What is unsupervised learning? Unsupervised Learning (see wikipedia) is used when no historical data is available that involves past faulty behaviour and/or ageing progression. Therefore,
Quality prediction and prescriptive maintenance for plastic press
Plastic injection press requires continuous interventions as the output quality declines overtime. These interventions can range from cleaning actions to change of mold and revision of the press.
Maintain the output quality above the required level by anticipating the type of maintenance to apply during the production process.
The press is equipped with a Variable Speed Drive delivering multiple data types while the press is at work (speed, torque, current, etc…). Moreover, a measure of quality is available through systematic inspection following each batch of production. These data are used by DiagFit to provide a Software Sensor of the output quality based on the VSD. Then DiagFit analyzes the deviation of features used in deriving the quality’s virtual sensor to predict the type of maintenance to apply to bring the quality back to its optimum level. Resulting prescriptive maintenance recommendations assume that history of maintenance operations is recorded and transmitted for data driven modeling.
Quality prediction using the speed profile of the press has an accuracy of 3% on the segment between 80 and 100%. 8 different maintenance operations were separated in 4 clusters and the deviation profile of the quality toward one or another cluster indicated the type of maintenance to apply. Benefits are: reject rate lowered by 50%+, press downtime reduced by 40%.