BOTTLE MAKING 17 PETplanet Insider Vol. 26 No. 11/25 www.petpla.net Intelligent inspection of recycled PET containers When to use AI – and when not to Bottles that use recycled PET have become a growing challenge for quality control: recycled material introduces greater optical and surface variability, making defects harder to detect with traditional vision systems. AND & OR tackles this problem with ACUITY, an intelligent inspection solution that combines classical image processing with deep learning to spot anomalies in both virgin and recycled PET containers in real time. Drawing on years of experience in advanced machine vision, AND & OR has developed ACUITY, a machine vision software designed to detect patterns, objects, and defects with millimetre-level accuracy. It is claimed to allow rapid identification of issues that may be missed by other systems. Through native integration via OPC UA, the company’s vision platforms can communicate directly with MES, SCADA, and PLC systems, enabling standardised data exchange to support monitoring and control across production lines. Inspection of virgin and recycled bottles via classical vision and deep learning A challenge for quality control is the increased use of rPET, which exhibits greater optical and surface variability and can make defect detection more difficult with conventional vision systems. To address this, the ACUITY system from AND & OR has been implemented on PET bottle inspection lines - both for virgin PET and for bottles containing varying amounts of recycled content - enabling real-time, comprehensive monitoring of production quality. ACUITY combines classical image processing algorithms with deep learning models trained to detect anomalies, leveraging the strengths of both approaches, says AND & OR. The classical processing component, running on CPU, handles tasks such as inspecting ovality, geometric measurements, injection points, black spots, opacity, and dimensional control, while the deep learning component, running on GPU, enables the intelligent detection of more complex defects and visual variations associated with the use of recycled materials. Additionally, the system integrates anomaly detection and one-shot learning, enabling it to identify new defect types without extensive prior training, learning from reference patterns of correctly manufactured bottles. Flexibility and advanced analysis The system’s algorithms allow inspection settings to be tailored to each client’s requirements. It can count black spots and calculate their coverage on the container surface with customisable filters, classify defects based on size and actual measurements, and measure ovality while performing geometric checks at the bottle neck. It can also evaluate injection points and surface defects, analyse material distribution to identify areas of irregular thickness, and inspect opacity or translucency uniformity - even in semi-transparent or coloured containers - using independently configurable zones. Operators can manage these settings through an intuitive interface or integrate them directly with the PLC for automatic recipe and format management, AI in industrial vision The use of AI in industrial vision is becoming increasingly common, but it is not always the most suitable solution, says AND & OR. The company applies AI selectively when defects are diffuse, irregular, or poorly defined, such as optical variations in rPET or contamination-related flaws. For repetitive, well-defined, or geometrically measurable defects, classical processing can achieve higher precision with lower computational load. So, AND & OR designed its systems modular and hybrid, enabling AI modules to be activated or deactivated depending on the product and the complexity of the defects. In testing, ACUITY demonstrated an inspection rate exceeding 40,000bph. The system’s self-adjustment capability and reduction of false rejects translated into a 15% improvement in operational efficiency, particularly on high-speed lines, says AND & OR. www.andyor.com BOTTLE MAKING
RkJQdWJsaXNoZXIy NTY0MjI=