Considering the state-of-the-art tools to support the manufacturing processes planning and the current state of the tool manufacturing process, we will develop an innovative software solution for improved estimation of durations for tool manufacturing operations in the company. Instead of statistics, the key step of the approach will be to use AI, in particular data mining techniques, for building predictive models capable of estimating the operation durations. These models will then assist and potentially replace the current subjective estimations in the planning phase. <br /> For devising the solution, two sources of information will be deployed: (1) the technological database available in the company that contains data on the characteristics of tool manufacturing operations performed in the past and their actual durations, and (2) expert knowledge on the importance of individual data items and their impact on operation durations. The former will be preprocessed by checking for missing data and outliers as well as normalizing the actual operation durations of the same kind on machines of different performance. Moreover, a specific step will be to appropriately exploit/combine the data available in various formats, for example geometry description of tool parts in a specific formal notation and their visual representation. The latter will be extracted through interactions of data mining specialists with process planners in the company and used to identify meaningful relationships between the operation specificities and the operation durations.

Due to the small size of the project the department was involved in all project steps. From collaborations, meetings and discussions with all involved partners, researching new features to be used for machine learning part, and in smaller part also in implementation. Big effort was also given to researching and experimenting on different machine learning approaches to be used for developing models for estimating the operation durations.

KET4C