This project involved developing a machine vision–based defect detection and alerting system for a PVC manufacturing environment. The objective was to automatically detect surface and color defects during production and immediately notify both on-site operators and office personnel to minimize material waste and downtime.

The vision system continuously analyzes the production output to detect anomalies such as stains, pigment stripes, miscolorization, surface damage, and other visual defects. Custom-annotated datasets were used to train the detection model specifically for PVC-related defect patterns, enabling reliable identification under real production conditions.

To support root-cause analysis, the system generates heat maps indicating where defects most frequently appear across the material surface (e.g. center, edges, or specific zones). This spatial insight helps operators and engineers quickly identify whether issues originate from tooling, material feed, temperature distribution, or alignment problems.

Before triggering an alert, detected anomalies are passed through an additional AI-based validation step, which evaluates the confidence and context of the defect to reduce false positives. Only verified issues generate alerts.

The alerting mechanism operates over a local wireless network, with remote access available from office workstations. In addition, a custom-built wearable alert device is used by the head operator. This wrist-mounted controller receives notifications via a radio-frequency communication module and displays defect warnings in real time, allowing immediate intervention without requiring constant visual monitoring of screens.