Design and training of AI models
Convolutional neural networks for image and video processing, including detection, classification and semantic segmentation applications in both 2D images and 3D information, such as point clouds or RGBD images.
Data analysis and machine learning: analysis of variables and search for indicators that can help improve production processes and extract value from data.
Generative models (Generative Adversarial Networks, stable diffusion, etc.) for content generation, style transfer, simulation, etc.
Detection of anomalies in sensor signals, to detect events with little or no examples labeled by humans, and that can be used immediately.
Reinforcement Learning: systems that learn control policies based on trial and error autonomously for use in control and planning tasks.
Training with synthetic data. Use of multi-physics 3D simulators to generate synthetic data and train more robust models in a short time.
GPU (Graphics Processing Unit): They offer a combination of flexibility and optimal computing power for most applications.
FPGA (Field-Programmable Gate Array): enables the combination of large number of image processing operations with Artificial Intelligence.
TPU (Tensor Processing Unit) and VPU (Vision Processing Unit): less flexible but the most efficient in terms of consumption and price.
Digitization and data capture
Industrial protocols: Profinet and Profibus.
Industry 4.0 Protocols: Logicmelt is an expert in the Industry 4.0 OPC-UA protocol, which will be the next industrial standard. Logicmelt has developed its own OPC-UA server and client, capable of transmitting complex objects over the network, facilitating connectivity and interoperability between devices.
IoT/Cloud Protocols: Logicmelt can send data to cloud platforms via IoT/Cloud protocols, such as MQTT and OPCUA PUB-SUB.
AI Product design
Data modeling: selecting the best data sources and sensors to solve the problem.
Training of Artificial Intelligence models: we select the best AI models to solve the problem, we test several, we optimize their parameters and select the best model for the application.
Advice on hardware selection: selection of the computing architecture and chip for the product that, meeting the requirements, minimizes its cost.
PCB Design: Design of the circuit board necessary to connect power, sensors, processing unit and communication ports to obtain a compact and small product.
Firmware programming: programming the processing and communications software necessary to have a 100% functional AI product.