Deep learning-based poka-yoke process monitor

An Italian SME has developed and patented an innovative system which makes any rule-based process error-free. The system has been designed for real-time checking of human-machine interaction or human and machine actions within manufacturing. The system guides human and/or machine actions with human-like collaborative vision in order to improve efficiency, production quality and safety. The firm is open to technical or commercial deals with technical assistance with end users coming from the industry having manual processes to be managed. Errors and anomalies are unavoidable in any manufacturing process. But if mistakes are caught at the right time, the consequences can be prevented. The initial removal of defects reduces errors costs, in many cases down to zero. This patent-pending solution uses 3D machine vision and artificial intelligence, applied in collaborative mode, to understand the events that occur within a certain workspace. Thus, the system puts a motion-step method-based poka-yoke in place, checking if all the tasks of a given process have been performed correctly, in the preset order and times. The solution is a real-time process monitoring system that helps users to achieve zero-defect manufacturing and maximize efficiency, quality and safety. Optimizing time-and-motion means improving efficiency and productivity, enabling the staff to manage and execute production processes correctly. This is the way to get great savings, eliminating all waste and minimizing costs. But the optimized work cycle is often difficult to respect, for humans and sometimes also for machines. The system works alongside those who work, whether this is the man or the machine or both at the same time. It collaborates with them, in a non-intrusive but inflexible way: it monitors and guides their work in order to maximize efficiency, quality, and productivity while preserving the safety of operators as well as the performance and integrity of the machines since it can intervene promptly in case of danger. The proposed solution takes furthermore advantage of an exclusive combination of cutting-edge technologies, making it unique in terms of performance and flexibility of use. Deep learning makes the system capable of recognizing multiple elements moving in the scene – including objects which can vary in shape – without getting confused by distortions, different perspectives, variable contrasts or objects which look not really sharp. Furthermore, the learn-by-example algorithms that characterize deep learning allow the AI quotient to increase indefinitely, with no need for human intervention. 3D machine vision grants the system the ability to see with maximum realism what happens at any moment, with the spatial perception typical of human vision. So, it is possible to obtain: - A real-time contactless control, free from changes in performance over time - A constant, impartial, evaluation capability that allows to classify both the quality and efficiency of your process. Setting up a process control with this system is easy and intuitive, in fact, a wizard suggests how to set up the system. It is not necessary to write even a line of programming code: the approach is totally visual and does not require specific competencies. After the configuration phase, the system turns the attention to the workspace and interacts with human operators, machines or both, providing directions, reporting any mistake in the performance of the process under control and nipping potentially dangerous situations in the bud. Thanks to the ability to analyze seamlessly, even multiple simultaneous events, this solution offers unprecedented reactivity in the event of anomalies.