Winery capacity planning using discrete-event simulation

‍

When talking about the largest and most complex logistical challenges, people often default to thinking about industries like container shipping, airlines, or electricity distributions. Although few would initially put the wine industry on this list, wineries are a treasure trove of exciting operational challenges. OSI has delivered projects along all facets of the winemaking process for some of the biggest wineries in the US.

‍

Our project developed an application that quantifies asset utilization under a variety of possible scenarios at a major US winery. Faulty demand forecasting had left our client with a costly gap between demand and their wineries’ capacity. The client wanted to have a better understanding of how changes to their winery assets, such as tanks, filters, and bottling lines, affect their utilization and overall capacity of their operations to better match their demand.

‍

Winery assets

‍

Wineries are a massively complex operation. The winery creates many hundreds of “base wines”, which are then combined as ingredients to the final wines you see on store shelves. Each base wine is distinguished not just by grape varietal and vintage but mainly by the process it has undergone: Different types of fermentations, different durations spent in barrels of different wood types, different flavorings that have been added, etc. Each of these hundreds of base wines has their own unique route through the winery, creating a complex network of batches using common resources.

‍

Our model captures this entire operation in detail, from the raw grapes arriving by truck to the end of the bottling line. This allows the model to return a detailed report on the utilization of key assets such as.

• Tanks
• Barrels
• Filters
• Presses
• Bottling lines
• Labor

‍

For example, the graph below shows the tank utilization for a group of tanks that is used primarily for fermentation. As expected, the utilization peaks during harvest season in the fall when arriving grapes start their fermentation upon arriving at the winery. The rest of the year, the tanks are used to a lesser extent for general storage.

The application is wrapped inside an user-interface that allows users to run the model with the click of a button and view the utilization of each of their assets using dashboards like the one below..

‍

Model inputs and scenarios

‍

In order to achieve such a detailed model, we worked closely with our client to develop a database structure of tables that serve as the input for our model. These tables give our client fine control over the supply, demand, asset makeup, and routing of the batches

‍

Changes to the input can represent all sorts of hypothetical scenarios for which the client would like to know the effects on capacity. These scenarios could include:

• Changes in demand.
• Losing or acquiring certain grower contracts.
• Unexpectedly low grape yield due to climate conditions, an important concern in the face of climate change.
• Closing one of the winery locations and shifting its grape supply to the remaining locations.
• Acquiring new tanks, bottling lines, filters, or any other equipment.
• Labor shortages.
• The addition of new base wines required for new products.

‍

Discrete-event simulation

‍

There exist many different mathematical frameworks for modeling logistical operations, each with their own unique strengths and drawbacks. The question of which framework to choose often depends on the characteristics of the particular project at hand and is a decision we deliberate carefully.

‍

For the winery capacity model, the underlying framework is called discrete-event simulation (DES). This framework has a number of important advantages compared to mathematical optimization and integer programming, another popular technology OSI commonly leverages in their projects. Below we summarize the advantages of DES.

1. Expressiveness: DES is incredibly flexible. Integer programming is subject to much tighter restrictions on what it can model. It is quite common to run into details that cannot be captured in the appropriate equations needed for integer programming to solve effectively.
2. Speed: When incorporating an immense amount of intricate detail, DES is usually much faster in both development time and in the run time of the final application. Especially in cases where the solution space is so small that optimization would be overkill. If the question is should I close this winery location, then the solution space is simply yes or no. For cases like these, it is easier to just try each option using DES rather than using mathematical optimization.
3. Cost: Even a medium-sized integer programming model will quickly become difficult to solve without an expensive commercial solver. Similarly, there are commercial tools available for DES, which OSI has leveraged in the past, but these expensive DES tools only become necessary when models become truly massive. For most cases, as in this one, open-source software easily suffices all requirements, saving money on costly software. The aforementioned quicker development time also heavily cuts down on labor costs.
4. Confidence: DES can readily capture the stochastic nature of a system. Discrete-event simulations don’t just give you an answer, they also tell you how confident you can be in that answer.

‍

OSI prides itself in having deep expertise in a wide array of tools and techniques. We make sure that our clients can feel confident that their business problems are solved using the best and latest available methods.

‍