PREDICT: a company specialized in robust and industrial predictive technologies
PREDICT is a company specialized in robust and industrial predictive technologies to improve the production efficiency, to tend to zero defect, to increase the machines performances and to decrease the maintenance costs.
PREDICT develops its own predictive digital platform, KASEM, which supports prediction of any kinds of drift of performances on the CNC machines, of product defects, of tools wear, of energy
overconsumptions, of dysfunctions and of breakdowns. KASEM integrates:
- Industrialized, robust and accurate prediction estimating in real-time
- diagnostic assistance to help users to find and eradicate the root cause of the abnormal behaviors,
- aggregated CNC machine health indices for the future days, weeks, and months allowing to
avoiding under performance, downtime and proposing tasks to do to be under control.
Role and challenge
PREDICT is involved in both InterQ concept:
- InterQ-Data (WP4) Challenge is to develop a historical Data validation service able to detect drifts on measurement before data are used to take decision or make prediction. Main contributions are:
- Formalization of machine tool concepts for historical data validation and the definition of set business rules for automatically generate data analysis pipeline.
- Development of a semi supervised approach for pattern recognition based on Dynamic Time Warping approaches, i.e., that only one labelled pattern is necessary to find most of the pattern occurrences.
- InterQ-Product (WP3) Challenge is to evaluate the risk of quality problem on part using process data. In addition, PREDICT offers its expertise in machine tools through its software tools CASIP for data collection and KASEM for data analysis and sharing. Main contribution is the development of a methodology based on cutting condition impact analysis assuming that these impacts should always the same as machined part is the always the same. For estimate cutting conditions impact it is first necessary to recognize process in times series where tools cut the raw part. After, impacts can be computed and modelized to finally use anomalies detection techniques.
Check out the video on historical data validation