The Research Of Fault Diagnosis Method Based On Ensemble Algorithm

Fault detection and diagnosis technology plays an important role in modern industry, which can ensure the reliability and safety of industrial systems. Nowadays, the industrial process is more complicated and is expanding in scale. Meanwhile, the types of faults in industrial process are variety and diversification. The classical single fault diagnosis methods can not meet the actual industrial need. Therefore, the ensemble fault diagnosis method that combining several single fault diagnosis methods has become the research focus and hot spot. In order to optimize the performance of fault diagnosis, based on analyzing and summarizing the relevant literature, the thesis proposed a framework of ensemble fault diagnosis, namely, "Denoising pretreatment based on lifting wavelet + recursive type feature extraction + fault diagnosis and classification". The main contents include the following aspects:(1) In complex industry, take the penicillin fermentation process as the research object, through fault settings and data collection, the adequate data for the research of the method are getted. In order to minimize the impact of noise on fault detection and diagnosis methods, the lifting wavelet(LW) method was adopted to denoise the noise data.(2) For complex industrial processes, a ensemble fault diagnosis method based on LW-RPCA-IPNN algorithm was proposed. The algorithm use the sample-wise update of Recursive Principal Component Analysis(RPCA) to extract features, by increasing the number of hidden layer neurons, then use Incremental Probability Neural Network to classify the training data. Take LW-IPNN and LW-RPCA-IPNN method as a comparison, after extracting features use RPCA, the fault diagnosis accuracy rate has raised obviously and the speed of diagnosis has improved greatly. Experiments show that the ensemble algorithm is effective.(3) Another ensemble fault diagnosis algorithm was proposed, namely, LW-RPCA –RLSSVM. By using the RPCA to extract the features of the preprocessing data and using the Recursive Least Squares Support Vector Machine(RLSSVM) to classify data diagnosis, the ensemble fault diagnosis method was resrached. Through comparing with the existing method, the result show that the ensemble fault diagnosis algorithm has very big promotion in the quickness and accuracy of diagnosis, the diagnosis effect is far superior to a single fault diagnosis method.(4) In the Lab VIEW environment, based on the ensemble fault diagnosis algorithm, the experiment application platform was set up, the theoretical methods and experimental validation were applied to the industrial platform, the proposed ensemble fault diagnosis methods were researched and compared.Theoretical analysis and experimental research show that the two proposed ensemble algorithms are well in fault diagnosis performance, they have increased significantly in the quickness and accuracy of diagnosis.They are superior to the single fault diagnosis method on the performance.