Early outlier detection in three-phase induction heating systems using clustering algorithms

Abstract

Induction heating (IH) devices transfer the electric power to the contactless cookware via the electromagnetic field. Therefore, the temperature of cookware is measured remotely, and the early detection of cookware overheating will ensure the user’s safety as well as extend the remaining useful life of electronic components. Therefore, this work presents a clustering model for outlier detection in IH systems based on clustering algorithms and measured data using two thermal sensors. First, a healthy dataset is collected for the temperatures of inverters and cookware under different sizes and materials of cookware items, different amounts of water in cookware, and different amounts of electrical power. After that, K-means and fuzzy c-means were utilized to cluster this normal dataset, where the maximum distance between their centers and data points was selected as a threshold. Finally, the clustered model is investigated using a testing dataset that includes outliers. According to the results, the K-means algorithm detected around 96% of the produced outliers, however, the fuzzy c-means algorithm detected around 68%. In conclusion, the deployment of the clustering model in outlier detection is simple and uses only the threshold and the cluster centers.

Keywords

• Outlier detection
• Clustering algorithms
• Induction Heating
• K-means
• Unsupervised machine learning