We Invest in Deep-Learning for Outgoing QC
Detecting quality defects is crucial to prevent customer dissatisfaction or even damages in a production line. To guarantee a consistent quality standard, we were continuously conducting a constant manual inspection. In a few cases, we even performed a 100% checking rather than random sampling to meet customer’s quality agreement.
Challenge We Face Everyday
In production, there are irregularities which can be tolerated and others that have to be separated. Shipping packaging material with surface/cosmetic defect will result in the complaint and rejection, which against our commitment to deliver quality products to customer. Moreover, cosmetic defects like dirt, black spot, faded/bleeded printing, flashing, scratch, and other surface-related defects are difficult to be detected. It requires an unending labor-intensive and time-consuming effort. Both cases are costly and require extensive quality assurance activities. On the one hand, this is often an unfulfilling, repetitive task which requires high concentration for long period of time. On the other hand, it comes with high costs, especially as humans can perform such inspection only at a certain rate and reliability.
Mere AI-Based Inspection is not Enough
AI technology and Machine Vision may help us control the quality effectively. A common and basic scenario is that a capturing device looking at the whole surface area of the packaging and detect sign of anomalies based on trained datasets from good and bad samples. However, one of the biggest challenges when dealing with cosmetic, or surface defects on consumer packaging is that they are dynamic. Typical defects like hits, scratches, or stains may be indiscernible during early production. These defects only become visible under specific lighting conditions later in the production process. While the cost of late detection can be painfully high, so can false rejects. This inspection is especially important because poor packaging quality can lead to recall events or customer complaints.
Conventional vision technology can often miss complex cosmetic packaging defects such as bubbles inside glass bottle blowing, color degradation, scratches, cracks, overprint, and other issues. These types of unpredictable defects or variations are easy to discern by human inspectors, but very difficult to program with rule-based machine vision algorithms.
We Invest in Deep Learning Technology
We at Dermapack have decided to invest in AI technology, specifically in Deep Learning Image Analysis to further detects cosmetic defects on rough, glossy, matte, and textured plastic/glass surfaces as reliably as human inspectors, but with the speed of a computerized system. The defect detection tool catches defects with standard illumination, even when image quality is poor, by forming a reliable model of the part’s shape and texture based on training images. Furthermore, with deep learning technology, we are only required to train the algorithm with small sample size (50-100 pcs), making it even more usable. From here, it identifies deviations in the surface as anomalies and uses a classification tool to classify them according to the common defect types. This investment enables us to:
- Lower the risk from performing an exhaustive and repetitive checking task
- Lower the cost incurred from recall/failed delivery and labor-intensive task
- Increase the reliability and speed of quality inspection
- Conduct a full quantity inspection for all packaging materials being delivered
- Avoid quality-related issue to deliver On Time & In Full