Cognitive geometrical control of machined forged parts based on big data from the manufacturing process

The main factor of competitiveness of the machining of forged parts relies on quality, productivity and costs` management. Extensive control of machined parts represents a bottleneck which lowers the manufacturing cell productivity and causes costs. Additionally, the productivity is lowered, and costs increased by tool breakup and consequent machine tool maintenance cost. The proposed project uses the benefits of the digitalisation process to tackle these problems; machine condition and process are monitored and stored in the cloud in the form of big data. The idea is to avoid 100 % part control, and to prevent tool failures by big data processing. The participants of the project are the cofounder and project partner company Marovt d.o.o., who is specialised in the forging and machining of parts for the automotive industry, and Inkolteh, who is the developer of a production control system named Ccleap. It collects process data from the forging and machining process for each part in the cloud. These data will be the source for cognitive prediction of the appropriateness (good/scrap) of parts based on process data. The cognitive prediction model will be made by the third partner, a group of researchers from the University of Maribor. The partners in the project are combining the problem, capabilities of data acquisition and knowledge of intelligent manufacturing for data processing. Project objectives are 90 % less automated control of machined parts and savings of up to 50 % on machine tool maintenance cost caused by tool failure. To achieve these objectives the following research objectives shall be addressed: (1) Development of a holistic procedure for the creation of a representative database with extracted features, manageable on the smallest possible representative dataset, (2) State-of-the-art ML and DL models` formulation, tailored to address the project objectives, (3) Reaching a breaking-point where the in-silico results are sufficiently close enough to the in-vivo results. Project outcomes will be a computer system for capturing and pre-processing of process data, Algorithms for Feature Extraction, Feature Fusion, Feature Selection and Machine Learning, software upgrade of Ccleap and a manufacturing cell with a system for quality prediction. These outcomes will enable long-term goals: The development of a system for processing of the big data from the technological process, prediction of quality for different manufacturing systems, prediction of tool wear and tool life for tool management, and prediction of failures for predictive maintenance. The project will be carried out in 3 phases, which are detailed further in Work Packages and activities: The first phase is Acquisition and pre-processing of process data, the second Data processing with model development and third, the Implementation in the production control software. Responsibilities are assigned to researchers and Work Package leaders are defined. Work packages are also defined for Project Management and dissemination.