Deep Convolutional Transfer Learning Network: A New Method for Intelligent Fault Diagnosis of Machines With Unlabeled Data

TLDR

Intelligent fault diagnosis requires labeled data and matching training–testing distributions, but many machines lack sufficient labeled data and suffer distribution shifts, limiting the applicability of existing methods. This study introduces a deep convolutional transfer learning network (DCTLN) to address these challenges. DCTLN combines a 1‑D CNN for automatic condition recognition with a domain adaptation module that maximizes domain recognition errors while minimizing distribution distance to learn domain‑invariant features. The approach was validated in six transfer fault diagnosis experiments, demonstrating its effectiveness.

Abstract

The success of intelligent fault diagnosis of machines relies on the following two conditions: 1) labeled data with fault information are available; and 2) the training and testing data are drawn from the same probability distribution. However, for some machines, it is difficult to obtain massive labeled data. Moreover, even though labeled data can be obtained from some machines, the intelligent fault diagnosis method trained with such labeled data possibly fails in classifying unlabeled data acquired from the other machines due to data distribution discrepancy. These problems limit the successful applications of intelligent fault diagnosis of machines with unlabeled data. As a potential tool, transfer learning adapts a model trained in a source domain to its application in a target domain. Based on the transfer learning, we propose a new intelligent method named deep convolutional transfer learning network (DCTLN). A DCTLN consists of two modules: condition recognition and domain adaptation. The condition recognition module is constructed by a one-dimensional (1-D) convolutional neural network (CNN) to automatically learn features and recognize health conditions of machines. The domain adaptation module facilitates the 1-D CNN to learn domain-invariant features by maximizing domain recognition errors and minimizing the probability distribution distance. The effectiveness of the proposed method is verified using six transfer fault diagnosis experiments.