Logistics Intelligence Series | Module 2 of 6

Tyre-wear prediction:
Wear is predictable

Tyres are one of the worst-predicted cost blocks in any fleet. A gradient-boosting model infers from route profile, load and telematics which tyre will fall below the wear limit in the next 4 weeks. Before the workshop notices.

Model scenario - fictional data

This module is part of a six-part worked example about a fictional mid-sized logistics company (150 trucks, 43 regular customers). All figures, datasets, model metrics and euro amounts are fully simulated - they illustrate the methodology and the order of magnitude of the achievable effect. They are not results of real clients and not a performance promise. What values are achievable on your real data is established only in a pilot project.

01 The tyre problem - reactive instead of predictive

Why the current process systematically burns money

In most fleets tyre management works like this: the driver reports "tyre looks bad", the workshop measures at the next inspection, and the change happens - either too early (wasted tread) or too late (breakdown, fine, safety risk). Both cost money.

What most people don't see: your telematics system records per-trip data that directly influences tyre wear - speed profiles, braking frequency, load weights, route types. Combined with workshop measurements, this creates a picture no fleet manager can see manually. The cost basis of this model calculation: 3.4 cents per kilometre - the ATRI operating-cost anchor.

Raw telematics data
Route profile features
Wear model
Change prediction
Workshop planning
€340
Avg. cost per tyre change
12 St.
Avg. tyres/truck per year
150 LKW
Fleet size
€612k
Total tyre costs/year

02 Data foundation - what your systems already record

Synthetic data based on real telematics and workshop structures

We simulate 150 trucks × 10 tyre positions × 52 weeks - approximately 78,000 data points with wear progressions. Each tyre starts at 8mm tread depth (new) and wears according to a physics-based model.

▸ Example values - method illustration, not a performed analysis
Dataset: 78.000 Messwerte
Fahrzeuge: 150
1.800 Reifenwechsel/Jahr (Modellannahme: 12 je LKW)
KWLKWPositionProfilkm/WoBeladungBremsenProfiltiefeAlter
12LKW-007VLMischverkehr2.34019.2t1.0536.42 mm14 Wo
12LKW-007HL1Mischverkehr2.34019.2t1.0536.78 mm11 Wo
12LKW-023VRStadt-/Verteiler1.87014.6t1.6824.15 mm28 Wo
12LKW-089AL1Bergstrecken2.81022.4t1.2643.21 mm34 Wo
12LKW-142HR2Autobahn-dom.2.58020.1t7745.93 mm18 Wo

03 Exploratory analysis - what eats the tyres

Route profile, position and load tell the story

Wear rate by route profile (mm / 1,000 km)
↳ Insight

Urban/distribution traffic wears tyres 57% faster than motorway traffic. Everyone knows this intuitively - but the real lever is not brand choice, it is tyre pressure and deployment profile: per the ATA TMC anchor, 20% under-inflation costs around 30% of tyre life. Managing pressure and route profile together puts tyre changes where they actually occur.

Wear rate by tyre position
Tread depth progression - 3 example trucks over 52 weeks (position VL)
↳ Pattern Identified

LKW-023 (distribution) hits the 3mm limit after just 24 weeks, while LKW-142 (motorway) achieves 38 weeks. That is a 14-week difference - almost half a tyre lifetime. This knowledge exists in your data, but nobody is using it for workshop planning.

04 Feature engineering - making wear predictable

13 features from telematics + workshop data

▸ Example values - method illustration, not a performed analysis
Feature-Matrix: 13 Features
Positive Klasse (Wechsel nötig): 8.7%

The key feature is verschleiss_accel: it measures whether wear is accelerating. A tyre that suddenly wears faster often signals a mechanical problem - misalignment, defective shock absorber, or changed driving behaviour. No workshop technician catches this in a visual inspection.

05 Model - gradient boosting + survival analysis

Two models, one goal: when does the tyre need to come off?

We use a hybrid approach: an XGBoost classifier for the binary question "change needed in 4 weeks: yes/no?" and a survival model for the precise question "how many more weeks?"

0.85
Precision
0.74
Recall
0.79
F1-Score
0.94
AUC-ROC
↳ What the Numbers Mean

Precision 85%: when the model says "change needed", it is correct in 85 out of 100 cases. Your workshop receives few false alarms. Recall 74%: the model identifies 74% of all genuinely needed changes in advance. The remaining 26% are mostly tyres with atypical damage patterns (nail, kerb strike).

Feature importance - what drives the prediction?

Current tread depth and wear trend dominate - logical. Tyre brand appears as a feature in 5th place, but it remains an explanatory factor, not a quantified procurement lever: the effective lever lies in pressure management and deployment profile - that is where wear optimisation starts.

06 Live view - the digital tyre status of your fleet

What the dashboard for your fleet manager would look like

Based on the trained model we show the current status of all 1,500 tyres (150 trucks × 10 positions) and predictions for the next 4 weeks:

1.176
Tyres OK (>5mm)
214
Monitor (3-5mm)
83
Change in 4 wks.
27
Immediate change (<3mm)

Example: LKW-023 - tyre status detail

This distribution vehicle (urban/mixed traffic) shows the typical pattern: front axle wears significantly faster than the trailer.

VL - Vorne Links (Lenkachse)
4.15 mm
Prediction: 2.8 mm in 4 Wochen → Wechsel einplanen
VR - Vorne Rechts (Lenkachse)
4.52 mm
Prediction: 3.3 mm in 4 Wochen → Beobachten
HL1 - Hinten Links 1 (Antrieb)
5.89 mm
Prediction: 5.1 mm in 4 Wochen → OK
AL1 - Auflieger Links 1
6.21 mm
Prediction: 5.6 mm in 4 Wochen → OK
HR2 - Hinten Rechts 2 (Antrieb)
3.21 mm
Prediction: 1.9 mm in 4 Wochen → SOFORT-Wechsel
AR2 - Auflieger Rechts 2
7.12 mm
Prediction: 6.5 mm in 4 Wochen → OK
↳ Anomaly Detected

HR2 on LKW-023 wears 40% faster than HR1 - at identical mileage. The model flags this as an anomaly. Probable cause: rear-axle misalignment or defective right-side shock absorber. Without the model this only surfaces at the next technical inspection - or as a breakdown on the A2.

07 Business impact - the euro calculation

Predictive vs. reactive: what your workshop actually saves

Annual savings potential by category
€98.820
Total savings / year
16 %
Reduction in tyre costs
Savings categoryMechanismAmount/yearConfidence
Wear/pressure optimisation12% of tyre costs (assumption; ATA TMC: 20% under-inflation = -30% tyre life)€73.440Medium
Breakdown prevention10 breakdowns at €1,800 each, 85% predictively avoidable€15.300High
Workshop scheduling36 avoidable unplanned visits at €280 each€10.080High
↳ The Hidden Lever

The largest item is wear/pressure optimisation (€73,440) - 12% of tyre costs as an assumption, deliberately below the ATA TMC anchor: 20% under-inflation costs around 30% of tyre life. The lever is pressure and wear management by deployment profile, not brand choice. Breakdown prevention and workshop planning come on top.

08 Next steps in tyre management

From proof-of-concept to live integration

This model can be implemented on your real data. Data requirements are minimal:

① Data sources

Telematics export (km, GPS, braking) + workshop records (tread-depth measurements, tyre-change data). Minimum 6 months of history.

② Pilot group

30 trucks with different profiles. Test phase. Weekly predictions vs. workshop validation.

③ Rollout

Dashboard for fleet manager with traffic-light system. Automated workshop scheduling. Email alert on anomalies.

All 6 modules: AI in logistics