Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
1Fraunhofer Institute for Production Systems and Design Technology, Germany; 2Institute for Machine Tools and Factory Management IWF - Technische Universität Berlin, Germany
Due to the increased digital networking of machines and systems in the production area, large datasets are generated. In addition, more external sensors are installed at production systems to acquire data for production and maintenance optimization purposes. Therefore, data analytics and interpretation is one of the challenges in Industry 4.0 applications. Reliable analysis of data (e.g. internal and external sensors) and information, such as system-internal alarms and messages produced during the operation, can be used to optimize production and maintenance processes. Furthermore, based on the data analytics, information and knowledge can be extracted from those raw data and used to develop data-driven business models and services, e.g. offer new availability contracts for production systems. This paper illustrates a concept for decentralized data analytics based on a smart sensor networks. The basic elements of this concept are the single-board computers, such as Raspberry Pi 3 and MEMS (Micro-Electro-Mechanical Systems) vibration sensors and standard communication technologies. Moreover, the decentralized data analytics by means of machine learning algorithms for data processing and pattern recognition, such as support vector machines will be presented using exemplary applications.
100. Mining shop-floor data for preventive maintenance management: integrating probabilistic and predictive models
Production processes are subject to degradation in their machinery and consequently loss of yield and quality may occur. Maintenance strategies and policies are constantly growing and developing to meet the requirements of high reliability and availability of resources in the production process. Among maintenance policies and strategies, it is possible to cite preventive maintenance, supported by the growing number of tools and information available that guarantee a reliable evaluation of the processes behavior. This type of maintenance covers regular inspections based on estimates of the machinery condition, preventing certain failures before they occur. Although literature and practical experience have proven that preventive maintenance is more effective than corrective maintenance, poorly determined intervals can still lead to high costs and a significant reduction in equipment availability. In order to avoid these issues, an optimized diagnostic analysis of the production process can be performed, allowing a better possible evaluation of the future behavior of the machinery involved. In this area, effective collection of shop-floor data is required, as well as adequate tools and techniques to transform this data into useful information for the manager or decision maker. Information extracted from event logs can be used in probabilistic and predictive models, effectively aiding in the evaluation of process behavior. The process mining techniques are approached in the present work to obtain a process model, used for the construction of a probabilistic model in Bayesian Networks (BN). The BN model outputs are frequency probabilities of the process activities that will feed the Autoregressive Integrated Moving Average (ARIMA) models. Preventive maintenance intervals are simulated between the production activities and the sum of the cycle times variations are compared until the best maintenance interval is found. The BN model also allows simulating variations in the productive activities frequency, re-feeding the ARIMA predictive models and providing new inferences for the preventive maintenance intervals. To validate the proposed methodology, an applied study is performed to a database collected from a lathe machine (CNC Turning) through a FIS, installed in an automobile industry. Simulations with increase and reduction in the machining activity frequency were performed and the values in the predictive models outputs are compared to the real values in the event log. For the application of this methodology is required a reliable collection of shop-floor data and a correct standardization of the event log, avoiding the existence of data that diverge from the real process behavior.
236. Data mining and machine learning for condition-based maintenance
Riccardo Accorsi1, Riccardo Manzini1, Pietro Pascarella2, Marco Patella2, Simone Sassi1
1Department of Industrial Engineering, Alma Mater Studiorum Bologna, viale del Risorgimento 2 – 40136 Bologna (Italy); 2Department of Computer Science and Engineering, Alma Mater Studiorum Bologna, viale del Risorgimento 2 – 40136 Bologna (Italy)
Complex production systems may count thousands of parts and components subject to multiple physical and logical connections and interdependencies. This level of complexity inhibits the traditional and statistically-based approach to reliability engineering, failure prediction and maintenance planning.
In the era of the Industry 4.0, emerging technologies, e.g. Radio Frequency Identification (RFID), Micro-Electro-Mechanical Systems (MEMS), Supervisory Control and Data Acquisition (SCADA) systems, Product Embedded Information Devices (PEID), represent more and more performing and available solutions to collect and monitor operating conditions of several components and functional groups, parts of such production systems.
The existing ICT solutions simplify the collection of large amount of data from on-field. The aim is to collect the right amount of data in order to predict in advanced the performance of the production system including the health/failure status. Is it possible to prevent an event of failure? Which is the role of data mining and machine learning techniques to support decision making in maintenance planning and execution? This paper introduces a number of state-of-the-art data analytics models and methods that can be profitably used for decision making in general, and, specifically, in maintenance engineering. Some numerical examples inspired to real case studies are illustrated demonstrating the effectiveness of such models and methods.