Fault diagnosis of stamping process based on empirical mode decomposition and learning vector quantization

Sheet metal stamping process is widely used in industry due to its high accuracy and productivity. However, monitoring the process is a difficult task since the monitoring signals are typically non-stationary transient signals. In this paper, empirical mode decomposition (EMD) is applied to extract the main features of the strain signals. First, the signal is decomposed by EMD into intrinsic mode functions (IMF). Then the signal energy and the Hilbert marginal spectrum, which reflects the working condition and the fault pattern of the process, are computed. Finally, to identify the faulty conditions of process, the learning vector quantization (LVQ) network is used as a classifier with the Hilbert marginal spectrum as the input vectors. The performance of this method is tested by 107 experiments derived from different conditions in the sheet metal stamping process. The artificially created defects can be detected with a success rate of 96.3%. The method seems to be useful to monitor a sheet metal stamping process in practice.