Probability of an egg cracking during packing can be predicted using a simple non-destructive acoustic test

Abstract

1. The aim of this investigation was to test the predictive power of the dynamic stiffness measurement to identify eggs which are most likely to crack under field conditions.

2. A representative sample of eggs (n = 1660) was collected from the front of the cages in a commercial battery unit. Egg weight,% damping and dynamic stiffness (K _dyn) were recorded using an acoustic crack detection device. Intact eggs were marked and replaced in the front of the cages. These eggs were subsequently passed through online collection, grading and packing machinery, along with a volume of unmarked eggs. At the end of packing the acoustic test was repeated on the marked eggs, and these were subsequently categorised as being either intact (0) or cracked (1).

3. A logistic regression of the probability of cracking vs K _dyn revealed that as the K _dyn measurement decreases below 15 000 N/m there is a rapid increase in the probability that an egg will crack during routine handling.

4. Additional variables (visit, egg weight,% damping and position in the house (battery [1 to 7], side [1, 2] and tier [1 to 8]) were also fitted to the model but only egg weight, visit and tier effects significantly improved the model fit.

5. This study confirms that the dynamic stiffness measurement can predict the probability of an egg cracking in the field and with high precision. As this measurement also has a high heritability, it could be incorporated into breeding programmes, where it would offer an excellent method to improve eggshell quality and reduce the incidence of cracked eggs.

Introduction

Cracked and damaged eggs can account for between 6 and 8% of total production (Hamilton et al., 1979) and can be particularly problematic in older flocks. This results in substantial economic loss to the egg industry (Hunton, 1995). The economic losses on a worldwide basis are difficult to estimate but for the UK industry an estimate can be made using the statistical data published annually by DEFRA. In 2005, for example, the total eggs passing through UK packing stations was estimated to be 24 741 000 cases (360 eggs per case). Assuming an average breakage of 5%, then the losses sustained by the UK industry in 2005 could have amounted to as much as £16·7 million, based on an average packer to producer price of 45 pence/dozen. Cracked eggs don’t just affect the economy however; they also provide a direct route for pathogenic organisms to contaminate the egg contents (Messens et al., 2005) and as a result constitute a risk in terms of food safety. It is therefore in the interests of both egg producers and consumers that an attempt is made to reduce the number of cracked and damaged eggs.

Obviously any egg will break if it is hit hard enough (Carter, 1970) and as a result considerable effort has been directed at ways of reducing the insults experienced by eggs during routine egg handling procedures (Hamilton et al., 1979), and more recently at ways of detecting cracked eggshells online in packing stations (De Ketelaere et al., 2004). A complementary strategy is to improve the strength and quality of the eggshell by breeding.

Both environmental and genetic factors affect the strength or quality of any eggshell, and hence its likelihood of cracking during normal egg handling processes. To some extent the environmental factors can be controlled, for example, through improvements in bird management and nutrition. Nevertheless genetics remains an important way to reduce eggshell breakage, as it is both permanent and cumulative. For decades, breeding companies have used laboratory-based measurements such as shell breaking strength, non-destructive deformation and specific gravity in their selection programmes (Preisinger and Flock, 2000). While these measurements have generally responded to selection, it has been notoriously difficult to prove that they directly relate to the incidence of breakage in the field (Thompson et al., 1985; Grunder et al., 1991).

Coucke (1998) and Coucke et al. (1999) devised a simple acoustic resonance test, which can be used to calculate the mechanical or dynamic stiffness of intact eggs. This measurement has subsequently been found to correlate well with other measures including static stiffness (r = 0·90) and eggshell thickness (r = 0·78) (De Ketelaere et al., 2002). Dunn et al. (2005) reported that the dynamic stiffness measurement also has both a high heritability (0·53) and a high genetic correlation with eggshell breaking strength (0·49). These results indicate that this novel measurement could be used successfully in a breeding programme to improve eggshell quality. Genetic progress in eggshell quality, however, depends not only on having a measurement whose variance contains a substantial genetic component; it must also be shown to relate to the incidence of breakages in the field (Grunder et al., 1991). Compared to other eggshell quality measurements, dynamic stiffness has the benefit of being a non-destructive measurement, which can be rapidly performed using a mobile and inexpensive piece of equipment. The aim of this investigation was therefore to directly test the possibility that the dynamic stiffness measurement can predict whether an egg will crack during routine egg handling procedures, and thus confirm the potential benefits of this measurement as a means of improving eggshell quality and reducing the incidence of cracked eggs by breeding.

Materials and methods

Animals and egg collection

On two separate occasions a representative sample of eggs were collected by hand from the front of the first two cages in a conventional commercial unit containing 135 000 Hyline brown layers with online egg collection, grading and packaging facilities. The unit consisted of 7 batteries in which the cages were arranged 8 tiers high with a walking platform between tiers 4 and 5. The birds each received a standard layer ration and a constant lighting programme of 15 h of light per day. There were 5 birds per cage and the birds were 69 weeks of age on the first visit and 70 weeks of age on the second visit. This particular age group of birds was chosen to ensure that there would be enough cracked eggs passing though the packing process to allow us to prove statistically the link between the dynamic stiffness measurement and % of cracked eggs, whilst sampling a manageable number of eggs. The incidence of cracking was reported by the producer to be between 5 and 7% in the house just prior to the first visit.

Source and treatment of eggs

On each visit between 135 and 376 eggs were taken from each of the 7 batteries, with roughly similar numbers from each side [1 and 2] and from each tier [1 to 8] (range 173 to 297 eggs). Each egg was clearly numbered with an indelible pen in such a way that it was possible to trace it back to its origin (battery, side and tier), then tested using the lab-scale version of the acoustic test equipment described by De Ketelaere et al. (2002). The acoustic test algorithm designed by these authors distinguishes between intact and cracked eggs and also records egg weight (g), dynamic stiffness (K _dyn, N/m) and % damping. For the purpose of this investigation only the intact eggs (total n = 1660) were replaced in the front of the cages. The egg collection belts were then run and the marked eggs, along with approximately 8 times the number of unmarked eggs, were conveyed from the production unit though the adjacent grading and packing station. Both the online dirt detection equipment and the online crack detector were turned off to prevent removal of marked or cracked eggs. All of the marked eggs were subsequently retrieved from the egg boxes at the end of each packaging line and the acoustic test repeated using the same apparatus. Each marked egg was subsequently categorised as being either intact (0) or cracked (1).

Statistical analysis

As a preliminary, we examined whether the K _dyn values measured at the cage front differed between eggs which remained intact, and eggs which cracked after passing through the collection, grading and packing chain. This analysis of variance also allowed for possible differences between the two visits. However, the main aim of our study was to determine if the K _dyn measurement could be used to predict the probability of egg cracking. To address this question a logistic regression of the probability of cracking (Pr(Crack)) vs K _dyn was fitted. This is a weighted regression analysis of the variable, y =log(Pr(Crack)/(1 − Pr(Crack)), and was fitted using Genstat version 6·1 (VSN International Ltd, Oxford, UK). Additional variables (egg weight,% damping, visit [1, 2] and position in the house, namely, battery [1 to 7], side [1, 2] and tier [1 to 8]) were also fitted to the model, and retained if they significantly improved the model fit.

The relationship of y and K _dyn was found not to be linear, and was better fitted by a quadratic curve. This curve decreased with increasing K _dyn and then increased again at higher values of K _dyn. This pattern did not seem plausible, so in order that the relationship should decline but flatten out for higher values, K _dyn was transformed to transK _dyn using the following expression: . This transformation was the negative integer power which gave the minimum deviance, or equivalently the most significant improvement in fit when accounting for K _dyn, rescaled by 10¹² to give values between 0 and 3·3. This improved the predictive power of the dynamic stiffness measurement whilst still fitting only a single parameter.

Predictions of the probability of cracking for both K _dyn and egg weight were calculated and compared, together with approximate confidence limits which were derived from models including either transK _dyn or egg weight, but not both. As neither K _dyn nor its transformed value (transK _dyn) showed any relationship with egg weight, the predictions of transK _dyn and egg weight were very similar in both the full and the reduced models.

Results

Incidence of cracks

Two hundred and forty-two eggs out of a total of 1660 test eggs cracked as a result of collection, grading and packaging in this study. The number of cracked eggs identified by the acoustic test equipment on both visits was consistently higher (10% for visit 1 and 15% for visit 2) than that reported by the producer (5 to 7%).

K _dyn measurements on intact eggs before and after grading

There was close agreement between the initial measurements of K _dyn taken at the cage front and the measurements taken after grading and packaging on both visits 1 and 2 (R ² = 85·4 and 93·8%, respectively).

Relationship between K _dyn and egg weight

There was no correlation between K _dyn and egg weight (r = −0·02, P = 0·49). This result validates the linear normalisation for egg weight in the dynamic stiffness calculation (Coucke et al., 1999).

K _dyn measurements on intact and cracked eggs

Analysis of variance revealed that in general the eggs which cracked as a result of the handling procedures, had significantly lower K _dyn values than the eggs which remained intact (t ₁₆₁₇ = 7·87, P < 0·001). The mean and standard deviations of K _dyn for intact eggs was 14 293 ± 1735 N/m (range 8405 to 19 829 N/m), and for cracked eggs was 13 055 ± 1943 N/m (range 6728 to 19 714 N/m). The K _dyn values were approximately normally distributed, and the variances for K _dyn were similar for intact and cracked eggs (Figure 1). The combined effects of cracked/intact eggs and visits [1, 2], however, accounted for only 6% of the overall variation in the K _dyn measurements.

Figure 1. Frequency distribution of initial K _dyn (N/m) for intact eggs and cracked eggs. Intact eggs white bars, cracked eggs solid bars.

Effects of K _dyn, egg weight, % damping, visit and position in the house (battery, side, tier) on the probability of an egg cracking (Pr(Crack))

The logistic regression analysis revealed that there were significant effects of K _dyn, egg weight, visit [1, 2] and tier [1 to 8] on the probability of cracking (Pr(Crack)). Battery [1 to 7], side [1, 2] and % damping did not have a significant effect. The estimated effects, along with their standard errors for the terms retained in the logistic regression model are summarised in the Table.

Table The estimated effects, along with their standard errors for logistic regression on the probability of cracking (Pr(Crack)) vs dynamic stiffness (K _dyn)

Effect	Estimate	SE	t (1645 DF)	P
Constant	−7·57	0·969
Visit (2 − 1)	0·659	0·178	3·70	<0·001
Tiers ([3, 4, 7 and 8] – [1, 2, 5 and 6])	0·516	0·147	3·50	<0·001
transK dyn (N/m)	3·449	0·388	8·89	<0·001
Egg weight (g)	0·052	0·013	3·87	<0·001

The tier effect (  = 19·3, P < 0·01) was found to be more succinctly summarised by the single degree of freedom comparison which was achieved by combining and averaging the effects of tiers [1, 2, 5 and 6] and tiers [3, 4, 7 and 8]. This simplification was based on the pattern of tier effects and the house layout. The walkway at the height of tiers 4 and 5 divided the vertical space of the house into two equal halves. Thus, tiers 1 and 2 in the lower half of the house were equated and combined with tiers 5 and 6 in the upper half of the house because of their relative height from either the floor or the partitioning walkway.

The probability of cracking was increased on the second visit; increased for combined tiers [3, 4, 7 and 8], decreased with increase in K _dyn and increased for heavier eggs. The estimates for the two covariates, transK _dyn and weight, were independent (r  = 0·02), as expected.

The parameters were estimated on the logistic scale of the probability of cracking (p). For q = 1 − p, this means log(p/q) = constant +visit (2 − 1) + tier([3, 4, 7 and 8] − [1, 2, 5 and 6]) + 3·4 * transK _dyn + 0·052 * egg weight. This expression allows a useful interpretation for the egg weight effect on p/q, the odds ratio of an egg cracking. That is, for a unit increase in egg weight the odds ratio increases by 5% (a multiple e ^0·052 = 1·05). No such interpretation, however, is possible with K _dyn because a unit increase in transK _dyn results in different increases in K _dyn depending on the level of transK _dyn.

The simplified tier effect allowed a clearer picture of the effects of K _dyn and egg weight on the probability of cracking (Pr(Crack)) to be established (Figures 2 and 3, respectively). Figure 2, for example, shows that the effect of K _dyn on Pr(Crack) is relatively small for a wide range of higher K _dyn values, but increases sharply for K _dyn in the range 10 000 to 15 000, depending on the visit and combined tier, namely, those with higher proportions of cracked eggs show this increase at values of K _dyn nearer to 15 000. In contrast, the change in Pr(Crack) is less marked over the observed range of egg weights (Figure 3). For visit 2 and combined tiers [1, 2, 5 and 6], the overall estimate of K _dyn and egg weight for the probability of cracking is shown as a contour plot in Figure 4. This figure shows that the effect of egg weight becomes more influential for higher values of K _dyn.

Figure 2. Graph of the fitted probability of cracking (Pr(Crack)) vs K _dyn (N/m) for the model excluding egg weight. The different visits are indicated by the colour of the lines: first visit, black lines; second visit, grey lines. The combined tier effect is indicated by the line style: tiers [1, 2, 5 and 6], solid lines; tiers [3, 4, 7 and 8], dashed lines.

Figure 3. Graph of fitted probability of cracking (Pr(Crack)) vs egg weight (g) for the model excluding K _dyn. The different visits are indicated by the colour of the lines: first visit, black lines; second visit, grey lines. The combined tier effect is indicated by the line style: tiers [1, 2, 5 and 6], solid lines; tiers [3, 4, 7 and 8], dashed lines.

Figure 4. Contours of the fitted probability of cracking (Pr(Crack)) vs K _dyn (N/m) and egg weight (g) for visit 2, and combined tiers [1, 2, 5 and 6]. Values on the contour lines indicate the probability of cracking.

Precision of the estimates of K _dyn and egg weight

Conservative 95% confidence limits for egg weight, using slope ±2* SE(slope), showed that the effect of a unit increase in egg weight gives an increase of between 3 and 8% in the odds of cracking (data not presented). Conservative confidence limits for changes in Pr(Crack) with K _dyn were considerably narrower than those for egg weight, and are shown for visit 2 and combined tiers [1, 2, 5 and 6], along with the deviance-based limits which are narrower still (Figure 5). Transforming K _dyn to transK _dyn resulted in markedly tighter confidence limits. Similar patterns to Figures 4 and 5 were observed for the other combinations of visit and tier.

Figure 5. Fitted probability of cracking (Pr(Crack)) for visit 2 and combined tiers [1, 2, 5 and 6] (solid black line), for the model excluding egg weight, with conservative confidence limits from slope ±2* SE(slope) [dotted grey lines] and deviance-based confidence limits [dashed grey lines].

Discussion

In this study we show that the dynamic stiffness measurement (K _dyn) can be used to predict the probability that an egg will crack (Pr(Crack)) during routine handling, and that our estimates have high precision. These results were achieved by measuring the dynamic stiffness of eggs at the cage front, and observing the fate of individual eggs as they passed through a commercial online grading and packaging facility. Previously, the relationship between eggshell quality measurements and the incidence of breakage in the field have been based on correlative observations rather than direct measurement. Thompson et al. (1985), for example, categorised the birds from two flocks into groups on the basis of 4 different eggshell quality characteristics, then correlated this with how subsequent eggs from each group fared during routine egg handling procedures. In general these authors found that the eggs from the highest quality group suffered the least amount of damage; however, there were still significant numbers (15%) of cracked eggs within their high quality group. Grunder et al. (1991) used a similar correlative method to reach the conclusion that specific gravity was the best eggshell quality trait for a selection programme based on heritability, genetic correlation with% intact eggs, and ease of measurement.

Figure 1 shows that the eggs which cracked in the current study had lower initial K _dyn values than those which remained intact after collection, grading and packing, an effect that was consistent over the two consecutive visits. A logistic regression was then used to try and explain the variation in the probability of cracking, looking at the effects of K _dyn,% damping, egg weight and position in the house (battery, side and tier). This showed that the likelihood of cracking was much higher for those eggs which had K _dyn values of less than 15 000 N/m (Figure 2). The precision of K _dyn as a predictor of cracking was also very high (Figure 4). By including information on the other variables known to influence the prediction, particularly the tier from which the eggs originated, the prediction was improved, although the effect of K _dyn on the probability of cracking remained the dominant effect in the model (Table and Figure 4).

One of the possible applications for the K _dyn measurement arising from this study is to use it as a means of identifying those eggs which are most at risk during normal handling practices. If, for example, we had set a minimum cut-off value of 11 000 N/m for K _dyn at the cage front then, according to our data, this would have reduced the number of eggs which subsequently cracked by approximately 12·2%. In an industrial context, removal of high risk eggs at the earliest opportunity would theoretically benefit producers as this would reduce the number of eggs which crack and cause further downgrading due to cross-contamination. Consumers would also benefit as they are less likely to receive a damaged product.

An increase in the probability of cracking for heavier eggs was also observed in this study (Figure 4). This seems plausible, as all other things being equal, heavier eggs will develop a greater momentum during collisions with either other eggs or part of the collection equipment, and as a result will have a higher likelihood of becoming damaged. In other words, for larger eggs the magnitude of the impacts will more often exceed the strength of the shell.

Our model predicts that eggs from the top two tiers in each level within the house, namely, tiers 3, 4 and 7, 8, have a higher likelihood of cracking compared to the lower two tiers in each level (tiers 1, 2 and 5, 6). This effect could be explained by the fact that the eggs in these locations are more likely to experience a higher incidence and magnitude of impacts due to the increased number of angles and turns encountered as they are transported through the vertical dimensions of the house.

The numbers of cracked eggs identified by the acoustic test equipment in this study were consistently higher than that reported by the producer (10% for visit 1 and 15% for visit 2, compared to 5% and 7%). One plausible explanation for this relates to the fact that we assessed our eggs after they had passed though the grading and packing machinery whereas online commercial crack detectors carry out their assessment before the eggs enter the egg grading equipment. Also, the figure provided by the producer represented the average % cracks for the whole house whereas our measurements were carried out on a relatively small sample from one particular area of the house. It is therefore possible that the eggs used in our study came from an area in the house which had a higher than average % crack rate. An adjustment to the operation of the collecting belts could equally explain the difference in % cracks recorded between visits 1 and 2.

Previously it was reported that K _dyn is not only non-destructive and easy to perform, it also has a moderate to high heritability and has a good correlation with quasi-static measurements (De Ketelaere et al., 2002; Dunn et al., 2005). The heritability and genetic correlation results, in combination with the data presented here, strongly support the contention that by incorporating the dynamic stiffness measurement into genetic selection programmes eggshell quality will improve and the number of eggs downgraded due to cracked and damaged shells will be reduced.

In conclusion this study confirms that the dynamic stiffness measurement (K _dyn) can predict with high precision whether an egg will crack during its sojourn through a commercial egg packing station and supports the hypothesis that K _dyn is an excellent tool for breeding companies to use to select for improved eggshell quality in the future. These findings are particularly timely given the move away from battery cages to alterative husbandry systems where the risk of cracked eggs becoming contaminated with potentially harmful bacteria is regarded as being higher.

Acknowledgements

This study was supported by EU project QLK5-CT-2001-01606 ‘EGGDEFENCE’ and FWO, project G.0221.00. The authors would also like to thank Glenrath Farms Ltd, Peebleshire, UK, for facilitating the studies on one of their farms.