# Vehicle Power Signatures in E/C Driving Signature

Topic radar · 43 documents · 2021-01-01–2026-06-07 · updated 2026-10-23

Canonical: https://slate.greyb.com/radar/vehicle-power-signatures-in-ec-driving-signature/innovations

## In one minute

### The route-to-cell calculation ships as separate programs, and the joins are an interface nobody owns
The most complete chain in this work runs from a recorded path to the power a single cell sees. It is assembled from parts built by different groups, each of which stops at a different place.

The Karlsruhe generator produces a speed profile and a slope profile for a predefined trip starting at a stated time, and says plainly that an EV simulator is required to turn those into a current or power profile and is not included in the package [1]. The Belfort-Montbéliard model begins where that ends: GPX files from round trips recorded with a Renault Zoe ZE40 and a 1 Hz GPS watch, run through the fundamental principle of dynamics, output as the power profile seen by a battery cell [2]. emobpy starts further back still, synthesising time series from empirical mobility statistics rather than from any vehicle's trace [3].

So the vehicle conversion step — mass, drag, drivetrain efficiency, auxiliaries — is the module each of these hands off. It is where the numbers diverge, and it is the piece with no reference implementation behind it.

1. [10.5281/zenodo.8205773](https://doi.org/10.5281/zenodo.8205773)
2. [10.25666/dataubfc-2022-04-08](https://doi.org/10.25666/dataubfc-2022-04-08)
3. [nature.com](https://www.nature.com/articles/s41597-021-00932-9)

### The electric-versus-combustion choice is merchandised as four variants of one car
BMW's current 7 Series is one body sold as a mild-hybrid six, a mild-hybrid V8, a battery-electric sedan and a plug-in hybrid, launched together and priced as a ladder. That is how the comparison actually reaches a buyer: not as two powertrains to be measured against each other, but as trim levels of one platform sharing tooling and development.


### Durability is being engineered to a rule that arrives next November
The most explicit commitment in this material is a supplier building a method to a deadline. Magna's complete-vehicle paper says plainly that legal minimums for high-voltage battery warranty turn lifetime into a whole-vehicle development target, then sets out how it will be developed: a representative monthly power and temperature profile generated from user routine, ambient temperature and charging behaviour, pushed through energy-flow and thermal simulation, then through a generic cell aging model to a state of health over life, with sensitivity studies on charging, climate and operating strategy.


### The route is the dominant term, which is why a vehicle-normalised signature is noisy
A simulation sweep across driving cycle, ambient temperature, load and initial state of charge found the route taken to have the highest impact on energy consumption and depth of discharge. A separate framework built on a three-dimensional Markov chain with machine learning, validated on Tianjin and Xining, puts the electrification benefit across those two cities between 16.4% and 77.8%.


### Battery health is now a number someone will insure, and it rests on a synthesised routine
The supplier-side paper in this work frames lifetime as a whole-vehicle development target forced by legal warranty minimums, and reaches its state-of-health prediction through a representative monthly power and temperature profile pushed through energy-flow simulation and a generic aging model.


### The tooling is the commitment: a generator given away, an optimiser bought
The upstream link is being released as a commons. Karlsruhe published a driving-profile generator in 2023 and again in late 2024, both times stating that an EV simulator is required to reach a current or power profile and is not in the package. emobpy turns mobility statistics into vehicle time series as an open tool, and Belfort-Montbéliard publishes an end-to-end model from GPX traces to cell power. A second release of the same generator eighteen months on is the maintenance behaviour of a group that expects others to build on it.


### The cell end is worked from both directions and the two do not meet
One dataset ages ten commercial INR18650 cells through an extended characterisation procedure: capacity, resistance, charging curves at several rates, and a WLTP-derived power profile replayed at different states of health. Measured forwards, that is what a drive profile does to a cell as it degrades.


### Driver behaviour is sold as event counts, and the automaker holds the pipe
The review of driver identification in electric vehicles concludes that on-board sensor data collected in natural driving is the objective, accurate source for the task, and that the trend runs through deep hybrid models rather than any single channel. What the insurance market does with that is narrower: mileage, time of day, hard braking, hard acceleration and phone handling, priced into premiums.


## Overview


The phrase in the title describes two quantities that differ by six orders of magnitude in timescale, and the split runs through everything below it. One is the voltage sag a processor leaves behind as an instruction executes, sampled at millions of readings a second. The other is the energy a whole vehicle draws over a drive of tens of minutes. Security work sits entirely on the first and drive-type comparison entirely on the second; nothing here tests whether a method built on one transfers to the other. Read carefully, the field has a name in common and no shared object.


## Research

### A bench can produce the power trace instead of observing it
At VNIT Nagpur a three-phase induction motor under current-sensorless indirect vector control is loaded against a dynamometer, with the resistive forces of a driving cycle emulated from equivalent rotational inertia over ECE R15 and NYCC. The stated output is the DC-link power profile for any driving scenario, from a battery-plus-ultracapacitor source.


### A recorded power profile is also being used to repair a test that does not conform
One document inverts the usual relationship. Traction-energy measurement for consumption tests, including compliance with SORT and E-SORT, is sensitive to the choice of road trace, and a trace long enough for SORT 3 is hard to make suitable for shorter tests as well. The remedy described is a correction of the trace using the electrical energy profile computed from the recorded power profile.


### Two different quantities share the name, and they differ by orders of magnitude in timescale
PHIDIAS fingerprints the voltage rail of an automotive microcontroller. It samples at 9.8 MS/s from the analog-to-digital converters of an Infineon AURIX TriCore, trains classifiers on normal and abnormal software behaviour, reaches 99.9% test accuracy with deep learning and 99.67% with gradient boosting, drives false positives to 6e-5 with availability-oriented training, and fits the whole pipeline into 382.58 kB of ROM and 16.83 kB of RAM. The mechanism is causal and short: an instruction executes, a current flows, a voltage sags.


### The hybrid split is a control law, so the DC-link trace becomes a software output
Sharjah's measurement on the New York City Cycle shows what the fast part of an urban load does to a battery bank, and proposes letting supercapacitors carry the motor through the accelerations while the battery is restricted to recharging them. The control work then specifies how the split is decided: an adaptive strategy reading the power profile and the two sources' dynamics, a non-linear state feedback controller with a disturbance observer on the traction motor, and a non-inverted buck-boost H-bridge for braking-energy recovery. Against a PI baseline the reported steady-state error falls from 0.43 to 0.0014 and peak overshoot from 12.07% to 3.51%.


### A charging curve carries manufacturer policy as well as cell physics
Idaho National Laboratory's model of session power recognises two distinct causes of a charge falling below the rated value: battery temperature away from the ideal, and manufacturer logic that throttles power in the cold or at high state of charge. The model is built from the Next-Gen Profile session data, and its purpose is grid load prediction.


### At fleet scale the modelled quantity is availability, not the trace
Multi-year measurements from 22 commercial EVs, combined with driving logbooks from more than 460 vehicles across 14 economic sectors, show pool power to be reliable during known idle times, with the uncertainty lowest exactly at the hours when the pool has most to offer; on the German frequency containment reserve market the estimate is €450 to €750 per vehicle per year for mid-2020 to early 2022.


### The drive cycle is crossing from output to input
A Technical University of Munich group argues that as battery-electric portfolios widen, each powertrain has to be dimensioned for its own vehicle concept and boundary conditions, with little carryover between models. Vitesco's team does the same one level down, sizing traction motor drives against recorded drive-cycle data with data-driven methods, motivated by development windows shorter than those for combustion engines and by tightening Department of Energy targets. AISIN, working with ESTECO, applies design optimisation to existing electric powertrain units and reports significant cost savings, without a number.


### Cross-powertrain claims should rest on the campaigns, not the headline efficiency number
The most quoted figure in this literature, a 30% energy-efficiency improvement, comes from dynamic skip-fire engine control combined with predictive eco-driving and powertrain-automation coordination, validated on a test track. Skip-fire is an engine actuator. No battery-electric vehicle has its equivalent, so the number describes a combustion powertrain and does not transfer.


## Market

### The only power figure settled in cash is the kWh at the charger, measured away from the vehicle
A charging session's power falls below the rated value for two distinct reasons — battery temperature away from the ideal, and manufacturer logic that throttles in the cold or near full charge — and the model that separates them is built for grid load prediction.


### The load-splitting argument gets paid where the duty cycle is known in advance
The clearest commercial result in this work comes from a vessel: an icebreaker simulation finds up to 16.3% fuel saving from a 2.5 MWh battery, with the saving driven by the shape of the load profile rather than by the energy stored.


### What is public is the method, and what is sold is the certificate
Across all of this, the pattern is consistent. The engineering that turns a recorded trace into a claim — a consumption estimate, a health figure, a driving score — is published in enough detail to be argued with, and the commercially valuable output is a number with someone's name behind it. The warranty figure comes from a synthesised routine, the driving score from event counts, the charging cost from a meter.


## Strategy

### A tier-one has built power-rail detection into an ECU, and left enrollment open
Bosch's PHIDIAS is the most industrially serious document in the security part of this work. It runs the full detection pipeline on an Infineon AURIX TriCore, sampling at 9.8 MS/s through the microcontroller's own converters, at 99.9% accuracy, in 382.58 kB of ROM and 16.83 kB of RAM. That is a detector characterised on a production-class part by a supplier that sells the part's surroundings, which is the shape of a feature rather than a study.


### The datasets being funded are operational, and identification will arrive as a by-product
Every large collection visible here is funded to forecast something other than the vehicle. Idaho National Laboratory's model separates battery-temperature effects from manufacturer throttling so that grid load can be predicted. RWTH and Jülich use multi-year measurements from 22 commercial vehicles and logbooks from more than 460 to price reserve capacity at €450 to €750 per vehicle per year.


### The supercapacitor line stops at simulation, while the same logic keeps going at sea
A line of work ending is worth naming. The UAE and Malaysia hybrid paper closes by saying its solutions will be implemented in a prototype EV system in future work, and nothing later in this material on a road vehicle follows it. The bench that came closest is a dynamometer built to produce a DC-link profile rather than observe one.


## Areas

- Vehicle Energy Consumption & Eco-Driving · 17 documents
- Energy Storage & Hybridization · 7 documents
- Power Electronics & Drivetrain · 5 documents
- Connected & Grid-Integrated Energy · 4 documents
- Simulation & Testing Infrastructure · 4 documents
- Security & Intrusion Detection · 3 documents