How do you find a similar problem's fix once its engineer left?

What do you need before you search for a similar past problem?

Before you search for a similar past problem, you need a written description of the current case on fixed similarity axes, and a list of every place past cases may be recorded. The axes describe the case and the conditions a fix depends on: symptom, substrate, chemistry, concentration and equipment, plus temperature and water where the product is water-mixed. Write the current value for each axis, with its unit and the record or reading it came from. An axis you cannot fill yet is marked unknown, not guessed. Outcome is checked on each past case: whether its fix held and for how long.

Case-based reasoning describes the same first step: retrieval starts by identifying the relevant problem descriptors, then matching, searching and selecting the best case (Aamodt and Plaza, 1994). Choosing the descriptors before searching keeps the search from drifting toward whatever the first hit happens to share with the current case.

The list of record places matters as much as the axes when the engineer who solved the old case has left. Their knowledge now sits in work orders, service and visit reports, complaint files, lab reports, supplier correspondence, certificates of analysis and email. The cost of losing that knowledge is covered in when your best technical person leaves; the steps below are about recovering what was written down.

How do you find a similar problem step by step?

Finding a similar problem takes five actions, each done before the next:

1. Fix the axes. Write the current case on the 5 core axes (symptom, substrate, chemistry, concentration, equipment) and add any condition that changed on site, such as a new water source.

2. Search every record type by axis values, not by product name: the alloy, the machine or line ID, the water source, the symptom in two or three wordings ("milky", "cloudy", "emulsion split").

3. Search for failures on purpose: reopened work orders, returned product, complaints closed with no cause found, and trials that were stopped.

4. Compare the closest few candidates on the same axes in one table, marking each axis match, differ or unknown.

5. Judge fit with the check table in the next section, and write the verdict: reuse, adapt or no match.

Step 3 is easy to skip, because a search for "what worked" returns only closed successes. Step 4 keeps the comparison honest: when the candidates are laid out on the same axes, a case that matched on symptom but differed on substrate and equipment stops looking like a precedent. In case-based reasoning, reuse of a past case focuses on the differences between the past and current case and on what part of it can be transferred (Aamodt and Plaza, 1994).

How do you check that a similar past problem really matches?

You check that a similar past problem really matches by testing it axis by axis against the current case and treating any unrecorded axis as unknown rather than as a match. A past case can be reused only on the axes where it passes. The table below decides whether a found past case matches the current problem on the fixed axes:

The key row is outcome: a past case that does not say whether the fix held is a lead, not a precedent, however close it is on the other axes. A case that passes on all six checks can be reused with a confirming test. A case that fails on one axis can still give you the method, the sequence of checks that found the cause, while the action itself has to be adapted. Write the verdict for each candidate under the table, with the date and the records used, so the next reader can see why a case was reused, adapted or set aside.

Why do keyword and product-name searches miss similar problems?

Keyword and product-name searches miss similar problems because records are filed by what was sold or by the words one person used, while similarity lives in the conditions. The same coolant failure can sit under three product names after a supplier change, and the same symptom can be logged as "foam", "aeration" or "pump cavitation" by three technicians. A product-name search finds the cases where the product matched and skips the cases where only the conditions did.

Case-based reasoning draws the same line. Some systems retrieve past cases largely on superficial, syntactical similarities among problem descriptors, while others retrieve cases based on features with deeper, semantic similarities, which needs general domain knowledge to explain why two cases match (Aamodt and Plaza, 1994). A keyword search is the first kind. Searching by axis values moves it toward the second: the alloy, the water source, the machine ID and the measured concentration are features that explain a match.

Indexing helps when it is done at entry. NASA's Lessons Learned system indexes approved lessons first, so they can be retrieved by simple or complex search criteria (NASA, accessed 2026). A site that files each case by the same axes gets the same benefit at a smaller scale.

Why do failed cases disappear from the record?

Failed cases disappear from the record because closing a job documents what finally worked, while the attempts that did not work leave no required field. A work order closes with the last action taken. A trial that was stopped often ends in an email, not a report. When the engineer leaves, the failed attempts leave with them, and the next search finds a clean success that hides two earlier dead ends.

The loss is not only a small-site problem. A 2012 audit found that only 12 of 28 NASA project managers surveyed (43 percent) contributed lessons to the agency's Lessons Learned Information System (NASA OIG, 2012). NASA's own definition counts negative experience, such as a mishap or failure, as a lesson (NASA OIG, 2012). Case-based reasoning also keeps failures: failure information may be retained as separate failure cases, so a later failure can be recognized and corrected (Aamodt and Plaza, 1994).

To stop the loss, record a failed attempt with the same axes as a success, and give it an outcome: did not work, made it worse, or held for a stated time. Reliable recommends recording failure mode, failure cause, corrective action and recommended follow-up on every corrective work order (Reliable, accessed 2026). Those fields apply to an attempt that failed as much as to one that worked.

What do you do when no past problem matches?

When no past problem matches, write a no-match verdict, list the axes and records you checked, and work the current case from its own evidence. A no-match verdict is a finding, not a failure of the search: it stops a near miss from being reused as if it fitted. Keep the closest candidates with the axes on which they differ, because a partial match can still give you a method.

Then treat the case as new. IEC 62740:2015 describes the basic principles of root cause analysis and specifies the steps that a process for it should include, for events that have occurred (IEC 62740:2015). Take full readings on every axis, so the current case becomes the record the next person can find. For water-mixed fluids, OSHA's metalworking fluids manual recommends collecting observations, laboratory analyses and data on additions, tabulated so relationships and trends can be spotted (OSHA Metalworking Fluids Best Practices Manual, accessed 2026). File the result with the outcome once it is known, including whether the fix held.

When should you call a specialist about a similar past problem?

Call a specialist about a similar past problem when no case matches and the root cause analysis stalls, when the only close match differs on chemistry, or when the old fix would change a hazardous process. The product supplier's technical service can often tell you whether a past case on a former product applies to the current one, because the supplier knows what changed in the formulation. Bring the axis table, the candidate cases with their match marks, and the no-match verdict if you wrote one. Handle any retained samples according to the product's safety data sheet (SDS) and your site's chemical safety rules.

A specialist also helps when the record holds many partial matches and no one can say which axis matters most for this chemistry. That judgment comes from knowing the mechanism, and it is the part a departed engineer took with them. Lubinpla compares successful, failed and repeated cases on the same axes and says so when no case fits.