International audience

The growing demand for lithium-ion batteries (LIBs) in industries such as electric vehicles (EVs) and renewable energy storage underscores the need for tools that ensure transparent, sustainable, and compliant production processes. This paper presents the Battery Production Ontology (BPO), a comprehensive framework designed to standardize the representation and traceability of LIB production lifecycles. By modeling key aspects such as material flows, energy consumption, carbon emissions, and production activities, the BPO supports environmental impact assessments, supply chain transparency, and process optimization. The BPO is aligned with existing standards, including the EU’s digital battery passport requirements, ensuring interoperability across diverse systems. Developed using a structured methodology, the ontology underwent rigorous validation. Real-world case studies demonstrated its capacity to model emissions, trace materials, and represent production sequences, while quantitative assessments confirmed its scalability, reasoning efficiency, and accuracy for industrial applications. Additionally, the ontology integrates seamlessly with external standards like BONSAI, FOAF, Schema, and OWL Time, fostering semantic reuse and interoperability. By addressing the critical need for transparency and sustainability in LIB production, the BPO provides stakeholders with a robust tool to drive the green energy transition and achieve global sustainability goals. • A novel ontology models the lifecycle of lithium-ion battery (LIB) production. • Enhances traceability, sustainability, and compliance with EU battery regulations. • Integrates material flows, energy use, and emissions for lifecycle assessment (LCA). • Validated through case studies and quantitative performance evaluations. • Aligns with BONSAI, FOAF, Schema, and the EU Digital Battery Passport standards.

Clustering time series reveals valuable insights into hidden patterns about evolving particular phenomena over time. While sharp clustering operates rigidly (each time series is assigned to one cluster), soft clustering offers a more realistic and informative representation of data points, which is relevant according to the complex and uncertain nature of the time series data. This paper focuses on evidential clustering techniques based on the Dynamic Time Warping (DTW) distance due to its efficiency in measuring dissimilarity between time series. We conducted several experiments on various datasets, including a real-world battery aging dataset. The results showed interesting performance of the Evidential C-Medoids algorithm based on DTW in comparison with several other clustering techniques both on public time series data sets and on aging battery data set.

Monitoring complex systems is essential for preventing failures and ensuring operational safety. This paper proposes a method that combines ontology which provide structured knowledge representation with stream reasoning to enhance anomaly detection in dynamic environments. Unlike monitoring systems, our approach focuses on interpretable and scalable analysis of continuous data streams, enabling systematic identification of deviations from expected behavior. We demonstrate the applicability of this framework in monitoring lithium-ion batteries, where early detection of thermal anomalies is critical. By integrating a knowledge-driven model with data stream analysis, our method improves the reliability and safety of complex systems while offering explainable insights into detected anomalies.

Conditional random fields (CRF) are widely used for sequence labeling such as Named Entity Recognition (NER) problems. Most CRFs, in Natural Language Processing (NLP) tasks, model the dependencies between predicted labels without any consideration for the syntactic specificity of the document. Unfortunately, these approaches are not flexible enough to consider grammatically rich documents like patents. Additionally, the position and the grammatical class of the words may influence the text’s understanding. Therefore, in this paper, we introduce SynCRF which considers grammatical information to compute pairwise potentials. Syn-CRF is applied to TRIZ (Theory of Inventive Problem Solving), which offers a comprehensive set of tools to analyze and solve problems. TRIZ aims to provide users with inventive solutions given technical contradiction parameters. SynCRF is applied to mine these parameters from patent documents. Experiments on a labeled real-world dataset of patents show that SynC RF outperforms state-of-the-art and baseline approaches.

Explainability is a fundamental principle in the field of Artificial Intelligence (AI), ensuring that AI models and systems are understandable and transparent to end-users. Specifically, it tackles the challenge of providing explanations for AI predictions. In interpretable machine learning, classification rules are regarded one of the most well-known explainability techniques, due to their expressive power and transparent structure. In this paper, we first show that computing classification rules is equivalent to mining disjunctive patterns from the corresponding transaction database. Second, we show that our approach provides a clear characterization of optimal classification rules, wherein disjunctive patterns satisfy the non-redundancy property in the target class and such patterns correspond to minimal generators in this class. Then, we propose a SAT-based solution of the problem for computing optimal classification rules using MaxSAT solvers, for which the optimality is a balancing between the accuracy and the size of the rules. Finally, we present an empirical evaluation on several representative datasets, showing that our approach achieves good performance in terms of accuracy and interpretability compared to existing baselines.

In today’s energy-dependent world, ensuring the safety and efficiency of lithium-ion batteries is crucial. Early representation of anomalies becomes essential for optimizing performance, reducing disruptions, and prolonging battery lifetime in electric vehicle applications. This objective necessitates the integration of data from distributed and heterogeneous sources, a challenge traditionally tackled by semantic web technologies. In response, this paper introduces an ontology-based model that capitalizes on representing anomalies in lithium-ion batteries. Ontologies play a vital role in representing knowledge in a machine-interpretable format. Our approach enriches sensor data with contextual information, employing structured concepts, rules, and semantics specifically designed for representing anomalies in lithium-ion batteries.

A critical aspect of managing lithium-ion battery packs in electric vehicle applications is accurately determining the State of Charge (SoC). There are several methods available to estimate it, including coulomb counting with direct evaluation, Open circuit voltage, kalman filter with adaptive approach, particle flter, as well as fuzzy logic and data-based approach. In this paper, we use the state of charge data already computed by a data-driven approach and combine it with an ontology of a battery pack. The built ontology models the battery pack, taking into account the topology, types of cells and their organization inside. To make an exact estimation, different strategies of balance control of the cells are considered. SWRL rules are used to compute the state of charge of the whole battery pack. Matlab Simulink multi-physics model of a lithium-ion battery is used to provide simulated data for the experiments. The given model is evaluated based on regression metrics showing its performance.

This paper presents a new approach for solving the electrical load sharing problem generally known in energy circles as the Duck Curve problem. The Duck Curve problem is a curve showing the difference between the total electrical load a utility serves to its consumers (energy from thermal power plants), and what that load looks like after wind and solar generation (or local generation) has served a portion of that load (renewable resources or green energy). This approach based on unsupervised learning Long Short Term Memory (LSTM), with the attention mechanism, aims to give a clear interpretation of the Duck Curve prediction, and to understand the clear reasons for this discrepancy which can help decision makers to better interpret the curve and solve the problem efficiently. Information and Communication Technology (ICT) and Internet of Things (IoT) are necessary for the deployment of green energies. Therefore, the data from the different sensors can be used as a support to validate the information at the local production level and contribute in an effective and targeted way to solve the problem of the ”Duck Curve”.

In recent years, Deep Learning methods have become very popular in Natural Language Processing (NLP), especially transformer-based architecture. NLP domain requires a high volume of annotated data to work. Unfortunately, obtaining high-quality and voluminous labeled data is expensive and time-consuming. One promising method which has singled out for its performance in the context of data deficiency is semi-supervised learning with Generative Adversarial Networks (GAN). In this paper, we propose a new approach called PaGAN which is a combination of a document classifier and a sentence-level classifier inside a GAN for patent documents understanding. The idea is to mine the patent’s motivating problem (aka contradiction in TRIZ domain) which is fundamentally important to understand the underlying invention and its originality. PaGAN is applied and evaluated on a real-world dataset. Experiments show outperforming results of PaGAN comparatively to baseline approaches.

Lithium-ion batteries are used in most battery powered devices. Today’s research on Lithium-ion batteries mainly focuses on better energy management strategies and predictive maintenance. In this paper, a new approach based on auto-encoders and long short-term memory neural networks applied to usage data (voltage, current, temperature) is used to make a State of Health prediction. Encouraging results are obtained when conducting tests on various battery ageing datasets published by Sandia National Laboratories, the Massachusetts Institute of Technology and NASA’s Prognostics Center of Excellence.

International audience

TRIZ theory is considered as a kind of innovative theory, which mainly functions in solving contradiction and aims to improve engineers creativity. It may reason on different support such as web and scientific documents. Most of the current knowledge, especially in industrial domains lies in patents. They are, nevertheless, an underused resource because of their complexity, their length and the need of domain knowledge to make use of patent. In this paper, we introduce an innovative application of deep learning to mine knowledge from patents. We show that our summarization based approach, called SummaTRIZ1, enables the extraction of these contradictions essential for TRIZ problem solving engine. A BERT based summarization approach is introduced to retain contradiction sentences. Our approach is experimentally evaluated on a real data set showing the performance of SummaTRIZ.

Lithium Ion batteries are a core component of many lately designed devices. It is of crucial importance to be able to fully master the behaviour of batteries in order to meet the requirements in terms of safety, and performances. Predicting the Remaining Useful Life of batteries can help preventing a failure before it occurs, with an increased safety for the user and reduced costs linked to maintenance.The work described in this paper is an attempt to accurately predict the Remaining Useful Life of Li-Ion batteries using machine learning regression methods applied to a new set of ageing data published by the department of chemical engineering of the Massachusetts Institute of Technology. By changing the usual approach applied to data and feature management, very encouraging results were obtained. Compared with previous approaches in the literature using linear regression or Convolutional Neural Networks on the same dataset, our work on how to build a more efficient representation of ageing data combined with Artificial Neural Networks leads to more accurate predicting performances.

As today’s manufacturing domain is becoming more and more knowledge-intensive, knowledge-based systems (KBS) are widely applied in the predictive maintenance domain to detect and predict anomalies in machines and machine components. Within a KBS, decision rules are a comprehensive and interpretable tool for classification and knowledge discovery from data. However, when the decision rules incorporated in a KBS are extracted from heterogeneous sources, they may suffer from several rule quality issues, which weakens the performance of a KBS. To address this issue, in this paper, we propose a rule base refinement approach with considering rule quality measures. The proposed approach is based on a rule integration method for integrating the expert rules and the rules obtained from data mining. Within the integration process, rule accuracy, coverage, redundancy, conflict, and subsumption are the quality measures that we use to refine the rule base. A case study on a real-world data set shows the approach in detail.

As a scientific discipline, systems biology aims to build models of biological systems and processes through the computer analysis of a large amount of experimental data describing the behaviour of whole cells. It is within this context that we already developed the Biomolecular Network Ontology especially for the semantic understanding of the behaviour of complex biomolecular networks and their transittability. However, the challenge now is how to automatically populate it from a variety of biological documents. To this end, the target of this paper is to propose a new approach to automatically populate the Biomolecular Network Ontology and take advantage of the vast amount of biological knowledge expressed in heterogeneous unstructured data about complex biomolecular networks. Indeed, we have recently observed the emergence of deep learning techniques that provide significant and rapid progress in several domains, particularly in the process of deriving high-quality information from text. Despite its significant progress in recent years, deep learning is still not commonly used to populate ontologies. In this paper, we present a deep learning-based NLP ontology population system to populate the Biomolecular Network Ontology. Its originality is to jointly exploit deep learning and natural language processing techniques to identify, extract and classify new instances referring to the BNO ontology’s concepts from textual data. The preliminary results highlight the efficiency of our proposal for ontology population.

Within manufacturing processes, anomalies such as machinery faults and failures may lead to the outage situation of production lines. The outage of production lines is detrimental for the availability of production systems and may cause severe economic loss. To avoid the economic loss that may be caused by the outage situation, the prediction of anomalies on production lines is a crucial concern for manufacturers. Recently, data mining techniques have been applied to the manufacturing domain for predicting occurrence time of anomalies, such as the moment of machinery failure. However, existing predictive maintenance approaches have been limited to the prediction of the time of occurrence of machinery failures, while lacking the capability for identifying the criticality of the failures. This may lead to inappropriate maintenance plans and strategies. In this context, in this paper, we introduce a novel ontology-based approach to facilitate predictive maintenance in industry. The proposed approach is a combination use of fuzzy clustering and semantic technologies, where fuzzy clustering techniques are used to learn the criticality of failures based on machine historical data, and semantic technologies use the results of fuzzy clustering to predict the time of failures and the criticality of them. As results, a domain ontology for modeling predictive maintenance knowledge is developed, and a set of Semantic Web Rule Language (SWRL) predictive rules are proposed to reason about the time and criticality of machinery failures. A case study on a real-world industrial data set is followed to evaluate the usefulness and effectiveness of the proposed approach.

Chronicles are a kind of sequential patterns that consider the time dimension to produce relevant knowledge for decision makers. Mined from pairs of event-time, chronicles are represented in graphs for which vertices are events and edges are labeled with intervals representing the time between the two linked events. Chronicle mining is interesting in several domains where predicting the time interval of an event is important, such as network failure analysis, pharmaco-epidemiology and human activities analysis. In this work, we are interested in predicting the failure time of monitored industrial machines. We introduce a new approach to mine the most relevant chronicles in an industrial data set. The extracted chronicles are then used to predict the failure time of a given machine. Our approach is validated through several experiments led on a benchmark data set.

In recent years, the mining of frequent itemsets from uncertain databases has attracted much attention. Several researches have been conducted using different uncertain frameworks as probabilities, fuzzy sets and, most recently, evidence theory. There is very little study paid to mining pertinent knowledge from data where reliability is questionable. In this paper, we study and extend the evidential database framework in accounting data reliability. We propose new measures of support and confidence under uncertainty that consider the reliability and extend the state-of-the-art works. The proposed framework is thoroughly experimented on a real case problem for developing classification model from a chemical database.

In the belief function theory, several measures of uncertainty have been introduced. One of their possible use is unreliable source discounting before the fusion stage. Two different measures of uncertainty exist which are the intrinsic and extrinsic ones. The intrinsic measure makes it possible to assess the source's confusion whereas the extrinsic one measures the contradiction between sources. In this paper, we associate both measures in order to estimate the global reliability of a source. This method, named Generic Discounting Approach (GDA), is proposed in two different versions: Weighted GDA and Exponent GDA. Those reliability measures are integrated into a classifier. The method was tested, against to some pioneer approaches, on several UCI datasets as well as on an urban image classification problem and showed very encouraging results.

Decision making by considering multiple information sources could provide interesting results. For that reason, fusion formalisms were a major concern in the belief function community. In this context, the Belief function theory allows information fusion thanks to its combinations tools that it integrates. Nevertheless, belief function theory highlights a limit in the merging of contradictory (conflictual) sources. Many authors tackled this problem offering contributions in this field. Unfortunately, no proposed operator has distinguished by its adequacy regardless the type of handled sources. In this paper, we demonstrate the limits of some referenced works and we diagnostic the issues origin. We propose a conflict management approach based on an extra-information that guides the treatment. We also integrate a generic associative base borrowed from the data mining domain in order to apply the adequate conflict management.