the record usable by another engineer or analyst without overstating permanence, coverage or certainty.
Apply the worksheet to this domain
Use the event record to align robot serial identity, original alarm timeline, approved replacement basis and observation window. Preserve the original state or value before interpreting it. Add a column for the competing explanation, the observation that could distinguish it and the authorised next check. Mark a change of configuration as a new comparison condition. An unresolved entry should identify the missing evidence and its effect on the conclusion, not simply say pending. The next reviewer can then continue a specific branch without repeating every earlier action.
Four routes and their tradeoffs
For exact part-number replacement, collect evidence for suffix, revision, provenance and permitted parameter handling. before closing a diagnostic branch. The route fits when the installed identity and replacement basis are already known. Retain this limit: A narrow scope still needs receipt and installed-task checks.
For documented alternative, collect evidence for mechanical, electrical, feedback and controller support separately. before closing a diagnostic branch. The route fits when an applicable maker or responsible party has confirmed the relationship. Retain this limit: More sourcing flexibility creates additional evidence to review.
For diagnosis before purchase, collect evidence for original logs, triggering task and authorised diagnostic route. before closing a diagnostic branch. The route fits when the symptom has not established the failed assembly. Retain this limit: Upfront investigation can reduce unnecessary component buying.
For interface and handover review, collect evidence for drawings, configuration, receiver inspection and unresolved conditions. before closing a diagnostic branch. The route fits when a proposed component exists but its evidence pack is incomplete. Retain this limit: A review clarifies scope but is not a supply or repair commitment.
Prepare the next review with original evidence
For the next diagnostic review, attach the first deviation and the original evidence, not only the final alarm or chart. State the leading explanations and the observation that could distinguish them. Ask for an authorised, task-specific next check and preserve its result even if it fails. The references below help interpret definitions; they do not establish this fault or discrepancy. Close only the branch actually supported, keeping the observation window and unresolved conditions explicit.
[S1] Harmonic Drive Engineering Data; [S2] UR-Series Service Manual.
Primary-source reading
Harmonic Drive Engineering Data
https://www.harmonicdrive.net/_hd/content/documents1/FBB_DifferentialGear.pdf
UR-Series Service Manual
https://www.universal-robots.com/manuals/EN/PDF/Shared/service-manual_ur-series/Service_Manual_UR-Series_en.pdf