Use forecasting models to predict product quality (protein powder) prior to production and allow operators to adjust the filtering loop accordingly. Result: Reduced product quality variation and improved yield.

Identify and rectify harmful user behaviors such as overriding safety measures. For instance, operators adjusting a temperature sensor post-CIP to fill storage tanks with cold milk sooner. Result: Prevents material fatigue and, in the worst-case scenario, the implosion of large silo tanks.

Identify and rank the most important contributors to variation in cheese by using historical data. In this case dry matter was the key quality metric but it could be any thing. Result: 3/4 of the hypothesis could be laid to rest and focus on solving the most important contributor for quality variation. The results were better and more convincing than a previous R&D project spanning over 5 years had accomplished.

Forecast potential malfunctions in large engines such as spray dryers using predictive models. Result: Early warning on bearing failure, which would causedowntime.

Apply anomaly detection models to alert operators and maintenance engineers about impending issues such as filter breakage, clogging and degradation in the milk treatment. Result: Prevents one major failure every 18 months, which is more than €1 million per annually.

Predict and adjust pH in the wastewater before it leaves the plant using forecast models. This model uses CIP data from a large number of lines and tanks. Result: Compliance with governmental regulations, contributing to environmental sustainability.

Reduce the volume of wastewater through detailed analysis and mapping of variations. For instance, discovering that UHT filtration had a higher contribution than expected. Result: Reduced water, energy, and chemical consumption (CIP).

Uncover where and how waste occurs, for example, by measuring dry matter in wastewater and identifying peaks. Result: Identified and rectified 3 main root causes in a day, resulting in a 60% waste reduction.

Waste reduction. A hypothesis was that there was too much product being sent to the waste buffer tank. This is milk, whey, and other products which are in the pipes or at the bottom of tanks after the batch is “completed”. Deep analysis on who are the main contributors revealed two main contributors with unexpected high volume. Result: Changed recipe in the automation layer gave approx. 30% reduction.

Identify energy wastage in idle operations to assess potential automation and routine modifications. For example, evaluating pasteurization running only on water circulation. Result: Identifies 800 kWh yearly over in a dairy wasted on heating water.

Forecast the temperature of hot water/steam returning to the boiler to adjust boiler settings based on projected rather than measured temperature. Result: Energy reduction and peak power reduction, as it will now only produce the energy needed and when it’s needed.

Leverage forecast models to optimize how heat exchangers are operated. Result: Reduced energy consumption (and CIP performed based on the condition of the heat exchangers).

Leverage data analytics tools to identify which production parameters affect capacity. Result: Overall throughput increased by over 20% while maintaining quality.