7. Outlook

Section 6.​1 illustrates that a wide range of methods and technologies is now available to master uncertainty through sufficient robustness. For both aspects, first uncertainty quantification, and second robust optimisation of components and systems, there is a need for multi-purpose, easy-to-use software frameworks. First and in more detail regarding uncertainty quantification: a software framework is needed supporting a consistent workflow from the quantification of uncertainty within the product design phase, to the propagation of uncertainty in the production phase and to the prediction of the system’s reliability in the usage phase. Within this framework, efficient probabilistic parameter calibration methods, e.g. in a Bayesian framework, shall be available to cope with the increasingly complex and computationally intensive models used in the further virtualised product design. Second and regarding robust optimisation of components and systems: available mathematical methods are currently not supported by general purpose software, and prior modelling based on human experience is needed before using it for practical problems. Both facts still inhibit Sustainable Systems Design.

We expect software technologies to close this gap in the near future. Yet, there are still open research topics. The first addresses mathematical research, the second engineering research. Above all, the mathematical tools that enable robust optimisation, as described in Chap. 6, all exploit the underlying structure of the problem in one way or the other. Extending these methods to systems for which the corresponding mathematical structure is different, lacks refined methods and therefore requires mathematical research. Thus, although robust optimisation has developed into a relatively mature field with many contributions, there are still many open research questions, in particular for problems of practical interest, such as dynamical, i.e. transient problems.

As has been seen in Chaps. 1, 3 and 6, Robust Design and the related Sustainable Systems Design as it is understood here, can be seen as solving a constrained optimisation problem with the specified systems function as one constraint to be solved for a given design space. The objective has three dimensions: effort, availability and acceptability. To master uncertainty in the customer expectation, in material or component properties, usually an increase of effort, e.g. regarding material consumption, is needed. Section 3.​5 listed seven inherent Robust Design and operating concepts that potentially offer additional freedom of design without additional cost and weight. Thus, tailored material or component behaviour could be one promising approach with graded material or component stiffness. Also the deliberate use of residual stresses potentially offers additional freedom of design without additional cost and weight.

When it comes to systems, the robustness of individual modules or components is strongly influenced by the behaviour of other components of the system. Thus, the application of Taguchi’s DoE method as one tool of the Robust Design methodology can become quite expensive. Therefore, a further important research area is the efficient and comprehensive validation of a module’s robustness under simulated realistically detuned operation and installation conditions.

In contrast to resilience, robust systems do not show recovery phases. Hence, robustness is achieved mainly by “smart” decisions made in the (A) product design prior to the product life sequence B, C, D, E. In the future, merging of data gained from experiences made in the product usage or a physical or cyber-physical product validation test with prior knowledge will become much more important. This merge will result in grey-box models enabling the mentioned “smart” decisions. Indeed, there is a need for integrating Bayesian methods with Robust Design and risk assessment in product design.

Section 3.​5 introduced the concept of flexibility and Sect. 6.​2 depicts promising approaches for mastering uncertainty by increased flexibility. At the same time, it also reveals the additional costs and complexity of design and production processes concomitant with higher flexibility. Further research work is necessary to provide either (i) smart modules or (ii) smart modularisation, cf. Sect. 3.​5, which allows for the highest flexibility at a specific minimised cost. One promising approach for future developments could be the application of lean design engineering principles to obtain flexible systems.

First, smart modularisation is an interesting application field for optimisation methods presented in this book. Second, smart modules usually incorporate semi-active or active components within complex technical systems. They offer a freedom in usage and by this cover different customer needs or expectations. However, the reliable mastering of uncertainty, e.g. by the methods presented within this book, is necessary to, on the one hand, legitimate the increased effort associated with the semi-active and active components and, on the other hand, increase the acceptability for the customer and within society.

The current driver for modularisation is the speed when scaling up as well as satisfying customer demands. Functional units are integrated into modular type packages fulfilling a functional requirement specification. Further open questions are automated documentation as well as the approval processes. From a Sustainable Systems Design perspective, as defined here, it is clear that the specified function will be a constraint, whereas the minimisation of social costs measured in energy or material consumption will be an objective. This demands the definition of metrics and the aggregation of quality KPI from the component level up to the business level. Thus, commissioning, approval and learning will be enabled by FAIR data management as specified above.

As discussed in Sect. 6.​3, robust systems do not show a recovery phase after a severe impact, whereas resilient systems do. Besides seldom exceptions, only smart agents, humans or cyber-physical modules enable a recovery phase being characteristic for dynamically resilient systems. Those agents heal severely experienced damage by having the ability to measure, react, learn and anticipate.

We can imagine that in a composed system, agents interact in such a way that each agent measures its surrounding and all agents together react in a self-organised manner. This vision can be seen as a biologicalisation of products and processes. The driving potential for the agents to act is the loss of functional quality, cf. Chap. 1. From this perspective, the recovery phase of the resilience triangle is a Continuous Improvement Process (CIP) of products or the product design phase. Only now, the latter takes place within the usage phase. Hence, in this picture of dynamic resilience (A) product design becomes integrated into the product life sequence B, C, D, E mentioned above.

There are few examples of self-healing materials, components or systems without cyber or real agents. In Sect. 3.​5, liquid sealant added to the inside of a tire was mentioned as one example: a puncture is self-healed by this sealant; a wooden boat seals itself by swelling the wood; a leather boot automatically seals small holes. In the named examples swelling is the basis for self-adaption or self-healing. In nature we observe stress-induced shape optimisation inline and online integrated in the life sequence B, C, D, E. Today, we use such shape optimisations offline in the (A) product design. Also here, the future task is to integrate (A) product design in the (C) usage phase as nature does. From this we conclude that the design of self-healing materials or self-repairing machines could be stimulated from nature. Their integration into technical systems could pave the way to the so far difficult to achieve recovery of structures.

In Sect. 6.​3, static resilience is distinct from dynamic resilience: static resilience is the property of a system predefined by the system’s design; dynamic resilience is the skill to react to a loss of functionality. The degree of static resilience e.g. of a water supply network is established in the (A) product design. The future will focus on the trade-off between static resilience and the costs achieving this static resilience. Our current research shows that there is a saturation of gained static resilience versus costs as one would expect. Still, there are open questions regarding the resilience of networks.

Mathematical tools to optimally design resilient systems have been developed for a long time, often under different names like network survivability, etc. The corresponding problems are inherently multi-level and an exploitation of the particular structure is necessary in order to be able to solve the corresponding optimisation problems. Similar to robust optimisation, the future is likely to see a refinement and extension of the available tools and hopefully software support. Moreover, incorporating learning into the systems poses interesting mathematical challenges.