Time-critical detection and machine learning scenarios, which run entirely on Leantegra UWB-BLE devices or industrial PPE appliance.
Integrated and customized TensorFlow Lite libraries for Leantegra UWB-BLE devices to enable on-device machine learning scenarios.
Large ecosystem of TensorFlow software and partner tools, such as Edge Impulse SaaS, become available for tiny low-power devices.
TensorFlow Lite supports powerful ML features from TensorFlow, such as Recurrent Neural Networks (RNN) and LSTM (Long Short-term Memory Cells), but with a much smaller memory footprint.
Tasks about building, training, deploying and executing ML models on Leantegra devices (or 3rd-party devices) become more organized and standardized than feature-specific coding of ML algorithms.
Supported hardware platforms (chipsets) by TagML firmware:
Nordic nRF52 and nRF91: nRF52832, nRF52833, nRF52840, nRF5340, nRF9160
STM32: STM32H7 (Cortex-M7), STM32F4
On-device classification of primitive motion patterns or complex activities for people or equipment (cranes, vehicles) using accelerometer and barometer data together with TagML.
Recognized activities can trigger the on-device alerts (e.g. haptic feedback) or can be visualized on the map of CVO Portal:
walking, running, climbing, inactivity, fall detection, forklift/truck driving etc
PPE detection is an important use case for industrial safety. Detecting presence of safety helmets or reflective vests can be implemented using Computer Vision or Wearable Devices.
The option about Wearable Devices is more suitable for tough environments with obstacles, dust, absence of light and enables detection in remote places without any network coverage – for example: without Ethernet or Wi-Fi networks.
Leantegra modules or UWB-BLE devices for PPE detection can be mounted externally or internally into the Smart Helmet appliance.
The main algorithm includes neural networks trained for classifying specific motion patterns of the helmet appliance.
Reliable fall detection is still a challenge considering high chances of false alarms.
ML model training for fall detection must include multiple scenarios with project-specific requirements (e.g. ladder falls).
It is crucial to have a straightforward option about configuring and training the fall detection system according to the project-specific requirements, instead of having a single universal device. Otherwise the rate of false alarms will be high.
Location details must be also supplied together with each fall detection alarm, otherwise the search process might consume time and efforts. This is where RTLS comes into play, considering that all Leantegra devices support RTLS too.
Real-time Sensor Classification
Augmenting the industrial RTLS systems with additional sensors, such as CO or methane or humidity sensors, enables more precise and contextual information about assets, personnel or environment.
Sensor data is processed directly on BLE beacons, UWB tags or mining locators for detection scenarios based on TagML.
Preventive alerts and actions can be applied before the actual disaster happens. Preventive maintenance for equipment is one popular scenario where this solution can be used.