A Novel Fault Diagnosis Method for Denoising Autoencoder Assisted by Digital Twin

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Intelligent manufacturing relies on accurate fault diagnosis, often hindered by insufficient data in dynamic production environments.
• Traditional fault diagnosis methods struggle with limited or unavailable fault state data, impacting model training and reliability.
• Digital twin (DT) technology offers a solution by simulating real-world systems to generate necessary data.

Purpose of the Study:

• To propose a novel digital twin-assisted fault diagnosis method.
• To address the challenge of limited or unavailable fault state data in dynamic production scenarios.
• To enhance the accuracy and feasibility of fault diagnosis in intelligent manufacturing.

Main Methods:

• Establishment of an ultrahigh-fidelity digital twin (DT) model for simulating machine operations and generating fault data.
• Development and training of a novel denoising autoencoder (NDAE) utilizing Mish activation function with DT-generated source domain data.
• Application and validation of the proposed DT-assisted NDAE method on a triplex pump fault diagnosis case.

Main Results:

• The DT model successfully generated diverse fault state data, mimicking real-world conditions.
• The NDAE, trained on DT data, demonstrated effective fault feature extraction and denoising.
• The proposed method achieved intelligent fault diagnosis even with limited or unavailable real-world fault data.

Conclusions:

• The digital twin-assisted fault diagnosis method provides a viable solution for scenarios with data scarcity.
• The integration of DT and NDAE enhances the robustness and applicability of fault diagnosis in intelligent manufacturing.
• This approach enables reliable intelligent fault diagnosis, overcoming limitations of traditional data-driven methods.