A5101459756- Machine Fault Diagnosis Techniques
- Fault Detection and Control Systems
- Gear and Bearing Dynamics Analysis
- Engineering Diagnostics and Reliability
- Adaptive Control of Nonlinear Systems
- Machine Learning in Bioinformatics
- Distributed Control Multi-Agent Systems
- Water Quality Monitoring and Analysis
- Structural Integrity and Reliability Analysis
- Robotic Locomotion and Control
- Advanced DC-DC Converters
- Rough Sets and Fuzzy Logic
- Multilevel Inverters and Converters
- Neural Networks and Applications
- Artificial Intelligence in Healthcare
- Imbalanced Data Classification Techniques
- Microgrid Control and Optimization
Foshan University
2016-2024
In real industry, due to changes in operating conditions and differences systems of interest, domain shift is a common problem, which results the degradation diagnostic performance. Moreover, insufficient labeled or unlabeled samples greatly limit adaptability fault diagnosis methods. To solve these problems, latest adversarial transfer learning techniques are studied. However, most existing studies just reduce distribution discrepancy two domains with fixed distance metrics, cannot adapt...
Achieving deep learning-based bearing fault diagnosis heavily relies on large labeled training samples. However, in real industry applications, data are scarce or even impossible to obtain. In this study, we addressed a challenging few-shot problem with few no samples of novel categories. To tackle problem, considered semi-supervised prototype network based pseudo-labels. The existing prototypical networks pseudo-label methods train pseudo label model unlabeled using high-dimensional data,...
A power converter is a time-varying system which featured with typical characteristic of nonlinearity. Therefore, it lacks systematic way to investigating the controllability due limitations linear circuit theory. In order study three-phase inverters in this paper, three-wire inverter and four-wire are modeled as switched systems. Then, state two studied based on criterion derived from Thereafter, canonical decomposition conducted when not completely state-controllable. The simulation...
Traditional deep learning-based fault diagnosis scenarios always rely on a certain type of neural networks (DNNs) to extract its inherent features, which can overcome the drawbacks shallow model methods. However, invariant features extracted have some inevitable limitations, lead performance degeneration and lower generalization ability these scenarios. To address this problem, we propose feature-integration-boosting-based discriminative stacked autoencoder (D-SAE) network for bearing...
In few-shot fault diagnosis tasks in which the effective label samples are scarce, existing semi-supervised learning (SSL)-based methods have obtained impressive results. However, industry, some low-quality hidden collected dataset, can cause a serious shift model training and lead to performance of SSL-based method degradation. To address this issue, latest prototypical network-based SSL techniques studied. most scenarios consider that each sample has same contribution class prototype,...
This paper focuses on research of an adaptive clustering algorithm and its application in medical diagnosis based probabilistic neural networks. A PNN-CAdaBoost model is proposed the standard AdaBoost together with algorithm. Both PNN-AdaBoost are established respectively. For testing our validness, experimental data collected from Wisconsin breast cancer set UCI database, computations comparisons multiple indicators. It proved that can effectively improve classification performance has good...
We propose an improved strategy for control of underactuated two-link manipulator named acrobots. Initial conditions nonlinear systems determine its trajectory even under same law, so we introduce a firing law to change initial conditions, which leads better performance. The motion space is divided into three areas: firing, swing-up and attractive; laws each are designed. For area, based on weak-control Lyapunov function(WCLF) employed increase the energy posture actuated link attractive...
This paper presents a rapidly and lower neural networks to treat those waste water index that is difficult be measured. Model called soft sensor composited two parts: one used estimate the principal linear output, other adjust estimated error obtain better accuracy. Selection of features effects greatly computation scale predict accuracy discussed also achieve result. Finally, an experiment provided for testing. Simulation results show method proposed here valid has good performance.