#SPAICER — The Vision

AI-based Resilience Management in Production Engineering

The Vision, Image: © WZL | Daniel Trauth & Semjon Becker
Co-Authors: Thomas Bergs (WZL), Andreas Feuerhack (senseering), Christian Gülpen (Time), Sabine Janzen (DFKI), Wolfgang Maass (DFKI), Patrick Mattfeld (WZL), Philipp Niemietz (WZL), Frank Piller (Time) — alphabetical order

What is the purpose of this?

The past has shown that a lack of resilience management can lead to high costs in the production industry. Supply bottlenecks or disputes with suppliers can lead to costs of up to EUR 100 million per week [1] or EUR 410 million per bottleneck [2]. Costs that ultimately have to be covered by customers or employees. A resilience management has the task to point out alternatives in time. However, there is no such resilience management for production industry.

The Vision

The vision of the #SPAICER research project is to develop a framework model for AI-based resilience management for production companies in production networks. On the basis of hybrid AI platforms and accompanying economic and legal usage concepts, the basis for a Smart Resilience Service ecosystem for different stakeholders in production networks will be created.

Timeline and Background

In Q1/2019 the Federal Ministry for Economic Affairs and Energy (Bundesministerium für Wirtschaft und Energie, BMWi) launched an Innovation Competition called Artificial intelligence (AI) as a driver for economically relevant ecosystems. Within that AI competition, DFKI x RWTH (TIME and WZL) submitted their concept idea

#SPAICER: Scalable adaptive production systems through AI-based resilience optimization

and won together with 34 other applicants. We are now asked to work out the concept in the next 4 months in such a way that we can then realize the concept in a three-year implementation phase. #SPAICER will be evaluated by an independent jury in August 2019. If successful again, then #SPAICER2 will start January 1st, 2020.

Resilience in a nutshell

Resilience describes the ability to withstands/resist life changing situations without lasting disruption. While in the past resilience was mainly used in psychology, our approach is to transfer the concept of resistance to technical processes. Technical processes are also subjected to fluctuations and uncertainties that affect process quality. Our key question is: How much disturbances can a production engineering system tolerate before lasting effects occur? And this question will be answered by #SPAICER.

The #SPAICER concept idea

Germany is a country of production. After services, manufacturing output accounts for by far the second highest share of gross value added, at just under 26 %. Compared to the 2014 reporting period, the gross value added of the relevant manufacturing industry increased from 593.6 to 674.3 billion euros in the 2017 reporting period. Compared to the previous year, there was an increase of 3 percent [3]. This makes the manufacturing industry one of the core guarantors of growth and prosperity in Germany. In the USA, USD 12 trillion in intermediate products were purchased by companies in 2007 [4].

Reliability and cost efficiency are key differentiators, especially in global value networks with multiple relationships and dependencies. Shorter technology and product life-cycles and rapidly changing customer requirements leave less and less time to build quality-optimized logistics and production chains. Convertible production in the sense of industry 4.0 leads to the necessity of faster changeover and learning processes. This increases the susceptibility to malfunctions and errors. For example, the loss of production caused by the earthquake in Fukushima led to a 1.2 % reduction in Japan’s gross national product [5].

SPAICER’s promise of benefits for production companies

This is exactly where the #SPAICER research project comes in, so that production companies can specifically apply artificial intelligence (AI) methods in heterogeneous production contexts in order to optimize resilience to internal and external disturbances and changes in the context of an industry 4.0 production network with the aim of being able to act flexibly and adaptively in global competition.

Vision of a Smart Resilience Management Ecosystem, Image: © DFKI | Wolfgang Maaß

Manage disturbances and changes

Hamel and Välikangas call the ability of a company to permanently adapt to large, internal and external changes and disturbances in complex, rapidly changing production networks the search for resilience [6].

  • Disturbances in production companies concern the supply of material of insufficient quality, leakage of lubricant lines, damage to machines (extension of predictive maintenance), power failure or illness of employees [7. Furthermore, Disturbance scan can be predictable or not predictable [8].
  • Changes usually affect companies from outside, such as systematic market changes in the form of innovative technologies (e.g. shared production lines or 3D printing), changes in demand behavior or abrupt changes in the political or financial system [9,10]. In addition, changes can lie in products themselves, such as their quality, branding, and manufacturing (in-)efficiency. Moreover, there are changes in political regulation, the labor market and the environment itself [11].

Bridging Psychology and Production Engineering

In order to optimize the resilience of production companies and networks, AI-based services, so-called Smart Resilience Services (SRS), are needed, which

  • predict disturbances and changes in extremely heterogeneous, distributed and permanently changing machines and technology environments close to real time,
  • identify optimized options for action and
  • propagate resilience-optimizing information in the production network.

Due to production-oriented requirements for latency times, data protection and performance, innovative hybrid architectures are required that optimize the execution of AI technologies in a cloud and decentralized environments (AI on the Edge) directly on different production machines. The technological side requires an integrative framework model for resilience management as part of production management. An ecosystem of Smart Resilience Management (SRM) emerges when manufacturing companies are able to search, obtain and use smart resilience services on demand, which in turn requires standardized data and programming interfaces (APIs) and open technological architectures.

What #SPAICER does

The automation, as well as the cooperative and collaborative support of analytical resilience management by means of AI methods, establishes adaptability and receptivity. Particularly suitable are methods of machine learning to derive forecasts and recommendations for action from data, and planning and inference methods to be able to use structured knowledge in a controlled manner.

In addition, there are hybrid approaches which combine symbolic knowledge representations and machine learning approaches [12]. Currently, AI applications are successfully used for speech recognition and translation in the cloud. The demand for AI systems that work on end devices or embedded systems (AI on the Edge) and in the cloud is also growing. Such hybrid architectures face several challenges:

  • (1) large performance differences between edge devices and cloud servers,
  • (2) edge devices are mostly heterogeneous (e. g. from ARM CPUs to GPUs), which complicates application development,
  • (3) hardware and software update cycles of edge devices are slower in the context of production than in data centers,
  • (4) reduction of the growth of storage capacities with simultaneous growth of generated data, data storage is generally cost inefficient [13].

At the level of AI research, #SPAICER will focus in particular on hybrid AI-oriented architectures, which distribute AI-based services to edge devices (e.g. production machines) and the cloud according to their performance and data protection requirements. Portable software designs for AI systems are developed for such heterogeneous system landscapes. For this purpose, innovative software stacks are to be developed which abstract from hardware functions and enable access via common APIs [14,15].

One challenge is the partitioned execution of machine learning models. In particular, a model composition is to be investigated, whereby the execution of lighter, but less accurate models on terminal devices and computationally intensive models is carried out with higher accuracy in the cloud. Since data capacities are strongly limited to edge devices, adaptive compression methods are analyzed. Explainable AI (human AI communication) is the main focus.

The #SPAICER Use Case

In #SPAICER, the fine blanking manufacturing system, the upstream and downstream process as well as the supply chain will be examined as examples. Fine blanking is a highly economical sheet metal cutting process for large series production, whereby fine blanked components have a very high surface quality. The use of precisely blanked components is often safety-critical, for example as brake caliper carriers or belt straps in automobiles. In total, however, there are up to 250 fine blanked components in a luxury car.

Finblanking press Feintool XFT 2500 speed. Image: © WZL | Winandy

The first step in this manufacturing system is to investigate resilience management with regard to internal faults. Single faults in the output of the fine blanking process, for example, caused by fluctuating material quality or wear, can propagate through the process chain to the finished product. Therefore, resilience management is subsequently investigated with regard to disturbances within the entire process chain of a component.

The cooperation of companies in the fine blanking industry in the resilience context depends above all on the ability of companies in value-added networks to understand each other and to use the functions of the other [16].

Conclusion

In concrete terms, resource manufacturers, for example, would be informed directly and retrospectively about fluctuating product quality, which could improve their product development for the future. Material suppliers could send materials that vary in their properties to targeted customers for whom the properties are still within the tolerance range. If resilience is understood as a function, resilient production networks, such as the fine blanking industry, are able to functionally propagate analyses and measures for optimizing resilience capability in networks, so that companies in the network can adapt to each other’s changes.

References

[1] https://www.welt.de/wirtschaft/article157764510/Lieferstopp-kostet-VW-100-Millionen-Euro-pro-Woche.html

[2] https://www.zeit.de/mobilitaet/2016-08/volkswagen-vw-golf-wolfsburg-produktion-stillstand

[3] Datenreport (2018) — Ein Sozialbericht für die BRD, Bundeszentrale für politische Bildung, 2018.

[4] Streitwieser, Mary L. (2009), A Primer on BEA’s Industry Accounts. Survey of Current Business, 40–52.

[5] Carvalho, V. M., Nirei, M., Saito, Y., & Tahbaz-Salehi, A. (2016). Supply chain disruptions: Evidence from the great east japan earthquake.

[6] HAMEL, G., & VÄLIKANGAS, L. (2003). The quest for resilience. Harvard business review, 81(9), 52–63.

[7] Selcuk, S. (2017). Predictive maintenance, its implementation and latest trends. Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture, 231(9), 1670–1679.

[8] Madni, A. M., & Jackson, S. (2009). Towards a conceptual framework for resilience engineering. IEEE Systems Journal, 3(2), 181–191.

[9] Kusiak A. Smart manufacturing must embrace big data. Nature 2017;544(7648):23–5.

[10] Weller, C., Kleer, R., & Piller, F. T. (2015). Economic implications of 3D printing: Market structure models in light of additive manufacturing revisited. International Journal of Production Economics, 164, 43–56.

[11] Yeung, H. W. C., & Coe, N. (2015). Toward a dynamic theory of global production networks. Economic Geography, 91(1), 29–58.

[12] Nickel, M., Murphy, K., Tresp, V., & Gabrilovich, E. (2016). A review of relational machine learning for knowledge graphs. Proceedings of the IEEE, 104(1), 11–33.

[13] Stoica, I., Song, D., Popa, R. A., Patterson, D., Mahoney, M. W., Katz, R., … & Goldberg, K. (2017). A berkeley view of systems challenges for ai. arXiv preprint arXiv:1712.05855.

[14] Garcia Lopez, P., Montresor, A., Epema, D., Datta, A., Higashino, T., Iamnitchi, A., … & Riviere, E. (2015). Edge-centric computing: Vision and challenges. ACM SIGCOMM Computer Communication Review, 45(5), 37–42.

[15] Dinh, H. T., Lee, C., Niyato, D., & Wang, P. (2013). A survey of mobile cloud computing: architecture, applications, and approaches. Wireless communications and mobile computing, 13(18), 1587–1611.

[16] D. Chen , G. Doumeingts , F. Vernadat , Architectures for enterprise integration and interoperability: past, present and future, Comput. Ind. 59 (7) (2008) 647–659 .

Get in contact

You have questions or want to join/contribute in any way?

DFKI

Ask for Sabine or write an E-Mail | Follow DFKI on Twitter

TIME

Ask for Christian or write an E-Mail | Follow Christian on Twitter

WZL

Ask for Daniel or write an E-Mail | Follow Daniel on Twitter

Footer, Image: © WZL | Daniel Trauth & Semjon Becker