Predictive maintenance is a critical element of the Industry 4.0 initiative for smart factories. By using smart sensors to monitor the current health and conditions of industrial equipment, with enough data, factories can predict when and where issues will occur in the future.
The sensor modules used in PdM typically contain an MCU and Bluetooth/BLE/Wi-Fi or other connectivity option along with sensors that are ideal for detecting various performance issues and anomalies.
Minor differences in the hardware setup and environment, such as mounting the sensors on a different surface material/location or the ambient temperature, can affect the accuracy of a machine learning model. For best performance, each individual machine needs to be equipped with a unique model that is periodically updated and improved.
With an intuitive UI, Qeexo AutoML allows for onsite data collection, automated model development, and swift library deployment so users can find the best model for each unique use case quickly and efficiently.
Governments all over the world are investing in machine learning and sensors attached to bridges and other structures to monitor their “health”. In Japan, for example, regular earthquakes and tremors increase the fatigue of bridges, railroads, roads, and buildings, which need to be monitored in the interest of public safety.
Sensors commonly used in SHM include high precision, low drift/bias accelerometers, strain gauges, displacement transducers, level sensing stations, anemometers, temperature sensors, and dynamic weight-in-motion sensors. They measure everything from tarmac temperature and strains in structural members to wind speed and the deflection and rotation of the cables, and any movement of bridge decks and towers.
These sensors are often combined with a low-power MCU and a low data bandwidth connectivity technology like LoRa.
Often, thousands of sensors are required to monitor a single structure. Traditionally, all of the raw sensor data are sent to a remote location for processing. Machine learning models built with Qeexo AutoML are optimized to run locally on MCUs, so decisions can now be made on-device. Only useful information such as alerts, not the raw data, are relayed to the monitoring station.
In addition, due to the remoteness of the sensors attached to structures in SHM, these sensors are typically required to run on limited battery power for extended periods of time. Qeexo AutoML produces extremely efficient models with low power consumption, thus extending the battery life of these devices.
Intelligent appliances and other smart devices around the home have gained popularity in recent years. Most of these are equipped with sensors but only using very primitive algorithms and logic. Machine learning can further increase the usability and functionality of these devices, making our daily lives more convenient.
Compared to traditional, threshold-based logic, devices equipped with machine learning can better make sense of the sensor data and make these appliances around the consumers’ homes even smarter.
In addition, since models built with Qeexo AutoML runs on-device, consumers’ data is not sent to the cloud, preserving their privacy and saving data and infrastructure costs.
Due to their compact form factor, wearable devices are highly constrained in processing power, memory size, and battery life. Balancing performance with battery life is a constant struggle for device manufacturers.
Often, wearables in the market are either extremely expensive or not very smart and needs to be connected to your smartphone for full functionality. Powered by a mid-to-low-tier MCU, the processing power and memory space available to wearables are limited, making developing TinyML on these constrained environments very difficult.
Optimized for the Edge, Qeexo AutoML produces lightweight models that are powerful but small in size and uses very little power, giving your devices high intelligence without the high costs.