Data Stream Classication for Structural Health Monitoring via On-line Support Vector Machines

Abstract—An important objective of building monitoring is to diagnose the building states and evaluate possible damage. This is a data classication problem. The building states come from many on-line sensors. Normal classication methods, such as support vector machine (SVM), cannot classify this large data stream. In this paper, the classical SVM is extended to an on-line classier (OLSVM). This SVM can classify large data stream directly. It is applied for on-line structural health monitoring. The experiment results of a lab scale prototype show the proposed algorithm can detect the damage with the data stream. This method can also be applied to big data classication, when the data set are transformed into a data stream. I