Sensing of weld pool surface using non-transferred plasma charge sensor

By W Lu¹, Y M Zhang¹ and John Emmerson²

¹Welding Research Laboratory, Center for Manufacturing and Department of Electrical and Computer Engineering, University of Kentucky, KY 40513, USA

²Magnatech Limited Partnership, East Granby, CT 06026, USA

E-mail: ymzhang@engr.uky.edu

Received 2 January 2004

Published 20 April 2004

Online at stacks.iop.org/MST/15/991 (DOI: 10.1088/0957-0233/15/5/031)

Abstract

Gas tungsten arc welding (GTAW) is the primary process for precision joining of metals due to its capability for accurate control of heat input.  As a close relative and modification of GTAW, plasma arc welding (PAW) has higher energy density and can penetrate thicker workpieces while maintaining the desired capability for accurate control of heat input.  In order to produce quality welds consistently using PAW, in addition to accurate control of heat input, the sensing and control of the weld pool surface are also critical.  It has been found that the non-transferred plasma arc, in the absence of the transferred arc or the main plasma arc for welding, can establish a plasma charge potential which is indicative of the arc length.  In order to take good advantage of this intrinsic characteristic of the non-transferred arc and eliminate the influence of the transferred arc in normal PAW process, a power module is used to cut off the main arc current periodically for a very short period of time to acquire accurate information for monitoring the weld pool surface.  A non-transferred plasma charge sensor is proposed based on this mechanism and experiments verified its effectiveness.

Keywords: surface, weld pool, plasma charge, arc, welding, control, penetration

1. Introduction

Gas tungsten arc welding (GTAW) is the primary process for precision joining of metals due to its capability for accurate control of heat input.  Although plasma arc welding (PAW) is a close relative and modification of GTAW and can accurately control the heat input like GTAW, it can achieve much deeper penetration.  In fact, its highly concentrated arc can easily melt the metal and form a deep weld pool, even a small funnel-shaped cavity referred to as the keyhole [1], just as laser welding does.  It is this keyhole that significantly increases heat transfers efficiency and hence penetration capability.  In order to maintain an appropriate state of the keyhole, the welding current must be adjusted at a suitable level so that the workpiece is fully penetrated as desired but the melted metal is not blown away by high arc pressure, a phenomenon known as burn-through.  This suitable level of welding current must be determined by monitoring a key parameter which can accurately depict the state of the process such as the weld pool surface or weld pool depth.  Hence, in order to achieve quality welds, a reliable sensor is critical.

    Detection of the weld pool has been attracting the interest of researchers since the beginning of arc welding applications.  Methods proposed include weld pool oscillation [2-4], ultrasound [5-6], arc light spectral lines [7] and infrared sensing [8].  At the Welding Research Laboratory at the University of Kentucky, a number of arc welding sensors have been successfully proposed and implemented, including weld sag sensing [9], weld pool vision sensing [10-12], arc light spectra analysis [13, 14], plasma cloud charge sensing [15], efflux plasma charge sensing [16], light beam aided sensing [17, 18], etc.  All these sensing mechanisms are based on the intrinsic characteristics of the welding process and have useful application in arc welding processes.  However, they all lack wide application either because of sensor complexity, high cost or unavailability for specific applications.  Thus, a reliable, robust and easy-to-use arc welding sensor is still actively being sought.

    In this paper, a new sensing technique for PAW is proposed, based on the space charge between the PAW torch nozzle and the workpiece when only the non-transferred plasma arc is present.  In PAW, two arcs are induced by two separate power supplies.  One is the pilot arc or non-transferred arc, induced by the power between the tungsten electrode and the plasma torch nozzle.  Another is the main arc or transferred arc, established by the power from the workpiece to the electrode.  It has been found that the plasma arc is not electrically neutral [19] and an electrical potential exists between the two ends of the plasma jet due to the difference in mass and thermal velocity between ionized electrons and positive ions [20].  A previous study [21] shows that when only the non-transferred arc is present, there will be a voltage drop between the torch nozzle and the workpiece and this voltage is a good indictor of the non-transferred arc length, thus the surface condition of the workpiece, in terms of sensitivity and reliability.  Because the non-transferred arc does not obey the minimum voltage principle (which governs all transferred arcs and commands them to follow the minimum path between the electrode and the workpiece), its flow is straight and its path is independent of workpiece geometry.  However, if a transferred arc is present so that the minimum voltage principle applies, the arc path will be affected by the geometry of the workpiece in that the arc will no longer be directly related to the shape of the weld pool surface.  Moreover, the electrical force from the main arc plays a much more important role in the movement of charged particles than thermal energy, resulting in a different pattern of plasma charge.  Therefore, the non-transferred arc sensing mechanism would become invalid.  In order to implement the mechanism in a normal PAW process, in which the transferred arc is always present, an insulated gate bipolar transistor (IGBT) power module is used in this study to isolate the main power for a very short period of time so that during this period, the transferred arc extinguishes and only the non-transferred arc is present.  In this way, the non-transferred arc sensing environment can be constructed artificially and the indicative arc voltage can be measured effectively.  Hence, the proposed sensing approach can be applicable in a real-time welding process.

    In this study, the electrical potential between the torch nozzle and the workpiece due to the non-transferred plasma charge is sensed by a resistor and a capacitor in series.  The sensor is referred to as the non-transferred plasma charge sensor (NTPCS).  Because this sensor requires no additional attachment, the torch's accessibility, an important factor in determining the practicality of welding sensing and control systems, will not be affected.

2. Non-transferred plasma charge sensor (NTPCS) description

The main power supply is connected to the tungsten electrode and the workpiece through an IGBT power module.  By controlling the voltage between the gate and emitter of the IGBT, one can control the on/off state of the IGBT, and, thus, the connection of the main power supply to the welding process.  Most of the time, the IGBT is on in order to provide welding power, as in a normal welding process.  For a very short period of time, normally of the order of ms, the IGBT is turned off so that the main power is separated from the welding system.  With the pilot arc power supply on all the time, only the non-transferred arc exists during this interval, which is exactly the same as non-transferred micro-plasma sensing [21].  By monitoring the nozzle-to-workpiece voltage, one can tell the length of the non-transferred arc, which is typically straight.  If the torch standoff is kept constant during the whole process, the weld pool surface, especially its depth, can be obtained by subtracting the standoff from the arc length.  Hence, this period is referred to as the sensing period.

    In order to sense the electrical potential between the nozzle and the workpiece, a low-pass filter (a resistor and a capacitor) is used.  The voltage across the capacitor is sampled as the output of the NTPCS.  In order to provide an appropriate load, the typical resistance and capacitance at the load terminal are chosen as 53 kW and 0.01 µF, respectively, giving a cutoff frequency of 2 kHz.  Thus, the resistor and capacitor circuit serves both sensing and low-pass filtering purposes.

3. Theoretical background

To understand the cause of the signal and its relationship to the weld pool surface, it is necessary to understand the physical phenomena involved in the welding process.  Because the sensing signal is the voltage between the torch nozzle and the workpiece due to electron and positive ion distribution in this region, anything that influences the charge distribution will affect the sensor output.

    Consider the non-transferred arc case first.  The inert gas is ionized and blown out of the torch nozzle chamber by the plasma gas, forming a plasma jet referred to as non-transferred plasma.  The non-transferred plasma is subject to considerably greater flux of electrons than of positive ions due to the much higher thermal velocity of electrons [20, 21].

    The mean electron velocity v_e is determined by the mean electron temperature T_e  and Boltzmann's constant k according to equation (1) [22]:

               n_e = (8kT_e / pm_e)^1/2                                                       (1)

where m_e is the mass of an electron.  For the typical plasma electron temperature of 5300-7000- K, the mean electron velocity is in the range of 4.5 x 10⁵ to 5.2 x 10⁵ m s^-1.

    On the other hand, the ion speed v_s is determined by equation (2) [23] given below:

                           n_s = (k[T_e + T_i] / m_i)^1/2                                                       (2)

where T_i and m_i are the temperature (K) and mass (kg) of the positive ions.  In the case where the temperatures of electrons and positive ions are both equal to the plasma temperature (i.e. T_e = T_i = T) and where the lightest species of positive ion has a mass of approximately 2000m_e, the typical ion velocity is in the range of 9.0 x 10³ to 10.0 x 10³ m ^-1.  Comparing the two velocities it is obvious that the flux of electrons greatly exceeds that of positive ions.  Thus, more electrons approach the bottom side of the pilot arc, i.e., the workpiece, while more positive ions accumulate at the top, i.e., the nozzle, resulting in an electrical field being established at the two ends of the plasma jet until it retards the approaching charged particles causing a charge balance.  The established electrical field points from the nozzle to the workpiece.

    According to [20], the plasma charge potential V₀ can be estimated as

                                        n 2.5kT / e                                                                (3)

where T is the temperature of the plasma jet, k Boltzmann's constant and e the electron charge.  For a given experimental condition, the plasma temperature can be considered constant, resulting in a nearly fixed plasma charge voltage independent of the arc length.

    The arc resistance varies with the arc length.  To determine the resistance of the non-transferred arc, the resistivity can be calculated as [24]

                                           ŋ = 6.53 x 10³(ln L / T^3/2)                                        (4)

where ln L is a constant depending on the electron density and the temperature.  Hence, the arc resistance R_a can be calculated as

                                    R_a = ŋ x l / S                                                       (5)

where l and S are the length and cross-section area of the arc, respectively.

    In stable plasma jet, the electron density and temperature are nearly fixed, implying a constant arc resistivity, as well as a constant cross-section area.  Hence, the non-transferred plasma arc resistance is variable and depends mainly on the arc length.

    Combining the voltage and resistance analysis, the non-transferred plasma arc can be modeled as a constant voltage source V₀ in series with a variable resistor R_a which is mainly determined by the length of the arc [21].  The Thevenin source model of the NTPCS system is shown in figure 2.  Hence, the sensor signal, the voltage on the capacitor, can be expressed in the frequency domain as

                                V_a(s) = V_o / s[(R + R_a)C_s + 1]                                      (6)

where V_a is the voltage on the capacitor, V₀ is the equivalent voltage source due to the plasma charge, R_a is the equivalent resistance of the arc and R and C are the sensing resistance and capacitance, respectively.

    V_a can be expressed in the time domain as

                                v_a(t) = V₀ x [1 - exp (-t / (R_a + R)C].                               (7)

Since the sensing period and the time constant (R_a + R)C are both of the order of ms, the exponential component can be expanded with only the first two terms, yielding

                        v_a(t) ≈ V₀ x [1 - (1 - t / (R_a + R)C)] = [V₀ / (R_a + R)C]t.            (8)

Thus, should the sampling time T be chosen, the signal v_a is inversely proportional to the arc resistance R_a, and in turn, the arc length l.  Equation (8) can be reorganized for the arc length measurement in terms of the sensed voltage:

                            l = (SV₀nT / V_a(nT)C) - (RS / ŋ) = (A / V_a(nT) -  B)                (9)

where A = SV₀nT / C and B = RS / ŋ are two constants, and nT represents the nth sampling time.

    Assuming the non-transferred arc approaches the weld pool surface and the torch standoff L is constant, the weld pool depth d can be calculated by subtracting the standoff from the arc length:

                                        d = l - L.                                                               (10)

If the standoff distance varies, a low current can be applied periodically to resume a flat weld pool surface for short periods of time.  The corresponding measurements will give real-time adjustments to the standoff distance L in equation (10).

    During the normal welding period, i.e., the IGBT is turned on and the main power is supplied, the workpiece becomes the positive terminal, resulting in a negative NTPCS signal.  For a data acquisition system taking only positive signals, the negative signal is read as zero.  When the IGBT is turned off, positively charged particles do not vanish immediately.  Instead, it will take some time to neutralize them, as well as the negatively charged capacitor.  Thus, for the first few samples in every sensing period, the signal will remain at zero until it rises gradually to the steady state.  Experimental results show that this time normally lasts about 20-30 ms.  However, one does not need to wait until the capacitor is fully charged.  Should the sensing period be fixed and greater than the charge recovery time, the maximum sample in every sensing period, normally exactly when the IGBT is turned back on, is able to reveal the weld pool depth during this period.  Hence, 10 ms is typically chosen as the sensing period and the signal sampled just at the end of the sensing period is taken as the output of the NTPC sensor.  Figure 3 shows a typical waveform of the sampled signal.

    Theoretically, equation (10) gives the absolute value of the weld pool surface depth.  However, due to the harsh nature of the welding environment and a great number of uncertainties involved in the arc and weld pool, especially when the main arc current is applied, the pool surface depth may differ from the prediction given by equations (1)-(10) considerably.  Thus, instead of calculating the absolute weld pool surface depth, a relative height is estimated with respect to a reference, which will be discussed in greater detail later.

4. Experimental procedure

In order to test the response of the proposed NTPCS at various welding conditions, a series of experiments have been conducted in PAW.  In these experiments, two inverters designed for gas tungsten arc welding and plasma arc welding are used as the main power supply and pilot arc power supply, respectively.  The IGBT used is CM300H-12A, rated at a maximum capacity of 300 A and 600 V.  The host computer adjusts the on/off state of the IGBT by controlling the voltage level between the gate and emitter of the IGBT, as well as the main welding current via the analogue output interface to the main power supply.  The torch, a regular commercial straight-polarity plasma arc welding torch, is attached to a manipulator.  The motion of the manipulator is controllable and the moving direction is parallel to the workpiece to maintain a constant torch standoff.

    All the experiments are bead-on-plate welds on stainless steel (type 304).  For simplicity, the pilot arc current is fixed as 15 A.  Table 1 lists additional common parameters used in the experiments.

5. Experimental results

5.1. Validity test experiment

In order to verify that the non-transferred plasma arc signal is not affected by the main power supply when the IGBT is turned off, the welding current is pulsed between 25 A and 10 A repeatedly every 400 ms.  After every 100 ms, the IGBT is turned off for 10 ms in order to sense the non-transferred plasma arc signal.  The torch travels at a fast speed so that though the welding current is applied, no metal is melted.  Figure 4 shows the sampled signal together with the welding current.  It can be seen that although the main current varies the signal remains nearly constant, implying that the non-transferred plasma arc signal does not depend on the external main power supply once it is isolated by the IGBT.

5.2. Pulsed current experiment

Pulsing current gives better process controllability and reduces heat input.  In order to test the validity of NTPCS under pulsed current, experiments are conducted on a 1.9 mm thick workpiece at the welding speed of 2 mm s^-1.  The welding current is pulsed between 25 A and 10 A every 400 ms so that the workpiece is just penetrated and the back side bead remains nearly constant.

    It is apparent that under pulsed current the weld pool grows deeper and larger during the peak current duration due to higher heat input and arc pressure, while it grows narrower and smaller during the base level, as shown in figure 5.  According to the inversely proportional relationship between the non-transferred plasma signal and the weld pool surface depth, the signal should be larger during the base period and smaller during the peak period.  Moreover, the weld pool grows and shrinks with the pulsed current's pattern repeatedly, suggesting periodical rises and falls of the weld pool and its surface depth.

    The sampled signal is shown in figure 6.  It can be seen that the non-transferred plasma arc signal increases gradually under base current and decreases under peak current, responding to the transition of welding current levels.  This behavior is exactly in accordance with the assumption of the weld pool surface variation.

5.3. Speed variation experiment

In order to verify that the non-transferred plasma arc signal can be used to detect the variation of the weld pool surface depth, the welding speed is varied in order to obtain different penetration and weld pool surface conditions.  The experiment is conducted on 2 mm thick plate with a welding current of 25 A.  At the beginning, the torch moves at 2 mm s^-1, causing insufficient heat input and a non-penetrated weld.  After 30 s, the speed is decreased to 1 mm s^-1 and full penetration is established.  Then after another 20 s, the speed is changed back to 2 mm s^-1 .  The original signal is shown in figure 7(a).  Figure 7(b) shows the plot of the maximum signal in every sensing period, i.e., the signal sampled just before the IGBT is turned back on.  Figures 7(c) and (d) show the front side and back side of the weld, respectively.

    As can be seen, during the first 30 s, because the workpiece is not penetrated, the weld pool is relatively narrow, corresponding to a higher non-transferred plasma arc signal.  When the torch moves more slowly so that the heat input increases, the backside beads appear and the weld pool becomes deeper.  This is reflected in a smaller non-transferred plasma arc signal.  After the speed is increased again, the signal rises because the weld pool gets shallower and narrower.  The weld pictures verify the variation of the weld pool.

5.4. Welding current variation experiment

As shown above, the variation in welding current does not impact the signal if the weld pool is not undergoing changes.  When the weld pool does have considerable variation, however, the NTPCS should be able to detect it regardless of what welding current is being used.  For verification, the welding current is varied every 10 s from 25 A, 45 A, 15 A to 40 A, with 1.7 mm s^-1 welding speed and 1.9 mm workpiece thickness.  Figure 8(a) shows the original sensor signals together with the welding current.  To examine the data more closely, the maximum during every sensing period is extracted from the original and redrawn in figure 8(b).  The front side and back side of the weld are shown in figures 8(c) and (d), respectively.

    Figure 8(d) shows that during the first 10 s, the workpiece is not penetrated due to the low welding current.  The non-transferred plasma arc signal is relatively high, implying that the arc is short.  When the higher welding current (45 A) is applied, the workpiece is penetrated quickly.  This is reflected in the signal by an immediate drop in the non-transferred plasma arc signal.  Moreover, because of the large heat input, the weld bead becomes bigger and bigger, as can be seen from the back side of the workpiece.  As a result, the signal dwindles gradually.  With the welding current decreasing and the backside beads disappearing again during the third 10 s, the sensor signal rises again.  Then the sensor signal descends as expected when the current becomes large enough for penetration in the final 10 s.  However, due to different thermal stress under different currents, the workpiece has been deformed during the third 10 s, resulting in a longer standoff distance.  This explains why the signal at the third current transit is not as distinct as those during the first two.

5.5. Quasi-keyhole plasma arc welding experiment

Quasi-keyhole plasma arc welding, pulsing the welding current between peak and base levels in order to open and close the keyhole repeatedly, has proved to be an efficient approach for plasma arc welding process control [25, 26].  With the existing keyhole sensor, for example, the efflux plasma charge sensor (EPCS) [16], one can detect the instant at which the keyhole is established, as well as the state of the keyhole.

    EPCS measures the plasma from the back side of the workpiece as shown in figure 9.  It gives a non-zero output when and only when the keyhole is established such that the efflux plasma charges a voltage V_e between the electrically isolated workpiece and the detection plate, typically 2 V in this experiment.  Should the keyhole not be established, there is no efflux plasma between the workpiece and the detection plate, resulting in a zero output signal V_e.  Moreover, a larger V_e implies a larger keyhole diameter and a wider backside bead.  Hence, the NTPCS signal can be compared with the EPCS signal during the same welding process to examine the response of the proposed NTPCS.

    Experiments are conducted on a 4.7 mm thick stainless steel workpiece with a welding speed of 2 mm s^-1.  The welding current is pulsed between the peak level and the base level periodically.

    Figure 10(a) and (b) show the resultant signals for two quasi-keyhole plasma arc welding cases, partial keyhole and full keyhole.  In case (a), because the EPCS signal never exceeds the threshold, the keyhole is not established due to the relatively low peak current.  In case (b), the keyhole is established during the peak level and closed during the base level.  It can be seen that if the keyhole has been established by the peak current, the signals sampled in peak and base levels differ by as much as 0.5 V or more.  However, if the keyhole is not established during the peak current period, the difference is much smaller.  This phenomenon can be reasonably explained by the distinction of keyhole profiles between these two cases, as shown in figures 10(c) and (d).  When the keyhole is open in the peak duration, the arc length becomes much longer than that in the base duration, during which the keyhole is supposed to be closed due to low heat input.  As a result, the voltage signal differs significantly.  On the other hand, if the keyhole has never been established during either the peak or the base level period, the difference in the weld pool surface depth is relatively small, resulting in a small NTPCS signal variation for the two durations.  Hence, by computing the signal difference between peak and base durations, one can detect the establishment of the keyhole.

    Figure 11 shows the plots of the EPCS signal V_e, the NTPCS signal V_a and the pulsed welding current I (shown as I/50) for a weld in which the keyhole has never been closed in either the peak or base duration.  As can be seen, during the peak duration a big keyhole is established, corresponding to high V_e and low V_a.  When the current is switched to the base level, the keyhole is still open but decays due to less heat input.  This is shown by the decreasing V_e and increasing V_a.  The weld pool dwindles until it approaches the minimum and then grows again.  With the growth of the weld pool, the EPCS signal V_e keeps increasing and the NTPCS signal V_a keeps decreasing at the same time until the welding current is switched to the peak level again.  Thus, the NTPCS signal is in good accordance with the weld pool pattern as well as the EPCS signal.

    As mentioned earlier, the weld pool surface depth can be estimated by comparing the signal with a reference level due to a possible uncertainty in welding conditions such as the standoff distance and distortion caused by workpiece deformation.  For quasi-keyhole plasma arc welding, the keyhole is open during peak durations and closed during base ones.  The weld pool profiles during the base level and the peak level are similar to those shown in figure 5(a) and figure 10(d), respectively.  As can be seen, the weld pool in the base duration is much flatter and more stable than it is in the peak.  This is because the base current has much weaker capability to melt the metal and thus forms a much narrower and smoother weld pool.  Hence, the signal sampled during the base duration is suitable for use as the reference for the calculation of the weld pool surface depth in the succeeding peak duration.

    In order to test the applicability of this mechanism, preliminary experiments are conducted on a 4.3 mm thick workpiece with pulsed welding current.  The base current is fixed at 30 A, while the peak current is varied every 10 s from 120 A, 90 A, 110 A, to 80 A in order to obtain different keyhole diameters.  The base duration and peak duration are 350 ms and 250 ms, respectively.  Figure 12(a) shows the original NTPCS signal together with the welding current.  The maximum in each sensing period is extracted and the signals sampled in base and peak durations are redrawn separately in figures 12(b) and (c), respectively.  Figure 12(d) shows the back side of the weld.

    It can be seen that the workpiece is fully penetrated during the whole process although the penetration level varies.  The signal sampled during the base duration centers at an average level but varies upward and downward shortly during the whole process, implying that the weld pool undergoes only a smooth variation.  By contrast, the variation in the sensor signal is considerably larger for different peak levels and the higher the current, the lower the signal.  This is because a higher welding current causes a bigger and deeper keyhole and weld pool as can be seen from the back side of the workpiece.

6. Conclusions

• NTPCS technology has been developed for plasma arc welding.  Experimental results verified that the NTPCS is capable of monitoring the weld penetration and providing reliable information about the weld pool surface, especially the weld pool surfaces depth.  The NTPCS signal, sampled when the main power supply is isolated from the process, decreases as the arc length or the weld pool surface depth increases.

• The NTPCS signal may be affected by parameters such as pilot arc current, plasma gas flow rate, etc.  To assure the effectiveness of the NTPCS technology, experiments must be conducted in order to obtain the optimal combination of these parameters and thus the most indicative sensing signal.

Acknowledgments

This work is funded by the National Science Foundation under grant DMI-0114982, the Magnatech Limited Partnership and the University of Kentucky Center for Manufacturing.  The authors sincerely thank Dr. Arthur Radun and Mr. Yuchi Liu for fruitful technical discussions.

References

[1]    O'Brien R L 1991 Welding Handbook vol 2 (Miami, FL: American Welding Society) pp 330-49

[2]    Renwick R J and Richardson R W 1983 Experimental investigation of GTA weld pool oscillations Weld. J. 62 29s-35s

[3]    Sorensen C D and Eagar T W 1990 Measurement of oscillations in  partially penetrated weld pools through spectral analysis Trans. ASME, J. Dyn. Syst. Meas. Control 112 463-8

[4]    Xiao Y H and den Ouden G 1993 Weld pool oscillation during GTA welding of mild steel Weld. J. 72 428s-434s

[5]    George M G and Ume I C 1997 Automated system for laser ultrasonic sensing of weld penetration Mechatronics 7 711-21

[6]    Carlson N M and Johnson J A 1988 Ultrasonic sensing of weld pool penetration Weld. J. 67 239s-246s

[7]    Martinez L F, Marques R E, McClure J C and Nunes J R 1993 Front side keyhole detection in aluminum alloys Weld. J. 72 49-51

[8]    Wikle H C, Kottilingam S, Zee R H and Chin B A 2001 Infrared sensing techniques for penetration depth control of the submerged arc welding process J. Mater. Process. Technol. 113 228-33

[9]    Zhang Y M and Kovacevic R 1997 Real-time sensing of sag geometry during GTA welding Trans. ASME, J. Manuf. Sci. Eng. 119 151-60

[10]    Zhang Y M, Wu L, Walcott B L and Chen D H 1993 Determining joint penetration in GTAW with vision sensing of weld-face geometry Weld. J. 72 463s-469s

[11]    Kovacevic R and Zhang Y M 1997 Real time image processing for monitoring of free weld pool surface Trans. ASME, J. Manuf. Sci. Eng. 119 161-9

[12]    Zhang Y M, Kovacevic R and Wu L 1993 Closed-loop control of weld penetration using front-face vision sensing Proc. Institution of Mechanical Engineers, Part 1, J. Syst. Control Eng. 207 27-34

[13]    Li P J and Zhang Y M 2001 Robust sensing of arc length IEEE Trans. Instrum. Meas. 50 697-704

[14]    Li P J and Zhang Y M 2001 Precision sensing of arc length in GTAW based on arc light spectrum trans. ASME, J. Manuf. Sci. Eng. 123 62-5

[15]    Zhang Y M, Zhang B M and Liu Y C 2001 Plasma cloud charge sensor for pulse keyhole process control Meas. Sci. Technol. 12 1365-70

[16]    Zhang S B and Zhang Y M 2001 Efflux plasma charge based sensing and control of joint penetration during keyhole plasma arc welding Weld. J. 80 157s-162s

[17]    Saeed G and Zhang Y M 2003 Mathematical formulation and simulation of specular reflection based measurement system for gas tungsten arc weld pool surface Meas. Sci. Technol. 14 1671-82

[18]    Saeed G, Lou M J and Zhang Y M 2004 Computation of 3D weld pool surface from the slope field and point tracking of laser beams Meas. Sci. Technol. 15 389-403

[19]    Dowden J. Kapadia P and Fenn B 1993 Space charge in plasma arc welding cutting J. Phys. D: Appl. Phys. 26 1215-23

[20]    Li L, Brookfield D J and Steen W M 1996 Plasma sensor for in-process, non-contact monitoring of the laser welding process Meas. Sci. Technol. 7 615-26

[21]    Zhang Y M, Li P J and Zhang S B 2002 Apparatus, system, and related method for sensing a characteristic of a workpiece in an automated process US Patent No 6437281

[22]    Dendy R 1993 Plasma Physics: An Introduction Course (Cambridge: Cambridge University Press)

[23]    Auciello O and Flamm D L 1989 Plasma Diagnostics vol 2 Surface Analysis and Interactions (Washington, DC: Academic)

[24]    Cambel B 1963 Plasma Physics and Magnetofluid-Mechanism (New York: McGraw-Hill)

[25]    Lu W, Zhang Y M and Lin W-Y 2004 Nonlinear interval model control of quasi-keyhole arc welding process Automatica 40 805-13

[26]    Zhang Y M and Liu Y C 2003 Modeling and control of quasi-keyhole arc welding process Control Eng. Pract. 11 1401-11