This paper presents an online data and knowledge-based diagnosis method. It leverages decision trees where decisions are informed by diagnosis meta knowledge, specifically focusing on the properties of diagnosis indicators. This knowledge is used at each node to articulate a symbolic classification problem, outputting discriminating functions. The outcome is a multivariate decision tree that produces a compact model for diagnosis. The use of decision trees increases the explainability of the outcome, all the more so as one discovers the explicit formal expressions of diagnosis indicators, structured in the form of analytical redundancy relations. This article centers on the essential components required for the method to be suitable for dynamic systems. It is tested on the well-known two-tank system, showing that the performances match those of model-based diagnosis.