Case Study: Dormer Pramet | Computer Vision for Quality Control

Introduction and Problem Formulation

Our client Dormer Pramet, the global manufacturer and supplier of tools for the metal cutting industry, faced a quality control problem. Their complex manufacturing process relied heavily on manual inspections, leaving the risk of undetected defects until a final inspection or even worse — by the customer. This approach not only affected the quality of the product but also limited the efficiency of production. Compounding the problem was their extensive portfolio. Tens of thousands of inserts with various features made thorough inspection a challenging task. To make things even more complicated, tiny defects, like microscopic cracks, invisible to the naked eye, required specialized equipment. Finally, the lack of detailed defect data hindered analysis and improvement. Our client was therefore looking for an AI-based visual inspection solution that would increase efficiency, reduce costs and deliver flawless products.

Moreover, the client’s extensive product portfolio, encompassing tens of thousands of insert variants with diverse shapes, sizes, materials, and surface finishes, complicated the quality control process. The sheer volume and variety of products made it challenging to implement a thorough inspection strategy, contributing to the need for a more sophisticated solution.

Manual inspection presented an additional hurdle due to the minute size of defects, often requiring the use of specialized equipment. For instance, cracks, a common type of defect, could be as small as 10 micrometers wide, making them impossible to identify without a microscope.

The absence of detailed defect statistics in the client’s information system further hindered the company’s ability to analyze and address recurring issues systematically. As a result, the client recognized the pressing need for a solution to enhance overall efficiency, reduce costs, and improve product quality. This realization marked the initiation of their journey towards implementing AI visual inspection in their manufacturing process.

Our Solution

To address the challenges faced in the manufacturing process, our company implemented a tailored end-to-end solution that integrates cutting-edge technologies, including high-resolution cameras, macro lenses, artificial intelligence (AI), and robotics. The strategic decision was made to initiate automated control in the press room, the starting point of production, where early defect detection proves most economically advantageous.

Given the microscopic nature of defects, some as small as 10 micrometers, a high-resolution camera and macro lens were deemed essential. However, the challenge arose due to the limited angle of view of this assembly (around 20x20mm), while inspected products could be up to approximately 70 mm in size. To overcome this limitation, the solution incorporates motorized stages that move the product under the camera, capturing several overlapping images. Software then combines these images into a single comprehensive view.

The core of the system involves AI deep-learning-based image analysis, which efficiently identifies defects in the product. This deep learning model offers unparalleled flexibility in conceptualizing and generalizing the appearance of various insert variants, accommodating the extensive product portfolio of our client. Unlike traditional machine vision, deep learning excels in handling natural variations in complex patterns, eliminating the need to define tolerance parameters individually for each variant.

Our solution features a clear and intuitive user interface, providing operators with tools for fast and accurate analysis. Operators are able to mark and classify the results of the visual inspection, if necessary, and provide feedback to the system, fostering a continuous learning process contributing to the training of the model and its ability to adapt to specific production characteristics.

The inspection station seamlessly communicates with a robotic arm responsible for product handling within the press. Upon completing the inspection, the station signals the robot for the removal of the piece and the insertion of another. Additionally, the system informs the robot whether the inspected piece is deemed good or defective.

A data management module allows the generation of detailed reports and statistics, providing valuable insight into the evolution of quality over time. Operators are able to easily access historical data, analyze visual inspection results and track the trend of defects on the production line.