A novel fault diagnosis method based on improved AdaBoost and Kernelized Extreme Learning Machine for industrial process

For the industrial process, fault diagnosis technology is an effective and important way to ensure the process safety and prevent the accidents. With the increasing complexity of modern industrial process systems, some tiny faults in the system should be detected early and eliminated in time. Otherwise, it may cause the failure and paralysis of the whole system. At the background, higher requirements are put forward for the accuracy and self-adaptive capacity of fault diagnosis. To solve the problem, a novel method based on improved AdaBoost and Kernelized Extreme Learning Machine (KELM) is proposed. Firstly, the error feedback mechanism is introduced to traditional Extreme Learning Machine (ELM) network. Though dynamically adjusting the hidden layer output, this neural network error can substantially decrease. Secondly, it is considered that AdaBoost algorithm can raise the data classification ability by adjusting the sample weights and weaken the classifier weights, then the established ELM with error feedback is used as weak classifier to increase the ability of characteristic extraction for AdaBoost. Thirdly, KELM carries out effective optimizing training on the data of the improved AdaBoost with extracted data features. Then, it performs fault diagnosis for optimizing data. The performance is tested and verified by standard UCI data sets and TE simulation process. The experimental results show that the proposed method achieves better performances in fault diagnosis than the traditional approaches.