Condition Monitoring using Marginal Energy and Hidden Markov Model

M. Ge, R. Du, and Y.S. Xu


Condition monitoring, wavelet packet transform, marginal energy, feature extraction, Hidden Markov Model (HMM), sheet metal stamping


In general, condition monitoring consists of two parts: extracting appropriate features from sensor signals and recognizing possible faulty patterns from the features. This article presents a new condition-monitoring method. First, it uses Time Marginal Energy (TME) and Frequency Marginal Energy (FME), which are acquired through wavelet packet transform of the sensor signal, as the features. Then, based on the features, it uses the Hidden Markov Model (HMM) to recognize the fault patterns. The extracted features have clear physical meanings, and the pattern recognition is effective. The new method is tested for condition monitoring of sheet metal stamping operations. Based on the tests (one is a simple blanking operation and the other is a progressive operation), the new method outperforms some of the commonly used methods, such as Artificial Neural Network (ANN). In addition, the new method does not require excessive computation and hence can be used for online monitoring applications.

Important Links:

Go Back