- Email: [email protected]
Wear@Work – A New Approach for Data Acquisition Using Wearables
Available online at www.sciencedirect.com
ScienceDirect Procedia CIRP 50 (2016) 529 – 534
26th CIRP Design Conference
Wear@Work – A new approach for data acquisition using wearables Dennis Bauer*, Rolf Wutzke, Thomas Bauernhansl Fraunhofer Institute for Manufacturing Engineering and Automation IPA, Nobelstrasse 12, 70569 Stuttgart, Germany * Tel.: +49 711 970 1355; fax: +49 711 970 1028. E-mail address: [email protected]
Abstract Smart data acquisition is an important tool for companies in international competition as it allows new ways of creating machine understandable knowledge as well as revealing unexploited optimization potential. Furthermore, the current movements in production to autonomous and decentralized intelligence are based on this acquisition of data. This paper deals with the challenge of developing a new approach for data acquisition in industrial environments, recognizing the need for a user-friendly designed system supporting the operator in his work rather than distracting him. Moreover, this gathered data is the basis for data-driven process optimization and creation of new knowledge. © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license © 2016 The Authors. Published by Elsevier B.V. (http://creativecommons.org/licenses/by-nc-nd/4.0/). Selection under and peer-review responsibility Lihui Wang. responsibility ofunder the organizing committeeofofProfessor the 26th CIRP Design Conference Peer-review
Keywords: Smart data acquisition; Wearable; Knowledge creation
1. Introduction Social megatrends such as globalization, urbanization, demographic changes, growth of population and sustainability have great impact on manufacturing enterprises [1]. Due to the trend to personalized production, the variety of products rises extremely while quantities per product and variant decrease, leading to a rise in complexity in manufacturing. A propagated solution within the manufacturing sector addressing the rising complexity by usage of technologies of the fourth industrial revolution is the smart factory. Key skill of the smart factory is decentralized and autonomous selforganization, based on data collected by cyber-physical systems and human intelligence [2]. Away from this future approach, reality at shopfloor areas around the world is in most cases less advanced: Processes that are mainly driven by humans are nowadays still underrepresented in terms of data acquisition. This can be seen as a potential drawback from several perspectives. On the one hand side, addressing the need for greater manufacturing flexibility can only be successful when human characteristics like intelligence and flexibility as well as their implicit knowledge are supported in an advanced mode. Furthermore, the Industrie 4.0 approach includes seamless
interaction between humans and the surrounding systems, no matter whether of virtual or physical kind. Without decent dynamic knowledge about human actions, there will be no advanced interaction with either virtual environments or the physical world. Besides, the approaching trend for big data, which will be a key concept in enabling the smart factory, starts with sufficient amounts of data from the field of interest. Google's customization of individual searches based on collected web data or personalized adverts within your facebook timeline have already been implemented because of the collection and analysis of vast data from millions of individuals over time. In order to transfer this data analysis concept to manufacturing a sufficient data base is necessary [3]. However, human work in manufacturing processes is likely the most difficult part where to gather data from because of reasons like regulatory constraints or the simple fact that a human being does not have a digital interface to connect with. Within this paper, a new approach for continuous data acquisition as a basis for data-driven process optimization and creation of new knowledge is stated, focusing on human work at shopfloor level.
2212-8271 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the organizing committee of the 26th CIRP Design Conference doi:10.1016/j.procir.2016.04.121
530
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
2. State of the Art Motivation for data acquisition results from legal requirements as well as intrinsic motivation. In the EU each manufacturer is obliged to prove that their products are developed and manufactured, based on state of the art technologies [4]. Data can serve as evidence in the documentation of methods and of gained insights in product development and manufacturing. In specific industry sectors e.g. pharmaceutical industry even stricter law requirements apply, forcing companies to collect and archive all raw data [5]. Also in research environments data acquisition and archiving for at least ten years is recommended [6]. Intrinsic motivation to collect and analyze data results from the need to optimize processes [5]. Furthermore, efficient production data acquisition is the backbone for smart factories [7]. In 1985 Michael E. Porter described the value chain, containing primary and secondary activities for companies to create value [8]. Based on Porter’s value chain Miller and Mork developed the idea of a data value chain (see Fig. 1). This data value chain aims at the management and coordination of data from data generation to decision making, accomplished through data discovery, data integration and data exploitation [9].
Fig. 1. Data value chain [9]
2.1. Production data acquisition Data acquisition is the first step in data discovery. In production environments various data types have to be collected such as product and process data, order data or data about the factory structure and resources [10]. Necessary tools are differentiated in manual, semiautomatic and automatic data acquisition tools [11, 12]. However, especially manual data acquisition is often limited to absolutely required data, due to timely and monetary effort [13]. State of the art production data acquisition tools are characterized by preliminary defined processes as well as structured and planned data. Data acquisition is triggered by defined events or planned times. In addition exact locations for data collection have to be planned in advance [11]. 2.2. Big data analysis Data exploitation containing activities in analyzing and visualizing data as well as decision making marks the other end of the data value chain. Due to enormously rising data amounts generated each year, new technologies are necessary. These datasets are often referred to as big data which describes datasets whose sizes and complexities are beyond the capabilities of traditional data analysis technologies [14]. The challenges of big data are often described by three dimensions: volume, velocity and variety [15]. Newer
publications are also listing the dimensions of data’s veracity and value [16, 17]. Furthermore, a common data representation (data integration in Fig. 1) becomes less important with big data technologies, because they are specialized in analyzing data with no regard to structuring levels or data representation [15]. Within the manufacturing sector a decrease up to 50 percent in product development and assembly costs as well as up to seven percent reduction of working capital by big data analysis is expected. Therefore, big data will become a key factor for competition and growth for companies of the manufacturing sector [14]. 2.3. Gap analysis According to a survey by PricewaterhouseCoopers (PwC) companies have already realized the importance of efficient data analysis and application for their future competitiveness and business model [18]. Various authors have shown the abilities of big data applications in manufacturing e.g. realtime optimization of supply chain networks, predictive maintenance or decision support [16, 17, 19]. But all these applications assume a suitable existing data base. While equipment depending on the industry e.g. in semiconductor manufacturing is sometimes highly integrated, data about human work is most likely not collected due to monetary or timely effort, data protection laws or missing suitable technologies. But for a vast of companies, production is dominated by human work and manual tasks [20]. Therefore, the presented approach needs to focus on data acquisition for manual tasks to address data discovery within the data value chain (see Fig. 1). In addition, collecting this data is necessary to close the gap between the current situation of data acquisition in production environments and the requirements of big data analysis. 3. Approach This section defines the requirements for the presented smart data acquisition approach and subsequently describes the approach named Wear@Work in detail. 3.1. Requirements State of the art in data acquisition shows that currently only data which has been structured and planned in advance is recorded in industrial production. Besides the actual data this also includes locations and times when to record data. Even in research laboratories data is mostly recorded for pre-defined purposes only. Furthermore, the introduction of additional data acquisition tools leads to significant timely and monetary effort. But data can only be analyzed when formerly recorded. Therefore, the presented approach requires data acquisition without any preliminary planning and structuring. In addition, a preferably high number of sensors should be included in order to create a suitable data base for various analyzing technologies. A high number of sensors combined with high
531
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
sample rates lead to the expectation of high data volumes. Thus, the presented approach also needs to deal with data transfer, data storage and data analysis. Data transfer includes the integration into the service-oriented architecture of smart factories. In exchange, the presented approach benefits from this integration by a reduced time needed to apply and integrate the solution. Besides the automated acquisition of sensor data a possibility for manual data recording also has to be included. The presented approach needs to aim at recording data which has not been or has only been partially recorded before. Existing data acquisition technologies should rather be complemented than replaced. Further data sources which are not in scope of the approach e.g. machine data or environment data could be included later. Out of the user perspective the data acquisition device has to be worn effortlessly on the body and needs to run continuously without any user interaction. Furthermore, the actual process of acquiring data should not distract the user from their main task for productivity and security reasons. 3.2. Wear@Work Approach Wearables are miniaturized computer and sensor devices, which are worn effortlessly on or in a further development in the body of the wearer [21, 22]. They provide information and interaction with information anytime and anywhere, which is often described as always on and always accessible [23]. Furthermore, wearables are designed for hands-free operation, which helps distinguishing wearables from portable devices like smartphones or tablets [22]. This allows the wearer to stay focused on their main task and be assisted by the wearable rather than be distracted. Therefore, wearables have been chosen as the technical basis for the presented approach in order to assist the wearer by gathering data without additional effort. Whenever a wearable needs to be aware of its wearer, location or activity, sensors are needed. This awareness is built-in through mechanical, acoustic, biological, optical and environmental sensors [21, 24]. Not all sensors are necessarily wearable, but every wearable is equipped with sensing capabilities [25]. Combined with the continuous operation of wearables they fit perfectly well the requirement of continuous data acquisition. In the presented approach, a preliminary planning and structuring of the data acquisition is not necessary because all sensor data is being recorded. Besides typical sensors like device’s acceleration or wearer’s pulse this also includes communication technologies signal strengths such as WiFi or Bluetooth, which can be used for localization purposes. All this data can be recorded continuously and userindependently. Additionally, a possibility for manual data acquisition by voice recording was integrated in order to enable the wearer recording data which is not covered by the wearables’ sensors. Depending on the number and diversity of integrated sensors, wearables are producing a wide variety of data
ranging from structured data like accelerometer values to unstructured voice or video recordings. Besides this variety, wearables are also generating data at high velocity. This means that values even in cheap sensors are collected with frequencies up to a few hundred Hertz. While one wearable is not necessarily generating high data volumes, imagine each worker within a factory being equipped with at least one data collecting wearable. All these wearables together are generating huge amounts of data (see Chapter 4). In order to handle this variety of at high velocity generated data, new approaches and technologies are necessary. These technologies to transfer, store and later analyze data are often referred to as big data technologies. For transferring high volumes of data at high velocity the wearable has to be integrated in the service-oriented architecture of smart factories. To overcome the limitations of traditional relational databases in storing data and handling requests, the technology of NoSQL databases has been chosen for the presented approach. With no need to change existing data but high rates of adventitious data the concept of NoSQL databases fits perfectly well [15]. Therefore, the time series database InfluxDB has been used within the presented approach’s implementation due to its data schema-free design, high performance as well as the possibility to store numeric values and strings which is often not included in other time series databases [26]. The following subchapters will show three key aspects of the presented approach in detail: the selection of a wearable type, the design of the user interface and approaches for data exploitation. 3.2.1. Selection of a wearable type According to Jiang et al. wearables can be classified by their product form [27]. Examples for each product form can be seen in Table 1. Table 1. Wearables classified by product form head-mounted
hand-worn
e.g. Google Glass, Oculus Rift
e.g. Apple Watch, Fitbit Charge
body-dressed
foot-worn
e.g. Ralph Lauren Polo Tech Shirt
e.g. Trumpf MagicShoe
Based on this classification by product form, commercially available wearables suitable for data acquisition have been analyzed. Only wearable types with significant market competition have been considered. Furthermore, the possibility to run vendor-independent applications and a screen for visual feedback to the user were required. The outcome of this analysis were three suitable wearable types: augmented reality glasses, smartwatches and activity trackers. According to this analysis the wearable type smartwatch has been chosen as an implementation for the presented
532
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
approach. Smartwatches are characterized by a high number of integrated sensors, sufficient computing power and the possibility to run vendor-independent apps on most smartwatches. Additionally the wearing position on the wrist is suitable for data acquisition in manual tasks. From the wearer’s perspective, the small size and the light weight as well as the similarity to traditional watches and, therefore, a high acceptance, are to be highlighted. In addition, main parts of the user interface are well known from smartphones. However, the presented general approach for data acquisition can be transferred to any other suitable wearable type. The usage of a smartwatch can be considered as an exemplary implementation.
well-known in scientific community and also increasingly in industrial applications suits this application very well. Its functionality could be expanded by plugins and the RHadoopimplementation opens the possibility to use R in distributed Hadoop environments. Subsequently, results of data analysis can be visualized by the use of dashboards. In addition, technologies out of the field of machine learning can be used for decision making. Detailed information about data analysis algorithms, R and machine learning technologies can be found in relevant literature [29, 30]. Further research needs to be done to address this part of the data value chain (see Chapter 5). 4. Results
3.2.2. User Interface design Special emphasis was placed on the design of the user interface, which includes all hardware and software components necessary for interaction between wearer and wearable. Due to user acceptance, the user interface is of particular importance and according to DIN EN ISO 9241 seven requirements have to be taken into account [28]. Amongst those the most important requirement for Wear@Work was the appropriateness of the user interface for its desired task. Especially in production environments or research laboratories it is crucial not to distract users from their main tasks due to security reasons. Instead, users have to be supported in their tasks to increase productivity and provide a safe and suitable working environment. In order not to distract the wearer, only applicable elements of the user interface are shown such as a button to start and stop data acquisition as well as a button to activate voice recording. On a second screen the data acquisitions status can be checked (see Fig. 2). Recorded sensor data is not shown, because there is no benefit for the user. The wearer only benefits from results of the data analysis, which does not necessarily have to be shown on the smartwatch.
Besides the requirements, three use cases have been selected as a basis to evaluate the approaches results. As the technical basis to implement the Wear@Work approach for evaluation the architecture of the manufacturing IT platform Virtual Fort Knox has been used [31, 32]. Within each test run the following data has been collected: acceleration, gravity, rotation, magnetic field, ambient light, pressure, GPS coordinates, WiFi signals, Bluetooth signals, wearer’s pulse and voice recordings. All sensors combined, a test run of 8 hours which complies to a default work shift in factories produced about 500 MB data exported to a text file. 4.1. Use Case 1 Use case one describes the localization of humans in the factory, which can be used to schedule maintenance personnel or to optimize routes of automated guided vehicles within the factory by avoiding humans. Tests for the first use case have been performed at a semiconductor manufacturer’s plant. For this, a machine operator and a shift supervisor have been equipped with Wear@Work to evaluate the ability to localize and track workers based on data collected by Wear@Work.
Fig. 3. Patterns in WiFi networks signal strength measured with Wear@Work Fig. 2. Wear@Work user interface
3.2.3. Data exploitation Data exploitation forms the value generating part of the data value chain by generating new insights and knowledge based on the acquired data. The presented approach is designed to include and link data from every source, structured and unstructured, for the exploitation. As the technical basis for data analysis algorithms are used to identify repetitions, regularities and similarities and aim to point out patterns and correlations. The R project which is
The basic approach was to search for patterns in measured WiFi networks signal strength and link these patterns to other collected data. In doing so, different patterns could be identified (see Fig. 3, each point represents a measured signal strength, various networks are differentiated by color). It was expected that the different patterns were based on time periods of movement (highly volatile signal strengths, periods 1+3 in Fig. 3) and stopovers (low volatile network strengths, periods 2+4 in Fig. 3). To prove this assumption the patterns were linked to other data sources e.g. GPS coordinates, ambient light and pedometer. Thus, periods of movement
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
could be clearly identified. In addition, weather data in form of sunshine hours was linked to check if the worker was inside production hall or on outdoor factory premises. An even more precise localization is possible when using algorithms based on fixed points e.g. triangulation to analyze WiFi signal strengths. But therefore preliminary definition of these fixed points is necessary, which somehow contradicts the Wear@Work approach. There are several advantages of Wear@Work over existing localization technologies based on mobile devices like smartphones. First, there is the continuous data acquisition because the smartwatch is worn on the body of the wearer and could, therefore, not be forgotten. Second, additional data sources like ambient light, which are often not included in smartphones or which could likely be blocked by carrying inside pockets are integrated. Third, not only a person can be localized but also individual body parts e.g. the wearer’s hand. 4.2. Use Case 2 Use case two aims at simplifying ergonomic analysis or documentation of work steps by detection of worker’s hand movements. Tests for this use case have been conducted at Fraunhofer IPA and have been divided in two stages. At the first stage the smartwatch was mounted to a robot arm to evaluate sensor capabilities, afterwards the movements were performed by a human. The basic approach was again to search for patterns in data, this time in movement data e.g. acceleration and orientation and link these patterns to specific movements. Gyroscope’s data showed clear patterns but the absolute values differed even though it was a programmed movement by a robot arm (see Fig. 4). Acceleration data showed a similar behavior but with even less precise patterns.
The evaluation of the sensors’ data showed that it is basically possible to identify patterns in movement sensor data for robot and human movements. Nevertheless, this is highly dependent on the sensors’ frequency and resolution. In addition, an extensive data pool is needed to link these patterns to specific movements. Further research is needed addressing this use case’s topic. 4.3. Use Case 3 Use case three aims to improve the documentation of work steps and results using voice recognition. Thereby, the wearer benefits from both their hands being still being available for their main task. Tests for this use case are still under execution in cooperation with students of the chemistry department of the University of Stuttgart. Within these tests students who are working in various laboratories are using Wear@Work to log observations, performed tasks and parameters as well as results by voice recording. The interim results are very promising so far. Every test person highlighted the benefit of hands-free operation and admitted that more data was recorded than usual due to the possibility to record data without interrupting work to take notes. This data could later be used as a data source for the laboratory notebook. Nevertheless Wear@Work is not (yet) a full substitution for laboratory notebooks but rather an enlargement. First there is not yet a possibility to take photos e.g. of experimental setups or excerpts of data sheets. Second the voice recording is performed by a Google technology at the moment. That means all recordings are transferred to Google servers for voice recognition. Furthermore, this voice recognition engine is not optimized for the language and terms used in laboratories. Therefore, further technical development is needed in order to establish Wear@Work as an electronic laboratory notebook. 5. Conclusion and outlook
Fig. 4. Patterns in gyroscope data
This was caused by the highly varying quality of sensors integrated in commercially available smartwatches (see Table 2). Only sensors with high frequencies and precise resolutions allowed a suitable tracking of movements. Table 2. Comparison of gyroscope sensors in smartwatches Comparison of gyroscope sensors
Device 1
Device 2
Sensor vendor
Motorola
InvenSense
Frequency (Hz)
25
200
Resolution (rad/s)
0,01
0,0011
This paper describes the Wear@Work approach for smart data acquisition of human work, representing the data discovery section within the data value chain, as a basis for data-driven process optimization and creation of new knowledge. Therefore, it is necessary to close the gap between limitations of current production data acquisition methods and emerging approaches from big data analysis. For data about human work which is currently most likely not collected in industrial production, this has been achieved by the presented approach for data acquisition without any preliminary planning and structuring. In addition, the approach has been implemented in the form of equipping workers with a Wear@Work smartwatch. Valuable insights for data-driven process optimization and knowledge generation can be gathered by big data analysis. However, research and development in this field is not finished yet. In order to achieve a complete digital representation of the factory further data has to be collected
533
534
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
and integrated into data analysis. This could be done by considering the following aspects: x Equipping the whole factory personnel with Wear@Work smartwatches. Considering this exponentially growing number of data sources the presented approach’s technologies have to be reviewed and evaluated again. x Besides workers also factory equipment has to be developed to cyber-physical systems collecting data about themselves and their environments as well as to be integrated in the smart factory’s service-oriented architecture. However, this equipment integration has to be done by other technologies than using wearables. x Further development of wearables: Miniaturization, decreasing cost, increasing battery life and flexible components allow an even more effortlessly wearing on and in a further development in the body of the wearer. x To fully address the data value chain (see Fig. 1) more research has to be done covering data integration and especially data exploitation by ideas of chapter 3.2.3. Furthermore, especially when collecting data about humans data protection laws as well as other regulatory restrictions have to be considered. Country-specific laws in a globalized world even complicate this task. But current consultations about data protection laws in the European Parliament are also offering an opportunity to negotiate new and Europe-wide compromises between protection of privacy rights and the need for further data analysis in the 21st century.
References [1] Bauernhansl T. Zukünftige Rahmenbedingungen und Entwicklungstrends in der Produktionstechnik. Galvanotechnik 2013; 104(11):2185–91. [2] Bauernhansl T. Die vierte Industrielle Revolution: Der Weg in ein wertschaffendes Produktionsparadigma. In: Bauernhansl T, Hompel M ten, Vogel-Heuser B, editors. Industrie 4.0 in Produktion, Automatisierung und Logistik: Anwendung, Technologien, Migration. Wiesbaden: Springer Vieweg; 2014. p. 5–35 . [3] Bauernhansl T. Von der digitalen Fabrik zum digitalen Schatten. Hamburg; 2015 Nov 10-11. (Digitale Fabrik@Produktion). [4] Produkthaftungsgesetz: ProdHaftG; 2015. [5] Atmosudiro A, Faller M, Verl A. Durchgängige Datenintegration in die Cloud. wt Werkstattstechnik online 2014; 104(3):151–5. [6] DFG e. Vorschläge zur Sicherung guter wissenschaftlicher Praxis: Empfehlungen der Kommission "Selbstkontrolle in der Wissenschaft". Ergänzte Auflage. Weinheim: Wiley-VCH; 2013. [7] Kippels D. Daten aus der Produktion stärken Industrie 4.0. VDI nachrichten 2014; 2014(43):14. [8] Porter ME. Competitive advantage: Creating and sustaining superior performance. New York, NY: Free Press; 1985. Available from: URL:http://www.loc.gov/catdir/bios/simon051/98009581.html. [9] Miller HG, Mork P. From Data to Decisions: A Value Chain for Big Data. IT Professional 2013; 15(1):57–9. [10] Lucke DM. Ad hoc Informationsbeschaffung unter Einsatz kontextbezogener Systeme in der variantenreichen Serienfertigung [Diss., Univ. Stuttgart, 2013]. Stuttgart: Fraunhofer-Verlag; 2014. (Stuttgarter Beiträge zur Produktionsforschung; vol 25). Available from: URL:http://nbn-resolving.de/urn:nbn:de:bsz:93-opus-90669.
[11] Meier M. Manufacturing Execution Systems for Micro-Manufacturing. In: Qin Y, editor. Micro-manufacturing engineering and technology. Oxford: William Andrew; 2010. p. 377–93 . [12] VDI 5600. Fertigungsmanagementsysteme: Mafucaturing Execution Systems - MES [Entwurf]. Berlin: Beuth Verlag GmbH; 2015-01. [13] Kletti J, Schumacher J. Die perfekte Produktion: Manufacturing Excellence durch Short Interval Technology (SIT). 2. Aufl. Berlin: Springer Vieweg; 2014. Available from: URL:http://ebooks.ciando.com/book/index.cfm/bok_id/1880176. [14] McKinsey Global Institute, editor. Big Data: The next frontier for innovation, competition, and productivity [cited 2016 Jan 20]. Available from: URL:http://www.mckinsey.com/~/media/McKinsey/dotcom/Insights%2 0and%20pubs/MGI/Research/Technology%20and%20Innovation/Big% 20Data/MGI_big_data_full_report.ashx. [15] Klein D, Tran-Gia P, Hartmann M. Big Data. Informatik Spektrum 2013; 36(3):319–23. [16] Demchenko Y, Grosso P, Laat C de, Membrey P. Addressing big data issues in Scientific Data Infrastructure. In: 2013 International Conference on Collaboration Technologies and Systems (CTS). Piscataway, NJ: IEEE; 2013. p. 48–55 . [17] Bauernhansl T. Industrie 4.0: Big Data als Treiber neuer Optimierungspotenziale. Wiesbaden; 2014 Sep 18. (Computer-Woche Konferenz Best in Big Data). [18] PwC, editor. Industrie 4.0: Chancen und Herausforderungen der vierten industriellen Revolution [cited 2016 Jan 20]. Available from: URL:https://www.pwc.de/de/publikationen/paid_pubs/PwC_Studie_Indu strie_4.0_141022_SCREEN_GESCHUETZT.pdf. [19] Lee J, Lapira E, Bagheri B, Kao H. Recent advances and trends in predictive manufacturing systems in big data environment. Manufacturing Letters 2013; 1(1):38–41. [20] Spath D, Ganschar O, Gerlach S, Hämmerle M, Krause T, Schlund S, editors. Produktionsarbeit der Zukunft - Industrie 4.0. Stuttgart: Fraunhofer-Verlag; 2013. [21] Barfield W. Wearable Computers and Augmented Reality: Musings and Future Directions. In: Barfield W, editor. Fundamentals of Wearable Computers and Augmented Reality. 2nd ed. Hoboken: CRC Press; 2015. p. 3–12 . [22] Huang P. Promoting Wearable Computing. In: Jin Q, Li J, Zhang N, Cheng J, Yu C, Noguchi S, editors. Enabling Society with Information Technology. Tokyo: Springer Japan; 2002. p. 367–76 . [23] Mann S. Humanistic computing: “WearComp” as a new framework and application for intelligent signal processing. Proceedings of the IEEE 1998; 86(11):2123–51. [24] Viseu A. Social dimensions of wearable computers: an overview. Technoetic Arts 2003; 1(1):77–82. [25] Park S, Chung K, Jayaraman S. Wearables: Fundamentals, Advancements, and a Roadmap for the Future. In: Sazonov E, Neuman MR, editors. Wearable sensors: Fundamentals, implementation and applications. San Diego, CA: Academic Press; 2014. p. 1–23 . [26] influxdata, editor. influxdb: key features [cited 2016 Jan 20]. Available from: URL:https://influxdata.com/time-series-platform/influxdb. [27] Jiang H, Chen X, Zhang S, Zhang X, Kong W, Zhang T. Software for Wearable Devices: Challenges and Opportunities. In: 2015 IEEE 39th Annual Computer Software and Applications Conference (COMPSAC). Piscataway, NJ: IEEE; 2015. p. 592–7 . [28] DIN EN ISO 9241-110. Ergonomie der Mensch-System-Interaktion Teil 110: Grundsätze der Dialoggestaltung. Berlin: Beuth Verlag GmbH; 2008-09. [29] Bishop CM. Pattern recognition and machine learning. Corr. at 8. print. New York, NY: Springer; 2009. (Information science and statistics). [30] Lantz B. Machine learning with R: Learn how to use R to apply powerful machine learning methods and gain an insight into real-world applications. Birmingham: Packt Publishing; 2013. [31] Holtewert P, Wutzke R, Seidelmann J, Bauernhansl T. Virtual Fort Knox: Federative, secure and cloud-based platform for manufacturing. Procedia CIRP 2013; 7:527–32. [32] Stock D, Stöhr M, Rauschecker U, Bauernhansl T. Cloud-based Platform to facilitate Access to Manufacturing IT. Procedia CIRP 2014; 25:320–8.
ScienceDirect Procedia CIRP 50 (2016) 529 – 534
26th CIRP Design Conference
Wear@Work – A new approach for data acquisition using wearables Dennis Bauer*, Rolf Wutzke, Thomas Bauernhansl Fraunhofer Institute for Manufacturing Engineering and Automation IPA, Nobelstrasse 12, 70569 Stuttgart, Germany * Tel.: +49 711 970 1355; fax: +49 711 970 1028. E-mail address: [email protected]
Abstract Smart data acquisition is an important tool for companies in international competition as it allows new ways of creating machine understandable knowledge as well as revealing unexploited optimization potential. Furthermore, the current movements in production to autonomous and decentralized intelligence are based on this acquisition of data. This paper deals with the challenge of developing a new approach for data acquisition in industrial environments, recognizing the need for a user-friendly designed system supporting the operator in his work rather than distracting him. Moreover, this gathered data is the basis for data-driven process optimization and creation of new knowledge. © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license © 2016 The Authors. Published by Elsevier B.V. (http://creativecommons.org/licenses/by-nc-nd/4.0/). Selection under and peer-review responsibility Lihui Wang. responsibility ofunder the organizing committeeofofProfessor the 26th CIRP Design Conference Peer-review
Keywords: Smart data acquisition; Wearable; Knowledge creation
1. Introduction Social megatrends such as globalization, urbanization, demographic changes, growth of population and sustainability have great impact on manufacturing enterprises [1]. Due to the trend to personalized production, the variety of products rises extremely while quantities per product and variant decrease, leading to a rise in complexity in manufacturing. A propagated solution within the manufacturing sector addressing the rising complexity by usage of technologies of the fourth industrial revolution is the smart factory. Key skill of the smart factory is decentralized and autonomous selforganization, based on data collected by cyber-physical systems and human intelligence [2]. Away from this future approach, reality at shopfloor areas around the world is in most cases less advanced: Processes that are mainly driven by humans are nowadays still underrepresented in terms of data acquisition. This can be seen as a potential drawback from several perspectives. On the one hand side, addressing the need for greater manufacturing flexibility can only be successful when human characteristics like intelligence and flexibility as well as their implicit knowledge are supported in an advanced mode. Furthermore, the Industrie 4.0 approach includes seamless
interaction between humans and the surrounding systems, no matter whether of virtual or physical kind. Without decent dynamic knowledge about human actions, there will be no advanced interaction with either virtual environments or the physical world. Besides, the approaching trend for big data, which will be a key concept in enabling the smart factory, starts with sufficient amounts of data from the field of interest. Google's customization of individual searches based on collected web data or personalized adverts within your facebook timeline have already been implemented because of the collection and analysis of vast data from millions of individuals over time. In order to transfer this data analysis concept to manufacturing a sufficient data base is necessary [3]. However, human work in manufacturing processes is likely the most difficult part where to gather data from because of reasons like regulatory constraints or the simple fact that a human being does not have a digital interface to connect with. Within this paper, a new approach for continuous data acquisition as a basis for data-driven process optimization and creation of new knowledge is stated, focusing on human work at shopfloor level.
2212-8271 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the organizing committee of the 26th CIRP Design Conference doi:10.1016/j.procir.2016.04.121
530
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
2. State of the Art Motivation for data acquisition results from legal requirements as well as intrinsic motivation. In the EU each manufacturer is obliged to prove that their products are developed and manufactured, based on state of the art technologies [4]. Data can serve as evidence in the documentation of methods and of gained insights in product development and manufacturing. In specific industry sectors e.g. pharmaceutical industry even stricter law requirements apply, forcing companies to collect and archive all raw data [5]. Also in research environments data acquisition and archiving for at least ten years is recommended [6]. Intrinsic motivation to collect and analyze data results from the need to optimize processes [5]. Furthermore, efficient production data acquisition is the backbone for smart factories [7]. In 1985 Michael E. Porter described the value chain, containing primary and secondary activities for companies to create value [8]. Based on Porter’s value chain Miller and Mork developed the idea of a data value chain (see Fig. 1). This data value chain aims at the management and coordination of data from data generation to decision making, accomplished through data discovery, data integration and data exploitation [9].
Fig. 1. Data value chain [9]
2.1. Production data acquisition Data acquisition is the first step in data discovery. In production environments various data types have to be collected such as product and process data, order data or data about the factory structure and resources [10]. Necessary tools are differentiated in manual, semiautomatic and automatic data acquisition tools [11, 12]. However, especially manual data acquisition is often limited to absolutely required data, due to timely and monetary effort [13]. State of the art production data acquisition tools are characterized by preliminary defined processes as well as structured and planned data. Data acquisition is triggered by defined events or planned times. In addition exact locations for data collection have to be planned in advance [11]. 2.2. Big data analysis Data exploitation containing activities in analyzing and visualizing data as well as decision making marks the other end of the data value chain. Due to enormously rising data amounts generated each year, new technologies are necessary. These datasets are often referred to as big data which describes datasets whose sizes and complexities are beyond the capabilities of traditional data analysis technologies [14]. The challenges of big data are often described by three dimensions: volume, velocity and variety [15]. Newer
publications are also listing the dimensions of data’s veracity and value [16, 17]. Furthermore, a common data representation (data integration in Fig. 1) becomes less important with big data technologies, because they are specialized in analyzing data with no regard to structuring levels or data representation [15]. Within the manufacturing sector a decrease up to 50 percent in product development and assembly costs as well as up to seven percent reduction of working capital by big data analysis is expected. Therefore, big data will become a key factor for competition and growth for companies of the manufacturing sector [14]. 2.3. Gap analysis According to a survey by PricewaterhouseCoopers (PwC) companies have already realized the importance of efficient data analysis and application for their future competitiveness and business model [18]. Various authors have shown the abilities of big data applications in manufacturing e.g. realtime optimization of supply chain networks, predictive maintenance or decision support [16, 17, 19]. But all these applications assume a suitable existing data base. While equipment depending on the industry e.g. in semiconductor manufacturing is sometimes highly integrated, data about human work is most likely not collected due to monetary or timely effort, data protection laws or missing suitable technologies. But for a vast of companies, production is dominated by human work and manual tasks [20]. Therefore, the presented approach needs to focus on data acquisition for manual tasks to address data discovery within the data value chain (see Fig. 1). In addition, collecting this data is necessary to close the gap between the current situation of data acquisition in production environments and the requirements of big data analysis. 3. Approach This section defines the requirements for the presented smart data acquisition approach and subsequently describes the approach named Wear@Work in detail. 3.1. Requirements State of the art in data acquisition shows that currently only data which has been structured and planned in advance is recorded in industrial production. Besides the actual data this also includes locations and times when to record data. Even in research laboratories data is mostly recorded for pre-defined purposes only. Furthermore, the introduction of additional data acquisition tools leads to significant timely and monetary effort. But data can only be analyzed when formerly recorded. Therefore, the presented approach requires data acquisition without any preliminary planning and structuring. In addition, a preferably high number of sensors should be included in order to create a suitable data base for various analyzing technologies. A high number of sensors combined with high
531
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
sample rates lead to the expectation of high data volumes. Thus, the presented approach also needs to deal with data transfer, data storage and data analysis. Data transfer includes the integration into the service-oriented architecture of smart factories. In exchange, the presented approach benefits from this integration by a reduced time needed to apply and integrate the solution. Besides the automated acquisition of sensor data a possibility for manual data recording also has to be included. The presented approach needs to aim at recording data which has not been or has only been partially recorded before. Existing data acquisition technologies should rather be complemented than replaced. Further data sources which are not in scope of the approach e.g. machine data or environment data could be included later. Out of the user perspective the data acquisition device has to be worn effortlessly on the body and needs to run continuously without any user interaction. Furthermore, the actual process of acquiring data should not distract the user from their main task for productivity and security reasons. 3.2. Wear@Work Approach Wearables are miniaturized computer and sensor devices, which are worn effortlessly on or in a further development in the body of the wearer [21, 22]. They provide information and interaction with information anytime and anywhere, which is often described as always on and always accessible [23]. Furthermore, wearables are designed for hands-free operation, which helps distinguishing wearables from portable devices like smartphones or tablets [22]. This allows the wearer to stay focused on their main task and be assisted by the wearable rather than be distracted. Therefore, wearables have been chosen as the technical basis for the presented approach in order to assist the wearer by gathering data without additional effort. Whenever a wearable needs to be aware of its wearer, location or activity, sensors are needed. This awareness is built-in through mechanical, acoustic, biological, optical and environmental sensors [21, 24]. Not all sensors are necessarily wearable, but every wearable is equipped with sensing capabilities [25]. Combined with the continuous operation of wearables they fit perfectly well the requirement of continuous data acquisition. In the presented approach, a preliminary planning and structuring of the data acquisition is not necessary because all sensor data is being recorded. Besides typical sensors like device’s acceleration or wearer’s pulse this also includes communication technologies signal strengths such as WiFi or Bluetooth, which can be used for localization purposes. All this data can be recorded continuously and userindependently. Additionally, a possibility for manual data acquisition by voice recording was integrated in order to enable the wearer recording data which is not covered by the wearables’ sensors. Depending on the number and diversity of integrated sensors, wearables are producing a wide variety of data
ranging from structured data like accelerometer values to unstructured voice or video recordings. Besides this variety, wearables are also generating data at high velocity. This means that values even in cheap sensors are collected with frequencies up to a few hundred Hertz. While one wearable is not necessarily generating high data volumes, imagine each worker within a factory being equipped with at least one data collecting wearable. All these wearables together are generating huge amounts of data (see Chapter 4). In order to handle this variety of at high velocity generated data, new approaches and technologies are necessary. These technologies to transfer, store and later analyze data are often referred to as big data technologies. For transferring high volumes of data at high velocity the wearable has to be integrated in the service-oriented architecture of smart factories. To overcome the limitations of traditional relational databases in storing data and handling requests, the technology of NoSQL databases has been chosen for the presented approach. With no need to change existing data but high rates of adventitious data the concept of NoSQL databases fits perfectly well [15]. Therefore, the time series database InfluxDB has been used within the presented approach’s implementation due to its data schema-free design, high performance as well as the possibility to store numeric values and strings which is often not included in other time series databases [26]. The following subchapters will show three key aspects of the presented approach in detail: the selection of a wearable type, the design of the user interface and approaches for data exploitation. 3.2.1. Selection of a wearable type According to Jiang et al. wearables can be classified by their product form [27]. Examples for each product form can be seen in Table 1. Table 1. Wearables classified by product form head-mounted
hand-worn
e.g. Google Glass, Oculus Rift
e.g. Apple Watch, Fitbit Charge
body-dressed
foot-worn
e.g. Ralph Lauren Polo Tech Shirt
e.g. Trumpf MagicShoe
Based on this classification by product form, commercially available wearables suitable for data acquisition have been analyzed. Only wearable types with significant market competition have been considered. Furthermore, the possibility to run vendor-independent applications and a screen for visual feedback to the user were required. The outcome of this analysis were three suitable wearable types: augmented reality glasses, smartwatches and activity trackers. According to this analysis the wearable type smartwatch has been chosen as an implementation for the presented
532
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
approach. Smartwatches are characterized by a high number of integrated sensors, sufficient computing power and the possibility to run vendor-independent apps on most smartwatches. Additionally the wearing position on the wrist is suitable for data acquisition in manual tasks. From the wearer’s perspective, the small size and the light weight as well as the similarity to traditional watches and, therefore, a high acceptance, are to be highlighted. In addition, main parts of the user interface are well known from smartphones. However, the presented general approach for data acquisition can be transferred to any other suitable wearable type. The usage of a smartwatch can be considered as an exemplary implementation.
well-known in scientific community and also increasingly in industrial applications suits this application very well. Its functionality could be expanded by plugins and the RHadoopimplementation opens the possibility to use R in distributed Hadoop environments. Subsequently, results of data analysis can be visualized by the use of dashboards. In addition, technologies out of the field of machine learning can be used for decision making. Detailed information about data analysis algorithms, R and machine learning technologies can be found in relevant literature [29, 30]. Further research needs to be done to address this part of the data value chain (see Chapter 5). 4. Results
3.2.2. User Interface design Special emphasis was placed on the design of the user interface, which includes all hardware and software components necessary for interaction between wearer and wearable. Due to user acceptance, the user interface is of particular importance and according to DIN EN ISO 9241 seven requirements have to be taken into account [28]. Amongst those the most important requirement for Wear@Work was the appropriateness of the user interface for its desired task. Especially in production environments or research laboratories it is crucial not to distract users from their main tasks due to security reasons. Instead, users have to be supported in their tasks to increase productivity and provide a safe and suitable working environment. In order not to distract the wearer, only applicable elements of the user interface are shown such as a button to start and stop data acquisition as well as a button to activate voice recording. On a second screen the data acquisitions status can be checked (see Fig. 2). Recorded sensor data is not shown, because there is no benefit for the user. The wearer only benefits from results of the data analysis, which does not necessarily have to be shown on the smartwatch.
Besides the requirements, three use cases have been selected as a basis to evaluate the approaches results. As the technical basis to implement the Wear@Work approach for evaluation the architecture of the manufacturing IT platform Virtual Fort Knox has been used [31, 32]. Within each test run the following data has been collected: acceleration, gravity, rotation, magnetic field, ambient light, pressure, GPS coordinates, WiFi signals, Bluetooth signals, wearer’s pulse and voice recordings. All sensors combined, a test run of 8 hours which complies to a default work shift in factories produced about 500 MB data exported to a text file. 4.1. Use Case 1 Use case one describes the localization of humans in the factory, which can be used to schedule maintenance personnel or to optimize routes of automated guided vehicles within the factory by avoiding humans. Tests for the first use case have been performed at a semiconductor manufacturer’s plant. For this, a machine operator and a shift supervisor have been equipped with Wear@Work to evaluate the ability to localize and track workers based on data collected by Wear@Work.
Fig. 3. Patterns in WiFi networks signal strength measured with Wear@Work Fig. 2. Wear@Work user interface
3.2.3. Data exploitation Data exploitation forms the value generating part of the data value chain by generating new insights and knowledge based on the acquired data. The presented approach is designed to include and link data from every source, structured and unstructured, for the exploitation. As the technical basis for data analysis algorithms are used to identify repetitions, regularities and similarities and aim to point out patterns and correlations. The R project which is
The basic approach was to search for patterns in measured WiFi networks signal strength and link these patterns to other collected data. In doing so, different patterns could be identified (see Fig. 3, each point represents a measured signal strength, various networks are differentiated by color). It was expected that the different patterns were based on time periods of movement (highly volatile signal strengths, periods 1+3 in Fig. 3) and stopovers (low volatile network strengths, periods 2+4 in Fig. 3). To prove this assumption the patterns were linked to other data sources e.g. GPS coordinates, ambient light and pedometer. Thus, periods of movement
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
could be clearly identified. In addition, weather data in form of sunshine hours was linked to check if the worker was inside production hall or on outdoor factory premises. An even more precise localization is possible when using algorithms based on fixed points e.g. triangulation to analyze WiFi signal strengths. But therefore preliminary definition of these fixed points is necessary, which somehow contradicts the Wear@Work approach. There are several advantages of Wear@Work over existing localization technologies based on mobile devices like smartphones. First, there is the continuous data acquisition because the smartwatch is worn on the body of the wearer and could, therefore, not be forgotten. Second, additional data sources like ambient light, which are often not included in smartphones or which could likely be blocked by carrying inside pockets are integrated. Third, not only a person can be localized but also individual body parts e.g. the wearer’s hand. 4.2. Use Case 2 Use case two aims at simplifying ergonomic analysis or documentation of work steps by detection of worker’s hand movements. Tests for this use case have been conducted at Fraunhofer IPA and have been divided in two stages. At the first stage the smartwatch was mounted to a robot arm to evaluate sensor capabilities, afterwards the movements were performed by a human. The basic approach was again to search for patterns in data, this time in movement data e.g. acceleration and orientation and link these patterns to specific movements. Gyroscope’s data showed clear patterns but the absolute values differed even though it was a programmed movement by a robot arm (see Fig. 4). Acceleration data showed a similar behavior but with even less precise patterns.
The evaluation of the sensors’ data showed that it is basically possible to identify patterns in movement sensor data for robot and human movements. Nevertheless, this is highly dependent on the sensors’ frequency and resolution. In addition, an extensive data pool is needed to link these patterns to specific movements. Further research is needed addressing this use case’s topic. 4.3. Use Case 3 Use case three aims to improve the documentation of work steps and results using voice recognition. Thereby, the wearer benefits from both their hands being still being available for their main task. Tests for this use case are still under execution in cooperation with students of the chemistry department of the University of Stuttgart. Within these tests students who are working in various laboratories are using Wear@Work to log observations, performed tasks and parameters as well as results by voice recording. The interim results are very promising so far. Every test person highlighted the benefit of hands-free operation and admitted that more data was recorded than usual due to the possibility to record data without interrupting work to take notes. This data could later be used as a data source for the laboratory notebook. Nevertheless Wear@Work is not (yet) a full substitution for laboratory notebooks but rather an enlargement. First there is not yet a possibility to take photos e.g. of experimental setups or excerpts of data sheets. Second the voice recording is performed by a Google technology at the moment. That means all recordings are transferred to Google servers for voice recognition. Furthermore, this voice recognition engine is not optimized for the language and terms used in laboratories. Therefore, further technical development is needed in order to establish Wear@Work as an electronic laboratory notebook. 5. Conclusion and outlook
Fig. 4. Patterns in gyroscope data
This was caused by the highly varying quality of sensors integrated in commercially available smartwatches (see Table 2). Only sensors with high frequencies and precise resolutions allowed a suitable tracking of movements. Table 2. Comparison of gyroscope sensors in smartwatches Comparison of gyroscope sensors
Device 1
Device 2
Sensor vendor
Motorola
InvenSense
Frequency (Hz)
25
200
Resolution (rad/s)
0,01
0,0011
This paper describes the Wear@Work approach for smart data acquisition of human work, representing the data discovery section within the data value chain, as a basis for data-driven process optimization and creation of new knowledge. Therefore, it is necessary to close the gap between limitations of current production data acquisition methods and emerging approaches from big data analysis. For data about human work which is currently most likely not collected in industrial production, this has been achieved by the presented approach for data acquisition without any preliminary planning and structuring. In addition, the approach has been implemented in the form of equipping workers with a Wear@Work smartwatch. Valuable insights for data-driven process optimization and knowledge generation can be gathered by big data analysis. However, research and development in this field is not finished yet. In order to achieve a complete digital representation of the factory further data has to be collected
533
534
Dennis Bauer et al. / Procedia CIRP 50 (2016) 529 – 534
and integrated into data analysis. This could be done by considering the following aspects: x Equipping the whole factory personnel with Wear@Work smartwatches. Considering this exponentially growing number of data sources the presented approach’s technologies have to be reviewed and evaluated again. x Besides workers also factory equipment has to be developed to cyber-physical systems collecting data about themselves and their environments as well as to be integrated in the smart factory’s service-oriented architecture. However, this equipment integration has to be done by other technologies than using wearables. x Further development of wearables: Miniaturization, decreasing cost, increasing battery life and flexible components allow an even more effortlessly wearing on and in a further development in the body of the wearer. x To fully address the data value chain (see Fig. 1) more research has to be done covering data integration and especially data exploitation by ideas of chapter 3.2.3. Furthermore, especially when collecting data about humans data protection laws as well as other regulatory restrictions have to be considered. Country-specific laws in a globalized world even complicate this task. But current consultations about data protection laws in the European Parliament are also offering an opportunity to negotiate new and Europe-wide compromises between protection of privacy rights and the need for further data analysis in the 21st century.
References [1] Bauernhansl T. Zukünftige Rahmenbedingungen und Entwicklungstrends in der Produktionstechnik. Galvanotechnik 2013; 104(11):2185–91. [2] Bauernhansl T. Die vierte Industrielle Revolution: Der Weg in ein wertschaffendes Produktionsparadigma. In: Bauernhansl T, Hompel M ten, Vogel-Heuser B, editors. Industrie 4.0 in Produktion, Automatisierung und Logistik: Anwendung, Technologien, Migration. Wiesbaden: Springer Vieweg; 2014. p. 5–35 . [3] Bauernhansl T. Von der digitalen Fabrik zum digitalen Schatten. Hamburg; 2015 Nov 10-11. (Digitale Fabrik@Produktion). [4] Produkthaftungsgesetz: ProdHaftG; 2015. [5] Atmosudiro A, Faller M, Verl A. Durchgängige Datenintegration in die Cloud. wt Werkstattstechnik online 2014; 104(3):151–5. [6] DFG e. Vorschläge zur Sicherung guter wissenschaftlicher Praxis: Empfehlungen der Kommission "Selbstkontrolle in der Wissenschaft". Ergänzte Auflage. Weinheim: Wiley-VCH; 2013. [7] Kippels D. Daten aus der Produktion stärken Industrie 4.0. VDI nachrichten 2014; 2014(43):14. [8] Porter ME. Competitive advantage: Creating and sustaining superior performance. New York, NY: Free Press; 1985. Available from: URL:http://www.loc.gov/catdir/bios/simon051/98009581.html. [9] Miller HG, Mork P. From Data to Decisions: A Value Chain for Big Data. IT Professional 2013; 15(1):57–9. [10] Lucke DM. Ad hoc Informationsbeschaffung unter Einsatz kontextbezogener Systeme in der variantenreichen Serienfertigung [Diss., Univ. Stuttgart, 2013]. Stuttgart: Fraunhofer-Verlag; 2014. (Stuttgarter Beiträge zur Produktionsforschung; vol 25). Available from: URL:http://nbn-resolving.de/urn:nbn:de:bsz:93-opus-90669.
[11] Meier M. Manufacturing Execution Systems for Micro-Manufacturing. In: Qin Y, editor. Micro-manufacturing engineering and technology. Oxford: William Andrew; 2010. p. 377–93 . [12] VDI 5600. Fertigungsmanagementsysteme: Mafucaturing Execution Systems - MES [Entwurf]. Berlin: Beuth Verlag GmbH; 2015-01. [13] Kletti J, Schumacher J. Die perfekte Produktion: Manufacturing Excellence durch Short Interval Technology (SIT). 2. Aufl. Berlin: Springer Vieweg; 2014. Available from: URL:http://ebooks.ciando.com/book/index.cfm/bok_id/1880176. [14] McKinsey Global Institute, editor. Big Data: The next frontier for innovation, competition, and productivity [cited 2016 Jan 20]. Available from: URL:http://www.mckinsey.com/~/media/McKinsey/dotcom/Insights%2 0and%20pubs/MGI/Research/Technology%20and%20Innovation/Big% 20Data/MGI_big_data_full_report.ashx. [15] Klein D, Tran-Gia P, Hartmann M. Big Data. Informatik Spektrum 2013; 36(3):319–23. [16] Demchenko Y, Grosso P, Laat C de, Membrey P. Addressing big data issues in Scientific Data Infrastructure. In: 2013 International Conference on Collaboration Technologies and Systems (CTS). Piscataway, NJ: IEEE; 2013. p. 48–55 . [17] Bauernhansl T. Industrie 4.0: Big Data als Treiber neuer Optimierungspotenziale. Wiesbaden; 2014 Sep 18. (Computer-Woche Konferenz Best in Big Data). [18] PwC, editor. Industrie 4.0: Chancen und Herausforderungen der vierten industriellen Revolution [cited 2016 Jan 20]. Available from: URL:https://www.pwc.de/de/publikationen/paid_pubs/PwC_Studie_Indu strie_4.0_141022_SCREEN_GESCHUETZT.pdf. [19] Lee J, Lapira E, Bagheri B, Kao H. Recent advances and trends in predictive manufacturing systems in big data environment. Manufacturing Letters 2013; 1(1):38–41. [20] Spath D, Ganschar O, Gerlach S, Hämmerle M, Krause T, Schlund S, editors. Produktionsarbeit der Zukunft - Industrie 4.0. Stuttgart: Fraunhofer-Verlag; 2013. [21] Barfield W. Wearable Computers and Augmented Reality: Musings and Future Directions. In: Barfield W, editor. Fundamentals of Wearable Computers and Augmented Reality. 2nd ed. Hoboken: CRC Press; 2015. p. 3–12 . [22] Huang P. Promoting Wearable Computing. In: Jin Q, Li J, Zhang N, Cheng J, Yu C, Noguchi S, editors. Enabling Society with Information Technology. Tokyo: Springer Japan; 2002. p. 367–76 . [23] Mann S. Humanistic computing: “WearComp” as a new framework and application for intelligent signal processing. Proceedings of the IEEE 1998; 86(11):2123–51. [24] Viseu A. Social dimensions of wearable computers: an overview. Technoetic Arts 2003; 1(1):77–82. [25] Park S, Chung K, Jayaraman S. Wearables: Fundamentals, Advancements, and a Roadmap for the Future. In: Sazonov E, Neuman MR, editors. Wearable sensors: Fundamentals, implementation and applications. San Diego, CA: Academic Press; 2014. p. 1–23 . [26] influxdata, editor. influxdb: key features [cited 2016 Jan 20]. Available from: URL:https://influxdata.com/time-series-platform/influxdb. [27] Jiang H, Chen X, Zhang S, Zhang X, Kong W, Zhang T. Software for Wearable Devices: Challenges and Opportunities. In: 2015 IEEE 39th Annual Computer Software and Applications Conference (COMPSAC). Piscataway, NJ: IEEE; 2015. p. 592–7 . [28] DIN EN ISO 9241-110. Ergonomie der Mensch-System-Interaktion Teil 110: Grundsätze der Dialoggestaltung. Berlin: Beuth Verlag GmbH; 2008-09. [29] Bishop CM. Pattern recognition and machine learning. Corr. at 8. print. New York, NY: Springer; 2009. (Information science and statistics). [30] Lantz B. Machine learning with R: Learn how to use R to apply powerful machine learning methods and gain an insight into real-world applications. Birmingham: Packt Publishing; 2013. [31] Holtewert P, Wutzke R, Seidelmann J, Bauernhansl T. Virtual Fort Knox: Federative, secure and cloud-based platform for manufacturing. Procedia CIRP 2013; 7:527–32. [32] Stock D, Stöhr M, Rauschecker U, Bauernhansl T. Cloud-based Platform to facilitate Access to Manufacturing IT. Procedia CIRP 2014; 25:320–8.