Process data fault classification method based on pseudo label method and weak supervised learning

Abstract

The invention discloses an industrial process data fault classification method based on a pseudo label method and weak supervised learning; a supervised classification network is composed of a multi-layer sensor, a Batch Normalization layer, a Dropout layer and a Softmax output layer, and the Gaussian mixture model is used for obtaining the inaccurate condition of a pseudo label. The multi-layer sensor can learn feature representation of the data from the labeled data; the BatchNormalization layer is used for accelerating convergence of a multi-layer perceptron model, the Dropout layer is usedfor preventing training overfitting of a multi-layer perceptron, and the Softmax output layer is used for carrying out fault classification according to fault sample features extracted by the multi-layer sensor. According to the invention, modeling can be carried out in a scene that the obtained labeled sample is inaccurate in label and has no label sample; label probability transfer matrix evaluation is carried out on a labeled sample label and a pseudo label predicted for a label-free sample based on a pseudo label method, and the label probability transfer matrix evaluation is used for correcting a loss function of a classification network to complete weak supervised learning, so that the classification precision of the model on the sample is improved.

Description

technical field [0001] The invention belongs to the field of industrial process fault diagnosis and classification, and relates to a fault classification method based on a pseudo-label method and weakly supervised learning. Background technique [0002] In industrial process monitoring, when a fault is detected, it is necessary to further analyze the fault information, and fault classification is an important part of it. Obtaining the type of fault is conducive to the restoration of the industrial process. [0003] In traditional fault classification, samples are required to have labels for model training. However, in industrial process data, the labels of labeled samples may be inaccurate and the sample labels are missing, that is, some samples are unlabeled. Pseudo-labeling methods are an effective way to utilize both labeled and unlabeled samples. However, the pseudo-label method does not consider the accuracy of labeling (pseudo-label) unlabeled samples. Putting samples...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems that the labels of the labeled samples obtained in the current industrial process may not be accurate and the pseudo-labels of the unlabeled samples may be mislabeled by the pseudo-label method, the present invention proposes a fault detection method based on the pseudo-label method and weakly supervised learning. The classification method, which is based on the classification network and Gaussian mixture model composed of MLP, BatchNormalization layer, Dropout layer and Softmax output layer, realizes the accurate classification of fault samples in industrial processes

Images