Most field systems today are maintained systems. Decisions about reliability of these systems are based on field or test failure data. For highly reliable and good quality systems the failures are few and far between. To understand their failure behaviour and reliability characteristics will take a long time. If the prototype systems are subjected to tests with higher levels of stresses leading to accelerated deterioration then the required information can be obtained in a shorter frame of time. Such tests are called accelerated tests and the failure data generated by these tests along with the stress levels are modelled by accelerated failure time models. These can then be used to detect latent failure modes, to demonstrate and improve their reliability, and further extrapolated to predict their expected failure behaviour and reliability under normal conditions. Very little work has been carried out in this area till date. In this paper an empirical imperfect repair accelerated failure time process based on times to failure and various levels of stress is proposed for a maintained system. The imperfect repair processes postulated are arithmetic reduction of intensity and arithmetic reduction of age processes with memory one. The models are presented along with their inferences and applications through a simple case study. These models play a vital role in predicting the reliability of maintained systems.