Keeping the Per-VIN Twin Useful When the Cloud Link Fails
A cloud-authoritative per-VIN twin is a defensible architecture when its outputs support workshop analysis and deferred maintenance decisions. Central execution preserves the full multi-body model, simplifies calibration governance and avoids duplicating numerical software inside the vehicle. We reject that option for a vehicle expected to continue producing qualified chassis diagnostics after losing connectivity.
The rejected option keeps fidelity in one place
Under the rejected design, the cloud owns both the calibrated vehicle model and the current diagnostic state, while the vehicle uploads observations and displays returned assessments. That arrangement makes computationally expensive suspension, tyre and compliance models practical without committing equivalent resources to every vehicle, and it keeps engineering changes under a single deployment authority. Its weakness is not simulation quality but the dependency between fresh evidence and a reachable service: after connectivity is lost, a previously valid assessment can remain visible even though the vehicle has entered an operating region that assessment never covered. Retaining the last answer preserves availability of a message, not validity of its physical interpretation. The decision would change if every consuming workflow could tolerate a diagnostic gap until reconnection and no onboard function required a current assessment, because the additional embedded model would then add lifecycle cost without satisfying a necessary availability requirement.
Validity determines availability
Accept a diagnostic output only while its required observations are no older than a proposed freshness limit of 500 ms and its operating conditions remain inside the validated model envelope; otherwise expose an explicit unavailable state instead of carrying the previous conclusion forward.
Keep the local model independently executable
The chosen architecture places a bounded diagnostic model and its accepted calibration inside the vehicle, while the cloud retains the high-fidelity multi-body twin used for parameter estimation, deeper investigation and candidate model evaluation. This local model is a separately validated executable projection of the physical model, with specified inputs, retained states and observable diagnostic targets, rather than an assumption that the full solver can simply run more slowly. Its accepted calibration package binds the VIN, vehicle configuration, parameter provenance, model revision and applicability envelope, so a physically meaningful estimate remains attributable to the chassis and software that produced it. Candidate packages are staged and checked before atomic activation, and the currently accepted package remains locally available without a remote entitlement check or a download at startup. When the cloud link disappears, local measurements continue driving the local estimator, while the interface publishes output validity separately from cloud connection status. The trade-off is explicit: local diagnostics retain only the capabilities demonstrated within that reduced model's envelope, and conclusions requiring the cloud solver become unavailable.
Reconnection must preserve physical meaning
The vehicle retains diagnostic events with their observation times, configuration identity and active model revision, allowing cloud analysis to distinguish evidence collected under different physical assumptions. After reconnection, delayed observations enter historical reconstruction rather than being injected as current measurements, and a newly computed cloud state does not overwrite the live local estimate without a validated state mapping and consistency check. Model updates need the same discipline: changing stiffness or damping parameters can change the interpretation of stored states and residuals, so an update may require explicit reinitialisation and temporary withdrawal of affected diagnostics. Before design freeze, acceptance testing should interrupt connectivity during representative manoeuvres and calibration transfers, then verify that eligible local outputs continue, stale conclusions disappear and reconnection introduces no unjustified state discontinuity. Automotive functional safety standards become relevant if these outputs influence a safety-related function, but continued diagnostic operation alone establishes no authority to command the chassis. Freeze the boundary with diagnostic outputs advisory; any later proposal to let the twin modify vehicle control must fund the additional validation and independent supervision that authority requires.