A Step-By-Step Guide

Building a complete solution from an existing science use case requires dissecting the science use case by the open or closed loop control problem or problems it contains. This Section describes the involved steps and discusses the individual decision parameters in more detail. The following Pattern Compositions Section discusses additional considerations when composing different design patterns, such as due to multiple loop control problems. Each loop control problem needs to be identified, including its properties and hardware/software architectural features. A step-by-step decomposition process would work as follows:

  1. Clearly define the experiment or experiments that are being performed

  2. Identify the loop control problem or problems that exist for each experiment.

  3. Classify each loop control problem by a strategic pattern.

  4. Identify the individual components of each loop control problem and associated strategic pattern.

  5. Classify each loop control problem by an architectural pattern that matches its strategic pattern.

  6. Match the identified components with the components of the architectural patterns.

  7. Design the hardware/software architecture of the solution based on the architectural patterns and the corresponding matched components, using the pattern properties as design guidelines.

What is the experiment?

It is important to clearly define the experiment or experiments, as the wrong definition ultimately leads the designer down the wrong path. It is often easier to think of an experiment as a concrete test process that demonstrates a specific known truth, examines the validity of a specific hypothesis, or determines specific properties of something. Clearly identifying the experiment devices, such as sensors, actuators, instruments and robots, is part of that definition as well. It is quite possible that one experiment in a laboratory tries to accomplish multiple objectives, in which case a single multi-objective experiment could be split up into multiple experiments, especially if it involves a workflow or completely separated loop control problems. There is no hard rule on this and any such split would be on a case-by-case basis.

Which loop control problems exist?

Separating out what is being controlled and how is the key to identifying the loop control problem or problems that exist for each experiment. In pretty much all cases, there is some type of simple open loop control, as described in the Experiment Control strategic pattern. Additional loop control problems may exist that may extend the simple open loop control, such as to the Experiment Steering strategic pattern, or uses/relies on the simple open loop control, such as with the Design of Experiments strategic pattern. There also may be multiple loop control problems for the same experiment, such as a combination of the the Experiment Steering and Design of Experiments strategic patterns. Similarly, a multi-objective experiment may have multiple loop control problems for different parts of the experiment, potentially requiring it to be split up into multiple experiment. Obviously, a multi-experiment workflow may have loop control problems for each experiment in the workflow. Pattern combinations that solve such issues are discussed in the Pattern Compositions Section.

Who is in control?

The science use case design patterns have one controller component and some have an additional planner component. These are not necessarily physical standalone components. Instead, an analyzer may already contain the decision-making logic and also act as a controller or planner. Similarly, the controller or planner may require human input or may be a human itself. While the goal is to reduce human-in-the-loop requirements with machine-in-the-loop capabilities, this may be a process that requires a transition and some human-in-the-loop requirements may not necessarily completely eliminated.

Which strategic pattern?

The key differences in features between the 4 strategic patterns are (1) no feedback, (2) feedback for the same experiment, (3) feedback for the next experiment, and (4) workflow of multiple experiments. If there is no feedback, then Experiment Control is the right strategic pattern. If there is feedback for the same experiment, such as changing a parameter based on a measurement to observe how that or another measurement changes, then Experiment Steering is the right strategic pattern. If there is feedback for the next experiment, such as to change the parameters and re-run the experiment, then Design of Experiments is the right strategic pattern. There are experiments, where the experiment plan constantly evolves as the experiment is performed, based on measurements. In this case, either Experiment Steering or Design of Experiments may be used, whichever is closer. In this case, using Design of Experiments splits the experiment into multiple separate experiments with different experiment plans. Multi-Experiment Workflow is used whenever there are multiple experiments without feedback. There could be a greater feedback loop over multiple experiments in a workflow. In this case, a separate strategic pattern is employed (see Pattern Compositions).

What is local? What is remote?

The architectural science use case design patterns distinguish between local and remote components based on communication delay. Any potentially significant communication delay to a component makes it a remote component. The term “significant communication delay” is purposely not clearly defined to give designers room for interpretation. There may be other reasons for defining a component as remote, such as when a component is physically located at an entirely different location that does not necessarily incur a significant communication delay but requires a special way of communication. A human that acts as a planner and communicates with the rest of the system via e-mail or a GUI would likely also be considered a remote component.