Axle Weight Distribution Estimator

Legal axle limits are enforced at the group level, not only as a single gross vehicle weight. A trailer can be under the GVW cap and still overload a steer, drive, or trailer axle group once freight is stacked toward one end of the deck. That mismatch is hard to see from a pallet list alone. Dispatchers and load planners usually discover it at a scale house, a roadside weigh station, or after a citation.

An axle weight distribution estimator closes that gap while the load plan is still editable. Machine learning takes the planned cargo (piece weights, positions, equipment type, and axle configuration) and predicts kilograms per axle group before departure. When a prediction exceeds a jurisdiction limit, the system raises a quality flag that names the axle group and the predicted weight. Incomplete piece or equipment weights produce no estimate rather than a false green light. The dispatcher still owns release; the model only surfaces risk early enough to re-stack or reassign freight.

This use case sits in the load stage of logistics operations. It complements spatial planning such as 3D load plan optimization, visual checks such as CV load completeness verification, rule checks such as hazmat segregation compliance, and paperwork such as bill of lading auto-draft. Weight legality is a distinct quality gate: space can look perfect while axle groups are already over limit.

Reliable axle estimates need more than a total shipment weight. At minimum the model should see:

• Equipment profile: tractor and trailer type, axle group layout, kingpin and fifth-wheel settings when they affect leverage, and any known empty or tare weights by axle group.
• Piece-level cargo data: mass for each handling unit, plus planned longitudinal and lateral placement on the deck (or equivalent bay/slot coordinates from the load plan).
• Operating assumptions: number of axles that will be on the ground, lift-axle state if applicable, and the regulatory template for the planned route (federal bridge formula style limits, state or provincial caps, or customer-specific stricter thresholds).

When piece weights are missing, estimated, or inconsistent with packing units, the correct behavior is an empty prediction set and a clear “insufficient weight data” condition. Guessing fills the screen with confidence the operation does not have. The same applies when axle configuration is unknown (wrong trailer type, missing lift-axle status, or mixed equipment not mapped in the master data). Empty output is a quality signal: do not treat silence as compliance.

Dispatchers benefit when the UI separates three states: estimate available and within limits, estimate available and over limit on named groups, and estimate blocked for incomplete inputs. Only the middle state should block or strongly warn on departure workflows, and even then human release remains possible after review.

The output that operations can trust is specific. Each flag should cite the axle group (for example steer, drive, trailer tandem, or trailer tridem) and the predicted kilograms for that group, optionally with the applicable limit and the overage. Vague alerts such as “weight risk” force staff back into spreadsheets. Named groups let the planner move heavy skids toward the opposite end of the trailer, split a shipment, or change equipment before the driver is staged.

A practical review loop looks like this:

1. Load plan commits piece positions and masses.
2. Estimator computes axle-group kilograms for the planned configuration.
3. Rules compare predictions to the selected legal or contractual limits.
4. Over-limit groups appear as quality flags with predicted kg and limit context.
5. Planner adjusts placement or composition; the estimate re-runs.
6. Dispatcher reviews remaining flags and releases or holds the load.

The model does not replace certified scales. It reduces the chance that the first hard measurement is already a violation. Where fleets already capture scale tickets or onboard weight sensors, those readings become training and calibration data so predictions track real equipment rather than generic textbook spreads.

Quality outcome for this use case means fewer illegal axle overages discovered after departure, and clearer remediation when risk appears. Success is measured in flagged overloads corrected before gate-out, not in model accuracy metrics alone. Accuracy matters only insofar as flags are precise enough that planners act on them.

Carriers and 3PLs already run overlapping systems. The estimator should plug into that stack rather than invent a parallel planning world.

Trimble ecosystems often hold dispatch, routing, and equipment context. Axle estimates belong next to load assignment and pre-departure checks so the person releasing the truck sees weight risk beside ETA and hours constraints.

Samsara and KeepTruckin (Motive) commonly surface vehicle telematics, ELD, and sometimes safety or compliance workflows. Predicted axle risk is not a substitute for onboard scales, but when sensor or scale data exists, pairing live readings with plan-based predictions helps spot plan-versus-reality drift (freight shifted, wrong trailer, lift axle up when the plan assumed down).

Paccurate and similar cartonization or 3D packing tools optimize cube and packing sequence. Axle distribution is the natural next constraint after volume: a dense, compact plan can still concentrate mass on one axle group. Feeding packing coordinates into the weight model (or feeding weight flags back into the packer as soft constraints) keeps spatial and mass planning in one loop instead of two handoffs.

Integration pattern that works: treat the estimator as a quality service on the load plan object. Packing, TMS, and telematics systems write plan and equipment state; the service returns axle-group predictions and flags; UI surfaces them in the same place planners already edit the load. Avoid a standalone portal that only compliance specialists open after the truck is sealed.

Jurisdiction rules change by corridor, and bridge formulas interact with axle spacing in ways that pure ML cannot invent from cargo lists alone. Keep hard limits in a maintained rules layer; use ML for the physics of how cargo mass maps onto axle groups given equipment geometry. That split keeps legal text auditable and predictions improvable without rewriting regulation tables every time the model retrains.

Also expect systematic blind spots: liquid surge, hanging meat, incomplete multi-stop unload sequences, and after-hours add-ons that never update the plan. Multi-stop routes should re-estimate after each planned drop if remaining cargo shifts the center of gravity. If the plan is not updated, flags will lag reality.

Governance should keep the dispatcher as final authority. Auto-blocking every soft overage can freeze yards when limits are conservative or data is noisy. Prefer: mandatory acknowledgment of named axle-group flags, optional hard block only for severe overages or regulated lanes, and an audit trail of predicted kg, limit used, and who released despite a flag. That record supports coaching and, when needed, shows that risk was visible before departure.

Training data hygiene matters. Prefer tickets tied to known load plans and axle configurations. Discard or down-weight events where the plan was edited after weighing or where equipment IDs do not match. Monitor false positives (flags that never confirm on scale) and false negatives (citations or scale failures with no prior flag) by lane and trailer type so the quality outcome stays honest.

Related load-stage work should stay linked in the same pre-departure checklist: spatial fit via 3D load plan optimization, presence of planned freight via CV load completeness verification, dangerous-goods placement via hazmat segregation compliance, and shipping documents via bill of lading auto-draft. Axle weight prediction is the mass-legality sibling of those checks. Together they reduce the class of problems that only appear after the truck is already committed to the road.