Department of Electrical Engineering (DEE)

The manufacturing industry represents a data rich environment, in which larger and larger volumes of data are constantly being generated by its processes. However, only a relatively small portion of it is actually taken advantage of by manufacturers. As such, the proposed Intelligent Data Analysis and Real-Time Supervision (IDARTS) framework presents a possible solution for the implementation of scalable, flexible and pluggable data analysis and real-time self-monitoring systems for manufacturing environments. IDARTS is aligned with the current Industry 4.0 trend, being aimed at allowing manufacturers to translate their data into a business advantage, combining distributed data acquisition, machine learning and run-time decision making support to assist in fields such as predictive maintenance and quality control, reducing the impact of disruptive events in production.

To meet flexibility and reconfigurability requirements, modern production systems need hardware and software solutions which ease the connection and mediation of different and heterogonous industrial cyber-physical components. Following the vision of Industry 4.0, the H2020 PERFoRM project targets, particularly, the seamless reconfiguration of robots and machinery. This paper describes the implementation of a highly flexible, pluggable and distributed architecture solution, focusing on several building blocks, particularly a distributed middleware, a common data model and standard interfaces and technological adapters, which can be used for connecting legacy systems (such as databases) with simulation, visualization and reconfiguration tools.

In the last decades, several research initiatives suggested new solutions regarding the interoperability and interconnectivity among heterogeneous production components and all the actors that somehow interact within the shop-floor. However, most of the proposed data representations are focused on the description of the production capabilities. In this paper, it is proposed a common data model focused not only in the production capabilities of the different components as well as the description of all the events, variables and resources that could indicate quality issues. Hence, the proposed data model describes all the information required by the GO0DMAN solution to reduce as much as possible, the defects, the respective causes and the strategies to avoid the propagation along the line. In order to increase the adoption of the proposed data model, it was developed using AutomationML. The proposed data model was designed and tested within the scope of the Horizon 2020 GO0DMAN project.

The proliferation of Information and Communication Technologies allowed the development of new solutions to be applied at the shop-floor and all the tools which helps the manufacturers. Hence, new solutions such as cyber-physical production systems, data analytics and knowledge management were developed and proposed to solve the well-known issues, such as quality control in multistage manufacturing systems. However, those solutions can only have a small contribution in solving that issues compared to an optimized and fully integrated approach. To allow the development of a fully integrated environment, it is necessary to deliver a standard way to communicate and interact with the different functionalities. The proposed research aims to provide an integration layer, capable of translating the rules defined at the knowledge management level, structured as Decision Model and Notation rules, into an AutomationML based language. This allows the cyber-physical production system the ability to apply these rules near the shop-floor. This article presents the template defined to represent the rules in AutomationML as well as the infrastructure developed to receive the rules from the knowledge management, translate them and deliver to the cyber-physical production system. At the end of the article is presented a test bed where the solution is instantiated with rules focused on quality control.

Product dimensional variability is a crucial factor in the quality control of complex multistage manufacturing processes, where undetected defects can easily be propagated downstream. The recent advances in information technologies and consequently the increased volume of data that has become readily available provide an excellent opportunity for the development of automated defect detection approaches that are capable of extracting the implicit complex relationships in these multivariate data-rich environments. In this paper, several machine learning classifiers were trained and evaluated on varied metrics to predict dimensional defects in a real automotive multistage assembly line. The line encompasses two automated inspection stages with several human-operated assembly and pre-alignment stages in between. The results show that non-linear models like XGBoost and Random Forests are capable of modelling the complexity of such an environment, achieving a high true positive rate and showing promise for the improvement of existing quality control approaches, enabling defects and deviations to be addressed earlier and thus assist in reducing scrap and repair costs. INDEX TERMS Machine learning, quality control, predictive manufacturing system, multistage, automotive industry, industry 4.0.

The time-varying microstructure of sleep EEG spindles may have clinical significance in dementia studies and can be quantified with a number of techniques. In this paper, real and simulated sleep spindles were regarded as AM/FM signals modeled by six parameters that define the instantaneous envelope (IE) and instantaneous frequency (IF) waveforms for a sleep spindle. These parameters were estimated using four different methods, namely the Hilbert transform (HT), complex demodulation (CD), matching pursuit (MP) and wavelet transform (WT). The average error in estimating these parameters was lowest for HT, higher but still less than 10% for CD and MP, and highest (greater than 10%) for WT. The signal distortion induced by the use of a given method was greatest in the case of HT and MP. These two techniques would necessitate the removal of about 0.4 s from the spindle data, which is an important limitation for the case of spindles with duration less than 1 s. Although the CD method may lead to a higher error than HT and MP, it requires a removal of only about 0.23 s of data. An application of this sleep spindle parameterization via the CD method is proposed, in search of efficient EEG-based biomarkers in dementia. Preliminary results indicate that the proposed parameterization may be promising, since it can quantify specific differences in IE and IF characteristics between sleep spindles from dementia subjects and those from aged controls.