Influence of the Robotic Milking System on Milk Production and Quality: Systematic Review

This review aims to report the direct influence of a robotic milking system (RMS) on milk production and quality. The Scopus, SciELO, and Web of Science platforms were used as search databases. We followed the PRISMA protocol for the identification and screening of articles. Initially, 336 articles were identified. We excluded 186 articles for duplicity, 53 after screening abstracts and titles, 20 for lack of access, and 58 articles based on the exclusion criteria. Nineteen articles from 2002 to 2021 from 10 different journals were selected. We observed an increase in publications related to RMS in recent years, and the Journal of Dairy Science gained prominence among the journals whose articles were used in the present study. After lexicographic analysis of abstracts, it was clear that there were five predominant classes, and the keyword RMS was more associated with factors related to cows than milk. The study results contribute to a greater understanding of RMS research, providing farmers and readers with clarification on the actual influences on the dairy chain system and future research projects.


Introduction
The robotic milking system (RMS) has represented one of the most remarkable advances in milk production techniques since the 90s, with rapid adherence by all livestock farmers in this area worldwide. By 2020, approximately 50,000 operating units were estimated to exist on the planet (Filho et al., 2020), located mainly in Europe and Canada (Cogato et al., 2021).
These systems have been popularized as they potentially provide more quality for workers, reducing labor and time effort compared to conventional milking systems. These systems also promise to optimize milk production rates, quality, and mammary gland health (Hovinen & Pyörä lä , 2011;Rodenburg, 2017;Hogenboom et al., 2019). Thus, some improvement in milk production by direct or indirect routes is unquestionable, given the growth of adherence to technology in recent decades. The significant difference in the system is the voluntary access of cows to the milking unit during lactation, as voluntary access can generate variation in the intervals between milking. This variation is higher in animals subjected to robotic milking than in animals milked using a conventional system, which is why they can provide higher milk production (Masí a et al., 2020). Nevertheless, a decrease in milking frequency from twice a day to once a day results in an immediate increase in the somatic cell count (SCC) (Stelwagen & Lacy-Hulbert, 1996).
In robotic milking, mechanical arms perform preliminary operations such as brushing and udder sanitation. Based on the identification of the animal, the robot adapts to the morphological characteristics of the cow (height, udder size, teat shape, and angle). However, there are criteria for excluding cows from the herd to achieve an efficient acceptance of the robot. Cows considered unsuitable for the system (Córdova et al., 2018b) revealed a possible failure in the system.
Considering the disagreement among several articles for the analyses of dairy production and robotic milking, and numerous conclusions suggesting premises for new research (Wagner-Storch & Palmer, 2003;Jacobs & Siegford, 2012;Córdova et al., 2020), the authors noted the need for a systematic review on the robotic milking system, as well as the evaluation of qualitative aspects. Therefore, our intent was to represent the direction of published research in this area and outline a qualitative level of the studies which have been addressed.

Methodology
This article presents a systematic review conducted according to the recommendations of the PRISMA protocol. We selected the studies in the SciELO, Scopus, and Web of Science databases using the keywords "Robotic milking," "Somatic cell count," "SCC," "Total cell count," "TBC," "Milk production," and "Milk quality." We used these keywords in both Portuguese and English. The keyword "Robotic milking" was integrated into the search with other keywords using the Boolean Operator 'AND,' as described in Table 1. We opted for complete articles in journals and reviews as the initial filtering method.
Two researchers independently evaluated the studies and discussed any doubts concerning the article selection until an agreement was reached. In cases of disagreement, a third evaluator was selected to decide on the article's inclusion in this review.
This systematic review aims to report on the influence of robotic milking on milk production and quality. So, we sought to answer the guiding question formulated by the Population Variable Outcome (PVO) strategy. 1 : What is the influence of robotic milking (variable) on milk production and quality (outcome) in dairy cattle (population)?
The inclusion criteria were as follows: 1) quantitative studies presented in the abstract, title, or keywords, 2) the characters "robotic milking" and the respective translation in the Portuguese language, 3) no year restrictions, and 4) publications in English, Portuguese, and Spanish. The exclusion criteria were as follows: 1) studies outside the objective, 2) qualitative studies on milk quality and production, 3) articles regarding other milking systems, and 4) review articles, letters to editor/editorials, personal opinions, chapters of books, textbooks, reports, and conference summaries.  ISSN 2166-0379 2022 After the including and excluding of articles, the remaining studies were subjected to risk and bias analyses. The list of criteria applied to each article is described in Board 1 based on the ideas established by Koutsos et al., 2019. We assumed three possible answers to the questions for each article, admitting the scale: Yes (Y) = 20 points, Inconclusive (I) = 10 points, and Not (N) = 0 points (Kitchenham et al. 2009). We calculated the final score by percentages of articles that resulted in 60% or more being included in the review. The bias assessment was described by Board 2. We observed that 19 of the 25 previously defined studies reached the desirable criterion of 60 %, confirming the credibility of the selected studies.

Journal of Agricultural Studies
The main information about the variables investigated in the selected studies was extracted for writing the results and discussion section of our review article. Therefore, the articles were analyzed specifically for the categories "milk quality" and "milk production." The text data were also processed and submitted for lexicographic analysis using the IRAMUTEQ 0.7 alpha2 software, aiming for qualitative analysis. Texts originally written in Portuguese or Spanish were translated into English for linguistic equalization. The keyword "Robotic Milking" was separated by "_", becoming "Robotic_Milking", to be read by the program as a single expression, avoiding parsing errors. Therefore, we used a descending hierarchical classification and similarity analysis

Results
Based on the combinations mentioned above, searching through keywords resulted in the initial identification of 336 studies. First, 186 articles were identified. Then, 150 articles were selected to read the titles and abstracts; 53 were excluded, resulting in 97 readable articles. We did not have access to the full text of the 20 articles. In total, 77 articles were read. After applying the exclusion criteria, 19 studies that answered the guiding question of our investigation were selected ( Figure 1). The selected studies were published in 10 journals over 12 years. The main characteristics of the studies and a summary of the results are presented in Table 2. It is important to note that the main conclusions and results described here always correlate with the variables studied in our review, thus increasing the reliability and highlighting the aim of the study Articles that considered milk production and quality essential characteristics in their study were presented between 2002 and 2021 (Graph 1). However, we observed that the theme of our investigation was still little explored, especially from 2002 to 2012, when the RMS was not yet popular and underwent some adaptations (Jacobs & Siegford, 2012).

Graph 1. Distribution of publications by years of selected articles
The journals of the selected articles were compared to the number of publications related to the review variables. The Journal of Dairy Science (Graph 2) shows the numerical superiority of articles, representing 47% of the articles chosen for data extraction. However, this data was not surprising because of the journal's reputation and its purpose of dealing directly with dairy chain issues Graph 2. Distribution of journals by publications of selected article The Descending Hierarchical Classification (DHC) method identified five classes of segments in the vocabularies (Figure 2). The corpus was divided into two sub-corpuses and partitioned to obtain Class 2. In the third stage, there were more partitions, which resulted in other classes. We observed a relationship between metabolic diseases and milk production and animal feeding in the system and the importance of milk management and quality in Classes 3 and 4. ISSN 2166-0379 2022

Figure 2. Descending Hierarchical Classification
A direct relationship between the five classes stood out in the similarity analysis. Classes with the terms "milk" and "cow" are entirely related. Still, some subclasses correlate with the class "cow." For example, the words "AMS" (entirely related to "robotic milking") and "farm" farm." On the other hand, we observed the words "health" and "udder" establishing a link with the class "cow," indicating that health is more related to cow and not milk in the selected articles. In the "milk" class, there is a coalition with the milk quality words "somatic," "count," "total" and "cells," indicating a result of a cohesive search in the articles.

Discussion
Based on the results of the 19 selected articles, we observed an interaction between the passage of years and an increase in studies on robotic milking. This interaction has been expected since the first commercial RMS was installed on a dairy farm in 1992 (Svennersten-Sjaunja & Pettersson, 2008). Despite being an innovative technology, the uncertainties of farmers and the low supply of representatives meant that, in the 2000s, the system was accepted in the USA and most European countries (De Koning, 2010).
At that time, research projects observed an increase in somatic cell count and a decrease in quality resulting from the new milking system (Klungel et al., 2000;Vorst, Y. Van der Hogeveen, 2000). This event triggered a warning in farmers that it was a new technology and a different method of conducting the business. Furthermore, this method depends on external factors, such as facility conditions, animal management, and staff rosters . A new wave of research on RMS began (Jacobs & Siegford, 2012) from these first complaints and the new challenges in the first decades of the Journal of Agricultural Studies ISSN 2166-0379 2022 2000s, corroborating with the data.
As shown in Graph 2, the Journal of Dairy Science stood out in the number of publications selected by the authors. These data converge because the journal is in the top 10 of the Scopus database in "Animal Science and Zoology" and is based on the CiteScore index, reaching a value of 6.2 (Scopus, 2021).
Considering the h-index by Scimago Journal & Country Rank, the journal becomes even more relevant in "Animal Science and Zoology" and stands out in the first place of the ranking (Table 3). Using the SRJ index as a reference unit, the Journal of Dairy Science remains in the top 10 rankings (Scimago Journal & Country Rank, 2021). There was a significant difference compared with the other selected journals. These were not even in the top 20 rankings.
This difference between rankings occurs in the way the indices are calculated. The calculation of CiteScore considers citations from many files such as articles, book chapters, reviews, and data articles. The journal performs the count over four years, dividing by the number of the duplicate files indexed in Scopus and published during the same period (James et al., 2019). The h-index is the reference that indicates the minimum citations referring to the total publications in a certain period, quantifying productivity and scientific impact (Bornmann & Daniel, 2007). Within the DHC (Figure 2), the sub-corpus refers to classes 1, 2, and 5 instead of classes 3 and 4, representing 64.35% of the total textual corpus, whereas classes 3 and 4 represented 35.65% of the results. The first set of data signaled the importance of the studied variables. This aspect can be observed through an analysis of the selected terms.
Through the analysis of Class 2, the relationship between metabolic diseases and animal rumination became clear. Milk production and rumination time are responsive variables to a cow's health status. However, cows with subclinical ketosis achieved low rumination rates in an RMS, although they produced the highest rumination rates . This may be related to the high production rates of the herd, indicating the need for energy supplementation in animals fed partial mixed rations (PMR). Regarding this class, the variations in beta-hydroxybutyrate (BHB) in the system did not vary significantly in the evaluated studies.
Classes 1 and 5 complement each other. Animal feeding has attracted the attention of researchers. The system's users know the use of the concentrate to acquire higher visitation rates for the robot. However, some studies have reported that the gains in production by using this technique are limited despite maintaining a constant dry matter intake (Lessire et al., 2017;Schwanke et al., 2019). Some researchers even suggest using pellets with neutral detergent fibers and high degradability rates in the robot to increase visits without affecting production and milk composition (Halachmi et al., 2009). However, these results may be controversial with respect to milk production. We observed more occurrences of animals with mastitis in the robotic system than in conventional milking, leading to decreased production rates (Stergiadis et al., 2012) Another essential factor in Classes 1 and 5 is the passage rate. It is essential to adapt a herd to voluntary milking, and this indicator is usually related to animal productivity (Borshch et al., 2020). Although voluntary milking can benefit animals, a higher visit rate was observed between 8 a.m. and 11 a.m. and between 3 p.m. and 6 p.m. This can be explained using the forced milking technique (Wagner-Storch & Palmer, 2003). Forcing animals to go through milking and increasing their frequency may decrease milk flow and increase milking time, milk production, and milk composition. Locomotion problems, lactation stage, and the production rate of each cow are also factors that influence visit rates (King et al., 2017;Córdova et al., 2018a).
In the second sub-corpus, Classes 3 and 4, the selected words were correlated with milk quality. Many studies have correlated the increase in somatic cell counts with the entry of animals into RMSs (Kruip et al., 2002;Avilez Ruiz et al., 2021;Van den Borne et al., 2021). However, some researchers have suggested that this high counting rate decreases after six months of herd adaptation (De Koning et al., 2003). The increase in SCC may be caused by failures in the fixation of teats, which can reach up to 7% of all milkings (Bach & Busto, 2005). Although there is an increase in SCC, there is also an increase in the milk fat of animals in RMS (Janštová et al., 2011;Avilez Ruiz et al., 2021;Matson et al., 2021).
An increase in SCC may also result from inadequate teat sanitation (Van den Borne et al., 2021). Cows with shallow and small udders are not recommended for robotic milking. Therefore, farms installing the system should select animals with greater udder depth to achieve better mammary gland health in this system (Córdova et al., 2018b;Tse et al., 2018).
Nevertheless, RMS is highly effective in terms of milk hygiene. Samples from cooled tanks were collected from facilities with an automated milking system, and the total bacterial count was below the minimum evaluation value (Janštová et al., 2011).
The similarity analysis unifies the terms "cow" and "milk," organizing a perception of the system. The term "AMS" (automatic milking system) is not directly associated with the word "milk." This indicates that milk production and quality are not related to the milking system but the animal. Thus, the system directly influences animal health and feeding (Figure 3). Milk production and somatic cell count are directly associated with "milk" and "cow." Thus, we can observe a link between animal welfare and better sanitary conditions in the final product.

Conclusion
This study demonstrated the influence of a robotic system on milk production and quality.
Analyzing the articles lexicographically, we can observe the similarity of their terms and glimpse at a new vision of the system and its variables for future decision making. SCC has significant similarities with "milk" and "cow." Therefore, farmers should act directly on the herd's health to modify the rates of somatic cells in milk, rather than modifying the milking system. As RMS is directly related to feeding, before deciding to change the property system, farmers should consider the nutritional supply of the herd