The present invention discloses a multi-source fault detection and diagnosis method and device. The method includes the following steps: obtaining historical data in a normal state of a system, constructing a multi-level knowledge map in a normal state, and mining a deep-level association path of the multi-level knowledge map in a normal state; obtaining data to be detected in a current to-be-detected state of the system, constructing a multi-level knowledge map under the to-be-detected state, and mining the deep-level association path of the multi-level knowledge map in the to-be-detected state; determining whether the data to be detected is in a fault state; if the system is in a fault state, a fault variable is determined, constructing a multi-level knowledge map fault model under the to-be-detected state; and performing multi-source fault diagnosis by using the multi-level knowledge map fault model and combing with a Bayesian theory.