The industrial processing and sorting of food, especially sensitive fresh produce such as fruit, places high demands on precision, speed and flexibility. Conventional sorting systems often operate mechanically or with basic optical sensors. However, they reach their limits when it comes to detecting subtle differences in quality, degree of ripeness or external defects.
Here, modern smart factory concepts come into play, intelligently linking artificial intelligence (AI), machine learning (ML), edge computing and powerful hardware platforms. One possible application would be sorting apples. The aim is to recognize different apple varieties, determine their degree of ripeness (e.g., green, red), and identify quality defects such as bruises, mold or rot. This information could be used for automatic sorting into defined quality classes. In addition, documentation could be created for quality management and traceability.
Possible applications for AI-supported image processing
At the core of this application is the AI system built on the Kontron K4021-Ux mSTX (Fig. 2). The integrated NPU (Neural Processing Unit) or AI co-processor enable accelerated processing of high-resolution image data captured by a 5GbE IP camera.
In a potential application scenario, the camera would capture the apples on the sorting belt in real time and deliver images of each individual product. A specially trained neural network would then evaluate these images. To achieve this, it is pre-trained on datasets containing thousands of images of different apple varieties, stages of ripeness and defects. This is how the system could hypothetically make decisions:
- Is the fruit in perfect condition?
- Is the apple ripe (red) or unripe (green)?
- Is there any external damage or are there any quality defects?
- Which quality or sorting class should the apple be assigned to?
These decisions are made with a speed and precision that would be impossible to achieve with manual methods. At the same time, each individual evaluation result is documented alongside the captured image data and can be used later for quality audits or process optimization.
Compact design for flexible integration
The K4021-Ux mSTX embedded system was designed for applications with limited installation space. It can be integrated either directly or in the SMARTCASE S502 package into existing sorting plants. The system is designed for real-time processing of image and sensor data, enabling process-integrated evaluation of captured characteristics during operation. Results can be transmitted directly to downstream systems via a dual network interface. Thanks to the long-term hardware availability of at least seven years, an additional Longevity Plus program and BIOS support for up to three years after the end of the product life cycle, the systems can be operated and maintained reliably over several years.
Data consolidation on the edge server
While the K4021-Ux industrial motherboard performs real-time analysis and sorting of apples on the line, all captured quality and process data is continuously transmitted to a central edge server. Here, the K3881-C µATX server motherboard (Fig. 3) is used and, acting as a local hub, consolidates various data streams from production. In addition to the sorting results and image data, the server also processes other production-related information. This includes environmental parameters like temperature and humidity, which are integrated via 5G-enabled sensors, as well as machine data for operational monitoring. This information is supplemented by batch indicators such as throughput, quantity distribution and sorting statistics. Product-specific characteristics, e.g. ripeness, sorting classification and documented deviations, are also recorded and evaluated. This delivers a consistent, up-to-date picture of the ongoing process, accessible to downstream systems and analyses.
Powerful infrastructure
The edge system based on the K3881-C µATX is built for continuous industrial operation in demanding conditions. It supports an extended temperature range, ensuring reliable operation even when the temperature changes. Furthermore, it enables both wired 10G connections for high-speed data transfer of large amounts of data to the cloud or to central production control systems and the integration of modern 5G interfaces for flexible connection of mobile units and wireless sensor systems. Remote management functions and an integrated board management controller (BMC) facilitate central administration and remote maintenance and contribute to reliability. Thanks to the long-term availability of the platform, the system is especially suited for applications with predictable product cycles in production.
The captured and processed data can be integrated into existing cloud infrastructures. There, it is available for various applications, such as analyzing sorting accuracy, investigating the causes of quality deviations or optimizing processes. In addition, this data enables seamless traceability of individual batches, monitoring of compliance with cold chain specifications and predictive maintenance through the analysis of machine-related operating data. The central provision of this information creates the basis for continuous monitoring not only of individual lines but also of entire plants or even across locations.
Fruit sorting 4.0 in practice
In a real-world practical scenario, a modern fruit sorting plant processes several tons of apples every day. Each apple is captured by an IP camera and analyzed by the K4021-Ux system within milliseconds. Automated sorting is performed using switches, wherein classification into different quality classes is based on predefined criteria. At the same time, all sorting results, defect images and sensor data on the environment (temperature, humidity) are transmitted to the K3881-C edge server. This edge server prepares the data, performs statistical evaluations and continuously monitors compliance with cold chain specifications. The results are stored in a central location and are also available in the cloud as complete batch reports. Both operations managers and quality managers therefore have access to current and historical data at all times.
Summary
The combination of AI, edge computing and networked automation solutions forms the technical foundation for further enhancing quality assurance, traceability and process stability in food processing. Systems like the K4021-Ux mSTX and K3881-C µATX demonstrate that these technologies are already available for practical deployment and can be integrated into existing production environments. In the fruit processing sector, in particular, this can significantly improve sorting quality, optimize yields and increase customer satisfaction through consistently high product quality. Simultaneously, the fully captured process data offers valuable insights for continuous improvements along the entire value chain.
This shows that data-based production processes in the food industry are already technically feasible and economically viable today.