Study: Machine learning used in feed mills can optimize pellet quality

10-Jan-2023 - Last updated on 10-Jan-2023 at 15:19 GMT

Factors related to the pellet manufacturing process, the feed formulation, and regional environmental data can be successfully used to build machine learning regression models for predicting pellet quality in a commercial setting, finds a study.

Such a model could be used within the feed mill to predict the pellet quality index for a given batch of feed, which in turn is a useful Key Performance Indicator (KPI) for efficient mill management, said the authors, based at the University of Guelph and Trouw Nutrition, Canada.

They say their paper, publsihed in the November 2022 edition of Animal Feed Science and Technology, outlines the first use of machine learning regression models to predict pellet durability index, based on a multifactorial dataset.

Manufacturing pelleted feeds involves a series of processes, including receiving feedstuffs, grinding, proportioning, mixing, conditioning, pellet conversion, cooling, and packaging. During these processes, several parameters play a crucial role in determining the pellet quality, including die specifications, ingredient particle size, conditioning time, and temperature.

“Pelleted feeds are widely used in monogastric animal production systems because they not only improve animal performance by increasing digestibility and feed consumption but are convenient to store and handle. However, pellet quality can be affected by many factors."

Machine learning

Machine learning (ML), as a sub-field of artificial intelligence (AI), is optimally positioned as a prediction tool that can consider large numbers of driving variables and complex inter-variable interactions, said the researchers.

Such ML models are able to learn from data, predict and generalize without explicitly being programmed to do so (Samuel, 1959), they noted.

In this current study, ML models were used to predict the pellet durability index (PDI) based on feed formulation, manufacturing process, and environment-specific factors associated with PDI in a commercial feed mill.

A dataset consisting of 2,471 observations describing the pellet manufacturing process, the feed formulation, and environmental conditions - outdoor temperature - were collected from two feed mill lines for eight months.

Sixteen features were used for building the regression models, and the output was the pellet durability index (PDI) of the pelleted feeds, explained the team.

Twelve regression algorithms were examined as part of the project, while analytical tools were used to identify what features were most relevant for each model.

Findings

For most algorithms, average outdoor temperature, bakery byproduct inclusion, wheat inclusion, and production line were deemed important and had an overall higher importance than all the other features, according to the authors.

“Interestingly, and perhaps in contrast with industry expectations, the fat added into the mixer was found to be less important than most features. However, controls in place at the mill, which place an upper limit on the fat that can be added into the mixer, might partially explain this outcome."

From a performance perspective related to pellet durability index predictions, one model, the Support Vector Regression, outperformed all the others, they said.

The researchers stressed though that while some factors related to feed formulation were included in the study, others such as nutrient composition of diets were also not available.

“These missing factors may provide more information compared with feed formulation details, as the nutrient composition of feedstuff can be variable and feed formulation can change with seasons. Adding such factors as features in future models might help to further improve the PDI prediction performance.”

While acknowledging that controlled research studies may have a limited ability to consider the numerous interactions present in a commercial feed mill setting, the authors said they believe their work demonstrates the potential practical utility of ML methods to address a common feed manufacturing challenge: pellet quality prediction and improvement.

“The ultimate goal of applying ML in feed mills is to ‘optimize’ pellet quality while considering trade-off factors such as the cost of feed ingredient, the mill usage of energy, mill efficiency, greenhouse gas (GHG) emissions and downstream animal performance. In the future, ML methods combined with optimization algorithms can help feed mills to achieve a sustainable and cost-efficient production of pelleted feed.”