week ELEVEN: Milking Performance and Udder Health of Cows Milked Robotically and Conventionally

week ELEVEN: Milking Performance and Udder Health of Cows Milked Robotically and Conventionally

By Misty A. Davis, Graduate Research Assistant & Douglas J. Reinemann, Professor
University of Wisconsin-Madison, Milking Research and Instruction Lab

Written for presentation at the 2002 ASAE Annual International Meeting / CIGR XVth World Congress


Introduction & Literature Review

Past studies comparing milking performance between cows milked robotically or in a milking
parlor show conflicting results. A recent Israeli study (Shoshani and Chaffer 2002) found
significantly higher daily milk yield (DMY) in a robotic milking system. Other studies have
found significantly less DMY (Kremer and Ordolff 1992; Wirtz et al. 2002) or no significant
differences (Svennersten-Sjaunja et al. 2000, Ordolff and Artmann 2000). Ordolff and
Artmann (2000) reported no significant changes in daily machine-on time (DMT) or milk flow
rate (FR) when robotic milking was compared to parlor milking. These conflicting results
justify further research to elucidate the factors contributing to or limiting potential benefits of
robotic milking technology.

Recent research have focused on improving the sensitivity of mastitis, or abnormal milk,
detection in robotic milking systems. Schemes to use deviations in electrical conductivity
(EC) and daily milk yield (DMY) lack the sensitivity to be used as the sole indicators of
mastitis detection (Batra and McAllister 1984; Chamings et al. 1984; Hamman and Zecconi
1998). The merits of the California Mastitis Test (CMT) in detecting mastitis have been
shown (Wesen et al. 1967) and is widely used. The status of udder health in this study was
quantified using changes in CMT status, EC, DMY and somatic cell count (SCC).


The objectives of this study were:
1. To determine if robotic milking has a significant effect on milking performance or
incidence of new mastitis infections;
2. To compare the following indicators of udder health: quarter CMT score, udder composite
EC, DMY and udder-composite SCC.

Materials & Methods

Measurements of milking performance and udder health were collected for a period of 30
weeks on 110 Holstein cows housed on either side of a 4-row, drive-through freestall barn
with center feed alley. Lights were turned off in the barn for six hours at night. All cows were
fed the same TMR ration. Robotically milked cows received a small amount of grain to
entice them into the milking stall and keep them quite during unit attachment.

The parlor group was milked in a double-6, herringbone parlor twice daily at 12-hour intervals.
The robot was a single-box system serving a single pen of cows. Cow movement in the robot
pen was voluntary with forced one-way traffic. The robot group could be milked at a minimum
interval of six hours (up to four times per day). The robotically milked group was inspected
twice per day to ensure that each cow was milked a minimum of twice per day.

The cows used in this study were selected from the available pool of cows from the UW
research herd (about 350 cows) using the following criteria:
• <200 days in milk (DIM),
• SCC <500,000 cells/ml at time of entry,
• no more than one previous monthly SCC >500,000 cells/ml

Cows admitted to the study were randomly assigned to either the robot or parlor group. The
study period began by introducing 15 cows to each group. The data analyzed in this paper
began on June 9th, 2001. Cows were added in groups of 4 – 12 cows every several weeks. Efforts were made to keep group size equal; however, some variation occurred individual
cows ended their lactation.

Parity and stage of lactation were recorded for each cow entering the study. Milk yield and
machine-on time were recorded automatically in the parlor at every milking. Milk yield,
machine-on time and udder-composite EC was recorded for robotically milked cows at each
milking. Individual udder-composite SCC was measured weekly for both groups of cows.
Milk samples for SCC enumeration were collected by milking staff during the morning milking
for parlor cows and by an automatic sampling device for robot cows.

A weekly CMT was conducted on each quarter of all cows for both groups. CMT scores were
coded as negative (CMT score negative or trace) or positive (CMT score one or greater).
The number of positive quarters was then summed for each cow to obtain the number of
CMT-positive quarters for each cow for each week. A variable called “new infection” was
created to describe the udder health of cows during each week over the test period. A “new
infection” was defined as a change in state from a CMT-negative to a CMT-positive. A
minimum of one CMT-negative was needed to “reset” the new infection score back to zero.

The distribution of all response variables except SCC appeared normal. A log10 transform
normalized the distribution of SCC. The single observations from each milking event were
assembled into a daily record for each cow, where milk yields and machine-on times were
summed for each day for each cow and daily averages were obtained for FR and EC for each

A mixed model with autoregressive correlation was used in SAS (SAS Institute, 2001) to
determine if milking method (robot or parlor) had an effect on milking performance. EC was
not tested between groups, as it was only available for the robotically milked cows. The
response variables DMY, DMT, FR and SCC were repeated measurements on the
experimental unit of cow so the statistical analysis used a split-plot design with a subplot error
of “cow” nested in “milking method”. The full model tested was:

Response = μ + Group + Parity + DIM + ε + DIS + (Group x DIS) + δ
Where: Response = DMY, DMT, FR, Log10(SCC)
μ = mean of response
Group = robot or parlor
Parity = 1 to 6
DIM = days in milk 10 to 452
ε = whole plot error
DIS = number of days each cow was in the study
δ = subplot error

Backwards elimination of non-significant terms and examination of Akaike’s Information
Criterion was used to determine the final model.

A generalized estimating equation (GEE) was used to test for a treatment effect on the
number of new infections at the quarter and cow level. This method was used due to the
combination of repeated measurements of a binomial response and both categorical and
continuous explanatory variables (Stokes et al.,2000). The GEE method was invoked by use
of the GENMOD procedure in SAS (SAS institute, 2001). The GEE model used a binomial
distribution, a type one autoregressive correlation structure and a “logit” link function since the
response was binary. The repeated measurement of “new infection” on the experimental unit
of cow was specified in the model. The full model tested was:

Response = βo + Group + Parity + DIM + DIS + (Group x DIS) + ε
Where: βo = intercept
ε =error

Pearson correlation coefficients were calculated between SCC and CMT score, EC, and

Results and Discussion

A summary of the data is presented in Table 1 and weekly averages of DMY and
Log10(SCC) for each treatment group are plotted in Figures 1 and 2.







It is important to note the difference in the method of measuring milking time for each milking
method. In the parlor group, a button pressed by the operator immediately preceding manual
cluster attachment initiates the milking time; the automatic cluster remover switchpoint was
set to 0.5 kg (1 lb) per minute plus a five-second delay. Teatcups are attached individually by
the robotic milking machine. The milking time robotic milking is initiated by milk flow sensed
in the first quarter. Teatcups are detached when milk flow ceases in each individual quarter
with the milking time ending at low flow for the last quarter.

The sample size, least square means (LSM), standard error (SE) and p-values for
measurements of milking performance and SCC are presented in Table 3. LSM, also known
as adjusted treatment means, were used to account for any bias in assignment of cows to
treatment group.

Figure 1. Daily milk yield averaged by the number of weeks each cow was in the study for
each treatment group.










Figure 2. Somatic cell count by weeks each cow was in the study for each treatment group.










Table 2. Least square means (LSM) and for continuous response variables.







The final model for DMY and SCC included terms for Group, DIM and DIS. The final model
for DMT and FR included terms for Group, Parity, DIM, DIS and (Group x DIS). Milking
method (Group) had a statistically significant effect on DMY, DMT, FR and Log10(SCC). The
robot group was milked 2.6 times daily compared to 2 times daily for the parlor group. The
robot group produced 0.5 kg (1.2 lb) more milk per day and spent 5.1 minutes more in contact
with the milking machine than the parlor group. The FR of milk harvested in the robotic
system was 0.5 kg/min (1.2 lb/min) slower than the conventional parlor. The SCC of the
robot group was 58,000 greater than the parlor group; however, the mean SCC for both
groups is in good standing.

As previously mentioned, the criteria for DMT is different between the groups. The greater
DMT for the robot group may have been influenced by differences in the measurement
criteria as well as delays in finding teats and attaching teat cups. Also important to note is the
calculation of FR, which is simply milk yield divided by milking time for each milking, or the
amount of milk harvested per unit of time. A change in FR, therefore, does not necessarily
imply any physiological changes to the teat or cow. Lower FR in cows milked more frequently
has also been reported by Stewart et al. (1993) due to the increased percentage of time
spent in low-flow conditions.

DMY was suppressed in the robot group during the first few days of adapting to the new
housing and milking system but soon recovered. The small overall gain in DMY in the robot
group was probably due to increased milking frequency, although this difference is smaller
than expected for a difference of 2 versus 2.6 milkings per day. It should be noted that no
special provisions were made to adjust the diet for the robot group and this may have
suppressed the gain in milk yield. The milking frequency of the robot group in this study
(2.6/day) was similar to that observed in other studies: Wirtz et al. (2002) 2.7/day with
significantly lower milk yield with robotic milking; Svennersten-Sjaunja et al., (2000) 2.4/day
with no difference in milk yield; and Shoshani and Chaffer (2002), 2.8/day with significantly
higher milk yield milk with a robotic system. Management factors other than milking
frequency are clearly important in order to realize increase milk yield through more frequent
milking with a robotic milking system.

A general decline in milk yield was observed for both groups as the study period and stage of
lactation progressed. The depression in milk yield in the parlor group during weeks 10
through 15 was influenced by heat stress during the months of July and August for many of
the cows in the study. It is interesting to note that the robot group did not appear to suffer as
large a decrease in milk yield during this period. The robotically milked cows had a slightly
higher SCC in our study. Though statistically different, neither of the average SCCs in this
study, 100,000 cells/ml (parlor) and 158,000 cells/ml (robot), are cause for concern.

No significant difference between the incidences of “new infection” was found between the
robot and parlor group at either the quarter or cow level. Model convergence was achieved
for each response. The percentage of quarters with “new infection” over the study is
presented in Figure 3. Though the treatment groups did vary slightly in percentage of “new
infections” observed over time, no significant differences were found between treatment

Figure 3. Percentage of quarters with new infection for each group over the study period.











• Cows milked with the robotic milking system produced significantly more milk per day
and spent more time in contact with the milking machine per day than cows milked in
a conventional parlor.

• Cows milked with the robotic milking system had a significantly higher SCC than cows
milked in a parlor. However; the SCC of both groups were quite low.

• No difference in number of “new infections” as indicated by CMT score was found
between the robot and parlor groups at either the quarter or cow level.

• The udder health measures of CMT score, EC and SCC were correlated but the
correlation coefficient was low.


This research was funded by a USDA Hatch / McIntire-Stennis grant. We are grateful to Ann Paulman, of the Emmons Blaine Dairy Cattle Research Center, for supervision of automatic collection and transport of milk samples for SCC enumeration.


Batra, T.R. and A.J. McAllister. 1984. A comparison of mastitis detection methods in dairy
cattle. Canadian Journal of Animal Science 64:305-312.

Chamings, R.J., G. Murray and J.M. Booth. 1984. Use of a conductivity meter for the
detection of subclinical mastitis. Veterinary Record 114:243-245.

Hamann, J. and A. Zecconi. 1998. Evaluation of the electrical conductivity of milk as a
mastitis indicator. Bulletin of the International Dairy Federation No. 334.

Kremer, J.H. and D. Ordolff. 1992. Experiences with continuous robot milking with regard to
milk yield, milk composition and behavior of cows. Proceedings of the International
Symposium on Prospects for Automatic Milking. Eds. A.H. Ipema, A.C. Lippus, J.H.M.

Metz and W. Rossing. Wageningen, Netherlands, November 23-25. EAAP
Publication No. 65. Pp. 253-260.

Ordolff, D. and R. Artmann. 2000. Surface temperatures of udder and teats in conventional
and automatic milking systems. Robotic Milking: Proceedings of the International
Symposium. Eds. H. Hogeveen and A. Meijering. Lelystad, The Netherlands, 17-19
August. pp. 301-302.

SAS Institute Incorporated. 2001. SAS Software, release 8.2. SAS Institute Incorporated,
Cary, NC, USA.

Shoshani, E. and M. Chaffer. 2002. Robotic milking: a report of a field trial in Israel.
Proceedings of The First North American Conference on Robotic Milking, March 20-
22, Toronto, ON, CA. Pp. III-56-III-63.

Stewart, S., P. Billon and G.A. Mein. 1993. Predicted maximum milk flowrates in milking
systems. Proceedings of 32nd Annual Meeting of the National Mastitis Council,
Kansas City, MO.

Stokes, M., C. Davis and G. Koch. 2000. Categorical data analysis using the SAS system.
SAS Publishing, chapter 13.4.

Svennersten-Sjaunja, K., I. Berglund and G. Pettersson. 2000. The milking process in an
automatic milking system, evaluation of milk yield, teat condition and udder health.
Robotic Milking: Proceedings of the International Symposium. Eds. H. Hogeveen and
A. Meijering. Lelystad, The Netherlands, 17-19 August. Pp. 277-288.

Wesen, D.P., L.O. Luedecke and T.L. Forster. 1967. Relationship between California Mastitis
Test reaction and bacteriological analysis of stripping samples. Journal of Dairy
Science 51(5): 679-684.

Wirtz, N., K. Oechtering, E. Tholen and W. Trappmann. 2002. Comparison of an automatic
milking system to a conventional milking parlor. Proceedings of The First North
American Conference on Robotic Milking, March 20-22, Toronto, ON, CA. pp. III-50-