Challenges in applying synthesis-based engineering
A synthesis-based engineering approach has many advantages over a more traditional engineering approach. However, there can be challenges when embedding such a new approach into industrial practice. It is essential to be aware of them, and manage them explicitly.
Most of the challenges that apply to a synthesis-based engineering approach also apply to other model-based engineering approaches, including verification-based engineering.
The following challenges are discussed:
Change in way-of-working
The use of model-based engineering, the modeling of behavioral specifications and control requirements, and the use of formal techniques such as supervisory controller synthesis, requires a certain mindset, knowledge and skills. It is important that personnel with the appropriate knowledge and expertise is present in a company. Having a team of properly trained and experienced experts that can assist with and steer the introduction of new techniques is essential. If a company does not have such experts, they could hire them. However, training and retraining for (part of) the existing personnel is often also required.
Furthermore, model-based engineering partly also requires a different way of working compared to traditional engineering approaches. It is important to understand the effects on the company’s development process, as well as its culture. The various pros and cons must be evaluated, and any impediments must be identified and addressed.
A transition like this will not happen in a day. Sharing experiences with other parties that have gone through a similar transition and/or are going through one can be of great benefit. Another way to reduce the risks is contracting an external party to help guide the process.
Furthermore, risks can be reduced by step by step introducing the changes to the development process, introducing more and more elements of model-based, verification-based and synthesis-based engineering. This way models become more and more leading, throughout the development process. For instance, you could follow these steps:
Start modeling (the requirements): Increase the quality of requirements by specifying them formally in a model-based way, during early development phases. This can already be combined with for instance simulation, to produce unambiguous specifications, leading to less mistakes and reduced rework. In this first step, the resulting requirements can still be put in a document and implemented manually.
Models as single source of truth: Formally but manually specify the controller model in a model-based way, based on the formal requirements. From the model, automatically generate the controller code. This is a step towards making the model the single source of truth.
Embrace formal methods: In this step, employ more formal methods to go beyond simulation and testing. Use formal methods that have more guarantees on completeness. For instance, use formal verification to guarantee that all specified requirements are satisfied in every conceivable situation. At the end of this step, you could fully adopt model-based and verification-based engineering.
Adopt synthesis-based engineering: Use supervisory controller synthesis, and fully adopt synthesis-based engineering.
Tool support
Synthesis-based engineering requires tool support to model plants and requirements, to synthesize supervisors, perform simulation, generate code, etc. Given that automation and computer-aided design are core principles, this is simply not feasible without appropriate tool support. Companies should consider various aspects regarding the tools they use, such as the following:
Tools ideally support as much of the development process of supervisory controllers as possible.
Consider how to integrate the synthesized supervisors into the system.
Consider how active the community around the tool is.
Consider whether commercial support is available.
The selected tools should also be used in the right way. For instance, naively applying synthesis and trying to obtain a single monolithic supervisor for larger and more complex systems will likely not scale very well. Employing the proper techniques for the given situation is essential to mitigate such concerns.