Kexxu Industrial Infrastructure
Kexxu can equip your factory, greenhouse, farm, or any other industrial equipment that may benefit, with AI Cameras.
But why would I need that?
Here are some example use cases to give you an idea of the possibilities:
1) Checking if cups are properly washed
Kexxu AI cameras can stear robots to pick up plastic reusable cups from industrial washing machines, saving personnel cost and simplifying the management of industrial scale washing facilities. Multiple cameras from the top each control one robot to identify the center and pick location of the cups real-time, guiding the robot with lightning speed towards the cups to suck them up, move them through the camera anomaly scanning area, and place them in the stacking automation.
The Kexxu camera solution used 6 cameras to check if cups are properly washed and clean by industrial washing machines using custom-made state of the art anomaly detection. No other neural network that we have tried that was available could match the AI performance of the cleansiness detector computer vision model created by Kexxu.
2) Measuring flowers.
Measuring flowers in greenhouses using a custom-made neural network developed in-house by Kexxu. Bigger flowers have another price than smaller ones, so action houses require flowers to be sorted by size. Our AI cameras can do this for you.
3) Facilitating steel factory automation
Our cameras provide feedback to automatic tapping systems in steel factories, making sure the automated tapping never poors too fast and over the edge, which would cause huge damage to the steel factory.
4) Checking the correct side of holes
Checking the correct side of holes punched in steel rods by a punching machine. The punching machine is the easiest part of the factory, so new employees often start there. They sometimes make mistakes, which can propagate to the supply chain. The Kexxu AI Cameras stop these mistakes early in the process.
5) Counting cattle
Compliance often requires strict counting of the whereabouts of cattle. This is a lot of work and is often skipped, because it's not the most fun job counting cattle multiple times a day. Our AI cameras can do this automatically and save it to a dashboard in the cloud.
We mainly use two types of cameras for factory automation, one being Power over Ethernet cameras with electronics by IDS-Imaging, which are connected and powered by ethernet cables of up to 100m long.
The other camera we often use which are cheaper so more suited for solutions with many cameras, is the custom Kexxu camera we developed inhouse with the Onsemi AR0234 sensor. This sensor is specifically suited for industrial environments and can measure at 120fps with a global shutter and in low-light environments. These qualities make it our work horse for computer vision tasks, and we employ them also for solutions where they are maybe overkill, to simplify our architecture and supply chain, eventually reducing cost and making spare parts management easier.
We can store and process many of the computer vision models locally in factories, which is specifically suited for real-time applications and the control of robots and machines based on AI Camera generated data. On the inside our edge processing devices are powered by Nvidia Jetson.
For horizontal scaling of systems with many cameras, we use what we've dubbed "ingesters". They connect to Kexxu cameras over usb, then "ingest" the images and transfer them to the AI edge devices for storage and AI inference. They can also listen to triggers, either from laser sensors in the factory or PLC signals from existing automation.
The Kexxu Camera infrasturcture can easily be deployed and adjusted for other use cases, because such infrastructure as devices setup using an App for iOS and Android using Bluetooth, uploading the data collected over a secure encrypted connection to a 100% European cloud, remote labeling, finetuning and updating AI models, and cloud access with user login have already been created and prepared after our years of experience deploying computer vision infrastructure in the real world.
Plant World Model
We are currently developing a Plant World Model, a foundation model just like GPT or DeepSeek, but specialised for plants. We are working with various partners to keep the Netherlands firmly at the number one spot globally in agriculture technology.