Modulated Broadband Mode Decomposition for the Feature Extraction of Double Pulse Metal Inert Gas Welding

A lot of adaptive signal decomposition methods have been applied for nonstationary DPMIG electrical signals as they are always affected by noise. Recently, to solve the problems of former methods caused by the Gibbs phenomenon and the calculation of extremas when dealing with broadband electrical signals such as square signals and sawtooth signals with “sharp corners”, broadband mode decomposition (BMD) method was proposed and the application of the algorithm showed a good performance. The main idea of BMD is searching in the associative dictionary contains both broadband and narrowband signals using a regulated differential operator as the optimal object. However, when applied to a broadband signal interfered by strong noise, as the relative bandwidth is not small enough, the BMD algorithm may treat the broadband signal to be several narrowband components. Therefore, modulated broadband mode decomposition (MBMD) is proposed in this paper to denoise broadband electrical signals based on modulated differential operator. By multiplying a high frequency mono-frequency signal, the relative bandwidth of the effective broadband signal is translated to be far less than 1, and the broadband signals are treated as approximate broadband signals to get more accurate decomposition results. For the further feature extraction of the electrical signals, locality preserving projection (LPP) is applied combined with MBMD. The effectiveness of the proposed method is testified by simulation and experimental signals, results show that it is effective when drawing broadband feature from noise, as well as is adaptive for the feature extraction of double pulse metal inert gas welding.