Hybrid Artificial Intelligence for Diagnosis in Industry 4.0

International STIC-AmSud project focused on hybrid artificial intelligence methods for industrial diagnosis and predictive maintenance.

Overview

Hybrid Artificial Intelligence for Diagnosis in Industry 4.0 was developed within the framework of the STIC-AmSud international cooperation program involving France, Colombia, Peru and Ecuador.

The project focused on the development of advanced diagnostic methodologies combining model-based reasoning, machine learning, clustering algorithms and knowledge extraction techniques to improve the monitoring and maintenance of complex industrial systems.

The research was carried out within the DISCO team of LAAS-CNRS and explored the integration of hybrid artificial intelligence approaches capable of exploiting both physical knowledge and operational data.

Scientific Context

Modern industrial systems generate large volumes of heterogeneous data originating from sensors, control systems, maintenance records and operational databases.

Although data-driven approaches have achieved remarkable success in anomaly detection and predictive maintenance, many industrial applications still require the integration of physical knowledge and expert information to ensure robustness, interpretability and reliability.

The project addressed this challenge by investigating hybrid diagnostic frameworks capable of combining model-based and data-driven paradigms within a unified architecture.

Main Objectives

The project pursued several complementary scientific objectives:

  • Develop hybrid diagnostic methodologies.
  • Combine supervised and unsupervised learning techniques.
  • Improve fault detection capabilities in complex industrial systems.
  • Enhance predictive maintenance strategies.
  • Increase model interpretability.
  • Explore adaptive diagnostic frameworks for Industry 4.0 applications.
  • Improve the robustness of intelligent monitoring systems.

Methodology

The project relied on several complementary methodological axes.

Axis 1 – Hybrid Diagnostic Architectures

Development of architectures combining:

  • Model-based diagnosis.
  • Data-driven diagnosis.
  • Expert knowledge.
  • Machine learning techniques.
  • Statistical monitoring methods.

The objective was to exploit the strengths of both physical models and artificial intelligence algorithms.

Axis 2 – Intelligent Data Structuring

The project investigated methods for restructuring complex industrial datasets to facilitate fault detection and anomaly identification.

Topics included:

  • Feature extraction.
  • Dimensionality reduction.
  • Data clustering.
  • Similarity analysis.
  • Pattern discovery.

Axis 3 – Advanced Learning Algorithms

Particular attention was given to advanced machine learning algorithms developed within the research consortium.

Examples include:

  • DyClee.
  • LAMDA.
  • Hybrid clustering techniques.
  • Semi-supervised learning.
  • Adaptive classification methods.

Axis 4 – Explainable Industrial Diagnosis

The project explored strategies to improve the interpretability of diagnostic systems.

This work later contributed to the development of explainable AI research activities in renewable energy systems and photovoltaic diagnostics.

Research Contributions

My contributions focused on:

  • Analysis of hybrid diagnostic architectures.
  • Evaluation of advanced clustering techniques.
  • Study of DyClee and LAMDA approaches.
  • Application of hybrid AI methods to energy systems.
  • Development of fault detection strategies.
  • Analysis of complex operational datasets.

Scientific Impact

The project contributed to:

  • Hybrid artificial intelligence.
  • Predictive maintenance.
  • Industrial monitoring.
  • Explainable diagnostics.
  • Data-driven supervision.
  • Intelligent industrial systems.

Many concepts investigated during this project later influenced research activities related to photovoltaic fault diagnosis, explainable AI and smart maintenance.

International Collaboration

Participating institutions included:

  • LAAS-CNRS (France)
  • EAFIT University (Colombia)
  • Pontifical Catholic University of Peru (Peru)
  • National Polytechnic School (Ecuador)

Expected Outcomes

The project generated:

  • International scientific collaborations.
  • Novel hybrid diagnostic methodologies.
  • Publications on intelligent monitoring.
  • Advances in predictive maintenance.
  • New perspectives for Industry 4.0 applications.

Keywords

Industry 4.0 · Hybrid Artificial Intelligence · Predictive Maintenance · Fault Diagnosis · Machine Learning · Explainable AI · Clustering · DyClee · LAMDA · Industrial Monitoring · Data Fusion

Project Status

Completed

References