Novel Discriminant Locality Preserving Projection Integrated With Sampling Method For Fault Diagnosis

In complex industrial processes,system failures may reduce product quality,cause serious damage to the industrial system,and even endanger the safety of workers.Fault diagnosis is an effective technology to ensure process safety.However,with the rapid development of science and technology,industrial processes have gradually become highly coupled and complex,making it difficult to process fault data directly.Nowadays,Discriminant Locality Preserving Projection(DLPP)has received more and more attention in data dimensionality reduction and feature extraction.The DLPP has shown excellent performance at data dimensionality reduction,and since the discriminative information of the data is added in the dimensionality reduction,the data after dimensionality reduction is convenient for data classification.However,due to the small sample size(SSS)problem of the DLPP,the performance will be greatly reduced when reducing the dimensionality of industrial data.In order to overcome this problem,this paper proposes a DLPP algorithm based on sampling optimization,using Monte Carlo method and improved Synthetic Minority Oversampling Technique(SMOTE).SMOTE samples the original data.When using the SMOTE to sample industrial data,the traditional SMOTE algorithm has been improved in response to the limitations of the traditional SMOTE algorithm when sampling industrial data.The improved SMOTE is used to samples the original data,generates similar data samples with the same data distribution,and increases the number of data sets involved in the calculation,so that the singular matrix when calculating the dimensionality reduction matrix becomes a non-singular matrix,thereby solving the small sample problem of DLPP.It also provides a new solution for other algorithms with the same small sample size problem.In order to test the effectiveness of the proposed algorithm,a simulation experiment based on Tennessee Eastman process data is carried out.The simulation results prove that the proposed method is effective and has the excellent performance in data dimensionality reduction,and demonstrated the improvement effect of the proposed algorithm on the DLPP algorithm.