Aspects related to prognostics are becoming a crucial part in the industrial sector. In this sense, Industry 4.0 is considered as a new paradigm that leverages on the IoT to propose increasingly more solutions to provide an estimate on the working conditions of an industrial plant. However, in context like the industrial sector where the number and heterogeneity of sensors can be very large, and the time requirements are very stringent, emerges the challenge to design effective infrastructures to interact with these complex systems. In this paper, we propose a full stack hardware/software infrastructure to collect, manage, and analyze the data gathered from a set of heterogeneous sensors attached to a real scale replica industrial plant available in our laboratory. On top of the proposed infrastructure we designed and implemented a fault prediction algorithm which exploits sensors data fusion with the aim to assess the working conditions of the industrial plant. The result section shows the obtained results in terms of accuracy from testing our proposed model and provides a comparison with a traditional Deep Neural Network (DNN) topology.

与预测有关的方面正在成为工业领域的关键部分。从这个意义上讲，工业4.0被认为是一种新的范式，它利用物联网来提出越来越多的解决方案，以估算工业工厂的工作条件。但是，在工业领域，传感器的数量和异构性可能非常大，并且时间要求非常严格，因此，设计与这些复杂系统进行交互的有效基础架构面临着挑战。在本文中，我们提出了一个完整的堆栈硬件/软件基础结构，以收集，管理和分析从一组异类传感器收集的数据，这些异类传感器连接到我们实验室中的真实规模副本工业工厂。在提议的基础架构之上，我们设计并实现了一种故障预测算法，该算法利用传感器数据融合来评估工业工厂的工作条件。结果部分以测试我们提出的模型的准确性显示了获得的结果，并与传统的深度神经网络（DNN）拓扑进行了比较。