Many approaches often handle fault detection and identification tasks separately and require retraining when new fault samples or classes are introduced. To address this, we propose an incremental support vector data description scheme using a Gaussian kernel function, integrated within a unified framework for industrial fault diagnosis. This approach incrementally updates the decision boundary using only the original support vectors and newly acquired samples. To enhance detection accuracy, an adaptive threshold and restructured radius are introduced. For identification, hyperspheres representing known classes are constructed using a decision tree, with new samples that do not match any existing class being identified as belonging to an unknown class. This method enables simultaneous fault detection and identification without the need for time-consuming retraining. Experimental results validate the effectiveness and superiority of this incremental learning approach in fault diagnosis.