Chercheur Doctorant H/F

Abstract

The aim of this PhD is to develop models and methods that integrate driver behavior and user profiles into charging scheduling and the intelligent management of electric-vehicle batteries. Despite significant progress in battery management systems (BMS), most current approaches treat drivers as homogeneous users and rely on predefined charging strategies that neglect behavioral variability. Yet empirical evidence shows that differences in driving style, charging frequency, and thermal sensitivity can lead to substantial divergence in battery degradation rates.

This research will analyze and model behavioral factors (driving habits, charging patterns, personal preferences) that influence battery lifetime and performance. Building on these insights, it will design clustering, stochastic modeling, and machine-learning methods to characterize drivers and predict their impact on battery state-of-health (SOH). The resulting behavioral models will inform multi-objective scheduling and control algorithms that personalize BMS parameters (charge/discharge cycles, thermal management, charging strategies) to jointly optimize SOH and user satisfaction. These solutions will be embedded in a real-time feedback loop, connected to the SHERPA-LAMIH simulator and the Dunasys Box, to deliver tailored recommendations and validate impacts on both user experience and battery durability.

This PhD lies at the intersection of optimization, AI (federated learning, reinforcement learning, metaheuristics), and behavioral sciences, and contributes to the objectives of BATTL-EU by proposing a reproducible methodology to extend battery life while improving user experience. The PhD is conducted within the BATTL-EU (ANR PRCE) project on the battery passport for electric vehicles, which combines AI, blockchain, and federated learning to ensure data traceability, privacy preservation, and improved lifecycle management, in collaboration with CESI (LINEACT), Dunasys, and Université de Valenciennes (LAMIH) and aligned with EU sustainable-mobility and circular-economy goals.

Scientific context

The electrification of mobility raises new requirements for battery lifecycle modeling that couple electro-thermal/aging dynamics with real-world usage variability and traceability constraints. Market growth and EU circular-economy ambitions make durability, second-life readiness, and trustworthy data a priority, motivating architectures that capture degradation drivers across the full lifecycle and ensure transparent tracking (battery passport). These policy and market drivers frame a technical need to embed user-induced variability directly into battery models and downstream decision-making, rather than relying on stylized duty cycles.

On the modeling side, recent health-estimation (State of health SoH) and RUL approaches move beyond static parameterizations toward sequence-learning and hybrid (physics-informed/data-driven) predictors that can encode operational history and context. For instance, recurrent generative models (e.g., VRNN) have been explored for RUL estimation under realistic usage variability, improving short-term prediction and capturing uncertainty. Such models provide a basis for incorporating exogenous behavior features (temperature exposure, C-rate patterns, dwell/soak times) that modulate lithium inventory loss and impedance growth trajectories over time.

To operationalize behavior, user/driver profiling pipelines extract features from telematics and charging logs, then apply unsupervised clustering or sequential modeling to derive representative archetypes. Public resources such as UAH-DriveSet and datasets capturing aggressive driving support feature design, benchmarking, and profile validation. Complementary driver-assistance studies demonstrate that control policies adapted to driving style and driver state can measurably alter vehicle-level dynamics—evidence that behavior-aware adaptation is both detectable and impactful in practice.

Given behavior-enriched models, scheduling and control naturally become multi-objective: maximizing SOH/RUL and energy efficiency while minimizing user disutility (e.g., time, inconvenience) and operational costs. This calls for optimization (including multi-objective/metaheuristic and learning-based control) that personalizes charge/discharge shaping, thermal set-points, and time-of-use strategies. Edge/cloud patterns for privacy-preserving analytics, notably federated learning for distributed model updates and blockchain for accountable passport records, enable fleet-wide learning without centralizing raw user data. Together, these elements outline a behavior-aware, closed-loop BMS paradigm in which user profiles inform predictive aging models and multi-objective schedulers, while secure, distributed data infrastructure sustains adaptation over time.

Subject

This PhD project aims to develop a behavior-aware battery management system by linking driver behavior profiles to adaptive, multi-objective charging plans. The research will create actionable user profiles from telematics and charging data, and a scheduler that customizes Battery Management System (BMS) settings. This system will also quantify in real-time how driving and charging choices affect battery State-of-Health (SOH).

The project primarily involves behavior and profiling development and optimization and control allowing the improvement of the BMS. It will interface closely with SOH-modeling (using predictors and feeding behavior covariates), data-collection (feature design and quality checks), and safety/trust activities (constraints, monitors, fallback policies, and explainability).

The research will be guided by the following questions, which structure the scientific inquiry and define the evidence to be gathered:

• RQ1: How can stable, privacy-preserving driver profiles (e.g., aggressiveness, trip patterns, charging habits, delay tolerance) be extracted to generalize across contexts and remain actionable for control?

• RQ2: How should these profiles parameterize prediction models and scheduling costs to make the SOH impacts of behavior explicit and quantifiable in real-time?

• RQ3: Which multi-objective charging/thermal scheduling strategies optimally balance SOH preservation, user satisfaction (time, convenience), and energy/cost, and how can they be adapted online to changes in profile and context?

• RQ4: What safety constraints, monitors, and fallback modes are necessary to ensure battery/thermal safety and user trust when deploying behavior-aware control?

• RQ5: How can the causal link between driving/charging behavior and SOH be effectively communicated via Human-Machine Interface (HMI) to promote interpretable recommendations and behavior change?

To address these questions, the PhD pursues the following objectives, each mapped to WP2/WP4 deliverables and integration milestones:

• O1: Design a profiling pipeline (feature engineering, clustering/sequence modeling) to generate portable driver archetypes with confidence measures (WP2).

• O2: Develop a real-time SOH-impact estimator, conditioned on profile and context, to reveal the marginal effects of behavior on degradation (WP2 ↔ SOH).

• O3: Create a behavior-aware, multi-objective scheduler (using Pareto optimization, scalarization, or Reinforcement Learning) that personalizes charging/thermal set-points while enforcing safety constraints (WP4).

• O4: Integrate and demonstrate the system in a closed loop (SHERPA-LAMIH + Dunasys Box) with live dashboards displaying instant and cumulative SOH impact and user-cost metrics (WP4).

• O5: Provide privacy-respecting learning mechanisms (e.g., federated updates) and explainable recommendations to support adoption (WP2 ↔ WP4 ↔ Safety/Data).

This PhD consumes the SOH/RUL predictors developed by the “SOH-modelling” PhD and, in return, supplies behaviour-conditioned covariates and stress proxies that improve model fidelity under heterogeneous usage. It co-designs acquisition protocols with the “data-collection” PhD to ensure that telemetry and charging logs expose the features needed for robust profiling, and it integrates runtime monitors and fallback strategies from the “safety” PhD to guarantee safe deployment of personalised schedules. Collectively, these interfaces operationalise BATTL-EU’s goals: enabling a user-aware, privacy-preserving battery-passport ecosystem that quantifies behaviour–SOH causality, delivers personalised charging policies with auditable safety and traceability, and demonstrates measurable gains in durability and user experience at the edge (Dunasys Box) and in simulation (SHERPA-LAMIH).

Are you the talent we are looking for?

• A Master’s degree in industrial engineering, computer science, or operations research.

Scientific and technical skills:

• Modeling, simulation, and optimization

• Strong understanding and development level in Python and C++

• Solid knowledge of machine learning methods

• Knowledge of stochastic modeling

• Report and scientific article writing; good communication skills

• (English: minimum B1 required; B2 preferred)

Soft skills: 

• Demonstrated autonomy, initiative, and intellectual curiosity.

• Strong teamwork and collaboration skills.

• Proficiency in both English and French.

To convince you a little more:

• CDD 36 mois

• 6 semaines de congés payés (au prorata du temps travaillé)

• 14 RTT (au prorata du temps travaillé)

• Tickets restaurant

• Mutuelle entreprise

• Prime participation/intéressement

• Charte du télétravail

• Ordinateur portable

If this profile suits you and you share CESI's values.

Don’t hesitate any longer and apply with us!