Warranty forecasting of repairable systems is very important for manufacturers of mass produced systems. It is desired to predict the expected number of failures after a censoring time using collected failure data before the censoring time. Moreover, systems may be produced with a defective component resulting in extensive warranty costs even after the defective component is detected and replaced with a new design. In this paper, we will present a method to predict the Expected Number of Failures (ENF) of a repairable system using record data over a time interval of observation. The record data is used to calibrate a Generalized Renewal Processes (GRP) model which accounts for different production and support patterns as explained below. Manufacturing of products may exhibit different production patterns with different failure statistics through time. For example, vehicles produced in a particular month may have a different failure intensity compared to vehicles produced at a different month. This may be due to supply chain differences, different skills of production workers, etc. In addition during the warranty period, there may be a time instance called “clean point” where a defective component or subsystem is detected and replaced with a new more reliable component or subsystem for all systems produced thereafter. This introduces different support (warranty) patterns before and after the detection of the “clean point.” We will present a modified generalized renewal process model for warranty forecasting of systems with or without a “clean point” behavior and demonstrate its capabilities using vehicle production data. The presented model will also account for production patterns with different failure statistics.