SCM, a global leader in woodworking machinery, and its partners, within project Edge AI, are developing an artificial intelligence application for the automated identification of defects in manufacturing processes, leveraging cutting-edge Edge AI technology.
At now, defect detection is performed manually by operators. The new AI system will enable real-time analysis, swiftly identifying anomalies and providing immediate feedback for more efficient defect management. This will optimize waste disposal and machine parameter adjustments, improving production efficiency, and reducing waste.
More specifically, the project involves developing a hardware/software platform based on AI-on-the-edge, incorporating computer vision algorithms and smart sensors capable of recognizing defects directly within the production flow. Functional and non-functional specifications have been defined to ensure a significant improvement in inline quality control, with precise verification and validation metrics.
Key Objectives
- Development of an AI-based edge computing platform for quality control.
- Training and implementation of AI algorithms specifically on embedded hardware.
- Use of distributed learning and transfer learning mechanisms for continuous model updates.
System Architecture
The system is structured in two main phases:
Data Acquisition and Preprocessing
- The first stage captures signals and preprocesses data through dedicated hardware, such as FPGAs, microcontrollers, and embedded CPUs.
- The data, in image or image-like format, is filtered and processed locally to reduce the amount of information transmitted.
Central Processing and AI Inference
- The preprocessed data is sent to a central unit that performs the final AI algorithm calculations.
- The central unit uses AI-dedicated hardware for deep learning, including GPUs, FPGAs with hardware accelerators, NVIDIA boards, and microcontrollers with integrated AI accelerators.
This edge-to-cloud architecture balances computational efficiency and processing speed, enabling real-time quality control directly on the production line. The AI-on-the-edge approach ensures reduced latency and greater operational resilience without relying on remote servers or complex cloud infrastructures.
TinyML and AI-on-the-Edge: A New Paradigm
In recent years, Artificial Intelligence has seen widespread applications, but these often require complex infrastructures and high energy consumption. Therefore, to overcome these limitations, the project integrates TinyML technologies, enabling AI models to run on resource-constrained devices, reducing energy consumption, and enhancing the responsiveness of embedded systems.
TinyML allows inferences to be executed on devices with just a few KB of memory and power consumption of only a few mW, making it ideal for IoT and industrial automation applications. By processing data locally, benefits such as reduced latency, increased privacy, and lower network costs are achieved. For example, the TensorFlow Lite (TFLite) framework optimizes AI models through techniques like pruning (removal of irrelevant weights) and quantization (conversion from 32-bit to 8-bit), ensuring efficient execution even on low-performance hardware.
The Role of RISC-V in Embedded AI Evolution
Despite its advantages, implementing TinyML poses significant technical challenges, requiring expertise in machine learning, software development, and hardware design. In this context, the RISC-V architecture, with its open-source nature and modularity, offers a flexible and efficient solution for edge devices. The ability to customize microcontrollers reduces costs and energy consumption, facilitating AI integration into resource-limited embedded devices.
The adoption of open-source hardware is also democratizing access to advanced technologies, allowing startups, researchers, and companies to develop innovative solutions without investing in proprietary tools. This collaborative approach fosters innovation and lowers entry barriers in the industrial AI sector.
Towards an Intelligent and Sustainable Future
Integrating TinyML and AI-on-the-edge represents a significant step forward in industrial innovation. The use of optimized algorithms, dedicated hardware, and distributed processing enhances production efficiency, reduces waste, and ensures automated, precise, and adaptive quality control. These technologies not only transform industrial processes, but also pave the way for widespread adoption of artificial intelligence in everyday devices, improving quality of life and reducing environmental impact.
Blog signed by: SCM team