Emanuele Caruso, Alessandro Simoni, Francesco Pelosin
Synthetic dataset generation in Computer Vision, particularly for industrial applications, is still underexplored. Industrial defect segmentation, for instance, requires highly accurate labels, yet acquiring such data is costly and time-consuming. To address this challenge, we propose a novel diffusion-based pipeline for generating high-fidelity industrial datasets with minimal supervision. Our approach conditions the diffusion model on enriched bounding box representations to produce precise segmentation masks, ensuring realistic and accurately localized defect synthesis. Compared to existing layout-conditioned generative methods, our approach improves defect consistency and spatial accuracy. We introduce two quantitative metrics to evaluate the effectiveness of our method and assess its impact on a downstream segmentation task trained on real and synthetic data. Our results demonstrate that diffusion-based synthesis can bridge the gap between artificial and real-world industrial data, fostering more reliable and cost-efficient segmentation models. The code is publicly available at https://github.com/covisionlab/diffusion_labeling.
In an industrial group like Safran, numerical simulations of physical phenomena are integral to most design processes. At Safran's corporate research center, we enhance these processes by developing fast and reliable surrogate models for various physics. We focus here on two technologies developed in recent years. The first is a physical reduced-order modeling method for non-linear structural mechanics and thermal analysis, used for calculating the lifespan of high-pressure turbine blades and performing heat analysis of high-pressure compressors. The second technology involves learning physics simulations with non-parameterized geometrical variability using classical machine learning tools, such as Gaussian process regression. Finally, we present our contributions to the open-source and open-data community.
Soft sensing infers hard-to-measure data through a large number of easily obtainable variables. However, in complex industrial scenarios, the issue of insufficient data volume persists, which diminishes the reliability of soft sensing. Generative Adversarial Networks (GAN) are one of the effective solutions for addressing insufficient samples. Nevertheless, traditional GAN fail to account for the mapping relationship between labels and features, which limits further performance improvement. Although some studies have proposed solutions, none have considered both performance and efficiency simultaneously. To address these problems, this paper proposes the multi-task learning-based regression GAN framework that integrates regression information into both the discriminator and generator, and implements a shallow sharing mechanism between the discriminator and regressor. This approach significantly enhances the quality of generated samples while improving the algorithm's operational efficiency. Moreover, considering the importance of training samples and generated samples, a dual data evaluation strategy is designed to make GAN generate more diverse samples, thereby increasing the generalization of subsequent modeling. The superiority of method is validated through four classic industrial soft sensing cases: wastewater treatment plants, surface water, $CO_2$ absorption towers, and industrial gas turbines.
In dynamic Industrial Internet of Things (IIoT) environments, models need the ability to selectively forget outdated or erroneous knowledge. However, existing methods typically rely on retain data to constrain model behavior, which increases computational and energy burdens and conflicts with industrial data silos and privacy compliance requirements. To address this, we propose a novel retain-free unlearning framework, referred to as Probing then Editing (PTE). PTE frames unlearning as a probe-edit process: first, it probes the decision boundary neighborhood of the model on the to-be-forgotten class via gradient ascent and generates corresponding editing instructions using the model's own predictions. Subsequently, a push-pull collaborative optimization is performed: the push branch actively dismantles the decision region of the target class using the editing instructions, while the pull branch applies masked knowledge distillation to anchor the model's knowledge on retained classes to their original states. Benefiting from this mechanism, PTE achieves efficient and balanced knowledge editing using only the to-be-forgotten data and the original model. Experimental results demonstrate that PTE achieves an excellent balance between unlearning effectiveness and model utility across multiple general and industrial benchmarks such as CWRU and SCUT-FD.
Industrial Vending Machines (IVMs) automate the dispensing of a variety of supplies like safety equipment and tools at customer sites, providing 24/7 access while tracking inventory in real-time. Industrial distribution companies typically manage the replenishment of IVMs using periodic schedules, which do not take advantage of these advanced real-time monitoring capabilities. We develop two approaches to optimize the long-term average cost of replenishments and stockouts per unit time: a state-dependent optimal control policy that jointly considers all inventory levels (referred to as trigger set policy) and a fixed cycle policy that optimizes replenishment frequency. We prove the monotonicity of the optimal trigger set policy and leverage it to design a computationally efficient approximate online control framework. Unlike existing methods, which typically handle a very limited number of items due to computational constraints, our approach scales to hundreds of items while achieving near-optimal performance. Leveraging transaction data from our industrial partner, we conduct an extensive set of numerical experiments to demonstrate this claim. Our results show that optimal fixed cycle replenishment reduces costs by 61.7 to 78.6% compared to current practice, with our online control framework delivering an additional 4.1 to 22.9% improvement. Our novel theoretical results provide practical tools for effective replenishment management in this modern vendor-managed inventory context.
Photo-realistic avatar is a modern term referring to the digital asset that represents a human in computer graphic advanced systems such as video games and simulation tools. These avatars utilize the advances in graphic technologies in both software and hardware aspects. While photo-realistic avatars are increasingly used in industrial simulations, representing human factors such as human workers psychophysiological states, remains a challenge. This article contributes to resolving this issue by introducing the novel concept of MetaStates which are the digitization and representation of the psychophysiological states of a human worker in the digital world. The MetaStates influence the physical representation and performance of a digital human worker while performing a task. To demonstrate this concept, this study presents the development of a photo-realistic avatar enhanced with multi-level graphical representations of psychophysiological states relevant to Industry 5.0. This approach represents a major step forward in the use of digital humans for industrial simulations, allowing companies to better leverage the benefits of the Industrial Metaverse in their daily operations and simulations while keeping human workers at the center of the system.
Recent research advances in Artificial Intelligence (AI) have yielded promising results for automated software vulnerability management. AI-based models are reported to greatly outperform traditional static analysis tools, indicating a substantial workload relief for security engineers. However, the industry remains very cautious and selective about integrating AI-based techniques into their security vulnerability management workflow. To understand the reasons, we conducted a discussion-based study, anchored in the authors' extensive industrial experience and keen observations, to uncover the gap between research and practice in this field. We empirically identified three main barriers preventing the industry from adopting academic models, namely, complicated requirements of scalability and prioritization, limited customization flexibility, and unclear financial implications. Meanwhile, research works are significantly impacted by the lack of extensive real-world security data and expertise. We proposed a set of future directions to help better understand industry expectations, improve the practical usability of AI-based security vulnerability research, and drive a synergistic relationship between industry and academia.
Cyberattacks on industrial control systems (ICS) have been drawing attention in academia. However, this has not raised adequate concerns among some industrial practitioners. Therefore, it is necessary to identify the vulnerable locations and components in the ICS and investigate the attack scenarios and techniques. This study proposes a method to assess the risk of cyberattacks on ICS with an improved Common Vulnerability Scoring System (CVSS) and applies it to a continuous stirred tank reactor (CSTR) model. The results show the physical system levels of ICS have the highest severity once cyberattacked, and controllers, workstations, and human-machine interface are the crucial components in the cyberattack and defense.
Yunpeng Ma, Khalil Younis, Bestoun S. Ahmed
et al.
Digital twins (DT) of industrial processes have become increasingly important. They aim to digitally represent the physical world to help evaluate, optimize, and predict physical processes and behaviors. Therefore, DT is a vital tool to improve production automation through digitalization and becomes more sophisticated due to rapidly evolving simulation and modeling capabilities, integration of IoT sensors with DT, and high-capacity cloud/edge computing infrastructure. However, the fidelity and reliability of DT software are essential to represent the physical world. This paper shows an automated and systematic test architecture for DT that correlates DT states with real-time sensor data from a production line in the forging industry. Our evaluation shows that the architecture can significantly accelerate the automatic DT testing process and improve its reliability. A systematic online DT testing method can significantly detect the performance shift and continuously improve the DT's fidelity. The snapshot creation methodology and testing agent architecture can be an inspiration and can be generally applicable to other industrial processes that use DT to generalize their automated testing.
Luigi Capogrosso, Alessio Mascolini, Federico Girella
et al.
Industry 4.0 involves the integration of digital technologies, such as IoT, Big Data, and AI, into manufacturing and industrial processes to increase efficiency and productivity. As these technologies become more interconnected and interdependent, Industry 4.0 systems become more complex, which brings the difficulty of identifying and stopping anomalies that may cause disturbances in the manufacturing process. This paper aims to propose a diffusion-based model for real-time anomaly prediction in Industry 4.0 processes. Using a neuro-symbolic approach, we integrate industrial ontologies in the model, thereby adding formal knowledge on smart manufacturing. Finally, we propose a simple yet effective way of distilling diffusion models through Random Fourier Features for deployment on an embedded system for direct integration into the manufacturing process. To the best of our knowledge, this approach has never been explored before.
High-entropy alloys are extensively studied due to their very promising properties. However manufacturing methods currently used to prepare HEAs are complicated, costly, and likely non-industrially scalable processes. This limits their evolution and poses questions regarding the material's applicability in the future. Considering the abovementioned point, we developed a novel methodology for efficient HEA production using a low ball-to-powder ratio (BPR). Using different milling times, we manufactured four HEA powder precursors using a BPR of 5:1, which were later sintered via the Spark Plasma Sintering technique and heat treated. Microstructural characterization was performed by optical microscopy, Scanning Electron Microscopy equipped with EDS and EBSD detectors, and X-ray diffraction. Mechanical properties were measured using nano and microhardness techniques. In this work, we follow the structural evolution of the material and connect it with the strengthening effect as a function of milling time. Furthermore, we discuss the impact of different sintering and annealing conditions, proving that HEAs characterized by high mechanical properties may be manufactured using low BPR.
Darius Sas, Ilaria Pigazzini, Paris Avgeriou
et al.
Architectural Technical Debt (ATD) is considered as the most significant type of TD in industrial practice. In this study, we interview 21 software engineers and architects to investigate a specific type of ATD, namely architectural smells (AS). Our goal is to understand the phenomenon of AS better and support practitioners to better manage it and researchers to offer relevant support. The findings of this study provide insights on how practitioners perceive AS and how they introduce them, the maintenance and evolution issues they experienced and associated to the presence of AS, and what practices and tools they adopt to manage AS.
Communication for control-centric industrial applications is characterized by the requirements of very high reliability, very low and deterministic latency and high scalability. Typically, IEEE 802.11-based wireless local area networks (WLANs), also known as Wi-Fi networks, are deemed ineligible for industrial control applications owing to insufficient reliability and non-deterministic latency. This paper proposes a novel solution for providing reliable and deterministic communication, through Wi-Fi, in industrial environments. The proposed solution, termed as \textsf{HAR\(^\text{2}\)D-Fi} (\underline{H}ybrid channel \underline{A}ccess with \underline{R}edundancy for \underline{R}eliable and \underline{D}eterministic Wi-\underline{Fi}), adopts hybrid channel access mechanisms for achieving deterministic communication. It also provides temporal redundancy for enhanced reliability. \textsf{HAR\(^\text{2}\)D-Fi} implements different medium access control (MAC) designs that build on the standard physical (PHY) layer. Such designs can be classified into two categories: (a) MAC designs with pre-defined (physical) time-slotted schedule, and (b) MAC designs with virtual time-slotted schedule. Performance evaluation, based on analysis and system-level simulations, demonstrates the viability of \textsf{HAR\(^\text{2}\)D-Fi} for control-centric industrial applications.
There has been an incredibly large investment in obtaining high-resolution stellar spectra for determining chemical abundances of stars. This information is crucial to answer fundamental questions in Astronomy by constraining the formation and evolution scenarios of the Milky Way as well as the stars and planets residing in it. We have just entered a new era, in which chemical abundances of FGK-type stars are being produced at industrial scales, where the observations, reduction, and analysis of the data are automatically performed by machines. Here we review the latest human efforts to assess the accuracy and precision of such industrial abundances by providing insights in the steps and uncertainties associated with the process of determining stellar abundances. To do so, we highlight key issues in the process of spectral analysis for abundance determination, with special effort in disentangling sources of uncertainties. We also provide a description of current and forthcoming spectroscopic surveys, focusing on their reported abundances and uncertainties. This allows us to identify which elements and spectral lines are best and why. Finally, we make a brief selection of main scientific questions the community is aiming to answer with abundances.
The cells and tissues that make up our body juggle contradictory mechanical demands. It is crucial for their survival to be able to withstand large mechanical loads, but it is equally crucial for them to produce forces and actively change shape during biological processes such as tissue growth and repair. The mechanics of cell and tissues is determined by scaffolds of protein polymers known as the cytoskeleton and the extracellular matrix, respectively. Experiments on model systems reconstituted from purified components combined with polymer physics concepts have already successfully uncovered some of the mechanisms that underlie the paradoxical mechanics of living matter. Initial work focussed on explaining universal features such as the nonlinear elasticity of cells and tissues in terms of polymer network models. However, living matter exhibits many advanced mechanical functionalities that are not captured by these coarse-grained theories. In this Review, we focus on recent experimental and theoretical insights revealing how their porous structure, structural hierarchy, transient crosslinking, and mechanochemical activity confer resilience combined with the ability to adapt and self-heal. These physical insights improve our understanding of cell and tissue biology and also provide a source of inspiration for synthetic life-like materials.
Lucas Goehring, Joaquim Li, Pree-Cha Kiatkirakajorn
Charged colloidal dispersions make up the basis of a broad range of industrial and commercial products, from paints to coatings and additives in cosmetics. During drying, an initially liquid dispersion of such particles is slowly concentrated into a solid, displaying a range of mechanical instabilities in response to highly variable internal pressures. Here we summarise the current appreciation of this process by pairing an advection-diffusion model of particle motion with a Poisson-Boltzmann cell model of inter-particle interactions, to predict the concentration gradients around a drying colloidal film. We then test these predictions with osmotic compression experiments on colloidal silica, and small-angle x-ray scattering experiments on silica dispersions drying in Hele-Shaw cells. Finally, we use the details of the microscopic physics at play in these dispersions to explore how two macroscopic mechanical instabilities -- shear-banding and fracture -- can be controlled.
Joshua E. S. Socolar, Tom C. Lubensky, Charles L. Kane
We present a model of a mechanical system with a vibrational mode spectrum identical to the spectrum of electronic excitations in a tight-binding model of graphene. The model consists of point masses connected by elastic couplings, called "tri-bonds," that implement certain three-body interactions, which can be tuned by varying parameters that correspond to the relative hopping amplitudes on the different bond directions in graphene. In the mechanical model, this is accomplished by varying the location of a pivot point that determines the allowed rigid rotations of a single tri-bond. The infinite system constitutes a Maxwell lattice, with the number of degrees of freedom equal to the number of constraints imposed by the tri-bonds. We construct the equilibrium and compatibility matrices and analyze the model's phase diagram, which includes spectra with Weyl points for some placements of the pivot and topologically polarized phases for others. We then discuss the edge modes and associated states of self stress for strips cut from the periodic lattice. Finally, we suggest a physical realization of the tri-bond, which allows access to parameter regimes not available to experiments on (strained) graphene and may be used to create other two-dimensional mechanical metamaterials with different spectral features.