FITOW - Detection of Flaws in Aluminum Die-Cast Parts forAutomotive Transmissions‍

‍

Case Overview

The automotive transmission is a crucial component of the vehicle's drivetrain, responsible for mechanical power conversion. During the manufacturing process, scratches can easily occur on the machined surfaces of the transmission, leading to improper sealing of the engine housing and potential oil leakage, posing significant safety risks.

FITOW Testing Technology Company utilizes the PaddlePaddle deep learning open-source framework to conduct pixel-level segmentation of multi-angle scratch defects on Volkswagen transmission aluminum die-cast parts. This approach significantly reduces the misjudgment rate compared to manual inspection, with a single machine capable of replacing 6 workers for 12 hours of work per day.

Scenario Analysis

Detection Standards

In the industrial sector, there are clear standards for product quality control. For example, if the defect area is smaller than a certain threshold, the product can be passed. The defect detection standards used in this case include the detection of discoloration spots, oil stains/dirt, and scratches, as illustrated below.

‍

Technical Solution‍

Limitations of Traditional Algorithms

Traditional methods for measuring product surface defects involve manually setting features and using threshold segmentation. When the background within the regression box is complex, traditional visual defect detection becomes inaccurate, often resulting in inflated statistical values for small defects.

‍

Optimized Solution

FITOW improved optical imaging by using a robotic small field-of-view shooting and photometric stereo method. They utilized Baidu's PaddleSeg image segmentation toolkit to achieve pixel-level segmentation of target defects, thus defining the size of the defects accurately. This improvement in defect detection rate helps clients maintain strict control over the quality of the products.

Implementation Results

Relative to manual inspection, the misjudgment rate decreased from 8% to 3%, and the missed detection rate decreased from 5% to 2%. A single machine can replace the work of 6 people for 12 hours per day.

‍

Company Profile

FITOW (Tianjin) Testing Technology Co., Ltd., established in 2013, is a comprehensive service provider focused on quality-centered supply chain management in the automotive industry. FITOW innovates third-party services in content and form, using software and intelligent hardware as carriers and intelligent data as the core to enhance customer productivity.

‍

‍