VisionMind Detection

The Challenge

The client, an automotive component manufacturer, inspected thousands of cast metal parts hourly. Their quality assurance (QA) pipeline depended on manual visual audits under high-intensity lights, which was slow and subject to fatigue errors, letting micro-cracks pass through to shipping boxes.

The manufacturing conveyor line moved rapidly, requiring components to be evaluated and classified in under 200 milliseconds. Uploading high-resolution images to the cloud for analysis introduced latency and was unreliable during local network dropouts, necessitating a localized edge computing solution.

The Solution

Ankur Weldtech India designed an edge-processing machine vision solution using Python, OpenCV, and TensorFlow Lite. We deployed optimized convolution models on industrial edge devices placed directly above conveyor lanes. High-speed cameras capture images of the moving components, feeding frames into OpenCV.

OpenCV pre-processes the frames by cropping the region of interest and adjusting contrast to highlight metallic surface details. The cropped images are evaluated by the TensorFlow Lite model, which infers defect tags locally. If a defect is detected, the device triggers an electronic signal to a mechanical sorting arm, routing the item to a reject bin.

Key Results

By deploying VisionMind Detection, the manufacturer automated the QA pipeline, achieving a near-zero defect escape rate. The edge processing system runs independently of cloud connectivity, keeping the manufacturing line operational 24/7.