Photovoltaic system faults diagnosis using discrete wavelet transform based artificial neural networks

Bengharbi, A. A.; Laribi, Souad; Allaoui, Tayeb; Mimouni, Aîda

doi:10.20998/2074-272x.2022.6.07

Cited by 4 publications

(1 citation statement)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Consequently, it aids in detecting abnormally high energy dissipation that occurs concurrentely with the birth of a defect. This indicator is given by the following formula [18]:…”

Section: Statistical Study Based On Rms Correlation Coefficient Energ...mentioning

confidence: 99%

Diagnosis and Monitoring Method for Detecting and Localizing Bearing Faults

Dahmane,

Berrabah,

Defdaf

et al. 2024

IJEEI

View full text Add to dashboard Cite

Induction motors in modern industry are becoming more and more functional and complex. Unfortunately, these machines are not free from damages what make their fault diagnosis the most critical aspect of system monitoring and maintenance. Vibrational signal data yields relevant information about the state of the entire system, as well as specifically about one of its components that makes its analysis quite interesting. For this effect, the current paper aims to propose an automatic diagnosis and monitoring method for detecting and locating bearing faults in an induction motor based on vibration signal processing. The suggested method combines the discrete wavelet transform (DWT) with the envelope spectrum (ENV) as advanced signal processing, incorporating a machine learning algorithm based on random forest classifier. The discrete wavelet transforms (DWT), using the Haar wavelet, decomposes the vibrational signal to provide both approximations and details. Each detail is then reconstructed to avoid any missing of information. To precisely select the reconstructed detail (𝑅𝑒𝑐𝑑 𝑘 ) that provides pertinent information about bearing faults, a statistical study is conducted. This study involves calculating four indicators (Root mean square (RMS), correlation coefficient (CC), energy coefficient (EC) and peak to peak (P2P) factor) is performed for each (𝑅𝑒𝑐𝑑 𝑘 ). These indicators are compared with threshold indicators, and this criterion is met by the reconstructed details 1 and 3. The obtained reconstructed details are then subjected to the spectral envelope analysis to detect the fault frequencies, which are considered as new features entering the random forest classifier model. This combination of approaches allows better feature extraction and structuring of the dataset, leading to improved accuracy of the random forest classifier, achieving a higher classification rate of more than 99,53 %. The proposed DWT-ENV-RF method indicates well its efficiency when compared to other recent works, and the attained results are all confirmed by the experimental tests conducted in the CWRU laboratory.

show abstract

“…Consequently, it aids in detecting abnormally high energy dissipation that occurs concurrentely with the birth of a defect. This indicator is given by the following formula [18]:…”

Section: Statistical Study Based On Rms Correlation Coefficient Energ...mentioning

confidence: 99%

Diagnosis and Monitoring Method for Detecting and Localizing Bearing Faults

Dahmane,

Berrabah,

Defdaf

et al. 2024

IJEEI

View full text Add to dashboard Cite

show abstract

A novel method for fault diagnosis in photovoltaic arrays used in distribution power systems

Zare,

Simab,

Nafar

et al. 2024

Energy Syst

View full text Add to dashboard Cite

This study addresses the critical issue of fault diagnosis in photovoltaic (PV) arrays, considering the increasing integration of distributed PV systems into power grids. The research employs a novel approach that combines artificial neural networks, specifically radial basis functions (RBFs), with machine learning techniques. The methodology involves training the RBF neural network using input features like voltage, current, temperature, and irradiance, derived from the PV array, to detect and classify various fault types. Notably, it comprehensively evaluates the accuracy of this approach, with a particular focus on detecting maximum power point tracking (MPPT) and mismatch faults. The findings reveal significant advantages, in which the proposed method outperforms existing techniques, achieving an approximately 20% increase in accuracy, with fault detection rates for specific faults ranging from 81.29 to 93.44%. Simulation results represent that by leveraging RBFs within neural networks, it offers improved fault detection and classification, making it a valuable advancement in the field of PV fault diagnosis.

show abstract