Implementation of Neural Network-Based Image Classification in Industrial Automation Systems

The integration of artificial intelligence (AI) into industrial automation has led to the development of edge-deployable inference platforms. This paper presents a deployment methodology and performance evaluation of two neural network models running on the SIMATIC S7-1500 TM NPU 2.0 module: one for the classification of Siemens automation components and another for fruit classification using a public ONNX model. The AI Model Deployer environment was used to convert and deploy both models, enabling automation engineers to implement machine learning solutions without prior AI knowledge. The system architecture includes a GigE Vision camera, MicroPython-based runtime logic and shared memory communication with a PLC. Experimental results show real-time inference with high accuracy and robustness under realistic conditions. This work demonstrates the feasibility and flexibility of using both industrial and generic pretrained models on embedded NPUs in manufacturing environments.