Preventive Maintenance at the world’s largest beer manufacturer
Machine Learning For Preventive Maintenance
Customer: World’s Largest Beer Company.
Partner: Pluto7 Consulting Inc.
Customers ‘Pre-Google’ Solution Challenges:
The brewery had issues predicting knockdown events before they occur. Like filter replacements during beer brewing, and classifying the actual knockdown events. The customer also wanted to know the specific factors that have a direct impact on the outlet turbidity.
Another problem was the prediction of events that are critical to improving production, reducing labor and filter costs, as well as drive better beer taste.
Actions(Preventive Maintenance Using Our Model):
Pluto7 provided a ‘proof of concept’ that demonstrated a ML classification model using TensorFlow. This model can categorize and identify the real knockdown events. Therefore reduce the number of times the filter has to be replaced. The TensorFlow regression model can explain the variability in the outlet turbidity of the beverage and the attributes that make up the turbidity values.
Pluto7 proved that ML can help improve productivity, save on costs, and improve the taste of the beer.
GCP Components Used:
TensorFlow ML model
Customer Acknowledgement of the Work Done:
Pluto7 finished in the top 3 out of 20 vendors who were invited for the events. Pluto 7 presented the journey on February 6th 2018 in US to the CFO of the brewery company.
Benefits Customer Has Seen so Far:
The preventive maintenance solution ML model identifies the real knockdown events with 92 percent accuracy and the classification model has an accuracy of 90.2%. The benefits are tremendous in terms of the cost savings, quality of beer manufactured, and reduced waste. Up to 40 percent variability of the outlet turbidity explained by the regression model has aided the operator in identifying attributes that cause rises in turbidity. This helps operators on the correct steps to identify attributes that cause turbidity increases and ensure smooth, clear beer. Pluto7 has recommended optimal ranges within which the filter should operate to increase its overall life time.
Click here for our use cases!