Neural Network to Retrieve Features of a Mechanical Component Acquired by a Robotic Inspection System

The concept of Industry 4.0 is closely associated with the development of new innovative computing methods such as Artificial Intelligence (AI), Machine Learning (ML) and Cloud Computing, and new visualization and measurement procedures that have enabled continuous control and monitoring in industry. This paper presents a “robotic inspection system” for geomet-ric inspection of the dimensions of a mechanical components. These systems, consisting of a robot and a line-triangulation laser sensor, support online quality control of the manufactured product with minimal assistance. The proposed work seeks to overcome and combine a traditional inspection system with neural network methods. In fact, both the acquired profiles and the geometric value expected from the profile measurement, are used to train a linear regression neural network model that in output provided the value of the measurement itself. In order to verify the performance, a numerical simulation model is implemented to provide different profile. This method enables the quality control of the component to be independent from the skill and competence of the human operator.