Case Histories
Predictive quality, food & beverage
Tasks
Data Analysis
Predictive quality
Anomaly detection
Industry
Food and beverage
OEM original equipment manufacturer
Technologies
Machine Learning
Regression techniques
Request
Predict the humidity of a product (pasta for food use) at the end of production, with only the machine’s input parameters available.
Starting point
Even with machines equipped with sensors, the client had to wait until the end of the entire production cycle (3-4 hours) before being able to obtain reliable measurements for the product, with the risk of wasting time, energy and raw materials.
Pasta cannot be sold with a moisture content over 12%, but as of writing there are no reliable methods or sensors available that can monitor moisture during the production process.
Not only that, but not all the correlations between production data were known, and therefore some were saved without having any effective impact on the production process.
Solution implemented
A system that identifies the relevant variables between several hundreds of them and use them within a regression model to predict the product’s final moisture level while it is being made, without waiting for the work cycle to be completed.
Predictions can be used to directly change the settings on the machines, optimizing the quality of the finished product.
Results
The system is being tested at the client’s facilities.
Through the software installed, it will be possible to consistently reduce the quantity of discarded products, which also means a reduction in wasted production time (about 3 hours saved for each batch potentially to be discarded) and the costs linked to the use of power and ingredients.
Current developments
The client has decided to add sensors to other parts of the production line and to apply additional anomaly detection models to the data collected, which will be managed through invariant.ai, in order to monitor the data in real time and predict possible deviations in judgement.