Cyber-Physical Systems (CPS for short) enjoy increasing popularity nowadays. Usually, they interconnect two worlds. One of them ("P") contains devices connected to the physical world for data collection and actuation. The other one ("C") is in charge of data processing and forwarding the results via the Internet or other networks. Applications of CPS include logistics, smart factories or even our homes.
CPS information processes are highly complex because they are typically geographically and functionally distributed. Their intelligence originates in the information fusion and co-processing from many subsystems. Linking to the Internet provides a basis for resource-intensive tasks (for example cloud) and on the other hand, enables the integration of data and knowledge of the Internet.
Architectural and functional complexity over a variety of devices and components characterize these systems. Many of the CPS applications execute safety-critical tasks because the interaction with the physical world can amplify the impact IT errors to a catastrophic level. Accordingly, dependability is of a top priority in their development.
My goal was to develop a model-based method checking the operation of CPSs by visually modelling their data processing and resource utilization. Visual exploration analysis is a useful engineering method for exploring and controlling the internal behaviour of complex systems. Adaptive exploration sequencing the individual steps of evaluation based on existing system models (error localization) as a guideline is an efficient method, for instance, for diagnostics.
Visual methods are also useful for fault detection and fault diagnostics in industrial processes; The required diagnostic depth depends on the steps to mitigate the problem and restore the continuity of manufacturing. For example, in a high-availability system, it is common to diagnose the fault roughly first and switch from a defective block to a faultless one. The next step is fine-granular diagnostics aiming at fault location within the previously separated defective block.
Visualization-based diagnostics should support each phase of fault management. For example, a CTO is interested in only what production line is defective to redirect production to a good line. Then the maintenance supervisor is looking for the exact location of the fault to start the repair. The diagnostic algorithm must be able to quickly locate the error with the diagnostic resolution corresponding to the actual demand.
To achieve this, I adapted the well-proven Integrated Diagnostics approach which enables to control the diagnostic of the data flows and local tests of the CPS. The nodes of the underlying test graph are the steps of the processes, the inputs, the outputs, and the tests, while edges between them represent the flow of information. Then following the actual error signal with the help of the graph, we can locate the error according to the required diagnostic depth.
Adaptive visual diagnostics guided by Integrated Diagnostics provides the basis for effectively check CPS systems and a quick fault localization with the help of existing and new visual tests.