The change of transformer vibration can be used to estimate its internal operation state. However

it isdifficult to extract effective features of vibration signals to monitor winding looseness state since it contains a largeamount of state information. Therefore

a method of monitoring looseness state of transformer winding based onchaos theory and kernel-based possibilistic c-means clustering algorithm KPCM is presented. Firstly

the mea-sured transformer vibration signal sare reconstructed in thephase space with the optimal embedding dimension andtime delay based on the chaotic featuresof vibration signals. Then

the KPCM clustering method is used to identifythe distribution mode according to high dimensional spatial distribution of reconstructed signals. Consequently

the looseness state of transformer winding can be analyzed successfully. Calculated results of measured vibrationsignals show that the largest Lyapunov exponent calculated by the Wolf algorithm is positive

which confirms thechaotic characteristics of transformer vibration signals. Winding loosing mechanical fault can be identified effec-tively based on the changes of displacement vector of cluster centers obtained by KPCM clustering analysis. Theobtained results provide a theoretical basis for the looseness state monitoring of transformer windings from the view of chaotic dynamics.