Optimal Chicken Project
This project involved, in a period from November 2003 to December 2004, the deployment of PIPA system modules in several chicken slaughterhouses at Perdigão S.A.Problem Description
The Poultry Production Chain at Perdigão is verticalized, including breeder, chick and broiler production and importing broiler chicken grandparents. Besides this, it is based on partnerships with third party chicken producers, which supply the necessary farming to raise the chicken flocks. According to this integrated scheme, Perdigão supplies the farmers with chicks and the necessary raw material for its growth, such as feed and technical assistance, financially compensating them according to chicken flock performance.
Agricultural Field Integration is the company’s department responsible for planning and control of all activities that lead up to the chicken supply at the slaughterhouses. The slaughter schedule, for example, must be consistent with the production needs, both in quantity and quality (in this case the weight of the chicken). This schedule must also seek to reduce feed consumption which is proportional, among other variables such as sex and lineage, to the chickens age. On the other hand, chick housing must guarantee the medium term animal supply. Lastly, breeder (for chick production) housing and disposal and, eventually, third party chick purchases, must be compatible with the needs established by the housing plans.
Synchronizing all phases of the Agriculture Field Integration chain is a complex task. In first place because it involves the concatenation of several temporal cycles with duration varying from 60 weeks for breeders life cycle with daily egg laying, 21 days of egg incubation and between 30 to 45 days of chicken growing. Besides this, productivity in each phase is subject to seasonalities and uncertainties due to biological variability’s. Statistical analysis shows that, for example, (a) average chicken flock weight for the same slaughter age varies according to the season, (b) the normal curve represents well the weight distribution of animal within the same flock and (c) the weight gain of each flock is strongly dependant on the farming quality as well as physical and sanitary housing conditions. Finally, there is the need to adapt plans to the frequent market oscillations, which ultimately change the chicken slaughter profile.
Once slaughtered, the chickens are gutted and cooled in chillers. Soon after, it can be directly packaged (which normally occurs with low weight exportation carcasses, called grillers) or it can be dismembered in to smaller parts in the butchery. Many products produced by cuts, and even the grillers, are specified according to weight ranges. Products are also specified according to:
» “chicken type”: such as, for example, vegetable chicken, fed only with non animal ingredients (ex: bone meal); another typical example is the Chester, consumed around Christmas;
» quality: small hematomas prevent the use of products for export, even though they don’t represent any kind of impact on human consumption.
Besides guaranteeing the “natural” products market, slaughterhouses must meet industrialized raw material demand. This is done, generally, through the use of cutting process sub-products, such as mechanically separated meat (MSM).
The variability among several characteristics of chickens belonging to the same flock (mainly weight) allied to diverse product specification lead to complex, but paradoxically more flexible, production planning tasks in the slaughterhouses. On the one hand, not all the carcasses or parts of a same flock can be used to produce the same product. On the other, flocks of different average weight can lead to similar weight individual carcasses, making it easier to meet the required volume by “weight ranges” demanded by final products.
In November 2003, Perdigão began the Optimal Chicken Project (Frango Ótimo in Portuguese), seeking to deploy computational tools to assist optimized planning for activities regarding field integration and slaughterhouse production. Due to its size and complexity, the Capinzal – SC unit, where 340 thousand chicken are slaughtered daily, was selected as the project pilot plant.
UniSoma investigated planning practices used at Capinzal that were based on the use of electronic spreadsheets and business knowledge non formalized in computer terms, in other words, restricted to the planners. For the slaughter schedule, for example, the following problems were identified:
» no decision support tool was available for the planners, which nearly exclusively based their plans on chicken stock reports found on the corporate database;
» chicken slaughter weight was predicted by rural technicians, without the use of statistical methodologies;
» generating slaughter scenarios took larges amount of time;
» many meetings were necessary to define the schedules;
» lack of visibility towards the whole set of housed chickens;
» there was no medium term tactical scheduling, which frequently lead to situations in which one week would over slaughter while the following week would be interrupted due to lack of chickens.
The consequences of these situations were diverse, such as:
» large errors in weight prediction for flocks ready to slaughter;
» high average weight variability of slaughtered flocks and feed conversion;
» difficulties in meeting weight profile demand by the slaughterhouses.
In terms of production planning at the slaughterhouses, the situation wasn’t different. No agile and robust support tools were available for the planners who suffered with the raw material variability, with direct impact on meeting demands.
The pilot project, in Capinzal, involved the deployment of several of the PIPA – Integrated Poultry Production Planning – modules, namely: ABATE/F, PLAMES/F, PLADIA/F, APANHA/F, PENDURA/F and Computational Viewer.
Through the use of ABATE/F a daily slaughter schedule is generated with a typical horizon of 30 days. Developed in AIMMS software, this module has the following macro-functions:
» a computational intelligence, in the form of a Mixed Integer Programming Model, for the automatic and optimized generation of slaughter schedules with minimal cost, subject to demand constraints (production plans) and others;
» a graphical user interface, through which the user interacts with the mathematical model, changing input data and visualizing the results, such as the schedule;
» a planning case/scenario manager;
» an interface with the corporate base, used to read information regarding housed flocks.
Amongst the input data considered by ABATE/F, weight prediction, feed consumption and poultry mortality models should be highlighted. These models are a result of statistical analysis (periodically updated) conducted using units slaughter history and considering several control variables such as, for example, age at slaughter, lineage, sex, density, season and the farmers grade performance. The weight models, in particular, make use of partial samples taken from flocks at ages 21, 28 and 35 days, given the level of precision demanded by the planning processes, and due to the effects caused by extra-seasonal climate changes and variable nutritional feed.
ABATE/F allows for the agile rescheduling of slaughter scenarios, caused by non predicted events such as, for example, sanitary status change of flocks, broken transportation ways, breeder flock slaughter and other operational constraints (ex: the need to fix a specific slaughter date for some flocks).
After concluding the pilot project, the team rolled-out the ABATE project to other units:Videira (SC), Marau and Serafina Corrêa (RS), Carambeí (PR), Rio Verde (GO) and Nova Mutum (MT).
Once the slaughter schedules were defined, chicken collecting scheduling was defined, through APANHA/F. This scheduling determines the initial loading times for each flock, so that the slaughter lines work without interruption throughout the day (except for planned maintenance or breaks) guaranteeing a safety stock.
On the unloading platform, the initial slaughter schedule for each load is registered in the PENDURA/F module. Before the chiller, the Computational Viewer Module estimates, through the use of bi-dimensional carcass images, the average flock weight and weight distribution throughout the slaughtered flock. The information on PENDURA/F and Computational Viewer Module are used by production supervisors for very short production scheduling.
Through the PLAMES/F and PLADIA/F modules, the monthly and daily production plans are determined, respectively. Both were developed in AIMMS, such as ABATE/F, and make use of BIBPRO, Slaughterhouse Products and Processes Library. The plan generated by PLAMES/F seeks to maximize the units net marginal contribution, according to minimal (firm sales) and maximum (potential market) sales limits for final products and according to internal slaughterhouse capacities (ex: labor, freezing tunnel, etc.). PLAMES/F establishes the ideal demand profile that will be followed by ABATE/F. On the other hand, PLADIA/F makes use of a daily chicken supply offer and a short-term (7 day) known demand, seeking to establish a compatible daily schedule with the product demand dates.
The main qualitative results from the Optimal Chicken Project are the following:
- Formalization of the planning processes;
- More agile plan generation;
- More field scenario visibility (ABATE/F);
- Higher level of confidence on automatically generated plans.
In terms of quantitative results, highlighted are:
- better weight prediction;
- Chicken feed conversion reduction, despite average weight variations;
- Reduction of average slaughtered chicken weight variability and average feed conversion;
- Delivery dates improvement;
- Maximization of the company’s net marginal contribution;
- Decrease of final products stock.
With a gross revenue of US$ 3,4 billion in 2007, Perdigão is one of the largest Brazilian companies in the food business. Its supply chain consists of production and marketing of natural poultry, swine and bovine meat products, as well as raw materials of animal origin (industrialized). Besides this, the company has a strong presence marketing processed products, pasta, frozen vegetables and soy derived products. In 2007, the Unilever margarine lines were acquired by the company.
In 2007, UniSoma developed and deployed for Perdigão a Turkey Slaughter Scheduling Module (ABATE/P), at the Carambeí (PR) unit.
In the turkey production case, the growth cycles are larger than that of the chicken. The males are slaughtered according to an average age of 140 days (always heavy). On the other hand, heavy and light females are slaughtered at 110 and 67 days, on average, respectively. The heavy weighted males and females are slaughter throughout the whole year, meeting exportation demands for natural meat as well as demand from the industrialized plants. Light weight females start to be produced in the middle of the year to guarantee the end of the year season festivities.
Similar to ABATE/F and also developed in AIMMS, ABATE/P contains a computational intelligence, in the form of a Mixed Integer Planning Model, which allows for automatic and optimized generation of slaughter schedule plans with minimal cost, subject to demand constraints (production plans) and others. The main difference between ABATE/P and ABATE/F is the following:
- Turkey production is done in 2 phases: final growth is conducted by producers denominated “terminators”, after the initial production done by “initiators”. Due to the typical schedule horizon of 60 days, ABATE/P considers all housed flocks, be it at “initiators” or “terminators”;
- ABATE/P allows partial flock slaughter. Although this is not desired, this is eventually necessary, since the number of daily slaughtered flocks is much smaller than that of chicken;
- Average weight changes demand a specific setup on the gutting lines. Due to this, ABATE/P seeks to avoid slaughtering, in the same day, flocks that will decrease productivity.
|AIMMS Service Partner UniSoma|