The recent push by data modeling companies is showing investment may be smaller than originally expected, results may be more immediate and operators may unlock higher throughputs and improved quality. This article will examine current wins and how leaders’ vision for data modeling in food and dairy is evolving.

Data Modeling Goes Mainstream

In early June, Siemens and HighByte partnered to accelerate data modeling for manufacturing, including food, beverage and dairy. HighByte’s Intelligence Hub will gather process or factory data from the Siemens Industrial Edge platform and conduct data modeling and contextualization.

The partnership provides the ability to connect PLC and SCADA data to HighByte’s Intelligence Hub via Siemens’ Industrial Edge. This turnkey solution, in partnership with Siemens and HighByte, accelerates food producers' ability to build AI models, agents and applications at scale. FOOD ENGINEERING spoke with HighByte’s Aron Semle for the June issue about collecting and standardizing data, which, on the surface, seems to be a costly and rigorous challenge.

“The reality in today’s manufacturing world is that data doesn't have any context,” Semle says. “Changing the context of that data isn't that expensive and companies need to start and iterate. The faster manufacturers contextualize data, expose it to AI and experiment with people on the factory floor, the better.”

A theme emerging from industry leaders is that operators need to be involved with these initiatives from the beginning. “Operators are running the systems,” Cayuga Milk COO Hugh Roddy said in a recent webinar on dairy modernization. “Companies need them involved with new technology and early.”

Data modeling and batch production require operators to be successful. “If you put a standardization layer on data, use AI to infer and take value from the contextualization, companies can start to drive interesting insights,” Semle says. “This helps operators on the factory floor, so they can start to have conversations with the data.”

The modeling works by taking factory batch information, such as manuals, work orders and specifications, and uploading it to the platforms.

And, most importantly, operators oversee AI outputs and guide the intelligence to reasonable outcomes. “You still need humans-in-the-loop, otherwise there are no guardrails,” Semle says.

“Focus just as much on the people as you will on the technology, and then put a roadmap together,” Roddy noted. “Make sure you're putting a roadmap together, not just a bunch of projects.”

Forget AI, Pets May Take Over

Lynn Dornblaser, client advisor at Mintel Solutions, discussed the unprecedented growth of the pet food industry in 2024 during a FOOD ENGINEERING interview. Dornblaser cited new product introductions outranked the regular food segment from 2019 to 2023.

The segment has many tailwinds, including the humanization of pets and consumers' move to feed their pets natural, clean-label ingredients. Nestlé’s recent earnings call revealed that the segment is now the second largest category in the business, accounting for just over 20% of sales.

Ingredients for clean-label pet food are costly, and optimizing batch processing for food majors is essential. Hill’s Pet Nutrition recently adopted a Software as a Service (SaaS) machine learning solution from TwinThreads and began running models on its processing data. The pet food producer pulled in data from its AVEVA Manufacturing Execution System (MES), Wonderware historian and many different control points on the processing line, including coating, drying and extrusion segments.

Hill Pet focused on optimizing quality and having the right moisture, fat and protein levels throughout processing. “We've been able to predict the quality of Hill’s Pet Nutrition at the end of the line, but with the timeliness of being able to make changes at the beginning of production,” TwinThread President Andrew Waycott said in a recent IT/OT Insider podcast.

The data modeling solution integrates Startup and In-Work processes to continuously optimize across four quality measures. In addition, the pet food producer scaled this solution across approximately 18 plants, Waycott said.

Scaling machine learning for predictive maintenance across plants is well documented, but moving modeling upstream is now possible. “Once you have examples of a batch, a changeover, a startup, you can apply the balance of your historical data and train models,” says Virag Vora, technical sales engineer, TwinThread.

Vora adds: “Everybody is searching for the holy grail, such as deploying a Unified NameSpace (UNS) initiative and context being captured at the edge. This is a great aspirational vision, but a complete overhaul of a plant is cost prohibitive.”

Midsize Food Operations

While data modeling discussions are usually reserved for food majors, mid-major companies are adopting digital tools, with an eye on establishing a foundation for metrics and OEE and increasing throughput.

“The results of a new dashboard may be scary, but it provides a roadmap to where you really need to spend money and focus,” Roddy says.

Recently, Orlando Baking adopted Redzone’s real-time OEE productivity module to provide operators with metrics. The midsize commercial baker has over 100 stock-keeping units (SKUs) and distributes to 32 states. However, the commercial baker's continuous growth put pressure on operators and managers to obtain accurate production line data without new tools.

“There is a lot of downtime associated with so many products, due to allergens, cleaning and so forth,” says Rocco Orlando, continuous improvement manager at Orlando Baking. “The OEE solution helped us start tracking downtime, streamlining and improving our process.”

Orlando Baking did not provide its OEE measure, but operators now track downtime events and track progress through plant-floor tablets. The productivity platforms calculate OEE based on availability, performance and quality, while displaying minute-by-minute counts, hourly OEE and downtime Pareto charts.

“Continuous improvement and collaboration tools not grounded in real OEE data have limited impacts on productivity,” says Katie Bellott, director of product marketing at QAD Redzone. “Problem-solving, communication, and Lean habits are beneficial, but without tangible data around the impacts, it’s very hard to prioritize problems or realize the value of the problems solved.”