A fault diagnosis of high voltage circuit breakers with small samples using a PCA-cascade forest algorithm

Recently, many successful stories in mechanical fault diagnosis of high voltage circuit breaker (HVCB) have been reported, though the diagnostic accuracy largely depends upon the sample size. On the other hand, a large training set of HVCB is hard to obtain due to the complexity of its working environment. An improved Deep forest (DF) which can produce a good classification result using only a small sample set is then proposed in this study. Firstly, the contact displacement under five different states are measured and used as the raw input features. Then, an improved multi-grained scanning method is introduced to pre-process the dataset, and a feature extraction algorithm based on Principle Component Analysis (PCA) is applied in the cascade structure of DF to reduce the interference of the meaningless data. The diagnostic result using the proposed technique shows that the classification accuracy can reach 94.9 % with only 6 samples in each fault class, which is significantly higher than the traditional machine learning algorithms. In summary, the proposed technique provides a significant innovation by enabling accurate HVCB fault diagnosis with minimal data, providing a practical solution for real-world applications where data collection is limited. © 2017 Elsevier Inc. All rights reserved.