My name is Kay Peeters. Within the Dynamics & Control research group of the Department of Mechanical Engineering, my PhD concerns the production planning and control of poultry processing plants. These typically produce fixed-weight batches of products such as filets. Any weight over the minimum is effectively a ‘giveaway’ by the plant. Meanwhile, retailers require their daily batch orders to be finished before their due dates. Operational costs, food waste and other performance measures can therefore be improved by optimizing poultry processing plants. In turn, this will reduce production costs and improve the reliability of deliveries, making it possible for supermarkets to offer reduced prices.
Batches are produced using a variety of batchers. After weighing, these decide whether to allocate the product to an available bin or let it pass. Allocating items may yield giveaway but letting too many through may cause the (desired) deadline to be missed. I conducted interviews with Marel Poultry employees and their customers and found that several aspects of their production environment make it hard to manage their operational process. Primarily, flocks of broilers are delivered to the plant throughout the day. Different weights occur with different frequencies, and a batcher must allocate a product without knowing the weight of future products. There is a strong relationship between weight distribution, batch weight and resulting giveaway. Additionally, the decision of when to produce different orders directly influences giveaway.
An important first step was the development of mathematical models for both the batchers and the scheduling problem. Such models allow you to formally describe a setting and make sure that everyone is talking about the same problem. I also created simulation models that mimic the behavior of different batchers in order to test different algorithms. Simulation models are a powerful tool as they allow you to understand the inner workings of different solution approaches and quantify their performance in different production settings. This is vital to convincing people of the value of your approach.
Poultry processing plants had previously received little attention in literature despite having a complex production environment. We quickly saw that existing models insufficiently capture their intricacies, so my supervisors and I had to develop new ones. Through simulation models, I quantified the benefits of novel algorithms for different batchers over existing approaches. For a type of batcher that receives products one-by-one, I developed a heuristic that reduces giveaway and improves due date adherence compared to current practice in all simulation experiments. Similarly, I studied a type of batcher that also processes products one-by-one but has information about the arriving stream of products.
Utilizing this extra information has proven particularly challenging. Ultimately, an algorithm has been developed which does this extremely well, reducing giveaway and significantly improving throughput control over current state-of-the-art methods. Finally, I’ve shown that giveaway can and should be explicitly included in a plant’s production scheduling. Doing so leads to significant giveaway and tardiness reductions in comparison to less advanced methods.
The research on batchers is directly applicable to poultry processing plants and the results indicate significant improvements in terms of throughput control and giveaway. This directly translates into reduced operational costs. Although the research on the production scheduling needs to be developed further, it provides an important starting point. In time, planners at poultry processing plants will simply be able to plug in information about arriving flocks and customer orders and receive an optimized production schedule. Ultimately, my research may also be applicable to other meat processing industries or industries with similar characteristics.