This paper proposes a bivariate process-based model for maintenance and inspection planning of a parallel system, consisting of two components whose states evolve in one of three possible states: normal (0), satisfactory (1) and failure (2). The changes of states driven by a non-homogeneous Markov process are detected only by inspections and repair actions are determined by the observed state of the bivariate process. Outperforming maintenance strategies and other classical maintenance policies, the paper aims at minimizing the long-run average maintenance cost per unit time by deriving optimal inspection intervals and a preventive replacement threshold. A numerical example is given to illustrate the proposed model and examine the response of the optimal solutions to system parameters. The model explored here provides the framework for further developments.