A decision support system for scheduling the harvesting and wine making processes at a winery

Van der Merwe, Adri (2009-12)

Thesis (MSc (Logistics))--University of Stellenbosch, 2009.

Thesis

ENGLISH ABSTRACT: Technological advances made over the past century have had a major impact on traditional wineries. Software solutions for management issues are widely available and give rise to the prospect of computerized decision support systems assisting in various aspects of managing a winery. The most popular applications seem to be concerned with supply chain management or harvest scheduling. Such projects are under way all over the globe and great success has been achieved to this e ect. However, prior to this study no such project has been considered in South Africa. The phrase active cellar scheduling problem refers to the assignment of grape batches to processors inside the cellar where bottlenecks often occur during the busy harvesting period. The phrase harvest scheduling problem, on the other hand, refers to selecting the best possible dates to harvest the respective vineyard blocks in order to preserve grape quality. A mixed integer programming model for the active cellar scheduling problem is derived in this thesis, but proves to be too time consuming to solve exactly via the branch-and-bound method. A meta-heuristic tabu search approach is therefore designed to solve the problem approximately instead. When applied to a small, ctitious cellar, it is found that the tabu search method often solves the problem optimally. The computer processing time associated with the tabu search approach also constitutes a signi cant (often thousand-fold) improvement over that of the branch-and-bound approach for realistically sized problem instances. A generic tabu search is also designed to solve the over-arching harvest scheduling problem for a general winery. This schedule is found by referring to the smaller tabu search of the active cellar scheduling in order to verify the impact that harvesting moves have on activities in the cellar. One harvesting schedule is considered a better schedule than another when it has a lower harvest evaluation score, determined by the placement of the vineyard blocks in the harvesting schedule. The harvest evaluation score takes into account the combination of vineyard blocks selected for harvesting on the same day (and their e ect on the active cellar) as well as the ripeness and quality of the grapes. Both tabu searches are nally included in a exible, computerized decision support system, called VinDSS. This system is found to produce good harvesting schedules when compared to an actual ve day schedule during the 2009 harvesting period at Wamakersvallei, a winery serving as case study for this thesis.

AFRIKAANSE OPSOMMING: Tegnologiese vooruitgang oor die afgelope eeu het 'n groot invloed op tradisionele wynkelders gehad. Sagteware-oplossings wat besluitsteun tot bestuursaangeleenthede bied, is algemeen beskikbaar en het gelei tot die rekenaarmatige implementering van besluitsteunstelsels vir wynkelders. Dit blyk dat die mees popul^ere besluitsteuntoepassings in die wynindustrie te make het met besluite rakende van voorsieningskettings en oes-skedulering. Sulke besluitsteunprojekte is w^ereldwyd onderweg en het alreeds groot sukses behaal. Daar is egter tot dusver geen so 'n projek in Suid-Afrika onderneem nie. Die frase aktiewe kelderskeduleringsprobleem verwys na die toekenning van druifvragte aan masjiene binne die kelder waar bottelnekke algemeen tydens die besige parstydperk voorkom. Die frase oes-skeduleringsprobleem, daarenteen, verwys na die seleksie van bes moontlike oesdatums vir elk van die wingerdblokke om sodoende druifkwaliteit te verseker. 'n Gemengde heeltallige programmeringsmodel is vir die aktiewe kelderskeduleringsprobleem ontwikkel, maar die rekenaaroplossingstyd van hierdie benadering blyk te lank te wees om die probleem eksak deur middel van 'n vertak-en-begrens metode op te los. 'n Meta-heuristiese tabu soektog is dus ontwikkel om die probleem benaderd op te los. Wanneer hierdie benadering op 'n klein, ktiewe kelder toegepas word, word optimale oplossings dikwels verkry. Verder toon die rekenaaroplossingstyd van die tabu soektog 'n groot (in sommige gevalle byna 'n duisendvoudige) verbetering op di e van die eksakte oplossingsmetode. 'n Generiese tabu soektog is ook ontwikkel om die oorkoepelende oes-skeduleringsprobleem vir 'n algemene wynkelder op te los. So 'n oes-skedule word gevind deur na die kleiner tabu soektog vir die aktiewe kelderskedulering te verwys om sodoende die e ekte van veranderinge in die oesskedule op die prosesse binne die aktiewe kelder na te speur. Een oes-skedule word beter as 'n ander skedule beskou wanneer dit met 'n beter oes-evalueringswaarde gepaard gaan, soos deur die plasing van die wingerdblokke in die skedule bepaal. Die oes-evalueringswaarde neem die moontlike kombinasies van wingerblokke wat op dieselfde dag geoes word, in ag (en ook die e ek wat dit op aktiwiteite in die kelder het), asook die rypheid en kwaliteit van die druiwe. Beide tabu soektogte word in 'n plooibare, rekenaar-ge mplementeerde besluitsteunstelsel, bekend as VinDSS, ingesluit. Daar word gevind dat hierdie stelsel goeie oes-skedules lewer wanneer dit vergelyk word met 'n werklike vyf-dag skedule tydens die 2009 parsseisoen van Wamakersvallei, die kelder wat as gevallestudie vir hierdie tesis gedien het.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/3046
This item appears in the following collections: