Data Platform

Unified data platform

scattered truck data → one governed golden source

Four sources flow through a Bronze → Silver → Gold medallion into curated entities. Surfaces marked “To confirm” are the open questions we close with you in the workshop.

Sources

Truck & context data feeds (toggle a card to mute its contribution)

Medallion pipeline

Raw → cleaned/time-synced → curated entities

Bronze
Raw
Rows ingested4,820,000
Raw size12.4 GB
Last runtoday 06:00
Data stored exactly as received.
Ingest raw sensor & telematics payloads with no transformation. Full fidelity, replayable.
Silver
Conformed
Conformed98.7%
Time-sync (µs NVH ↔ 1s telem.)Aligned
Duplicates removed2.1%
Cleaned, de-duplicated, aligned to one timeline.
Schema-normalise, dedupe, and align µs NVH with 1 s telematics onto a single trip timeline.
Gold
Curated
EntitiesVehicle · Trip · Road-Segment · Sensor-Reading · Route
Records128,400
Curated, use-case-ready entities.
Materialise business entities the use-cases consume directly (map, drive, review, KPIs).

Data quality

Trip-level checks

Completeness97.9%
Range checks passed99.2%
Dropouts0.8 %
Clock drift12 ms

Data acquisition · CTP

Transmitting vibration over the Connected Truck Platform

CTP feasibility: To confirm with TT/S
≈ 4–12 GBper truck / day
PayloadAll channels at source rate

Full-fidelity µs vibration streamed to the cloud. Maximum signal — nothing thrown away before analysis.

Trade-off: May exceed CTP payload / bandwidth limits.

Illustrative: 4 vibration channels × 3 axes × 1 kHz × 2 bytes/sample × 24 h.

Time synchronisation

Aligning two streams onto one trip timeline (Silver layer)

NVH vibrationµs-level
Telematics · CTP~1 s
Common timelineresampled ✓

Every telematics tick anchors a window of thousands of NVH samples. The real NVH rate is to confirm — it sizes the Bronze volume and the sync strategy.

Sensor placement

Where each candidate signal would physically sit on the Actros — the sprung/unsprung question, made visual

1
2
3
1Longitudinal-G (TCM-G)To confirm
Cab
2Vibration (NVH)To confirm
Chassis or axle — location TBC
3Vertical acceleration (Z) — axle candidateProposed
Axle / wheel hub

Marker 1 (cab) and marker 3 (axle/wheel) are the two ends of the sprung-vs-unsprung question — see the Sensor inventory footnote below for why that placement choice matters more than any other spec on this page.

Sensor inventory

Candidate road-quality signals on the Actros — spec confirmed in the workshop

SignalSensor / makeAxisSampling rateMountMount typeRange / unitStatus
Longitudinal-G (TCM-G)Torque Control Module · G-sensorX (long.)To confirmCabSprung± g · to confirmTo confirm
Vibration (NVH)Tri-axial accelerometer · SiemensX / Y / Zµs-level · to confirmChassis or axle — location TBCTo confirmg · to confirmTo confirm
Vertical acceleration (Z) — axle candidateTri-axial accelerometer (proposed)Z (vert.)≥100 Hz · to confirmAxle / wheel hubUnsprung± g · to confirmProposed
Axle displacementSuspension displacement sensorZ (vert.)To confirmPer axleUnsprungmmTo confirm
Wheel speedABS / ESP wheel-speed sensor~1 Hz (CTP)Per wheelkm/hConfirmed
Vehicle speed · RPMPowertrain telematics~1 Hz (CTP)ECUkm/h · rpmConfirmed
GPS positionTelematics GNSSlat / lng~1 Hz (CTP)Telematics unitdegreesConfirmed
OxTS GNSS/INS (VOLTS)OxTS unit — AccelX/Y/Z, IsoIsVerticalAcceleration (ISO 8855)X / Y / Z + vertical (derived)~10 Hz · sporadic in catalogueVehicle body (test rig) — exact location TBCSprungg · scaling to confirmTo confirm

Sprung vs. unsprung is the biggest open architecture question here: a sprung-mass sensor (cab/chassis, e.g. today's longitudinal-G TCM-G) only sees vibration after the suspension has filtered it; an unsprung-mass sensor (axle/wheel hub) sees the road directly. Confirm which mount points actually exist on the Actros before committing to a signal.

Ground truth & validation

How we know a grade is correct — the open item this workshop must close

BSCB manual scoring
Domain experts ride and judge a segment by feel — today's baseline method, and the definition RQM has to match or beat.
To confirm
Laser profilometer IRI
A dedicated survey vehicle measures true IRI (m/km) as an instrument-grade reference for calibration.
To confirm
Repeat-pass consistency
The same segment driven multiple times, loads and speeds — the model should agree with itself before it's asked to agree with a human.
Proposed
Avg. RQI spread across repeat passes
± 4.1

Illustrative — computed once real repeat-pass data exists (see Segment.passCount / rqiStdDev).

No labelled dataset yet

No finalised labelled dataset yet. Validation depends on BSCB expert review plus a ground-truth deep-dive — this section is exactly the open item that review is meant to close.

Use case 02 · WarrantyIllustrative · mock

Predict part wear from the vibration you’re already collecting

The same axle vibration that grades the road also stresses the truck. One signal, read twice — how rough the road is, and which parts it’s wearing out.

Input · axle vibration
Axle RMS
0.34g
Road-residual
+0.18g
model predicts
Predicted component wear
Drive-axle bearingSuspensionCab mountsSteer-axle bearingChassis frame
Component health
GoodFairPoorSevere

Warranty reads the vibration the road doesn’t explain. Residual = measured axle vibration − what RQM's road model predicts for this road & speed. Above its own road-normalised baseline, that residual is a signal about the truck, not the road.

Component breakdown

Each component, the vibration feature that wears it, and its predicted health

Drive-axle wheel bearingPoor
High-freq band energy · 2–5 kHz
Bearing spalling / raceway wear
Predicted wear78%
Est. remaining ~35,000 kmProposed
Rear suspension (springs & dampers)Fair
Cumulative shock dose · g·s
Spring fatigue / damper fade
Predicted wear54%
Est. remaining ~90,000 kmProposed
Cab mountsFair
Sustained low-freq RMS
Elastomer hardening / creep
Predicted wear41%
Est. remaining ~140,000 kmProposed
Steer-axle wheel bearingGood
High-freq band energy · 2–5 kHz
Bearing wear
Predicted wear22%
Est. remaining ~210,000 kmProposed
Chassis frameGood
Peak shock count · > 2 g
Weld / rail fatigue cracking
Predicted wear15%
Est. remaining > 400,000 kmTo confirm

Warranty ontology extension

How the warranty domain hangs off RQM's Gold layer — the RQM output is the bridge

Segment-Assessment
RQM output · road-normalised
aggregates
Road-Wear-Index
cumulative exposure per vehicle
exposure of
Part
one per installed unit
fails
Failure-Event
one per component failure
triggers
Warranty-Claim
one per filed claim
covered byPart Warranty-Policy · coverage per part type
From RQM (bridge)New · warranty use case

What we still need

What warranty needs beyond the RQM sensor data — the open items for this use case

Failure & claims historyTo confirm

Which part failed, on which VIN, at what mileage — the warranty ground truth. A new data source (Daimler's claims system), not sensor data.

Parts / BOM master dataTo confirm

Maps a vibration channel to a specific physical component instance and its install date — asset data not in VOLTS or CTP today.

Diagnostic sampling rateTo confirm

Bearing-fault frequencies can sit above the rate spec'd for road roughness — same axle mount, possibly a higher rate.

Lifetime exposure aggregationProposed

A cumulative per-vehicle wear signal rolled up over the truck's life — a different feature-store pattern to RQM's per-trip view.