Distributed Abductive Reasoning for Self-Explaining Digital Twins

Edtlab

Nice

Sur place

EUR 18 000 - 30 000

Plein temps

14 jours+

Recevez plus de réponses des employeurs

Envoyez un CV adapté au poste en quelques minutes.

Avantages offerts par ce poste

Competitive PhD stipend

Social security and health coverage

Research travel support

Résumé du poste

Edtlab is seeking a motivated PhD candidate to join the Engineering Digital Twins program at Université Côte d’Azur. The position involves research on explaining discrepancies between digital models and real-world behaviors through abductive reasoning and developing distributed architectures.

The successful candidate will benefit from a collaborative research environment and opportunities for international partnerships in the realm of cyber-physical systems.

Qualifications

Master’s degree in a relevant field is required.
Strong background in reasoning systems or distributed computing.
Programming experience to implement algorithms.

Responsabilités

Conduct research on abductive reasoning for digital twins.
Develop architectures for explanation generation.
Collaborate with interdisciplinary researchers in AI.

Connaissances

Strong analytical and problem-solving skills

Good written and oral communication skills in English

Programming experience

Interest in cyber-physical systems and digital twin technologies

Formation

Master’s degree in Computer Science, Artificial Intelligence, or related field

Outils

Distributed systems

Abductive reasoning

Description du poste

We are seeking a motivated PhD candidate to contribute to the Engineering Digital Twins (EDT) program within Catalyst: the Reliable Hybrid Model Forge. The research focuses on developing explainable and distributed reasoning mechanisms for hybrid digital twins, with the goal of interpreting discrepancies between model predictions/simulation and real-world observations.

Digital Twins are increasingly used to design, operate, and maintain complex Cyber-Physical Systems (CPS). Modern implementations rely on hybrid models that combine physics-based models with data-driven components. This hybridization enables digital twins to operate in environments where physical models are incomplete, evolving, or partially unknown.

However, as digital twins interact with their physical counterparts, discrepancies inevitably arise between predicted and observed behaviors. This phenomenon, known as the reality gap, may result from model abstraction, sensor drift, environmental variability, or system degradation.

The goal of this PhD is to develop methods enabling digital twins to interpret and explain these discrepancies, transforming them from simple prediction errors into actionable knowledge about the system and its requirements.

Research Focus

Current approaches often treat discrepancies between digital twins and physical systems as numerical errors that must be minimized through reactive model adaptation. While effective in some cases, these methods present several limitations.

First, they rarely provide human-interpretable explanations for the causes of the observed mismatches. Second, reasoning processes are typically implemented in centralized architectures, preventing contextual reasoning near data sources. Third, adaptation mechanisms often ignore the broader system requirements, potentially improving local model accuracy at the expense of global objectives.

This PhD proposes to address these challenges by introducing abductive reasoning as a central mechanism for interpreting the reality gap in hybrid digital twins.

Unlike deduction (deriving predictions from physical models) or induction (learning models from data), abduction focuses on generating plausible explanations for unexpected observations. It is therefore well suited to identifying the causes of discrepancies between model predictions and real-world data.

Explanation Architecture

We will explore a hierarchical and distributed explanation architecture where:

Local explanations are generated at the edge, close to sensors and data sources
Intermediate reasoning layers at the fog level refine and combine explanations
Global reasoning at the cloud or system-level twin validates explanations and evaluates their impact on system requirements

This hierarchical process creates a traceable chain of explanations, linking low-level sensor anomalies to high-level system behaviors.

Research Directions

Abductive reasoning for hybrid digital twins

Designing reasoning mechanisms capable of explaining discrepancies between physics-based and data-driven models.

Distributed explanation architectures

Developing reasoning frameworks operating across edge–fog–cloud infrastructures.

Hierarchical explanation generation

Building multi-level explanation processes connecting local observations to system-level interpretations.

Requirement-aware reasoning

Ensuring that model corrections and explanations preserve global system requirements.

Ultimately, the goal is to transform digital twins from reactive error-correction systems into proactive, self-explaining companions capable of supporting trustworthy decision-making in complex cyber-physical environments.

Key Responsibilities

Conduct research on abductive reasoning methods for digital twins
Develop distributed architectures for explanation generation
Design algorithms linking local anomalies to global system interpretations
Implement and evaluate prototypes in cyber-physical system scenarios
Collaborate with interdisciplinary researchers in AI and digital twin engineering
Publish research results in international conferences and journals
Participate in EDT program meetings and collaborative activities

Research Environment

The PhD will be conducted at Université Côte d’Azur, within a research environment specializing in artificial intelligence, cyber-physical systems, and digital twin technologies.

The candidate will collaborate with researchers working on: