SYSE.15:5 - Worked Case: Methods for a Heat-Pump Controller Change
An engineering team is changing the controller of an occupied-building heat-pump plant. Its current process description points to a V diagram, a Systems Engineering handbook, a simulation repository, an issue tracker, a hardware-in-the-loop bench, an AI coding service, and a release checklist. The V diagram, handbook, and checklist are descriptions. The repository and issue tracker store or carry records. The bench is a platform System. The AI service may be an Agent, provider resource, or tool depending on the arrangement and the agency the service has. Calling this heterogeneous set our model-based process does not make it one Method.
The engineering lead acts as the Agent authorized to choose Methods for this controller project. Release authority remains with another Agent. The team starts from recurring results rather than the list of assets:
| Needed result | Candidate or retained Method | Evidence and current gap |
|---|---|---|
| Operating situations and protected room-temperature and equipment conditions | Recover operating situations, affected Systems, and required effects before choosing a controller concept. | Existing operating records cover normal weather; rare grid events remain an explicit gap. |
| Plant-response prediction | Identify physical parameters and simulate the heat-pump, building, sensor, and controller interaction. | Earlier predictions omitted measured sensor latency. |
| Controller implementation candidate | Implement the controller change from an existing template, or generate a bounded implementation candidate with AI. Independent implementation review remains a separate Method. | Use AI is not a Method. Recover the reusable implementation ways before comparing them; review supplies its own result. |
| Integration observation | Exercise the actual controller interface and plant model on the hardware-in-the-loop bench. | Bench time is scarce; occupied-building trials require separate safety and release conditions. |
| Configuration, assurance, and release decisions | Use the corresponding configuration, assurance, and release Methods for the identified controller and plant conditions. | None of these decisions follows from the V diagram or a green code check. |
The hardware-in-the-loop trial Method meets both experimental-engineering and cyber-physical-integration Method criteria. The team also uses it as a part of two stated whole Methods: controller integration and release-evidence development. Those classifications and whole–part relations are recorded separately. The modeling result informs the controller choice, but that result use does not make modeling a stage of one universal lifecycle.
In a representative three-day slice, an AI coding Agent performs bounded generation Work, a controls engineer reviews the candidate, and a test engineer performs the bench trial. If the coding service lacks enough agency to perform the assigned Work, it is recorded as a tool used by the controls engineer instead. The trial reveals sensor latency that destabilizes the candidate despite an acceptable simulation prediction. The release Agent withholds the candidate, and the controls engineer revises the physical model. Faster code generation did not shorten the limiting integration and assurance Work.
The next choice is how to produce a corrected implementation candidate for the revised latency model. The lead compares continuing with the team’s established implementation from a controller template and using bounded AI generation. Both must serve the same controller interfaces and required response, and both feed the same independent review and hardware-in-the-loop qualification. Neither option resolves scarce bench time.
For this illustrative comparison, assume a qualified controls engineer is available and the template’s applicability to this correction is supported. The team already knows how to use it. The AI alternative is also available, but requires preparing its context, examining unfamiliar generated constructs and adapting them to the controller interfaces. The lead’s qualitative estimate is that this additional preparation, review and integration work exceeds the generation effort saved for this correction.
The lead’s rule protects the required candidate quality and physical-evidence conditions, then prefers the option with lower total engineering effort for this correction. On the stated basis, the lead retains implementation from the existing template and defers AI generation. The accepted loss is slower initial code production in exchange for less preparation and checking of unfamiliar implementation choices. Independent review and latency-sensitive integration checks remain necessary for either option. A changed template, loss of the qualified engineer, or a better-supported AI alternative with lower full continuation cost reopens this choice. The corrected implementation still has to be produced and qualified.
The repertoire account keeps operating-situation recovery and linked concept development for their named results, carrying forward the rare-grid-event gap. Physical modeling and hardware-in-the-loop trial remain needed to address the exposed latency interaction. Configuration, assurance and release still supply their separate decisions. The account retains an occupied-building trial only under the stated safety, plant-state, weather and release conditions. The V diagram remains an overview. The phrase AI workflow Method remains unresolved; the identified bounded generation Method is a deferred alternative.
Scarce bench time leaves several possible repairs. The team can use OPS.11.1 to compare available time with
already assigned work and revise allocation under the existing authority. It can also compare trial Methods
that occupy the bench for less time while preserving the required integration result and evidence conditions.
A shorter trial that misses the sensor-latency failure would not serve the present use. New bench provision or a
different way of obtaining the result belongs to SYSE.12 or SYSE.24 when that contribution is needed. None of
these options is selected merely by observing scarcity.
Simultaneous modeling, generation, review, trial, and assurance conflicts go to SYSE.20. Evidence of wider use,
selection, retention, or loss can later inform SYSE.21; this project choice does not establish a cultural trend.
When the full pattern is unnecessary. If one identified Method already supplies a well-understood result for the same System and conditions, and no interaction, specialization, capability, provider, or currentness question can change the decision, use that Method and its direct evidence patterns.