MONITORING STABILITY OF KEYHOLE FORMATION IN VPPAW G. Tao, H. Wang and R. Kovacevic* ABSTRACT There are a number of factors that influence the quality of the weld formation in variable polarity plasma arc welding (VPPAW). Stability control of weld formation in VPPAW is required. One of the control strategies is to apply feedback control. The key for the successful application of feedback control is to acquire a characteristic signal that reflect the status of the controlled object. Airborne sound generated in the keyhole mode of VPPA welding process is used to characterize the presence of the keyhole and its size. Short-time spectrum analysis is used to study the frequency characteristics of the sound signal during the welding process. Short-time spectra are integrated along the frequency domain to obtain the power at different frequency intervals. The power is then used as the input to an artificial neural network (ANN) model to classify the keyhole diameter. KEYWORDS Sound signal, Time-frequency analysis, ANN, VPPAW INTRODUCTION Variable polarity plasma arc welding (VPPAW) is a promising welding technique for joining aluminum alloys. It has been used successfully in production, such as in the fabrication of the space shuttle external tanks (Ref. 1-3). VPPAW can generate high welding quality and high productivity at relatively low cost. However, the keyhole molten pool is very dynamic, and it can collapse and generate burn-through holes during welding, especially when welding plates with thicknesses above 4.0 mm. Thus, the selection of the welding parameters and the implementation of controller to ensure the stability of the weld formation in real-time remain a challenge. Recently, it was found that the presence or absence of a keyhole could be determined by measuring the ratio of hydrogen to argon in the plasma arc column with an optical spectrometer (Ref.4). However, the size of the keyhole cannot be determined and the welding process cannot be distinguished from the cutting process according to that signal. At present, two problems are associated with sensing of the keyhole weld pool in VPPAW: 1) Inaccessibility from the frontside of the weld pool because of the limited torch standoff-distance and the interference produced by the strong arc light. 2) Detection of the keyhole from the back-side of the workpiece is related to the complexity of the weld structure, for instance, in the welding of pressure vessels. Research Center for Advanced Manufacturing, Southern Methodist University 1500 International Pkwy, Suite 100, Richardson, Texas 75081 * corresponding author Tel: 1-214-768-4865, Fax:1-214-768-0812, Email: kovacevi@seas.smu.edu Previous work shows that keyhole size captured by a machine vision system can be used as a feature signal to the feedback control of the weld formation in VPPAW (Ref.5). However, it is difficult to precisely monitor the keyhole size by machine vision when the keyhole is very small or very large. A new approach to characterize the presence and the size of the keyhole is needed. One of the potential solutions is to relate the signature of the airborne sound signal during welding to the geometrical characters of the molten pool. The acoustic signal has been used successfully in the monitoring of different welding processes. For example, the time-frequency characteristics of airborne signals were investigated in laser welding (Ref.6). It was found that the acoustic spectrum of good-quality, full-penetration welds could be differentiated from the spectra of poor-quality welds, defined as either partially penetrated welds or welds having a gap between the sheets being joined. Welding sound signature in GTAW was also studied to reveal the conditions that generate weld defects (Ref.7). It was found that the sound signature produced by GMAW contained information about the behavior of the arc column, the oscillation of metal in the molten pool, and the metal transfer mode. The analysis of the acoustic signal from arc welding, oxy-flame cutting, and water jet cutting were carried out also (Ref.8). In addition, acoustic emission monitoring of laser beam welding was investigated to detect laser misfiring, loss of power, and improper focus (Ref.9). In this paper, the sound generated in the VPPAW is studied. The investigation is focused on the following: 1) Frequency and time-frequency analyses of the sound signal generated in VPPAW to study the welding status. 2) Establishment of the model to describe the relationship between the sound signal and the keyhole diameter of the weld pool. backside of the workpiece as a reference. The image acquisition and sound signal acquisition is synchronized. The frequency of image acquisition is 2 Hz while the frequency of the sound signal acquisition is 50 kHz. Experimental Procedures The experiments are performed in several steps: 1) The keyhole is generated by a stationary torch in order to study the sound signal at different welding modes. 2) A copper plate with pre-drilled holes is used to study the relationship between the size of the hole and the generated sound signal by the plasma. 3) A set of experiments is performed in order to study the relationship between the keyhole diameter and the level of sound signal simulating the real welding conditions. SPECTRUM ANALYSIS AND TIME-FREQUENCY ANALYSIS Spectrum analysis is used to study the process characteristics. In this paper, an averaged periodical diagram is used to obtain the spectral analysis to speed up the calculation and to increase the resolution. The data is separated into several segments with overlapping. For each data segment, p, the periodicity can be calculated by equation (1): M jrk/M Im,p (∞, ) = 2 / M 1⁄2 x piTMm (ie ̄/M2, k=1, 2,...M/2 or (M-1)/2 (1) i=1 The Hann window is used for truncation to decrease the power leakage and to get a smoother result. Finally, the average periodicity is obtained by averaging the periodicity of each segment, p, for p=1: L IΜ(Wx) = 1 / LŹ! M‚p(Wx) p=1 (2) Thus, the expectation of the standard deviation of the spectral analysis can be decreased to 1/L. Due to the variation of the VPPAW welding process, the acoustic signal is dynamic and time dependent, so the time-frequency distribution is an effective tool for studying this process. The time-frequency distribution can be viewed as a transform which represents the energy or density of a signal simultaneously in both time and frequency domains. It reveals the time-dependent features that the ordinary spectrum analyses fail to show. In this paper, a short-time spectrum analysis is used to study the time-frequency character of the VPPAW. Equation (3) shows the expression of the short-time spectrum analysis: dt SSPEC, (1,ƒ) = |STFT,(1,1)\2 = |{x(1')w(1-1)e ̄123a de (3) Where x(t) is the acoustic signal windowed by a suitable function w(t) (in this paper, it is the Hann window), and T is the interval of integration. Equation (3) can be expressed in a discrete version of the short-time spectrum analysis: The short-time spectrum at time i(the i the second of the sound signal) is obtained by averaging the periodical diagram. The results of the short time spectrum are integrated in different frequency intervals from f to f2 in order to present the frequency properties around the central frequency, f, in the same time (equation (5)): INSSPE ̧(i, f) = SSPEC ̧(i,k) (5) k=fkl In this paper, the real-time spectrum analysis is performed for the duration of one second by averaged periodicity. The next second, the previous data should be discarded in order to generate data for a new short-time spectrum analysis. SOUND GENERATION FOR DIFFERENT VPPAW MODES During welding, the pulsation of the arc plasma, and the oscillation of the weld pool are the major sources of the sound signal. The formation of the weld pool in VPPA welding passes through three stages: a no-keyhole weld pool or a fusion weld pool, a transition weld pool, and a keyhole weld pool. In the case of the no-keyhole weld pool, there is a layer of unmelted metal below the weld pool that allows for strong stirring of the weld pool. So, the weld pool generates slightly stronger oscillations at low frequencies. In the case of the transition mode, the keyhole is not formed, but the weld pool is generated along the entire thickness of the welded material. In this weld mode, the natural oscillation frequency of the weld pool is low. So, resonance could occur and the sound signal at the low frequency should be the strongest among the all three welding modes. In the case of the keyhole mode, there is a keyhole at the center of the weld pool and the stirring effect of the jet is decreased. So, the intensity of the sound signal at the lower frequencies is less pronounced than in the case of the transition mode. It is expected that the analysis of the intensity of the sound signal will be useful in distinguishing the modes of the weld pool in VPPAW. In the case of keeping a fixed position of the VPPAW torch with respect to the aluminum plate with a thickness of 3.18mm, all three modes of the molten pool are formed by controlling the intensity of the welding current. The signature of the sound in the time domain (Fig.2 a) does not show too much variation and only a number of randomly occurring singularities are present. Figure 2 b shows the power spectrum of the sound signal. There are several dominating spectra Figure 3. Diagram of the welding current (a) and short-time spectrum analysis results of peak values at the frequencies such as: 2.2 kHz, 3.9 kHz, 8 kHz,12 kHz. The sound signal is studied by integrating a short-time spectrum analysis around these peak values. The time-frequency analysis around the frequency of 2.2 kHz shows that there is a strong relationship between the power spectrum of the sound signal and the phases of the molten pool formation (Figure. 3). In this experiment, based on the changing of the welding current, the transition mode and keyhole mode will happen at 1.6 and 2.5 seconds respectively. Figure 3 shows the changing of the welding current and the short-time spectrum analysis results during the stationary torch welding at the frequency of 2.2 kHz. It is shown that the power spectrum at the frequency of 2.2 kHz will increase during the transition mode and then will drop back down when the keyhole is generated. It is clear that the spectral analysis of the sound signal provides a useful tool in distinguishing the welding pool modes in VPPAW. Simulation of the Keyhole Size Effect on the Sound In order to study the relationship between the intensity of the air-borne-generated sound by plasma passing through pre-drilled holes and the diameters of these holes, a number of experiments are performed. For these experiments, a special device is designed and constructed. a) 28 Frequency (kHz) b) Figure 4. Copper plate with cooling channels (a) and power spectrum of |