The process industry is confronted with rising demands for flexibility and efficiency. One way to achieve this is modular process plants, which consist of pre-manufactured modules with their own decentralized intelligence. Plants are then composed of these modules as unchangeable building blocks and can be easily re-configured for different products. Condition monitoring of such plants is necessary, but the available solutions are not applicable. The authors of this paper suggest an approach in which model-based symptoms are derived from a few measurements and observers that are based on the manufacturer’s knowledge. The comparisons of redundant observers lead to residuals that are classified to obtain symptoms. These symptoms can be communicated to the plant control and are inputs to an easily adaptable diagnosis. The implementation and validation at a modular mixing plant showcase the feasibility and potential of this approach.