Fault detection of non-Gaussian processes based on modified independent component analysis

A new process monitoring method based on noisy time structure independent component analysis

“…The mixing matrix is then estimated by  = (UQ) − 1 . To conduct fault detection, the two monitoring statistics are defined as follows [9,11,[15][16][17][18]24]…”

“…In the NoisyTSICA-based monitoring method, the number of the dominant ICs is selected as c = 7 so that the cumulative sum of the dominant ICs' kurtosis absolute values is also above 90% of the cumulative sum of all the extracted ICs' kurtosis absolute values. Both the choices are on the basis of the commonly used cumulative percent variance (CPV) criterion [11,33]. The parameter settings for Eq.…”

“…Different from PCA, ICA can further utilize the higher-order negentropy statistic [11] or the second-order time-delayed covariance statistic [12] to recover mutually independent non-Gaussian latent variables called independent components (ICs) from the original measured variables and can be applied to deal with non-Gaussian processes which are more practical in the real-world manufacturing environment [13]. Kano et al [14] proposed an ICA-based univariate statistical process monitoring (USPM) method and demonstrated its superiority over the PCA-based MSPM.…”

A Review of Industrial Fault Diagnosis Based on Data-driven Methods

“…This method is applied to the fault diagnosis in the TE process, which can effectively capture the nonlinear relationship in the process variables and show the superior fault detection ability. The [23] first analyzes some shortcomings of the original ICA method, and then proposes a ICA method based on particle swarm optimization algorithm (PSO-ICA), the fault diagnosis method is applied to the TE process, showing that PSO-ICA method can effectively capture the independent component of process variables.…”

Nonlinear Diversified Processes Monitoring

“…Statistical-based multivariate monitoring algorithms [5][6][7][8][9] such as principal component analysis (PCA) [10,11], partial least squares (PLS) [12], and independent component analysis (ICA) [13][14][15] have been extensively used to analyze the process data and find the further relationship between variables [16].…”