IOT- INDUSTRIAL DEFECT DETECTION SYSTEM USING AN EMBEDDED CAMERA

Abstract

Quality inspection is an essential step in manufacturing, because defective parts that reach customers lead to returns, warranty costs, safety risks, and loss of reputation. In many small and medium industries, inspection is still carried out by workers who look at each product on the conveyor and remove the ones that appear faulty. Manual inspection is slow, tiring, and inconsistent, particularly over long shifts and at high production speeds. This paper presents an IoT based Industrial Defect Detection System that uses an embedded camera and on-device image analysis to inspect products automatically and report quality data in real time. The proposed system places a camera module connected to a Raspberry Pi or ESP32- S3 board above a conveyor belt, with controlled LED lighting to give uniform illumination. An infrared proximity sensor detects each product as it arrives under the camera and triggers image capture. A lightweight convolutional neural network, trained on images of good and defective parts and optimised for embedded hardware, analyses each image to identify defects such as cracks, scratches, dents, missing components, and colour variation. When a defect is found, the controller activates a servo-driven or pneumatic rejection mechanism that pushes the faulty item off the belt into a reject bin, and it records the defect type, image, and timestamp. A stack light and buzzer show the inspection status to workers nearby. Because inference runs on the embedded device, the inspection decision is made within the time the product takes to pass the camera, without depending on a remote server. The rejection timing is calculated from the measured belt speed so that the pusher acts exactly when the faulty item reaches the reject point. Inspection results are published through MQTT over Wi-Fi to a cloud platform, where a dashboard displays production count, defect rate, defect types, and sample images for each shift. Supervisors receive alerts when the defect rate rises above a set limit, which may indicate a problem with a machine, tool, or raw material upstream. The stored images and statistics can also be used to retrain the model and to support root cause analysis. Shift summaries can also be exported for quality audits and customer reports. A prototype tested with metal washers and plastic bottle caps on a small conveyor achieved high classification accuracy and inspected items at a rate comparable to manual inspection while maintaining consistent performance throughout the test period. The system offers small manufacturers an affordable path towards automated quality control. Future work includes multi-camera inspection of all product faces, anomaly detection for unseen defect types, and integration with factory execution systems.