Practical A.I. and Machine Learning for Food Industrial Potato Baking

The challenge

Food processing in general and industrial potato baking in particular are quite challenging. The most important challenges are the variability of the quality of the raw materials – like the country of origin and the sugar content - and of the processing parameters, like humidity, ambient temperature, time in tank and temperature in the oven.

Irrespective of these inputs, we need to produce beautiful golden-brown fries with a minimum of waste. So the business question is: “for any feedstock given batch, what should be my process parameters to achieve golden-brown color and minimal waste after oven?”

The solution

Our partner AS4 Point sent us an batch data set from a food production process, using 5 feedstock parameters, 12 process parameters and 2 production targets: waste and color after oven.

We used our multi-target optimization feature to analyze the data and build a predictive model for both subjects at the same time. We concluded that the color after oven was driven by just 4 inputs and waste after oven by 3 inputs, and that no more than 5 inputs in total impacted the targets.

Then we analyzed in real time how changes in these inputs would vary the outputs using our optimizer tool. We selected the input constraints (e.g. sugar value and days to harvest) and identified the process parameters that would produce optimal targets. This allowed us to conclude what parameters gave the ideal color and minimal waste and thus see what configuration should be deployed in real production and what kind of raw materials should be bought to achieve these results.

The impact we created

Our predictive model allows:

• Transparency: insight on five drivers impacting two KPIs.
• Predictability: real-time prediction of targets for any given feedstock and operating regime.
• Optimization: for any given feedstock quality, we can give recommendations for optimal processing conditions to achieve the desired targets
• Insights for sourcing department: on what type of raw materials should they target to buy?