Canadian Poultry Magazine

Artificial intelligence and the future of feed

By Lilian Schaer   

Features Emerging Trends New Technology

Machine learning enhances feed precision, sustainability, and efficiency in Ontario poultry production.

Researchers from the University of Guelph partnered with Trouw Nutrition on a three-year project looking at the use of machine learning models to predict feed pellet quality. Photo: Mark Malpass

Next to genetics and management, nutrition is one of the biggest factors influencing successful poultry production. In order to be productive, birds need high quality, consistent and nutritionally balanced feed – and to support producer profitability, that feed also needs to be cost-effective. 

It’s estimated that about 70 per cent of poultry production costs are linked to feed, so getting it right matters. A feed mill is a complex environment, however, with many different factors influencing the quality and quantity of the feed that’s produced. 

That’s why the industry is actively looking for ways to increase the precision and uniformity of feed being produced, reduce errors and harness the skills and experience of its older workforce approaching retirement for the next generation of employees. 

Advertisement

Exploring machine learning
Dr. Mark Malpass, director of poultry nutrition at Masterfeeds, and Dr. Jennifer Ellis, Associate Professor in Animal Systems Modelling at the University of Guelph, believe machine learning, which is a form of artificial intelligence, could provide the solutions the feed industry is looking for. 

“A feed mill is not a perfect laboratory; it’s a real world environment where there will be variation in pellet quality from time to time between formulations, ingredients and even different operators,” says Malpass. “So can we use artificial intelligence to investigate that variability and help teach and manage feed mills in the future?”

It’s an idea that began to form when Malpass and Ellis, former work colleagues at Trouw Nutrition Canada, got into a conversation at a conference about a modelling trial she had just completed. The trial wasn’t successful, but the idea of using machine learning models to predict feed quality stayed with her as she joined the faculty at the University of Guelph in 2019. 

“There were a lot of discussions about what the future of modeling represents in agriculture and how we can we best position ourselves to ride the digital wave – not just collecting data, but using it to bring value,” Ellis says. 

“My background is in modelling and Mark has an ear to the ground in terms of what will work in the industry, so this partnership came together.” 

Modelling the prediction of pellet quality
With funding from the Ontario Agri-Food Innovation Alliance and a partnership with Trouw, Ellis, University of Guelph computational biology professor Dan Tulpan, and PhD student Jihao You embarked on a three-year project on the use of machine learning models to predict feed pellet quality. 

The team collected manual and automated data from the Trouw mill in St. Marys on manufacturing conditions, nutrition formulation and the environment inside and outside the mill, and then developed a series of machine learning algorithms to determine the best predictive model they could build. 

The project is finished, resulting in a proof of concept model with reasonable prediction accuracy. Although work had previously been done on the link between pellet quality and one or two variables, no real research had to date ever been done on integrating all the factors in a commercial setting that can impact pellet quality.

“What happens in a mill and how you make feed is different in January than it is in July in the middle of a heat wave, so how does this impact pellet quality?” Malpass says. “We know it’s challenging when it’s hot, but there are currently no clear guidelines around the thresholds of when we change what we do, and this will let us be more precise about it and know exactly the most impactful thing to do to take action.” 

Optimizing feed for sustainability
While the immediate focus of the work is being able to predict pellet quality, the end goal is to be able to make those predictions accurately and then use that information to optimize feed production for efficiency or for sustainability parameters, such as a certain level of greenhouse gas emissions per tonne of feed, for example. 

Long-term, this information could be tied into on-farm precision feeding systems, waste reduction and optimizing instead of maximizing production. To date, the work has only focused on pelleted feed for monogastric livestock, but the team hopes to expand into both other types of feed and ruminant animals in the future. 

Expanding data collection
New funding has just been secured to expand the work to more Trouw locations, as well as facilities owned by Masterfeeds and Molesworth Farm Supply. As each location will have slightly different processes and methods, collecting a broader spectrum of data will help make the models more robust and accurate. 

“Machine learning is only as good as the data that it learns off of, and their strength is when they’re built on large volumes of data,” Ellis says. “Partnerships like this with industry are so important and by expanding the work and having multiple feed companies collaborate on the overall goal, we can support the Ontario agriculture industry in feed manufacturing.”  

“As far as we understand, we’re the first lab group anywhere in the world to have published research on this in land animal feed,” adds Malpass. “That means Ontario is leading the way in the world on this topic. We also appreciate the funding from the Alliance; they shared our vision and took a gamble on us.” 


Print this page

Advertisement

Stories continue below