Novel Manifold Autoencoder for Industrial Process Fault Diagnosis

Abstract

Fault diagnosis plays a pivotal role in ensuring the safety of industrial processes. In the realm of fault diagnosis, stack autoencoders (SAE) have gained widespread popularity for their robust nonlinear feature extraction. Nevertheless, the unsupervised training mechanism of SAE tends to capture information unrelated to the underlying data structure and data type. In response to this issue, this article introduces a novel approach called the manifold stack autoencoder (MSAE). Within the proposed MSAE framework, the feature extraction capabilities of SAE and the manifold learning abilities are functionally integrated. This innovative MSAE method effectively extracts both the data type and the manifold structure, thereby enhancing fault diagnosis accuracy. To assess the practicality and effectiveness of the proposed MSAE, simulations are carried out using the Tennessee Eastman dataset, employing a random forest classifier for fault classification. The simulation results conclusively demonstrate the outstanding performance of the MSAE in terms of fault diagnosis accuracy.