Model-Based System Engineering for Diagnostics

One of the goals of the CareFree 2025 project has been the development of a methodology that connects Model-Based Systems Engineering (MBSE) to diagnostics, which is the process of identifying faulty hardware components in case of a system failure. For this purpose, we started with a methodology developed earlier in de SD2Act project, and applied it to a large and complex module in a realistic industrial system: a high-volume printer developed by Canon Production Printing. We identified various opportunities for improvement, found appropriate solutions, and accordingly enhanced the methodology and its supporting tooling, the open-source library MBDlyb. The resulting methodology takes as a starting point a system model expressed using a common MBSE tool, such as Capella. Such a model describes the component hierarchy in the system, the functions that the various components implement, and the dependencies between these functions. This system model can be extended with diagnostic information, such as observables, error messages, diagnostic tests, and prior hardware fault rates. MBDlyb can then import this extended system model, construct a probabilistic model for the system, and use that to generate diagnostic procedures at design time, indicating potential observability limitations of the design. To diagnose a troubled system, observations and error messages are entered into the tool, which then calculates which components are most likely to be faulty and which tests are most useful to do next. By applying the methodology to a realistic system, we learned that it is possible to construct a sufficiently detailed system model, complete with diagnostic information, with a reasonable effort. The system model only needs to contain direct interactions among components and functions, because the more remote interactions are calculated by evaluating the probabilistic model. It is also relatively easy to adapt the model when the system design changes or when new insights are gained. In this sense, we conclude that our methodology is promising as a replacement for more traditional approaches, such as FMEA and FMECA, which require systematic manual analysis of text-based information, and are therefore laborious and error-prone. Besides the conclusions of our research, this report also contains guidelines for system developers who want to apply our methodology.