Component | Technique | Method | Condition monitoring | Advantages | Problems and challenges |
---|---|---|---|---|---|
Transformer | IoT [120] | A sophisticated monitoring system that will send a notification to a specific device for further action | Successful | Two-way communication within connected devices is enabled | Creates a complex array of connected devices which may become difficult to control |
SVM, K-Nearest Neighbor (KNN), Naive Bayes (NB), Random Forest (RF), Artificial Neural Network (ANN), Adaptive Boosting (AdaBoost), and Decision Tree [121] | To handle the issue of reliability and uncertainty of Health Index of power transformer using Artificial Intelligence based algorithms | Successful | Weak classifiers with less accuracy are strengthened for improved results | The classifiers should be high quality classifiers | |
IoT [122] | A health monitoring system that uses temperature sensor to monitor transformers and send data for remote analysis | Successful | Two-way communication within connected devices is enabled | Creates a complex array of connected devices which may become difficult to control | |
Adaptive Neuro FuzzyInference System (ANFIS) [123] | A comparison between fuzzy model and adaptive neuro fuzzy model to the Health Index (HI) of transformers based on various parameters | Successful | Able to capture nonlinear structure with high adaptability and learning capability | Faces difficulty in handling large input datasets | |
Inverter | SVM [124] | A condition monitoring system gained by training an SVM model with characteristics of DC-link capacitors in a three-phase inverter | Successful | Efficient handling of nonlinear data samples | Requires extended memory and longer training time |
Convolutional Neural Network (CNN) [125] | A condition monitoring system gained by using an CNN model to analyze characteristics of DC-link capacitors in a three-phase inverter | Successful | Automated detection of unique features in training data samples | Huge quantity of data samples is needed for completing analysis | |
PV | Machine learning and deep learning models [126] | Different condition checking systems based on machine learning models | Successful | Integration of different data processing tools bears strong output | Requires large amount of nonlinear abstractions for meaningful representation |
Deep learning, Reinforcement learning, Transfer learning, Ensemble learning [126] | Use of deep learning models to tackle different issues of condition monitoring | Successful | Presents an accurate, stable and robust algorithm | Interpretation ability is reduced in this method | |
Wind turbine | ANN, Bayesian network, Support vector regression, RF, KNN [127] | Data on vibration taken from wind turbine is combined with data acquired from supervisory control and data acquisition systems (SCADA) which is analyzed using machine learning methods building a condition monitoring system | Successful | Able to develop individual prediction using historical data samples | Needs high computational power |
Bidirectional gated recurrent unit (BiGRU), CNN [128] | CNN and BiGRU methods are used on data acquired from supervisory control and data acquisition systems (SCADA) for condition monitoring | Successful | Quick response while using very little memory | Slight inaccuracies due to quick processing of data | |
Deep convolutional generative adversarial networks (DCGAN) [129] | A health condition monitoring (HCM) system for wind turbine using DCGAN | Successful | Generates high quality artificial data which further enhances the training sequence | Requires large quantity of data and is also difficult to train |