Fault detection and diagnosis of photovoltaic array: State of the art 24 Introduction Visual inspection for fault diagnosis

Get Complete Project Material File(s) Now! »

Power loss categorization

Another important factor in assessing the impact of PV faults is the power loss level, which directly determines the functionality of the PV system. The power loss levels describe the evolution of the power variation between the initial value and the one produced over time. In most cases, this difference between the reference values can lead to inconsistent results, as the power indicated in the datasheet can deviate significantly from its actual initial value.
For legal application, to evaluate the power loss, the power printed on the PV module datasheet could be used as the reference value. For practical application, the initial power could be adopted as a reference value. For various PV array faults, the power loss level is not easy to evaluate as it could vary from case to case and is mainly determined by the fault severity and spread. However, the common trend of the power loss due to these faults can generally be categorized. In Table I-6, five classic trends of power loss over time are defined [27].

Visual inspection for fault diagnosis

Visual inspection is a quick and effective method to identify faults of PV modules. It could be performed before and after the module has experienced environmental, electrical or stress test in the laboratory or operation in the field. The common detectable PV faults by visual inspection are listed in Table II-1.
Besides, PV images captured by various PV imaging techniques (electroluminescence, infrared, etc.) can also be examined by humans. These types of images could reveal some mechanical or electrical failures inside PV modules. A detailed presentation of these imaging techniques will be given in Section II. Despite its effectiveness, visual inspection is time-consuming and requires expensive equipment and personnel. According to [57], visual inspection is more appropriate for small-scale facilities where they can be frequent and cost-effective.
Some PV module faults that lead to performance degradation (e.g., induced partial discharge, broken internal interconnection) are usually invisible to the naked eye and cameras or sensors adopted for visual inspection. Detection of these types of faults requires other information and the implementation of more advanced methods.

Automatic information analysis for fault diagnosis

Automatic information analysis is playing an increasingly important role in PV FDD techniques. On the one hand, this is due to the increasing availability of PV data with the rapid development of the PV industry and, on the other hand, to the growing demand for automated and accurate monitoring of the health status of PV systems.
Automatic information analysis for PV FDD relies on a great variety of methods and techniques. The methods can be decomposed into four steps: modelling, pre-processing, feature extraction and feature analysis [58], as shown in Figure II-1.

Physics-based modelling

The models of a PV module are based on the description of the photoelectricity at a cell level. The cell models are combined in series and parallel to obtain the model of a module. They represent the relation between environmental inputs (irradiance and temperature) and the electrical signals, output current and voltage. The most common models are electrical equivalent circuit (one or two diode model [61], Merten’s model [62]), semi-empirical “point” model [63], Evans model or called simple efficiency model [64] ). Thanks to their simplicity and efficacity, the one or two diode models are the most used ones to estimate the operational parameters and characterize PV array. The electrical equivalent circuits for the two models are displayed in Figure II-2 and Figure II-3, respectively. It should be noted that the accuracy of the models depends on the accuracy of the model parameters, which can be extracted by various estimation methods [65], [66].
In the single-diode model: Iph is the photocurrent, Id the diode current representing the diffusion phenomenon, Rsh the shunt resistance representing leakage current path caused by the distributed manufacturing defects inside the solar cell, Rs the series resistance representing the power dissipation caused by the thermal effect in the whole junction substrates, and the electrical contacts.

Data augmentation

Statistical and more generally machine learning techniques are more effective when they have a large amount of data that is sufficiently representative of all operating modes. However, this is a serious challenge for measurement systems, even more so for image measurements.
There are two main obstacles to construct a large PV image dataset: 1) insufficient quantity of images due to the limited number of PV modules or the complexity of imaging procedure; 2) unbalanced dataset, i.e., the number of images of healthy modules and faulty modules is different [91]. These two obstacles can significantly hinder the learning performance of machine learning models. Therefore, data augmentation [92] is usually adopted to increase the number of images or balance the number of images among the different classes (operating conditions). The most common techniques include rotation, flip, clip, blurring addition, and adjustment of the brightness [93].

READ  Statics: surface deformation and contact angle on elastic materials

Format transformation

The format transformation of PV data is sometimes performed before analysis. These transformations are mainly done for two reasons: 1) to find a more appropriate representation for FDD analysis, 2) to combine different PV data format into an identical one. There are various transformations that are realized either by a simple rearrangement of data or signals into images or matrixes, or by special techniques. Some examples are presented in Table II-4.

Table of contents :

Table of contents
List of figures
List of tables
List of abbreviations
List of publications
General introduction
Background and motivation
Problem statement
Thesis outline
Contribution of thesis
Chapter I Photovoltaic array faults: State of the art
I.2.1. Defect due to human error
I.2.2. Environmental factors
I.2.3. Material interaction
I.2.4. Cause-effect circle
I.3.1. Fault classification
I.3.2. PV array faults
I.3.2.1. Cell-level faults
I.3.2.2. Module-level faults
I.3.2.3. Array-level faults
I.4.1. Safety hazard categorization
I.4.2. Power loss categorization
I.4.3. Summary of fault impact
Faults cases studied
Chapter II Fault detection and diagnosis of photovoltaic array: State of the art 24 Introduction Visual inspection for fault diagnosis
Automatic information analysis for fault diagnosis
II.3.1. Modelling
II.3.1.1. Physics-based modelling
II.3.1.2. Data-based modelling
II.3.2. Pre-processing
II.3.2.1. Format unification
II.3.2.2. Data cleaning
II.3.2.3. Data augmentation
II.3.2.4. Format transformation
II.3.3. Feature extraction
II.3.3.1. Statistical parameters
II.3.3.2. Signal transformation methods
II.3.3.3. Image processing methods
II.3.3.4. Multivariate transformation techniques
II.3.3.5. Estimation and control techniques
II.3.4. Feature analysis for FDD
II.3.4.1. Threshold analysis
II.3.4.2. Statistical analysis
II.3.4.3. Machine learning techniques
II.3.5. Illustration of the four-step automatic PV FDD scheme
II.4.1. Summary of fault diagnosis methods
II.4.2. Description of the proposed FDD strategy
Chapter III Correction of PV I-V curve measured under faulty condition 
Preparation of I-V curves for correction
III.2.1. PV array modeling
III.2.1.1. Cell-level modeling
III.2.1.2. Module-level modeling
III.2.1.3. Array-level modeling
III.2.2. Environmental settings
III.2.3. Configuration of faults
III.2.4. Impact of faults on I-V curves
III.3.1. Usual correction procedures
III.3.1.1. Procedure 1 (P1)
III.3.1.2. Procedure 2 (P2)
III.3.1.3. Procedure 3 (P3)
III.3.2. New correction procedure
Metrics for the evaluation of correction performance
III.4.1.1. Metric for the evaluation of correction of the entire curve
III.4.1.2. Metric for the evaluation of correction of single parameters
III.5.1. Performance of correction procedures using single I-V curve
III.5.1.1. Selection of G and Tm based on field-measurements
III.5.1.2. Correction performance with constant fault severity
III.5.1.3. Correction performance with varying fault severity
III.5.2. Performance of correction methods using multiple I-V curves
III.5.2.1. Selection of G and Tm for reference curves
III.5.2.2. Correction performance with constant fault severity
III.5.2.3. Correction performance with variable fault severity
Chapter IV PV fault diagnosis using I-V curves and machine learning classifiers
Configuration of the simulated dataset
IV.2.1. PV array model configuration
IV.2.2. Generation of dataset
IV.3.1. Correction of I-V curve
IV.3.2. Resampling of I-V curve
IV.4.1. Feature transformation
IV.4.1.1. Recurrence Plot (RP)
IV.4.1.2. Gramian Angular Difference Field (GADF)
IV.4.2. Dimensionality reduction of features
IV.5.1. Analysis techniques-machine learning classifiers
IV.5.2. Diagnosis results using simulated data
IV.5.2.1. Performance of fault classification
IV.5.2.2. Robustness to additional disturbance
IV.5.2.3. Influence of PCA
IV.5.2.4. Influence of transformation
IV.5.3. Diagnosis results using experimental data
IV.5.3.1. Description of experimental platforms
IV.5.3.2. Experimental test result
IV.6.1. Methods for comparison
IV.6.1.1. Methods based on partial usage of I-V curves
IV.6.1.2. Methods based on complete usage of I-V curves
IV.6.2. Comparison results
Conclusion and perspectives
Résumé en francais


Related Posts