Currently, there are many tribological components that require long life. In the case of unique machines such as satellites and wind turbines, parts replacement and maintenance require huge costs. Generally, machine surfaces get damaged (e.g. wear scar) during operations, which impedes stable sliding [1,2]. We establish a new surface which can avoid the damaged part in order that maintenance-free or energy-efficient mechanical components can be realized. The previous researchers have developed a method which actively controls friction by using a morphing surface. The morphing surface consists of morphing diaphragms and some supportive parts. Inspired by previous research, this study established a new friction stabilization method that controls the contact position when partial damage occurs on the counter surface. We applied genetic algorithm (GA), one of the artificial intelligence technology, to realize efficient contact point control. In order to stabilize friction, we searched for an optimum pattern of contact points that could avoid the damaged parts by applying GA. By repeating this searching method, a contact point pattern that avoids the damaged parts can be found. As results of experiments, we have succeeded in stabilizing friction by searching for contact point patterns that automatically avoid damage in the friction test using GA. However, even if the search is performed efficiently using GA, the calculation times is necessary about 10–20 times. Therefore, it is necessary to shorten the time before the damage is avoided. We report on a method to shorten the search time for the optimal contact point pattern using GA as the latest progress. In this method, some simple contact point patterns are used to roughly identify the locations where damage might be present during some rotations. At the section of creating the 50 contact patterns, this probability of damage existence is used to create the 50 contact patterns that are more likely to avoid the damaged parts. By adding this process before the friction test using GA, we succeeded in reducing the number of calculations from 10–20 times to 2 times.