Abstract: To identify the droplet transfer modes in a strong noise-interfered MIG welding process, a welding arc sound pattern recognition method based on human auditory model is proposed. An external and middle auditory canal transfer function is adopted to dispose the welding arc sounds and depress the low frequency noise. Through simulating the function of cochlear to decompose the disposed welding arc sound signals into different frequency bands, and the loudness in the each of frequency band is acquired through the Moore loudness model. Based on the loudness in different frequency bands a feature vector is constructed and a support vector machine model is adopted to identify the droplet transfer modes. The experimental results show that the proposed method significantly depresses the low frequency welding arc noise aroused by fluctuation of welding currents, and the correct rate for droplet transfer modes identification is higher than 98%. To verify the anti-noise interference ability of the proposed method, a series of white noise and environment noise in different signal noise ratios are added to the original welding arc sound signals. The identification results show that the proposed method has excellent anti-noise interference ability. This research provides a new method for the welding quality online monitoring.

Key words: auditory mode, droplet transfer, state identification, auditory perception, welding arc sound