Chaotic information-geometric support vector machine and its application to fault diagnosis of hydraulic pumps

Fault diagnosis of rotating machineries is becoming important because of the complexity of modern industrial systems and the increasing demands for quality, cost efficiency, reliability, and safety. In this study, an information-geometric support vector machine used in conjunction with chaos theory (chaotic IG-SVM) is presented and applied to practical fault diagnosis of hydraulic pumps, which are critical components of aircraft. First, the phase-space reconstruction of chaos theory is used to determine the dimensions of input vectors for IG-SVM, which uses information geometry to modify SVM and improves performance in a data-dependent manner without prior knowledge or manual intervention. Chaotic IG-SVM is trained by using the dataset from the normal state without fault, and a residual error generator is then designed to detect failures based on the trained chaotic IG-SVM. Failures can be diagnosed by analyzing residual error. Chaotic IG-SVM can then be used for fault clustering by analyzing residual error. Finally, two case studies are presented, and the performance and effectiveness of the proposed method are validated.