Transient faults belong to a wide-spread class of faults typical for control systems. These are the faults that only appear for a short period of time and might reappear later. However, even by appearing for a short time, they might cause dangerous system errors. Hence, designing mechanisms for tolerating and recovering from the transient faults is an acute issue, especially in the development of the safety-critical control systems. In this paper we propose formal development of a software-based mechanism for tolerating transient faults in the B Method. The mechanism relies on a specific architecture of the error detection actions called the evaluating tests. These tests are executed (with different frequencies) on the predefined subsets of the analyzed data. Our formal model allows us to formally express and verify the interdependencies between the tests as well as to define the test scheduling. Application of the proposed approach ensures proper damage confinement caused by the transient faults. Our approach aims at the avionics domain by focusing on formal development of the engine Failure Management System. However, the proposed specification and refinement patterns can be applied in the development of control systems in other application domains as well.