Papers
A Novel Architecture for Cyber-Physical Production Systems in Industry 4.0
The Hyperconnected Architecture for High Cognitive Production Plants (HyperCOG) project aims at the process industry’s complete digital transformation through an advanced Industrial Cyber-Physical Infrastructure. It is based on advanced technologies that allow a hyperconnected network of digital nodes to be created improving the classic automation hierarchy of communication layers. The nodes will collect data streams in real-time, offering cognitive sensing and information along with high performance computing capabilities making the process industry businesses solid in different scenarios. The system is validated in three fields of the process industry: steel, cement and chemical where optimisation in the use of energy and raw materials is obtained, among other benefits.
Hyperconnected Architecture for High Cognitive Production Plants
The HyperCOG project addresses the full digital transformation of process industry through an innovative Industrial Cyber-Physical System and Data Analytics. It is based on advanced technologies that enable the development of a hyperconnected network of digital nodes. The nodes can catch outstanding streams of data in real-time, which together with the high computing capabilities, provide sensing, knowledge and cognitive reasoning, making companies robust in the face of variant scenarios. The breaking-edge system proposed in this work is validated on productivity, environmental and replicability aspects on three use cases of three different sectors: steel, cement and chemical.
Clustering at the Disposal of Industry 4.0: Automatic Extraction of Plant Behaviors
For two centuries, the industrial sector has never stopped evolving. Since the dawn of the Fourth Industrial Revolution, commonly known as Industry 4.0, deep and accurate understandings of systems have become essential for real-time monitoring, prediction, and maintenance. In this paper, we propose a machine learning and data-driven methodology, based on data mining and clustering, for automatic identification and characterization of the different ways unknown systems can behave. It relies on the statistical property that a regular demeanor should be represented by many data with very close features; therefore, the most compact groups should be the regular behaviors. Based on the clusters, on the quantification of their intrinsic properties (size, span, density, neighborhood) and on the dynamic comparisons among each other, this methodology gave us some insight into the system’s demeanor, which can be valuable for the next steps of modeling and prediction stages. Applied to real Industry 4.0 data, this approach allowed us to extract some typical, real behaviors of the plant, while assuming no previous knowledge about the data. This methodology seems very promising, even though it is still in its infancy and that additional works will further develop it.
Identifying the Behaviors of an Industrial Plant: Application to Industry 4.0
We go deeper and deeper in automation: for decades, and even centuries, we have been dreaming of a completely automated world with the capacity to self-adapt to any imaginable context. A very promising paradigm to do so is Machine Learning, which allows an automatic and often reliable modeling of any unknown system, by analyzing its data and by deriving knowledge from them. The first tasks of such an approach are data preparation and knowledge extraction, which are areas of research on themselves; as such, experts are generally required in order to manually identify and characterize the studied systems, what can be a long, exhaustive and expensive task. Wouldn’t be great if this step could be executed automatically? In this paper, we will present some avenues of reflection in order to achieve such a cognitive system, by digging into the data through clustering approaches and by extracting any knowledge possible.
Characterizing N-Dimension Data Clusters: A Density-based Metric for Compactness and Homogeneity Evaluation
The new challenges Science is facing nowadays are legion; they mostly focus on high level technology, and more specifically Robotics, Internet of Things, Smart Automation (cities, houses, plants, buildings, etc.), and more recently Cyber-Physical Systems and Industry 4.0. For a long time, cognitive systems have been seen as a mere dream only worth of Science Fiction. Even though there is much to be done, the researches and progress made in Artificial Intelligence have let cognition-based systems make a great leap forward, which is now an actual great area of interest for many scientists and industrialists. Nonetheless, there are two main obstacles to system’s smartness: computational limitations and the infinite number of states to define; Machine Learning-based algorithms are perfectly suitable to Cognition and Automation, for they allow an automatic – and accurate – identification of the systems, usable as knowledge for later regulation. In this paper, we discuss the benefits of Ma chine Learning, and we present some new avenues of reflection for automatic behavior correctness identification through space partitioning, and density conceptualization and computation.
Deep Orientation-Guided Gender Recognition from Face Images
In the recent decade, gender recognition and face analysis has been one of the most researched issues in computer vision. Although several solutions have been provided to the problem of gender recognition from face images, nonetheless, it is regarded as a difficult issue. Deep learning has been proven to solve challenging problems. On the other hand, several existing works have proven their ability to accurately predict the head orientation angles. The remaining error in gender prediction models requires novel solutions to try to improve it further. In this work, we present a novel deep learning-based method to predict gender using both the face image and the head orientation angles. We show that the use of head orientation information consistently boosts the accuracy of gender prediction models. We achieve this by increasing the representational power of deep neural networks by introducing a head orientation adapter. It takes the head angles as input and outputs a vector that is used to recalibrate the deep learning neural networks. The proposed method was tested on a large-scale dataset called AutoPOSE, which has sub-millimetre-accurate head orientation angles. We show that using the head orientation adapter consistently boosts the gender prediction models’ accuracy, and reduces the error by 20%.
An approach of decision support system for drift diagnosis in cyber-physical production systems
Despite the development and application of new digital solutions in the production industry, the human operator is still essential in the production chain monitoring and control processes. In this context, some activities can be crucial for the human operator like, for example, drift diagnosis in production control process. It requires attention and experience and can be assisted by Decision Support System (DSS) to guide operators in decision-making in industrial production process control. Drift diagnosis process is a challenging problem in this context and artificial intelligence technologies are promising to tackle this issue. In this paper, we propose a new approach of DSS for drift diagnosis. The proposed approach is built upon a literature review on drift concept, drift detection methods and failure diagnosis approaches. This multi-model approach is designed to address all the diagnostics tasks of production systems and is based on Machine Learning (ML) algorithms to model the behavior of production systems, a knowledge-based model to integrate human experiences and a data-driven model to combine historical data from sensors. When the drift occurs, the proposed DSS can help human operator to determine drift causes and to suggest corrective actions. This article also provides guidelines about the design of a decision support system to support human operators in complex decision activities.
Modelling of CPPS: proposal of a generic model
In recent years, the introduction of new technologies, in particular cyber-physical systems (CPS), into industrial production has been at the heart of many research and economic growth prospects around the world. These developments are often called Smart Factories or Factories of the Future, and are characterised, among other things, by interoperability between its endogenous subsystems, assistance in decision-making, interaction between humans and machines, etc. In this context of industrial change, modelling appears complex and crucial for the implementation of CPS. This article, based on previous research carried out in the field of modelling of complex systems, aims to contribute by proposing a generic model of CPS. This generic model characterises the main components of a CPS, and we present an example of instantiation of this model in the context of chemical production. The originality lies in the integration of humans as an element at the centre of concerns for CPS modelling, which, to our knowledge, has rarely been done in the literature. Thus, this generic model constitutes a guide for the modelling of production systems based on CPS, notably taking into account the human in the loop and the management of the system through decision support
BSOM: A Two-Level Clustering Method Based on the Efficient Self-Organizing Maps
At the very beginning of the Industry 4.0 era, automated systems and automatic knowledge conceptualisation are becoming more and more essential. The ever faster, ever more resource-demanding processes are raising a pressing need for highly efficient handling of the systems. While the industries produce ever more, there is less and less time for back-up and quality assessment; a piece of solution may come along a real-time and automated monitoring, based on sensors’ data so as to assess products’ validity: this is the areas of Behaviour Identification and Anomaly Detection. In this paper, we propose to use Machine Learning and data-driven clustering to automatically identify the real behaviours of a system, and therefore its possible anomalies. More than the methodology, we mostly propose a more stable clustering method based on the Self-Organising Maps to achieve that purpose. We apply both methodology and that improved clustering method to real industrial data, and we show that it is more efficient, more stable and more relevant to dynamic systems.
ECD Test: An Empirical Way Based on the Cumulative Distributions to Evaluate the Number of Clusters for Unsupervised Clustering
Unsupervised clustering consists in blindly gathering unknown data into compact and homogeneous groups; it is one of the very first steps of any Machine Learning approach, whether it is about Data Mining, Knowledge Extraction, Anomaly Detection or System Modeling. Unfortunately, unsupervised clustering suffers from the major drawback of requiring manual parameters to perform accurately; one of them is the expected number of clusters. This parameter often determines whether the clusters will relevantly represent the system or not. From literature, there is no universal fashion to estimate this value; in this paper, we address this problem through a novel approach. To do so, we rely on a unique, blind clustering, then we characterize the sobuilt clusters by their Empirical Cumulative Distributions that we compare to one another using the Modified Hausdorff Distance, and we finally regroup the clusters by Region Growing, driven by these characteristics. This allows to rebuild the feature space’s regions: the number of expected clusters is the number of regions found. We apply this methodology to both academic and real industrial data, and show that it provides very good estimates of the number of clusters, no matter the dataset’s complexity nor the clustering method used.
Unsupervised Clustering at the Service of Automatic Anomaly Detection in Industry 4.0
Industrial processes are among the most complex systems, for they are dynamic, nonlinear and comprise many interdependent parts. In the scope of the contemporaneous fourth industrial revolution, known as Industry 4.0, the trend is to integrate Arti cial Intelligence and hyperconnectivity to intelligently exploit any available system’s resources. A key issue for system management is the control of anomalies, which may cause severe failures if not corrected rapidly, or aect product quality; de ning and knowing how to handle them is thus of major importance. In this paper, we propose to apply Machine Learning-based unsupervised clustering to industrial data to automatically identify the anomalies historically encountered; this is expected to help understand the system and de ne a framework for diagnosis of failures. We show that unsupervised clustering is able to detect salient groups of data, which can therefore be classi ed as anomalies by comparing them to the regular system’s behaviors, obtained using another round of unsupervised clustering.
Behavioral Modeling of Real Dynamic Processes in an Industry 4.0-oriented Context
With the Industry 4.0, new fashions to think the industry emerge: the production units are now orchestrated from some decentralized places to collaborate to improve efficiency, save time and resources, and reduce costs. To that end, Artificial Intelligence is expected to help manage units, prevent disruptions, predict failures, etc. A way to do so may consist in modeling the temporal evolution of the processes to track, predict and prevent the future failures; such modeling can be performed using the full dataset at once, but it may be more accurate to isolate the regions of the feature space where there is little variation in the data, then model these local regions separately, and finally connect all of them to build the final model of the system. This paper proposes to identify the compact regions of the feature space with unsupervised clustering, and then to model them with data-driven regression. The proposed methodology is tested on real industrial data, obtained in the scope of an Industry 4.0-oriented European project, and its accuracy is compared to that achieved by a global model; results show that local modeling achieves better accuracy, both during learning and testing stages.
Identifying the Regions of a Space with the Self-Parameterized Recursively Assessed Decomposition Algorithm (SPRADA)
This paper introduces a non-parametric methodology based on classical unsupervised clustering techniques to automatically identify the main regions of a space, without requiring the objective number of clusters, so as to identify the major regular states of unknown industrial systems. Indeed, a useful knowledge on real industrial processes is the identification of their regular states, and their historically encountered anomalies. Since both should form compact and salient groups of data, unsupervised clustering generally performs this task fairly accurately; however, this often requires the number of clusters upstream, knowledge which is rarely available. As such, the proposed algorithm operates a first partitioning of the space, then it estimates the integrity of the clusters, and splits them again and again until every cluster obtains an acceptable integrity; finally, a step of merging based on the clusters’ empirical distributions is performed to refine the partitioning. Applied to real industrial data obtained in the scope of a European project, this methodology proved able to automatically identify the main regular states of the system. Results show the robustness of the proposed approach in the fully-automatic and non-parametric identification of the main regions of a space, knowledge greatly useful to industrial anomaly detection and behavioral modeling thereafter.
HyDensity: A Hyper-Volume-Based Density Metric for Automatic Cluster Evaluation
As the systems produce ever more, becoming more and more complex and universal, the time spent on product validation narrows. The main alternative to the examination of every nal product is a statistical study, with a con dence interval; another solution may be an automated validation, possibly integrated to the system’s core. Some works addressed this topic using unsupervised learning, and especially clustering, but one challenge remains, that of supporting the information brought by these blind, automatized techniques. In this chapter, we address this problem of cluster validation using a data-driven metric, based on the hyper-volume theory, to estimate the clusters’ density, so as to estimate how representative they are. We apply this metric to real industrial data, and show how resilient and representative this metric is, and we extend it to two other metrics from literature to provide a hybrid quantifier, which allows to automatically validate or reject a cluster. The accepted clusters can eventually be assimilated to true regions of the feature space, whose representativeness and meaningfulness are given by this hybrid quantifier.
