A Fault Classification Method Based on One-Dimensional Multi-way Convolutional Neural Network

Abstract

The invention discloses a fault classification method based on a one-dimensional multi-channel convolutional neural network, belonging to the technical field of industrial process monitoring. This method improves the traditional two-dimensional convolutional neural network, uses multi-channel parallel one-dimensional convolutional neural network along the variable direction, deconstructs the mutual correlation between variables, and performs convolution on each variable independently , pooling and extracting time series feature information, the extracted features are more diversified and more robust, overcoming the sensitivity of the traditional two-dimensional convolutional neural network to the prior arrangement of variables in the input data, and more suitable for complex, high-level industrial process data; experiments show that the fault classification model based on the one-dimensional multi-channel convolutional neural network training provided by the present invention can effectively classify the faults of industrial process data, and has a higher generality than commonly used models. ability.

Description

technical field [0001] The invention belongs to the technical field of industrial process monitoring, and more specifically relates to a fault classification method based on a one-dimensional multi-channel convolutional neural network. Background technique [0002] Fault classification technology plays an important role in root cause diagnosis and troubleshooting after industrial process accidents. It has become a cross-cutting technology, mainly involving the fields of statistics, applied mathematics, signal analysis and machine learning. With the in-depth research and continuous development across fields, the different fault classification methods that many experts and scholars have applied can be summarized into two categories: mathematical modeling and data-driven. Among the data-driven methods, the deep learning method has achieved performance beyond the traditional signal analysis, multivariate statistical learning and other algorithms and has been widely used. [000...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the application defects of the existing two-dimensional convolutional neural network technology, the purpose of the present invention is to deconstruct the correlation between variables by using the multi-channel operation of one-dimensional convolution and pooling, and independently analyze the local timing characteristics of process variables For fault classification, it aims to solve the problem that the two-dimensional convolutional network is sensitive to the order of variables, resulting in low accuracy and precision of fault identification

Images