Intermittent process fault detection method based on double-weight multi-neighborhood preserving embedding algorithm

Abstract

The invention provides an intermittent process fault detection method based on a double-weight multi-neighborhood preserving embedding algorithm, and the method comprises the following steps: (1) collecting intermittent process data of a plurality of batches under normal working conditions, and forming three-dimensional training data X; (2) expanding the acquired three-dimensional training data Xinto two-dimensional data and performing standardization processing; (3) establishing a double-weight multi-neighborhood preserving embedding model, solving a mapping transformation matrix A, and obtaining a dimension reduction data matrix Y according to Y = ATX; (4) establishing statistics of a Hutlin statistical model T2 and a square prediction error statistical model SPE under normal data, andsolving control limits of the statistics; (5) collecting online intermittent process data to form test data Xtest, and carrying out expansion and standardization processing on the test data Xtest; (6)projecting the test data through the mapping transformation matrix A to obtain a dimension reduction data matrix Ytest; and (7) solving the statistical magnitude T2 of the Hutlin statistical model and the statistical magnitude SPE of the square prediction error statistical model of the test data, and judging whether a fault occurs.

Description

technical field [0001] The invention belongs to the technical field of industrial process monitoring and relates to an intermittent process fault detection method based on a double-weight multi-neighborhood preservation embedded algorithm. Background technique [0002] The intermittent process data has dynamic characteristics and does not satisfy the independent assumption of process sampling. If traditional methods are used for monitoring, it will inevitably lead to the existence of high false negative rate. In response to this problem, Ku first proposed the Dynamic Principal Component Analysis (DPCA) algorithm, which established a fault detection model through an augmented matrix containing sampling values ​​at the current moment and historical moments, taking into account the time series correlation between process data, and improved the accuracy of fault detection. detection rate. Since then, a series of multivariate statistical methods based on augmented matrices have ...

AI Technical Summary

Problems solved by technology

However, the above method reveals the global structure of the data, ignoring the local structure of the data, resulting in poor feature extraction effect, which directly affects the fault detection effect of the algorithm

[0003] In recent years, the NPE algorithm has been widely used in process monitoring because it can mine the local structure of the data while reducing the dimension. The matrix is ​​applied to the NPE algorithm, and the DMNPE algorithm is proposed, and it is successfully applied to the fault detection of the intermittent process. However, the data of the intermittent process is abundant, and the method of using the augmented matrix reduces the computational efficiency of the algorithm.

Miao proposed a time-series extended NPE (TNPE) algorithm, which reconstructs sample points by selecting k temporal neighbors for each sample point, so as to dig out the relevant information between sampled data. 2 and SPE statistics to monitor the performance of the process, which can better solve the problem of data dynamics. However, this method only mines the data in time, ignoring its spatial structure, and the NPE algorithm when calculating the reconstruction weight , does not consider the size order of the neighbors, resulting in the loss of some important information

Images