Unlike the major equipment used in power plants, auxiliary equipment usually does not possess a real-time system to analyze the machine condition. Therefore, detecting the fault of such auxiliary equipment in advance is difficult. Thus, the diagnosis of auxiliary equipment at a less cost is important for minimizing the downtime due to the fault of the equipment. In this paper, we introduce a diagnosis method for auxiliary equipment in power plants using rule-based and deep-learning algorithms. First, we calculate the probability of cause of a fault from current symptoms by using the rule-based algorithm. The rule used in this algorithm is established based on expert experience. We then conduct orbit detection using a convolution neural network. This algorithm self-learns the filter to classify orbit images as normal, rubbing, and unbalanced. The weakness of the deep-learning algorithm can be compensated by combining the results of the aforementioned methods.