WO1998045078A1

WO1998045078A1 - Weld quality measurement

Info

Publication number: WO1998045078A1
Application number: PCT/AU1998/000244
Authority: WO
Inventors: Stephen Walter Simpson
Original assignee: University of Sydney
Current assignee: University of Sydney
Priority date: 1997-04-08
Filing date: 1998-04-08
Publication date: 1998-10-15
Anticipated expiration: 1999-10-08
Also published as: AUPO607397A0; ATE249906T1; PT1007263E; EP1007263A4; DE69818275D1; DK1007263T3; ES2202826T3; EP1007263A1; US6288364B1; DE69818275T2; EP1007263B1; KR20010006180A; CA2285561C; CA2285561A1; KR100503778B1

Abstract

This invention concerns weld quality measurement. In particular it concerns an apparatus and a process for measuring on-line, while the welding process is under way, the quality of the resulting weldment. The invention is applicable to spray-transfer gas-metal arc welding, short-circuiting transfer gas-metal arc welding, pulse welding, radio-frequency resistance welding and submerged arc welding. It involves the measurement of voltage or current and the generation of an artificial signal for the other. A two dimensional signal analysis then produces data for comparison with data obtained from a high quality weld.

Description

WELD QUALITY MEASUREMENT

Technical Field

This invention concerns weld quality measurement. In particular it concerns an apparatus and a process for measuring on-line, while the welding process is under way, the quality of the resulting weldment. The invention is applicable to gas-metal arc welding, tungsten-inert gas welding, pulsed welding, resistance welding, submerged arc welding and to other welding processes where there is an arc plasma.

Background Art

The study of welding and cutting arc phenomena, involves the observation of both voltage and current signals having periods of milliseconds to seconds, or even micro-seconds. One way of monitoring these signals involves the use of high speed photography, and another is the use of oscillograms. The limitations inherent in the observation techniques and the difficulties in analysing the resulting data, make it difficult to provide a weld quality measvirement in real time.

Summary of the Invention

In a first aspect, the invention is an apparatus for measuring the quality of a weld. The apparatvis comprises: sampling means to sample either the welding current or the welding voltage to provide a series of values for a first signal. A second sampling means may be employed to measure the other variable to provide a series of values for a second signal. Alternatively, a signal generating means uses the first signal to generate a series of values for an artificial second signal, which depends upon at least some values of the first signal either explicitly or through a recurrence relation. For example, where voltage V is measured, an artificial current V can be mathematically generated using:

r„ = e-^Δt/T_n„, - V_n (1)

where τ is a constant which may be selected, and n is the sample number. This approximation can model the usual inductive-resistive circuit of a power supply but need not be an accLirate model since the artificial signal need only provide information about the time history of the sequence.

Using the symbols D_n for the real data sequence and A_n for the artificial sequence, two useful possibilities are

A_n = D_n._k (3)

where integer k>0. The first of these is similar to equation (1). The second possibility is a simple return variable.

Pairing means identify corresponding values of the first and second signals. Collection means collect pairs of values which are useful for quality monitoring into groups or regions. The pairs collected could be visualised to be those that wovύd fall within selected regions of a two dimensional scatter plot of the values of the first and second signals. The regions could be drawn on to such a visualisation. The regions need not be of equal size, and they may be smaller where population density is greatest and may be exponentially greater in dimension, in both the voltage and current direction, as they progress away from the point of greatest population density. Once the regions are chosen they are fixed during the monitoring process. In the case of 'dip' or short circuiting metal transfer in gas metal arc welding, there are large oscillations in voltage and current.

The regions selected will usually be those around the area of greatest density of sample points. However, the regions selected need not be contiguous. The population of sample points for each selected region can be represented by a two dimensional population density function ξ. for a set of regions r= 1 to m.

Multiplication means multiply the set of populations L by weights w_qr defined for the same set of regions, and sum means then sum the products to produce a set of new values for G_r, where G_r = ∑ _qr f_q ... r= l to m (4) q=l

To produce the final adj isted region populations P_r a function F is applied to each of the G_r values:

P_r = F(G_r) ... r= l to m (5)

F is a single-valued monotonic function.

The complete set {P_α....P_m} of the P_r collected is the welding signature. The weights w_qr are chosen to produce a welding signature which contains as much information about the properties of the final weld as possible for a given sampling rate and size. This may be done experimentally, by trial and error adjustment or by knowledge of the physical process. Since there is some statistical noise in the sample, it is useful to choose the w_qr to smooth the welding signature. The function F is chosen to maximise the sensitivity of the welding signature to changes in the quality of the final weld.

The sampling means repetitively provides a series of values and a new welding signatvire is produced for each series. Memory means retain a welding signature R= {P₁...P_m} collected under welding conditions known to be satisfactory and producing a high quality weldment. This may be reference data saved for some time, or could be data collected at the start of a welding run. In the case of a robotic welding, where a sequence of welds is carried out under conditions which may vary, a sequence of reference signatures may be stored and recalled when needed.

The reference signature can also be calculated continuously during welding from previous sampling. In this case the reference is a weighted average of the x signatures S^ S₂, S₃ ....S_x where S_t is the most recent signature calculated, S₂ is the signature calculated before that and so on. The reference signature R is determined from the weighted average

r_j=W₁s_lj+W₂s_2j+W₃s_3j+ +W_xs_xj j= l to m (6)

where T- becomes the adjusted region population numbered j in the reference signature R; s_αj to s_xj are the adjusted region populations numbered j in the signatures S_t to S_x calculated from previous sampling ; and W_x to W_x are signature weighting factors. The choice of the signature weighting factors W_α to W_x determines whether the reference represents an average of weld signature behaviour over a relatively long period of time or represents recent welding behaviour.

When signatures are multiplied or divided by a number, it is understood that every adjusted region population in the signature should be multiplied or divided by the number to produce a new signature. Similarly when signatures are added or subtracted, the matching adjusted region populations of each signature are added or subtracted, that is, the adjusted region population numbered j in one signature is added or subtracted from the adjusted region population numbered j in the other signature for j= 1, 2 up to m. The equation above can then be written more succinctly as

R=W₁S₁+W₂S₂+W₃S₃ + +W_XS_X (7)

Weld quality result calculation means then compare the welding signatures with the reference welding signature to produce a measure of weld quality. The part U of a welding signature S which is does not match the reference signature R is given by

υ = s - [R. R) (8)

where A.B is the inner product of two signatures A and B. If U is zero there is a perfect match.

The quality factor q may be defined by

U. U

(9) S. S

The quality q will be unity if U is zero and zero if U=S and S.R=0. A value of q=l would indicate perfect quality. As welding conditions deviate from ideal due to any faults in the welding process, S will no longer match R and q< l.

The inner, or dot, product of any two signatures A and B is defined by:

m A.B = 2_ ^a> ^{x b}> (11)

1=1

where a_s and b_j are the adjusted region populations P₁ of the signatures A and B respectively.

In another aspect, as currently envisaged, the invention provides a method of measuring weld quality comprising the steps of:

Sampling either the welding current or the welding voltage to provide a series of values for a first signal.

Sampling the other variable to provide a series of values for a second signal. Alternatively, generating a series of values for the second signal, which depends upon at least some values of the first signal either explicitly or through a recurrence relation.

Pairing corresponding values of the first and second signals. Collecting pairs of values which are useful for quality monitoring. This step could be visualised to involve plotting a two dimensional scatter plot, of the values of the first and second signals. The plot is then divided into regions. The regions need not be of equal size, and they may be smaller where population density is greatest and may be exponentially greater in dimension in both the voltage and current direction as they progress away from the point of greatest population density. Once the regions are chosen they are fixed during the monitoring process. The regions selected will usually be those around the area of greatest density of sample points. However, the regions selected need not be contiguous.

The process then continues by representing the population of sample points for each selected region by a two dimensional population density function f_r for a set of regions r= 1 to m.

Multiplying the set of populations f_q by weights w_qr defined for the same set of regions.

Summing the products to produce a set of new values for G_r, where G_r = ∑ _qr f_q ... r=l to m (12) q=l

Applying a function F to each of the G_r to produce the adjusted region populations P_r:

P_r = F(G_r) ... r=l to m (13)

Identifying the complete set {P ...P_m} of the P_r collected as the welding signature.

Repetitively sampling series of values to provide successive welding signatures.

Storing a welding signature R= {P₁...P_m} collected under welding conditions known to be satisfactory and producing a high quality weldment, or, alternatively, calculating a weighted average reference from previous signatures. Comparing the welding signatures with the reference welding signature to produce a measure of weld quality

Brief Description of the Drawings

An example of the invention will now be described with reference the accompanying drawings, in which:

Figure 1 is a two dimensional scatter plot of measured voltage and synthetic current;

Figure 2 is a population density distribution of selected regions of Figure 1; Figure 3 is a weighted version of the population density of Figure 2;

Figure 4 is an adjusted version of the population density of Figure 3 and shows a welding signature; and

Figure 5 is a graphical representation of a two dimensional subspace of reference signatures.

Best Modes for Carrying Out the Invention

In short-circuiting transfer in gas-metal arc welding, short circuiting occurs repetitively. The voltage wave form will as a result show generally rectangular pulses at about 20 volts separated by regions where the voltage falls to near zero. This voltage is sampled (s) 1024 times with a sampling time interval (Δt) of 0.5 milli-seconds over a sampling period (T) of 0.5115 seconds (T= (s-l).Δt).

It may not be convenient to collect current signals as well because measuring the arc current would require costly hardware such as a current transformer.

A synthetic current signal A_n is generated from the sampled voltage signal D_n that has been measured using the following formula:

A- = e-^A*(A_»-_ - D-.J (14)

where τ is the welding circuit time constant selected to be equal to four times the sampling interval; about 2 milli-seconds. The first artificial value A₁ may be set to zero.

The artificial current values may then be plotted against the voltage values as shown in Figure 1 in order to visualise the process.

The plot shown in Figure 1 is then divided in to rectangular regions of varying sizes.

The regions are chosen adaptively, based on the data itself. From a given set of reference data, a point (D₀ A is chosen to be at the mode of the sampled distribution; that is the point where the most data lies. The regions are chosen to have a width in the D direction which is smallest near the mode and tends to infinity at the edge regions. This means that resolution is improved where many data points are present. The width function is such that the width of the interval located at D_w is proportional to:

where ΔD is the standard deviation of the D_n, the set of values of the sampled voltage signal, and α is constant (set to unity in this example). An analogous procedure is used to set the height of the rectangular populations in the A direction, and there are a total of m (= 27x27) which equals 729 rectangular population regions.

The raw population density distribution resulting is shown in Figure 2. Once the population regions are chosen, they are fixed during the monitoring process, the same for both the reference signature and the monitored signatures. Weights w_qr are set to smooth the population distribution of the chosen regions in two dimensions, and the weights are selected according to:

w, _ _Θ -β (D_q - D_r) ²/(ΔD) ²-β(A_q - A_r)²/(ΔA)² (16)

where ΔA is the standard deviation of the A-, the set of calculated values of the artificial current, and β is a constant which determines the degree of smoothing (set to 25 in this example). (D_q , A_q) and (D_r , A_r) are the locations of the two regions q and r. Once the weights are chosen, they are fixed during the monitoring process, the same for both the reference signature and the monitored signatLire.

The smoothed distribvition, the set of region populations {G^.-G_u is shown in Figure 3. The set of adjusted region populations P_r is then calculated by applying the function F to the G_r values:

P, = F(G_X) ... r=l to m (17)

The single-valued monotonic function F is chosen to maximise the sensitivity of the welding signature to changes in the quality of the final weld. The function F may be chosen be a power law: F(x)=x^λ where λ is fixed, with the additional special value F(0)=0. If 0<λ<l, regions with low populations are emphasised in the welding signature, which can improve the sensitivity of the technique. λ=0.6 has been found to be a suitable choice. The set of adjusted region populations {P₁...P_m}, which is the final welding signature is shown in Figure 4. This can be compared with the signature from a reference weld.

Welding quality can be ascertained by an inspection of Figure 4, but it is convenient to calculate a quality indicator q, defined by:

R. S q = (18)

VRTRXVSTS

where R is a reference weld signature, S is a measured weld signature, and the inner, or dot, product of any two signatures A and B is defined by: m

i=ι where a_j and b_j are the adjusted region populations of the signatures A and B respectively.

The weld quality measurement could be fed back in some way to control the welding operation if required.

The invention may also be applied to situations using multiple references.

Suppose R_α and R₂ are the two reference signatures, recorded and stored during a welding run which produced a high-quality weldment. They should reflect the range of expected normal variation during the welding run.

Examples are:

(a) changes in joint geometry/surface condition and welding head orientation s ich as might happen with robotic welding of a complex work piece. (b) programmed lead in or shut down sequences, or programmed changes in the operating conditions.

(c) unintentional changes in welding, for example due to work piece heating through a run, which nevertheless do not degrade weld quality.

A signature S has been collected during another run and is to be compared with R_α and R₂.

Definition:

A normalised or Linit signature U' is calculated from any signature U as

U' = -^= (20)

Vuϋ

A signature P orthogonal to R_t is given by

P=R--^ )

Ri. R₁ (21 Ri

P will be nonzero provided R_t and R₂ are independent signatures, that is, not the same signature or signatures whose elements, the adjusted region populations, differ from each other by a constant multiplying factor. P' and R.,' are orthogonal unit signatures in the two-dimensional linear signature SLibspace defined by R_α and R₂. The component B of S¹ which lies in the subspace is

B = (S'.R'JR'α + (S'.P'JP' (22)

The component C of S' which is orthogonal to the subspace is

C = S' - B (23)

If C is non-zero, then the welding conditions deviate from the ideal and C can be used as part of a measure of weld quality.

Figure 5 illustrates the two dimensional subspace of the reference signatures with three possible locations for B. If B is inside the angle subtended by R- and R₂' (position 2), then the quality of the weld is satisfactory as far as the component B is concerned since the normalised welding signature is expected to move between Rj' and R₂' during normal welding as conditions change. However of B is in either of positions 1 or 3, this represents a discrepancy from ideal in addition to the discrepancy associated with non-zero C. Using the fact that the adjusted region populations making up the signatures are never negative, the condition that B lies between R_x' and R₂' can be written

(RVR'iMB'-R'j) > 0 and (R R'zMB'-Ry > 0 (24)

If this condition is satisfied then C gives the part of the measured signature which differs from the reference signatures and the quality q becomes

q = Vl - C.C (25)

Conversely, if the above conditions are not both satisfied, then either R or R₂' will be closest to S¹ and the quality q should be taken as the larger of R^.S¹ and R₂'.S'. These are the same values as would result from using a single reference of R^ or R₂ respectively. The above considerations can be extended to a three-reference system, R_x, R₂, and R₃. The component of S' in the three-dimensional subspace generated by R R₂, and R₃ is

B = (S^l.R^, ₁)R'₁+(S'.P_j)P₁+(S'.P'₁₂)P^l ₁₂ (26) where P_α = R₂ - (R'α.R₂)R'

with

Again the component of S' orthogonal to the subspace is C=S'-B. An approximate set of conditions for vector B' to lie within the solid region formed by R_t, R₂, and R₃ is

(R'₃-R^, ₁).(R^,2-R^, ₁)

(S'-R'J. R',-R'ι- (R^*2-R'ι) >o (27)

and

(R^l _rR^l ₂).(R^,3-R'₂)_ι

(S'-R'_z). R'ι-R'2 (R'3-R'₂) >0 (28)

(R'₃-R'₂)^.(R'₃-R'2) and (R'z-R^XR^-R^)

(S'-R's). RS-RS- (R'l-R's) >0 (29) (R^-R'aMR^-R'a)

If these conditions are all satisfied then the quality q is given by

q = Vl-C.C (30)

Conversely, if the three conditions are not all satisfied then the quality should be calculated successively for the three pairs of references R_u and R₂ ; R₂ and R₃ ; and R_α and R₃, using the method already defined for a pair of references. The largest of the three resultant quality estimates is taken to be the final quality.

It will be appreciated by persons skilled in the art that numerous variations and/or modifications may be made to the invention as shown in the specific embodiments without departing from the spirit or scope of the invention as broadly described. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive.

Claims

CLAIMS:

1. An apparatus for measuring the quality of a weld, comprising: sampling means to sample either the welding CLirrent or the welding voltage to provide a series of values for a first signal; a second sampling means to sample the welding voltage or the welding current, whichever is not sampled by the sampling means, to provide a series of values for a second signal; collection means to collect pairs of values of the first and second signals which are useful for quality monitoring into groups; calculation means to multiply values of the two dimensional group populations for each group by a weighting factor and to sum the products for each group; memory means to store a reference set of the resultant adjusted group populations collected for a high quality weldment; and comparison means to compare the reference set of adjusted group populations with another set produced by the calculation means to provide a measure of weld quality for the weld from which the other set was produced.

2. An apparatus for measLiring the quality of a weld, comprising: sampling means to sample either the welding current or the welding voltage to provide a series of values for a first signal; a signal generating means to use the first signal to generate a series of values for an artificial second signal, which depends upon at least some values of the first signal either explicitly or through a recurrence relation; collection means to collect pairs of values of the first and second signals which are useful for quality monitoring into groups; calculation means to multiply values of the two dimensional group populations for each group by a weighting factor and to sum the products for each group; memory means to store a reference set of the resultant adjusted group populations collected for a high quality weldment; and comparison means to compare the reference set of adjusted group populations with another set produced by the calculation means to provide a measure of weld quality for the weld from which the other set was produced.

3. Apparatus according to claim 2, wherein the values of the artificial signal are given by

A_n = e ^τA_n._τ - D_n where D_n are the values of the first signal where τ is a time constant and n is the sample number.

4. An apparatus according to claim 2, wherein the values A_n of the artificial signal are given by:

A_n = e^"Δ _n._t - D_n-1)

where τ is the welding circuit time constant and n is the sample number.

5. An apparatus according to claim 2, wherein the values A_ of the artificial signal are given by:

where D_n are the values of the first signal and k>0.

6. An apparatus according to any preceding claim, wherein the groups of pairs collected by the collection means are those that fall within selected regions of a two dimensional histogram of the values of the first and second signals.

7. An apparatus according to claim 6, wherein the regions are not of equal size.

8. An apparatus according to claim 7, wherein the regions are smaller where the density of sample points is greater.

9. An apparatus according to any one of claims 6, 7 or 8, wherein the regions selected are those having the greatest density of sample points.

10. An apparatus according to claim 1 or 2, wherein the calculation means sums the group population densities ^ weighted by weighting factors w_qr for a set of m regions to produce a set of weighted group populations G_lt G₂ to G_m according to:

G = ∑ w _qrf_q r= 1 to m q = l

11. An apparatus according claim 10, wherein the calculation means also applies a monotonic single-valued function F to each of the G_r values to produce the set of values P_α, P₂ to P_m according to:

P_r = F(G_r) ... r= l to m and the set P-,, P₂ to P_m is the welding signature.

12. An apparatus according to claim 11, wherein the comparison means compares a welding signature produced from a reference set of weighted group populations with a welding signature produced from another set.

13. An apparatus according to claim 12, wherein there is a single reference signatLire R, a welding signature S of a weld and a quality factor q of the weld is defined to be

R. S

VR ΪXVSTS

where the inner, or dot, product of any two signatures A and B is defined by:

m

}=ι

and ^ and b_j are the P_r of the signatures A and B respectively.

14. An apparatus according to claim 12, wherein there are two reference signatures R₁ and R₂, a welding signature S of a weld and a quality factor q of the weld is defined as follows a normalised or unit signature U' is calculated from any signature U as

υ Vu ^uTJ

a signature P orthogonal to R} is given by

P=R₂-^ _Rl Ri .Ri

where the inner, or dot, product of any two signatures A and B is defined by:

m

}=ι and ^ and b_j are the P_r of the signatures A and B respectively, P will be nonzero provided R_α and R₂ are independent signatures, that is, not the same signature or signatures whose elements, the adjusted region populations, differ from each other by a constant multiplying factor; P' and R-,¹ are orthogonal unit signatures in the two-dimensional linear signature subspace defined by R₁ and R₂, and the component B of S¹ which lies in the subspace is

B = (S'.R'_jJR'i + (S'.P')P¹

the component C of S¹ which is orthogonal to the SLibspace is

C = S' - B

the condition that B lies between R_α' and R₂' can be written

(R'a-R'iMB'-R'i) > 0 and (R R'_.MB'-R' > 0

q = Vl - C.C

conversely, if the above conditions are not both satisfied, then the quality q is taken as the larger of R^.S' and R₂'.S'.

15. An apparatus according to claim 12, wherein there are three reference signatures R R₂ and R₃, a welding signature S of a weld and a quality factor q of the weld is defined as follows a normalised or unit signature U is calculated from any signature U as

U' ^U

Vϋu

a signatLire P orthogonal to R_α is given by

where the inner, or dot, product of any two signatures A and B is defined by:

m

A.B = ∑a, x b, ]=i and a_j and b_j are the P_r of the signatures A and B respectively, the component of S' in the three-dimensional subspace generated by R_α, R₂, and R₃ is

B = (S'.R'₁)R'₁+(S'.P₁)P₁+(S^,.P^I ₁₂)P₁₂ where

with

P₂ ⁼ R₃ ^■ ( ι-R₃)Rι

the component of S' orthogonal to the subspace is C = S'-B, and an approximate set of conditions for vector B' to lie within the solid region formed by R-,, R₂, and R₃ is

(S'-R'X "R'₃ -R', - ^(R'^{3 '}^ ^{1 (R}'^{2 ~R>1 )} (R'₂ -K ) > 0 ^{3 l} (R'₂ -R'₁ ).(R'₂ -R'₁ ) ^{2 l} . and

and (R'₂ -R'₃ ).(R'₁-R'₃ )

(S'-R's). R -R' r (R'l-R's ) > 0

(R^-R'a ^R'i-R'a )

if these conditions are all satisfied then the quality q is given by

q = Vl - C.C

conversely, if the three conditions are not all satisfied then the quality is calculated successively for the three pairs of references R_α and R₂ ; R₂ and R₃ ; and R₁ and R₃, using the method of claim 13 and the largest of the three resultant quality estimates is taken to be the final quality.

16. A method of measuring weld quality comprising the steps of: sampling either the welding CLirrent or the welding voltage to provide a series of values for a first signal; sampling the welding voltage or the welding current, whichever has not already been sampled, to provide a series of values for a second signal; pairing corresponding values of the first and second signals, collecting pairs of values which are useful for quality monitoring into groups; calcLilating the products of the two dimensional group populations with fixed weights defined for the same groups and summing the products for each groLip to produce a new set of weighted group populations; storing a reference set of adjusted group populations obtained from producing a high quality weldment, or, alternatively, determining a reference set from a weighted average of sets of adjusted group popLilations calculated previoLisly; and comparing the reference set of adjusted group populations with another set to produce a measure of weld quality for the weld from which the other set was produced.

17. A method of measuring weld quality comprising the steps of: sampling either the welding current or the welding voltage to provide a series of values for a first signal; generating a series of values for an artificial second signal, which depends upon at least some values of the first signal either explicitly or through a recurrence relation; pairing corresponding values of the first and second signals, collecting pairs of values which are Liseful for quality monitoring into groups; calcLilating the products of the two dimensional group populations with fixed weights defined for the same groups and summing the products for each group to produce a new set of weighted group populations; storing a reference set of adjvisted group populations obtained from producing a high quality weldment, or, alternatively, determining a reference set from a weighted average of sets of adjusted group populations calculated previously; and comparing the reference set of adjusted groLip populations with another set to produce a measure of weld quality for the weld from which the other set was produced.

18. A method according to claim 17, wherein the values A_n of the artificial signal are given by: = e ^A _, - D_n

where D_n are the values of the first signal where τ is the welding circuit time constant and n is the sample number.

19. A method according to claim 17, wherein the values A_n of the artificial signal are given by:

where τ is the welding circuit time constant and n is the sample number.

20. A method according to claim 17, wherein the values A_n of the artificial signal are given by:

where D_n are the values of the first signal and k>0..

21. A method according to any one of claims 16 to 20, wherein the groups of pairs collected by the collection means are those that fall within selected regions of a two dimensional histogram of the values of the first and second signals.

22. A method according to claim 21, wherein the regions are not of equal size.

23. A method according to claim 22, wherein the regions are smaller where the density of sample points is greater.

24. A method according to any one of claims 21, 22 or 23, wherein the regions selected are those having the greatest density of sample points.

25. A method according to claim 16 or 17, wherein the calculating step sums the group population densities f_q weighted by weighting factors w for a set of m groups to produce a set of weighted group populations, according to:

∑ _qrf_q r=l to m q=l

26. A method according to claim 25, wherein the calculating step also applies a fixed monotonic single valued function F to each of the weighted group popLilations G_r to produce a new set of adjusted group populations P_r, according to:

P, = F(G_r) ... r=l to m

and the set P_t, P₂ to P_m is the welding signature.

27. A method according to claim 26, wherein the comparing step compares a welding signature produced from the reference set of adjusted group populations with a welding signature produced from another set.

28. A method according to claim 27, wherein there is a single reference signature R, a welding signature Sof a weld and a quality factor q of the weld is defined to be

where R is a reference weld signature, S is a measured weld signature, and the inner, or dot, product of any two signature A and B is defined by:

m

A.B = ∑ a, x b₍ ]^'=ι where a_} and b_j are the adjusted region populations of the signatLires A and B respectively.

29. An method according to claim 27, wherein there are two reference signatLires R₁ and R₂, a welding signature S of a weld and a quality factor q of the weld is defined as follows a normalised or unit signatLire U is calculated from any signature U as

u. ^u

Vϋϋ

a signature P orthogonal to R_α is given by p=R_-________ _Rl

Ri Ri

where the inner, or dot, product of any two signatures A and B is defined by:

m

j=ι and a_j and b_j are the P_r of the signatures A and B respectively, P will be nonzero provided R₁ and R₂ are independent signatures, that is, not the same signature or signatures whose elements, the adjusted region populations, differ from each other by a constant multiplying factor; P¹ and R-,' are orthogonal unit signatLires in the two-dimensional linear signature subspace defined by R_x and R₂, and the component B of S' which lies in the subspace is

B = (S'.R' R¹ _! + (S'.P')P'

the component C of S' which is orthogonal to the SLibspace is

C = S¹ - B

the condition that B lies between R_α' and R₂' can be written

(R R'iMB'-R'i) > 0 and (R R^MB'-Ry > 0

q = Vl - C.C

conversely, if the above conditions are not both satisfied, then the quality q is taken as the larger of Rj .S¹ and R₂'.S'.

30. An method according to claim 27, wherein there are three reference signatures R_1? R₂ and R₃, a welding signature S of a weld and a quality factor q of the weld is defined as follows a normalised or unit signature U¹ is calculated from any signature U as U

U'

VtJU

a signature P orthogonal to R_t is given by

_P=_R2-___ ^_Rl Ri Ri

where the inner, or dot, product of any two signatures A and B is defined by:

m

j=ι and Ά_] and b_j are the P, of the signatures A and B respectively, the component of S¹ in the three-dimensional subspace generated by R,, R₂, and R₃ is

B = (S'.R'₁)R'₁+(S'.P₁)P₁ + (S'.P₁₂)P₁₂ where

Pl2 ⁼ ?2 " (Pl-PzJ l with

the component of S' orthogonal to the subspace is C=S'-B, and an approximate set of conditions for vector B¹ to lie within the solid region formed by R_1; R₂, and R₃ is

(S'-R'J. fR'₃-R'ι- ^(R'^{3"Rl1 ) (R}'^{2"Rl1 )} (R'₂-R'ι ) >o ^lJ l (R'₂-R'₁)(R'₂-R'₁) and (R',-R'₂)(R'₃-R'₂)

(s'-Ry. R'ι-R' T (R'3-R'₂) >0 (R'₃-R'₂)(R'₃-R'₂) and

if these conditions are all satisfied then the quality q is given by

q = Vl - C.C

conversely, if the three conditions are not all satisfied then the quality is calcLilated successively for the three pairs of references R_t and R₂ ; R₂ and R₃; and R₁ and R₃, using the method of claim 27 and the largest of the three resultant quality estimates is taken to be the final quality.