PROPHET: PReference-based OPtimization for Human-cEnTric visual inspection
Part of the project SHOP4CF EU Horizon 2020
Visual inspection is a key operation in the production process to ensure the quality of the product.
The expert human operator is typically involved in difficult inspection processes, while an automatic vision system is involved in easier tasks. Indeed, complex inspection tasks require a set of decisions from the operator based on its experiences and preferences. This is an extremely repetitive activity that requires a high-quality level during the entire work. For these reasons, this activity is very stressful for the human operator.
Automatic inspection systems can relieve humans from these operations. However, such systems have different parameters that are difficult to tune without the intervention of an expert operator. Indeed, it is difficult for humans to translate their own experience into a parameter configuration. For this reason, the fine-tuning of automated inspection systems requires a lot of time and multiple trial-and-error iterations.
The expert human operator is at the center of the project because the component that will be developed exploits the knowledge of the expert to automatically configure the system.
It provides an intuitive and easy-to-use interface with the operator so that even an operator without programming skills can use. Currently, the operators are subjected to stressful shifts in which they must repeatedly carry out the same operation and at the same time always guarantee the same levels of quality.
This project aims to help both operators who specialized in quality control and can be reemployed in more gratifying tasks and SMEs technicians who can rapidly deploy and configure automatic control systems.
The main objective of PROPHET is to automatize the configuration of an inspection system learning from an expert operator.
The core of the system is an AI-based component that exploits the knowledge of the expert operator to set suitable values for the control parameters. The knowledge transfer will be done through a preference-based optimization process in which the expert operator provides a preference by selecting the output between two different experiments.
The AI module is very suitable in the control quality process in which it’s necessary to judge the quality of the output process for different experimental trials. In such a way, the quality feedback provided by the expert is used to optimize the set of control parameters intuitively and easily.
The results of this project will serve as the backbone of a family of new products and services.
The application of preference-learning technologies in the visual inspection has a strong impact from many points of view, such as simplifying and accelerating the deployment, allowing non-skilled operators to train the system, and increasing production efficiency, enabling in this way a wide adoption of such technology.
The developed component will be easily adaptable to a wide range of applications and rapidly transferred to other industrial fields thus greatly improving its potential impact.