composición y mandatos de la c omisión y la c orte

Grunow et al. [81] considered parts of the Venezuelan supply network, where most farms (ha- ciendas) are owned by the same business entity that owns the mill. According to Grunow et al., there is no paper in literature that deals with decisions across levels from strategic through tactical to operational. The main objective of their proposed mixed integer linear programming (MILP) model was to secure a minimum supply of cane to the mill. Their step-by-step, hierar- chical modelling and solution procedure starts at the strategic level by assigning a cultivation date to each hacienda, after which the solution for that model is implemented. When it is time to start harvesting, a tactical level model is solved in order to determine optimal haciendas to start harvesting for every day for the following two weeks. This harvesting schedule is used at a third modelling level to generate crew and equipment dispatch schedules. One assumption in their model is that once the harvest of a particular hacienda has begun, it only stops when completed. This constitutes an important difference between the South American and South African harvesting circumstances, since in South Africa farms are not by rule harvested in a single effort, but rather on a field-by-field basis with the harvesting effort tempered to adhere to a daily rateable delivery1 _{(DRD). In addition to securing supply to the mill during normal}

conditions, Grunow et al. claim to address the occurrences of unforeseen events such as criminal fires by accommodating data changes at the tactical level (harvest day).

A modelling approach by Higgins et al. [101] and Muchow et al. [158] incorporated six years of block productivity data in an attempt to optimise the harvest schedule of an entire mill region in Australia. The model parameters were supported by a number of one-way analyses of variance (ANOVAs) which were used to find significant differences in sugar yield due to geographical, cane variety, crop class and other circumstances. The authors of [101] concluded that there is opportunity to increase the profitability of the investigated mill region as a whole.

Higgins [97] adopted a MILP formulation towards modelling the transportation problem of sug- arcane in Australia. He emphasised the importance of using a participatory research approach which he attributes to Martin et al. [148]. The model was used to some extent by transport planners in the Maryborough mill region.

A review by Weintraub and Murray [227] shows how the forest harvest scheduling area addresses many of the issues found in sugarcane harvest scheduling. The mathematical formulations pre- sented by Weintraub and Murray include retaining a “mosaic” of forest patches analogous to retaining a healthy age or variety difference between adjacent sugarcane fields. The best solution approach towards solving real problems, according to Weintraub and Murray, are metaheuris- tics such as genetic algorithms, tabu search and simulated annealing. A model by Karlsson et al. [117], similar to the models reviewed by Weintraub and Murray, was designed to sched- ule harvesting crews while obeying a number of demand, capacity and transportation (road)

1

The term daily rateable delivery is taken throughout this dissertation to retain the same meaning as the term rateable delivery defined in the sugar industry agreement (SIA) [49]. The SIA term means that the total cane delivered by any grower shall be delivered to the mill in a rateable manner throughout the entire season.

2.2. OR approaches to modelling sugarcane flow 13 constraints. Karlsson et al. compared a number of solution approaches and also preferred a metaheuristic approach. R¨onnqvist [177] gives an overview of optimisation problems found in the forest industry.

Guilleman et al. [83] investigated several mill-scale harvesting schemes using CIRAD’s2 _simula-

tion tool named MAGI3 _{[134] to determine whether harvesting sugarcane without the require-}

ment of adhering to the DRD could increase revenue for all parties at the Sezela mill in South Africa during the 2000 season. Results indicated gains in RV % valued between approximately 4–15 million Rand, depending on scenario. The study was followed up by an investigation by Le Gal et al. [135] addressing essentially the same question, assuming that RV-prediction functions change significantly from year to year. Their conclusion was that the Sezela mill region could gain up to 7.44 million Rand, given an average sugarcane crush rate of 2.23 million tonnes per year. The authors in [135, p. 66] suggest several general views on modelling sugarcane supply, and claim that

“These optimisation models are not directly applicable in the South African case. Firstly, they need a detailed data set including the field level. Secondly, they as- sume either that a central controller implements the calculated solution by deciding precisely which fields have to be harvested within the season and by allocating both harvest and transport capacities between farms, or that all the stake-holders agree with the optimal solution and are ready and able to implement it at their own level. Thirdly, these models are based on complex algorithms which make them difficult to explain in a decision support perspective. Fourthly, they give little room for discussion or compromise as they provide ‘one best way’ solutions.”

In 2004, Higgins and Postma [102, p. 237] pointed out the importance of allowing each grower fair harvesting time windows throughout the season, which “. . . prevents any grower from being unfairly exposed to wet weather risks towards the start or the end of the harvest season, as well as other seasonal effects.” The results of Higgins and Postma’s work on the optimisation of siding rosters4 _{which entailed a solution to the staff roster problem, was implemented by}

the participating mill and cost savings were estimated to approximately 150 000 Australian dollars/year. Higgins and Postma [102, p. 248] claim—in obvious contrast to the above quoted Le Gal et al. [135, p. 66] claim, that

“Given the complexities of producing a staff roster acceptable for use in the Aus- tralian sugar industry, we would argue that traditional scientific research and de- velopment principles would not have achieved implementation. The siding rostering model of this paper is adaptable [to] all other sugar mill[s] in Australia and any sugar producing country that uses mechanical harvesters, such as Brazil, South Africa and the United States.”

In a 2002 paper, Higgins [96] placed focus on the participative nature required to gain success in developing and implementing operations research-based optimisation models or tools for the Australian sugar industry. This time Higgins developed an optimisation model for mechanical harvester scheduling, which eventually led to the highly successful work on the optimisation of siding rosters mentioned above.

2_{Centre de Cooperation Internationale en Recherche Agronomique puor le D´eveloppement.} 3

The meaning or origin of the acronym is not explained in [134].

4_{Siding rosters are the schedules used by Mackay Sugar to distribute time-slots for cane deliveries to railway}

Iannoni and Morabito [106] formulated a discrete event simulation model of the mill yard of a large sugarcane mill in Brazil, assuming that the arrival rate at the mill is a result of the number of long-range transport vehicles in the system combined with the transportation vehicles’ cycle time. The cycle time was defined as the sum of waiting and unloading time in the mill yard and the time to travel to the field, load and return to the mill yard. The authors ran three scenarios which showed promise in that it was, according to the model, possible to decrease the amount of sugarcane in queue and to increase the utilisation of the mill. The scenario that showed promise consisted of changing dispatching and queueing rules for vehicles of a certain type, effectively increasing the amount of unloaded sugarcane. The model represents the sugarcane supply system, which is, according to Iannoni and Morabito, a closed system since long-range transport vehicles loop. A problem with this approach, should it be applied to the South African case, according to the author of this dissertation, is that a closed system assumes that the amount of sugarcane available for pick-up per time unit is governed by the cycle time of the long-range transport vehicles. A more accurate way to model the available cane for the South African case is to consider the mill crush rate, the DRDs of all growers and the cutting capacity of all growers. However, Iannoni and Morabito provide validation of their model towards answering the questions they put forth.

Higgins and Davies [100] built a stochastic simulation model for the purpose of long-term trans- port system capacity planning and considered several scenarios developed together with rep- resentatives from growers, harvesters and the miller in the Mourilyan mill region in Australia. One of the scenarios was deemed promising and was later implemented, the results of which were evaluated by the group of representatives to be of benefit to the miller, while growers and harvesters were not negative. In the same paper, Higgins and Davies developed a second simulation model to optimise the starting time for each harvester in the promising simulation scenario. The new harvester schedule indicated a reduction of the 1.5 hour average waiting time at the railway sidings for harvesters by 70 %. The new harvesting starting times schedule was adopted by the group of representatives in 2004, not without harvesters being unhappy with the optimised schedule, raising some objections.

Le Gal et al. [133] presented a method for conceptually integrating three levels of modelling of the sugarcane supply chain, namely the operational, tactical and strategic levels. The framework that they proposed is an idea of how to combine the three software packages CIRAD’s MAGI [134], Rockwell Automation’s ARENA [7] and Weintraub and Epstein’s decision support system for forestry transportation scheduling ASICAM [226] and constitutes, according to the authors, a step towards an operational support method. Grunow et al. [81] as well as Le Gal et al. point out that little or no research is aimed at addressing decision support on all levels across the supply chain in a single approach. The author of this dissertation believes that there are no such actual decisions to be supported, mainly since the ownership structure of the sugarcane supply chain is too diverse, at least in South Africa. Before models can really be applied to all three levels of the same supply chain, the various owners must somehow embark on joint ventures under which such decisions are actually likely to occur. The highly successful project during which the decision support system ASICAM was developed was, incidentally, a project where such decision existed [226].

In Thailand, a recent effort by Piewthongngam et al. [170] points out that communication be- tween growers and millers in the northeast of Thailand is poor. Here, the amount of cane delivered each day is not determined by contract as in South Africa, which leads to problems such as under-utilisation of the mill and sugar losses. Piewthongngam et al. developed a frame- work for cultivation planning on the field level, in order to provide possible alleviation of the problem, using crop growth simulation models for a yield estimation module and mathematical

2.3. Supply chain improvement methodologies and philosophies 15