SUMMARYThis paper considers a two-unit system for improving reliability and discusses a model that systematically performs diagnosis to detect faults. The system performs comparison-checking diagnosis while it is operating as a two-unit system, and performs self-checking diagnosis when it operates as a single-unit system. It is assumed that loss due to performance degradation, for example, differs depending on whether comparison-checking diagnosis or self-checking diagnosis is performed. In a model performing comparison-diagnosis at periodic intervals for each operation, the time from the beginning of system operation to detection of faults in both units is defined as a cycle. The inspection strategy of reliability theory is applied to such a model, and the expected time and total expected cost are derived analytically. The optimal diagnosis interval that minimizes the expected cost per unit time is determined. Then, assuming that the fault distribution follows the exponential distribution, the optimal diagnosis interval and the expected cost are calculated numerically, with various discussions.