Abstract: On the basis of the distribution characteristics of elevator data in high-dimensional space, an elevator fault detection and diagnosis model, which was based on the description of positive and negative double hypersphere support vector data, was proposed. The real-time elevator data were divided into healthy samples, faulty samples, and unknown drift anomalies. Support vector data description (SVDD) was used to improve the discrimination accuracy of unknown drift anomalies caused by aging of elevator equipment. SVDD was combined with the convex bilevel quadratic programming method to form a bilevel SVDD. First, the method implemented the convex interval processing of elevator operation data. The sphere and radius of hyperspheres were updated through five iterations. Then, the distance of the data to the positive or negative hyperspheres was calculated to discriminate the category of the unknown drift anomalies. In addition, the models were stacked to divide the faulty sample space to diagnose four common elevator faults. The experimental results show that the discrimination of unknown drift anomalies can reach 98. 3% of the average classification accuracy, which confirms that the proposed method can rapidly and effectively diagnose elevator faults.