Jorde Daniel, Kriechbaumer Thomas, Berger Tim, Zitzlsperger Stefan, Jacobsen Hans-Arno
Department of Computer Science, Chair for Application and Middleware Systems, Technical University of Munich, 85748, Garching, Germany.
Sci Data. 2020 Dec 17;7(1):441. doi: 10.1038/s41597-020-00767-w.
Monitoring the internal conditions of a machine is essential to increase its production efficiency and to reduce energy waste. Non-intrusive condition monitoring techniques, such as analysing electrical signals, provide insights by disaggregating a composite signal of a machine as a whole into the individual components to determine their states. Developing and evaluating new algorithms for condition monitoring and maintenance-related analysis tasks require a fully-labelled dataset for a machine, which comprises standard industrial components that are triggered following a typical manufacturing process to produce goods. For this purpose, we introduce CREAM, a component level electrical measurement dataset for two industrial-grade coffeemakers, simulating industrial processes. The dataset contains continuous voltage and current measurements provided at 6400 samples per second, as well as the product and maintenance-related event labels, such as 370600 expert-labelled component-level electrical events, 1734 product ones and 3646 maintenance ones. CREAM provides fully-labelled ground-truth to establish a benchmark and comparative studies of manufacturing-related analysis in a controlled and transparent environment.
监测机器的内部状况对于提高其生产效率和减少能源浪费至关重要。非侵入式状态监测技术，如分析电信号，通过将机器整体的复合信号分解为各个组件来确定其状态，从而提供见解。开发和评估用于状态监测及与维护相关分析任务的新算法需要机器的完全标记数据集，该数据集包含标准工业组件，这些组件在典型制造过程中被触发以生产产品。为此，我们引入了CREAM，这是一个用于两台工业级咖啡机的组件级电测量数据集，模拟工业过程。该数据集包含以每秒6400个样本提供的连续电压和电流测量值，以及与产品和维护相关的事件标签，如370600个专家标记的组件级电事件、1734个产品事件和3646个维护事件。CREAM提供完全标记的真实数据，以便在可控且透明的环境中建立制造相关分析的基准和比较研究。
