Rousopoulou Vaia, Nizamis Alexandros, Vafeiadis Thanasis, Ioannidis Dimosthenis, Tzovaras Dimitrios
Centre for Research and Technology Hellas-Information Technologies Institute (CERTH/ITI), Thessaloniki, Greece.
Front Artif Intell. 2020 Nov 16;3:578152. doi: 10.3389/frai.2020.578152. eCollection 2020.
The exploitation of big volumes of data in Industry 4.0 and the increasing development of cognitive systems strongly facilitate the realm of predictive maintenance for real-time decisions and early fault detection in manufacturing and production. Cognitive factories of Industry 4.0 aim to be flexible, adaptive, and reliable, in order to derive an efficient production scheme, handle unforeseen conditions, predict failures, and aid the decision makers. The nature of the data streams available in industrial sites and the lack of annotated reference data or expert labels create the challenge to design augmented and combined data analytics solutions. This paper introduces a cognitive analytics, self- and autonomous-learned system bearing predictive maintenance solutions for Industry 4.0. A complete methodology for real-time anomaly detection on industrial data and its application on injection molding machines are presented in this study. Ensemble prediction models are implemented on the top of supervised and unsupervised learners and build a compound prediction model of historical data utilizing different algorithms' outputs to a common consensus. The generated models are deployed on a real-time monitoring system, detecting faults in real-time incoming data streams. The key strength of the proposed system is the cognitive mechanism which encompasses a real-time self-retraining functionality based on a novel double-oriented evaluation objective, a data-driven and a model-based one. The presented application aims to support maintenance activities from injection molding machines' operators and demonstrate the advances that can be offered by exploiting artificial intelligence capabilities in Industry 4.0.
工业4.0中大量数据的利用以及认知系统的不断发展，极大地推动了预测性维护领域的发展，以实现制造和生产中的实时决策和早期故障检测。工业4.0的认知工厂旨在实现灵活、自适应和可靠，以制定高效的生产方案、应对意外情况、预测故障并协助决策者。工业现场可用数据流的性质以及缺乏带注释的参考数据或专家标签，给设计增强型和组合式数据分析解决方案带来了挑战。本文介绍了一种具有工业4.0预测性维护解决方案的认知分析、自学习和自主学习系统。本研究提出了一种针对工业数据的实时异常检测的完整方法及其在注塑机上的应用。集成预测模型在监督式和非监督式学习器之上实现，并利用不同算法的输出达成共同共识，构建历史数据的复合预测模型。生成的模型部署在实时监控系统上，检测实时传入数据流中的故障。所提出系统的关键优势在于其认知机制，该机制基于一种新颖的双导向评估目标（数据驱动和基于模型的目标），包含实时自我再训练功能。所展示的应用旨在支持注塑机操作员的维护活动，并展示在工业4.0中利用人工智能能力所能带来的进步。
