The technology of multi-agent systems has been proven to be suitable for advanced distributed industrial control solutions. Many advantages are gained by distributing the control capabilities to autonomous agents, including handling more complex tasks, robustness and partial fault tolerance. However, every complex system needs flexible diagnostics, especially in the case of large industrial control systems in order to avoid large losses due to interruption of production. This paper examines flexible and complex distributed diagnostic system and the potential utilization of semantics for diagnostics. In terms of contribution, it describes fundamental concepts and approaches within HMM-based diagnostic system and semantic knowledge representation that are basis for proposed automatic HMM compilation process. This approach is illustrated on MAST platform.