3D Industrial Inspection
Description
For this strategic line, the work is focused on applying artificial intelligence in industrial environments, especially for quality control through 3D inspection. It aims to improve industrial processes such as in-line inspection, early detection of defects in the production environment, increasing product quality, and reducing material waste. Machine learning, pattern recognition, and metrology techniques are applied to achieve these objectives, ensuring quality and accuracy standards beyond the reach of traditional industry.
Industrial inspection for quality control is a task that is carried out during the manufacturing process. This approach facilitates the resolution of problems that cause defects in production just after being detected during the manufacturing process itself. It minimizes costs by avoiding the generation of large quantities of unusable products. It is helpful in any factory to improve productivity, decrease defect rates, and reduce additional work and wasted raw material.
Mass production has established processes for creating parts with identical dimensions and design, but these processes do not produce precisely identical parts, and this causes problems, resulting in user rejection. Quality control (QC) separates the action of testing and validating products to discover defects from the decision as to whether these products should be approved or rejected as part of the final delivery, which could be determined by pre-established constraints (Shewart et al., 1939). For example, for the performance of work contracted by government agencies, quality control issues are among the first reasons that may lead to a cancellation or non-renewal of a contract (Usopm, 2012).
Previous experience and research in the field of machine learning conducted by PRAIA have been vital in the development of the foundations of industrial 3D inspection technology for different "core" processes of the system. For instance, the data structures used for the storage of textures corresponding to each 3D point of a model (based on kd-trees) and the classification algorithms based on the nearest neighbor (supervised geometric classifier), as well as in the automatic learning of a reference model by compositing several acquisitions on the exact correct fundamental part, or even in the iterative alignment procedure based on point-to-point distance minimization techniques. Therefore, it is emphasized that Machine Learning is a discipline framed in the field of Artificial Intelligence, but with influences and contributions from other areas such as Theoretical Computer Science, Decision Theory, and general Statistics (Duda and Hart, 1973), (Gonzalez and Woods, 1992). It is necessary to apply techniques that generally require learning from real-world data to address tasks that require simulating behaviors that in a person would be interpreted as intelligent. This information is usually not structured as in other areas of Computer Science but is presented in the form of examples. These examples contain physical measurements, variables obtained from different data sources, user interactions, etc. The final objective is to construct an Inference model.