Multi-Fault Diagnosis of Rotational Machinery based on AIS and BSS

The multi-fault diagnosis of rotational machinery is one of the most difficult problems in machine fault diagnosis. When various kinds of faults co-exist, different fault symptoms are combined each other, therefore it is usually difficult to construct the mathematical model for faults and to perform reliable diagnosis. This paper proposes a multi-fault diagnosis method for rotating machinery coupling the advantages of artificial immune system (AIS), blind source separation (BSS) and data fusion theory. First, the mixed vibration signals measured by several sensors are separated into some single fault signals and noise by blind source separation. A major drawback of conventional BSS is that the contributions of acquired separation signals for source signals are undeterminant, i.e., the sum of separated signals is different from the source signal. This paper suggests an enhanced blind source separation matching source contributions of separated signals and finds new separation signals matching source contributions of separated signals. In addition, the immune detectors for diagnosing single fault diagnosis are constructed using AIS and the standard single fault samples presented by previous researchers. In order to improve the convergence of AIS and increase the variety of solution, a new objective function combining affinity information and density information is proposed. All separated signals are passed through immune fault detectors and the possibilities of the fault generation are found. Finally, the contributions of the separated signals for the source signals are found, and multi-fault is diagnosed by these contributions and data fusion theory. The comparison of presented and conventional method is performed. The effectiveness of the diagnosis method presented in this paper is verified by simulation and experiment.