Intelligent Real-Time Quality Inspection in Manufacturing Using Artificial Intelligence

To realize ambitious goals, such as zero-defect manufacturing in precision manufacturing in Industry 4.0, smart and autonomous quality control has to replace manual and error-prone quality control. We present Smart Inspect, a system that utilizes artificial intelligence (AI) to autonomously detect, classify, and segment defects with extremely high speed and accuracy. Unlike existing solutions, Smart Inspect fits smoothly into existing production lines, enabling manufacturers to scale inspection throughput with their production needs.

The Smart Inspect model is based on an advanced computer vision pipeline. The first step of the pipeline consists of the acquisition and preprocessing of each image. Several standard operations are available through OpenCV to achieve uniformity within our model's input. The array processing operations are performed using the array-processing library NumPy, and the defect detection is enabled and powered by a deep learning model created with the TensorFlow deep learning framework. We employ MobileNetV2, a compact convolutional neural network, and we utilize transfer learning to take advantage of the feature extraction capabilities of pre-trained networks on the task of manufacturing defect detection. Transfer learning improves the time required to train the model, and the number of images required. Its classification head is designed to prevent overfitting by applying a GlobalAveragePooling2D layer and a Dense layer with softmax activation function, allowing it to classify several kinds of surface defects and anomalies. To satisfy the real-time demands of the production environment, the architecture of the system is threaded. The threading separates the image capture and the model inference, so they can be run concurrently, and also prevents the inspection system from becoming a manufacturing bottleneck. JSON is also used to produce structured logs that show the identified defect and metadata associated with it (defect type, defect location, high-precision timestamp, etc.). This creates a clear and auditable history of quality control, which can be analyzed to improve the production process. Smart Inspect makes measurement and inspection automatic, allowing manufacturers to eliminate human inspection errors and increase quality assurance rates. It reduces subjectivity and inspection fatigue for speed and efficiency. It provides a highly accurate, low-cost, scalable, and flexible solution to manufacturers on their adventure to a zero-defect strategy and those wishing to differentiate their products through quality assurance, against manufacturers in an increasingly data-driven era.