General information about each work package is presented next. The scope of each one of them was designed so that the main issues related to ISI were covered. Their summaries contain a general description of the main problem that will be treated. Work package leaders will provide updated information about tasks in their reports to the Scientific Committee.
Work Package 01 (WP1)
TITLE: Dimensionality reduction of very large datasets
SUMMARY: We consider a scenario of Smart Industries with massive datasets and a very large number of variables. Model induction in such a situation may not be feasible due to sparseness of data and curse of dimensionality. In this package we aim at developing new methods to identify the most relevant variables for a given task so that the dimension is reduced and the induction problem becomes treatable. Conventional methods for feature selection may not be appropriate for Smart Industries datasets, which may have hundreds of thousands of variables and terabytes of data. We target the development of new univariate and multivariate approaches to reduce dimension and to provide visualization of data under the perspective of such Smart Industries datasets.
GOALS: Develop new methods of dimensionality reduction (e.g. feature extraction and selection) in the scenario of very large dimensions and huge datasets.
WORK PACKAGE LEADER: Carlos Eduardo Pedreira (UFRJ)
BRAZILIAN RESEARCHERS INVOLVED: Marcelo Azevedo Costa (UFMG), André Paim Lemos (UFMG), Antônio de Pádua Braga (UFMG), Thiago de Souza Rodrigues (CEFET-MG), Carlos Eduardo Pedreira (UFRJ), Carlos Eduardo Mello (UFRRJ), Adria Ramos de Lyra (UFRRJ), Cristiano Leite de Castro (UFMG), Alessandro Beda (UFMG), Danilo Melges (UFMG), Ricardo Hiroshi C. Takahashi (UFMG).
INTERNATIONAL PARTNERS INVOLVED: Plamen Angelov (University of Lancaster), Yaser Abu-Mostafa (CALTECH), Tobias Glasmachers (University of Ruhr-Bochum), Vasile Palade (University of Coventry).
PRIVATE COMPANIES: EMC, Algar, ENACOM, Vallourec, A3Data, OPENCADD, Neocontrol.
RESEARCH AXES: Data analysis and visualization (3), Modeling of large datasets (4).
Work Package 02 (WP2)
TITLE: Learning from Large Industrial Datasets
SUMMARY: Model learning from data aims at selecting a representation, typically a function, which is valid in the whole input-output domain. To ensure convergence of the empirical risk and to induce such a function, the dataset must be large enough and representative. However, despite huge datasets that prevail in the Smart Industry scenarios, it may not be computationally feasible to induce a model using all data available if we consider current induction/learning methods. Under these circumstances, new methods should be developed capable to either use the complete dataset (low computational cost methods), or able to identify the most relevant samples (e.g. selective sampling) or even search for subspace representation of the whole dataset. The first approach may demand parallel/multicore computing machinery, the second one appropriate discriminative approaches and the third one subspace grouping methods. This approach shall benefit from the approaches developed in work package 1.
GOAL: Develop new model induction/learning approaches and learning methods capable to handle large datasets.
WORK PACKAGE LEADER: Antônio de Pádua Braga (UFMG)
BRAZILIAN RESEARCHERS INVOLVED: Antônio de Pádua Braga (UFMG), Walmir Matos Caminhas (UFMG), André Paim Lemos (UFMG), Fernando Gomide (UNICAMP), Luiz Affonso Guedes (UFRN), Bruno Sielly Jales Costa (IFRN), Aluízio Fausto Ribeiro Araújo (UFPE), Hasenclever da França Bassani (UFPE), Carlos Pedreira (UFRJ), Felipe França (UFRJ), Priscila Lima (UFRJ), Marcelo Azevedo Costa (UFMG), Rogério Martins Gomes (CEFET-MG), Rodney Rezende Saldanha (UFMG), Cristiano Leite de Castro (UFMG), Thiago de Souza Rodrigues (CEFET-MG), Alessandro Beda (UFMG), Eduardo Carrano (UFMG), Felipe Campelo (UFMG), Lucas de Souza Batista (UFMG).
INTERNATIONAL PARTNERS INVOLVED: Plamen Angelov (University of Lancaster), Vasile Palade (University of Coventry, UK), Yaser Abu-Mostafa (CALTECH), Tobias Glasmachers (University of Ruhr-Bochum), Ryszard Kowalczyk (University of Swinburne).
PRIVATE COMPANIES: EMC, Algar, ENACOM, Vallourec, A3Data, Bluelux, OPENCADD, Ferrous, Neocontrol.
RESEARCH AXES: CPSU with embedded autonomy (1), Data analysis and visualization (3), Modeling of large datasets (4).
Work Package 03 (WP3)
TITLE: Online modeling and control of dynamic industrial processes
SUMMARY: Dynamic process modeling is a major challenge in Smart Industry because, in addition to the high dimension and large volume of data, time lags should also be considered. Even when the dimension is reduced, estimating functions from temporal data is a challenge due to the enormous amount of data and possible oversampling effects. In such a situation, viable approaches are to estimate predictive functions using sliding windows instead of considering the whole sampling range and to use granular modeling. These approaches are also useful for non stationary processes, as is the case of many, if not most of the industrial processes.
GOALS: Develop process data summarization techniques using time windows, and online learning, process modeling and control approaches capable to handle high frequency data streams.
WORK PACKAGE LEADER: André Paim Lemos (UFMG)/Cristiano Leite de Castro (UFMG)
BRAZILIAN RESEARCHERS INVOLVED: Antônio de Pádua Braga (UFMG), Walmir Matos Caminhas(UFMG), André Paim Lemos, (UFMG), Cristiano Leite de Castro (UFMG), Fernando Gomide (UNICAMP), Luiz Affonso Guedes (UFRN), Bruno Sielly Jales Costa (IFRN), Eduardo Gontijo Carrano (UFMG), Danilo Melges (UFMG), Alessandro Beda (UFMG), Aluizio Fausto Ribeiro Araújo (UFPE), Rodney Rezende Saldanha (UFMG).
INTERNATIONAL PARTNERS INVOLVED: Vasile Palade (University of Coventry, UK), Plamen Angelov (University of Lancaster, UK).
PRIVATE COMPANIES: Vallourec, TNK, Ferrous.
RESEARCH AXES: CPSU with embedded autonomy (1), Modeling of large datasets (4).
Work Package 04 (WP4)
TITLE: Embedded Intelligence in Cyber-Physical System Units (CPSU)
SUMMARY: Cyber-Physical System Units in a smart industrial environment require the ability to interact with other instances of CPSU as well as to perform pre-analysis of data acquired. The implementation of this intelligence, though, is constrained by processing performance, power consumption, and memory size. Consequently, this work package focuses on exploration of efficient algorithms for massive data analysis on limited resources, signal processing, methods of interaction among CPSUs and higher instances, and service-based production environments. The latter requires development of methods to engineer CPSU functions as services, enable its detection, and define strategies on how other services can be used to complement the execution of tasks.
GOALS: Develop new solutions for embedding autonomy abilities into CPSUs.
WORK PACKAGE LEADER: Cristiano Leite de Castro (UFMG)/André Paim Lemos (UFMG)
BRAZILIAN RESEARCHERS INVOLVED: Paulo Molck (Nitryx), Alexandre Tazoniero (Nitryx), Rodrigo Almeida Gonçalves (FACAMP), Fernando Gomide (UNICAMP), Cristiano Leite de Castro (UFMG), Antônio de Pádua Braga (UFMG), André Paim Lemos (UFMG), Wallace do Couto Boaventura (UFMG), Flávio Henrique Vasconcelos (UFMG), Hilton de Oliveira Mota (UFMG).
INTERNATIONAL PARTNERS INVOLVED: Prof. Ryszard Kowalczyk (University of Swinburne), Michael Hubner (University of Ruhr-Bochum).
PRIVATE COMPANIES: Vallourec, Nitryx, Neocontrol.
RESEARCH AXES: CPSU with embedded autonomy (1), Data acquisition and storage (2), Data analysis and visualization (3), Modeling of large datasets (4).
Work Package 05 (WP5)
TITLE: Efficient Data Communication Infrastructure for Smart Industries
SUMMARY: Efficient data communication and storage is of paramount importance for Big Data applications such as Smart Industries. The challenges raised by Big Data in Smart Industries are related not only to the sheer volume and complexity of information but also to the fact that the underlying communication infrastructure has grown considerably in scale, heterogeneity, and administrative decentralization. More specifically, the communication infrastructure supporting Smart Industries will likely to be comprised of a variety of networking technologies such as wired-, infrastructure-based and infrastructure-less wireless networks. Sensor networks is a notable example of infrastructure-less wireless networks with interesting applications in Smart Industries including monitoring different phases in the manufacturing processes, tracking inventory, assisting with employee safety and security of the manufacturing facilities, etc. The focus of this research thrust is to develop new data storage schemes, network protocols and services to enable efficient data communication in Smart Industry applications. The resulting framework will allow us to effectively control Smart Industries’ communication infrastructure in order to guarantee adequate performance.
GOALS: Develop a novel Software-Defined Networking framework to ensure efficient data communication in Smart Industry applications.
WORK PACKAGE LEADER: Felipe França (UFRJ)
BRAZILIAN RESEARCHERS INVOLVED: Flávio Henrique Vasconcelos (UFMG), Wallace do Couto Boaventura (UFMG), Felipe França (UFRJ), Adria Lyra (UFRRJ), Ahmed Esmin (UFLA), Denilson Pereira (UFLA), Priscila Lima (UFRJ).
INTERNATIONAL PARTNERS INVOLVED: Katia Obraczka (UCSC)
PRIVATE COMPANIES: INOVAX
RESEARCH AXES: Data acquisition and storage (2).
Work Package 06 (WP6)
TITLE: Smart energy management
SUMMARY: In this work package we aim at investigating intelligent decision support systems for decentralized coordination, optimization and control of energy use, generation, co-generation and storage. The package shall include intelligent profiling and forecasting, supply and demand response optimization and management, add-value applications and services. Solutions may include socially intelligent computing, i.e. peer-to-peer interactions among consumers (represented and supported by intelligent software agents) to guide their individual control decisions and, by aggregating the decisions, produce effective control of system demand. We seek to develop efficient consensus and incentive-based computational mechanisms for decentralized control of demand that respects system-wide objectives and individual preferences. Intelligent energy management solutions expected outcomes include minimized energy consumption and greenhouse gas footprint.
GOALS: To conduct leading-edge research and development of intelligent software solutions for efficient energy management in smart grid.
WORK PACKAGE LEADER: Fernando Gomide (UNICAMP)/Antônio de Pádua Braga (UFMG)
BRAZILIAN RESEARCHERS INVOLVED: Antônio de Pádua Braga (UFMG), Fernando Gomide (UNICAMP), Carlos Pedreira (UFRJ), Rogério Martins Gomes (CEFET-MG), Rodney Rezende Saldanha (UFMG), Cristiano Leite de Castro (UFMG), André Paim Lemos (UFMG), Wallace do Couto Boaventura (UFMG).
INTERNATIONAL PARTNERS INVOLVED: Prof. Ryszard Kowalczyk (University of Swinburne)
PRIVATE COMPANIES: EMC, ENACOM, A3Data, Vallourec, TNK
RESEARCH AXES: CPSU with embedded autonomy (1), Data acquisition and storage (2), Data analysis and visualization (3), Modeling of large datasets (4).
Work Package 07 (WP7)
TITLE: Data-driven control
SUMMARY: The increasing complexity of industrial processes and various control systems causes further difficulties to propose models using classical system identification principles. In Smart Industries many processes produce a huge amount of relevant data, over time, about the state of the process and equipments. Hence, one can use such data online, offline or both, to directly design a controller, to predict and reach states of the system, to evaluate system performance, to take decisions, or to detect faults. Data-driven Control (DDC), a new branch of research, is regarded as the set of control theories and methods in which the controller is designed directly using sampled data of the controlled system or knowledge about the processed data. Therefore, control systems using neural networks or fuzzy systems are considered examples of DDC.
GOALS: To develop new control strategies based on sampled data in Smart Industry scenarios.
WORK PACKAGE LEADER: Aluízio Fausto Ribeiro de Araújo (UFPE)
BRAZILIAN RESEARCHERS INVOLVED: Jaelson Castro (UFPE), Paulo Maciel (UFPE), Fernando Gomide (UNICAMP), Eduardo Mazoni Mendes (UFMG).
INTERNATIONAL PARTNERS INVOLVED: Oscar Pastor López (Polytechnic University of Valencia)
PRIVATE COMPANIES: Vallourec, TNK, Ferrous.
RESEARCH AXES: CPSU with embedded autonomy (1), Data analysis and visualization (3), Modeling of large datasets (4).
Work Package 08 (WP8)
TITLE: Smart Logistics
SUMMARY: Markets and customer behavior are getting more and more volatile, forcing current supply chains to react more flexibly to highly diversified customer demand and on the same time be cost efficient in all of its operations. A further challenge for current supply chains is their global dimension, connecting suppliers and customers in a world-wide network. Current research is tackling these challenges on different levels. On a technological level passive and active sensors being able to communicate with other devices make the supply chain transparent by making the objects flowing in it identifiable. Global coordinating information systems may be redundant when local intelligence, within the logistics equipment enable them to carry out local planning tasks. Restricted computational power, fast reaction capabilities to environmental changes by re-planning are the challenges to be addressed. These ideas are followed in the context of the research streams of “Internet of Things” and “Industry 4.0”. A second level is utilizing the data stemming from the technological level. Sensors provide a multitude of data, which can be utilized and analyzed in many different ways in order to measure the performance of the supply chain, to detect new dependencies in market and supply chain behavior and furthermore enable an improved forecasting of their behavior. The results of all of these activities lead to a tactical and strategic planning of the supply chain. The analysis and evaluation of the collected data from different sources relates to the methods being developed within the area of “Big Data”, while the planning of global supply chains for a strategic and tactical time horizon demands improved planning and optimization methods. The third level considers the processes taking place in the supply chain, which are integrating and coordinating the different actors within the companies. Overall, there is a demand for highly flexible planning methods utilizing only minimal computational resources (mainly on the technological level), efficient methods for the flexible analysis of large data sets and related forecasting methods, utilizing the analysis results. Ability to deal with high number of actors and deciding about a multitude of parameters, new planning and optimization methods are needed for tackling the multi-objective problems in a paralleled way. Aiming at the process conceptualization with decentralized nature of the processes, new technologies from the area of Computational Intelligence mainly stemming from the area of Intelligent Agents and Metaheuristics will be conceived and tried out.
GOALS: Development of a decision platform incorporating a method set for the local and global planning and optimization of smart global supply chains including the coordination of the planning processes.
WORK PACKAGE LEADER: Fernando Buarque (UPE)
BRAZILIAN RESEARCHERS INVOLVED: Fernando Buarque (UPE), Celso G. Camilo Jr. (UFG).
INTERNATIONAL PARTNERS INVOLVED: Bernd Hellingrath (University of Muenster), Herbert Kuchen (University of Münster)
PRIVATE COMPANIES: GoGeo, Red&White.
RESEARCH AXES: Modeling of large datasets (4).
Work Package 09 (WP9)
TITLE: Cyber-physical Systems Engineering
SUMMARY: Differently from software and knowledge engineering design principles, methods and tools currently available, cyber-physical systems (CPS) still need core science, principles, and solid engineering design foundations. Design of CPS expand the notion of designing agents because CPS must include capabilities to negotiate, forecast, communicate, coordinate and process data in a multiagent environment.
GOALS: Develop the foundations needed to engineer cyber-physical systems (CPS). From abstractions of specific industrial system environments and applications, develop fundamental scientific and engineering principles that underpin the design and integration of cyber and physical elements across application sectors. Identify research, education, and training needs in CPS.
WORK PACKAGE LEADER: Fernando Gomide (UNICAMP)
BRAZILIAN RESEARCHERS INVOLVED: Rodrigo Almeida Gonçalves (FACAMP), Fernando Gomide (UNICAMP), Marcelo Azevedo Costa (UFMG).
INTERNATIONAL PARTNERS INVOLVED: Prof. Ryszard Kowalczyk (University of Swinburbe, Australia)
PRIVATE COMPANIES: Nitryx
RESEARCH AXES: Modeling of large datasets (4).