Embedded Files
PURPOSE
Develop predictive models based on artificial intelligence and implement nondestructive techniques to assess the quality of ceramic production..
Typology: Industrial Research
Final Result: FR2 | Study, definition and implementation of models for quality prediction and classification of ceramic materials based on Artificial Intelligence methods and non-destructive diagnostic techniques.
Responsible Partner: University of Calabria
Duration: 36 months
Operative Units: Arcavacata di Rende (CS)
DESCRIPTION
WP 2 focuses on developing artificial intelligence (AI)-based predictive models and creating nondestructive measurement systems optimized for the ceramic industry.
The goal is to assess the quality of ceramic production using production, process and analysis data.
Artificial Neural Networks (ANN) will be used to develop models that identify nonlinear relationships between input parameters and output variables, despite the presence of noise in the data. Laboratory experiments will be conducted to produce controlled ceramic samples and measure quality indicators. Non-destructive measurement techniques such as thermography, ultrasound, acoustic emission and capacitive imaging will also be developed and implemented to evaluate quality indicators and detect defects.
The ultimate goal is to correlate the composition of ceramic materials, process parameters and defects in products.
The innovation lies in combining predictive models, nondestructive measurements and data analysis to improve the efficiency and quality of the ceramic process, making companies more competitive.
ACTIVITIES
In order to achieve the purpose of the WP, the activities described below have been planned.
2.1 Models for product quality prediction based on input and process data from domestic raw materialsDefinition of an architecture of product predictive models and feature extractions useful for classification from NDT analyses.2.2 Models for product quality prediction based on input and process data from imported raw materials and raw material mixesExtension and testing of the predictive model.2.3 Process quality control and defect detection methods using NDT and ML techniquesDevelopment of NDT systems for ceramic products and machine learning algorithms for defect detection.. 2.4 Modeling of the combined ceramic product-laying surface systemData acquisition on various types of combined systems, correlation analysis using ML and multivariate statistics between mechanical tests and NDT.2.5 Sensitivity analysis of footprint assessment modelsTesting of various sensitivity analysis algorithms and a posteriori verification.
RESULTS
Below are the results achieved to date:
2.1 Models for product quality prediction based on input and process data from domestic raw materials
The activities of WP 2.1 led to the development of predictive models for the quality of ceramic tiles.
Following an initial selection and chemical-physical-mineralogical characterization of the national commercial raw materials, mixtures were prepared and the first samples were produced.
The characterization and process data, combined with those deriving from the Gresmalt historical series of the last 6 years, were normalized and used as input to train machine learning (ML), deep learning (DL) and multivariate statistical models.
Furthermore, developing algorithms based on non-destructive techniques (NDT) allowed us to start the search for possible correlations between extracted parameters and the data traditionally acquired.
These processes led to the definition of predictive tools with accuracy greater than 70%, which will improve with further training.
The activities demonstrated how integrating traditional analyses with new AI-based predictive methodologies can significantly improve quality control in the production of ceramic tiles.Google Sites
Report abuse